JP2023160986A - Ultrasonic diagnostic device and analysis device - Google Patents

Abstract

To accurately obtain a discrimination result for a feature part.SOLUTION: An ultrasonic diagnostic device comprises a storage unit and a discrimination unit. The storage unit stores each of a plurality of learned models that corresponds to each of a plurality of stages and is used when executing each of a plurality of pieces of discrimination processing for outputting a discrimination result for a feature part of a subject drawn in input medical image data. The discrimination unit derives the discrimination result for the feature part of the subject by executing the discrimination processing corresponding to at least one learned model by using at least one learned model of the plurality of learned models. The discrimination unit executes the different discrimination processing from the discrimination processing executed in other stages in each of the plurality of stages.SELECTED DRAWING: Figure 1

Description

Embodiments of the present invention relate to an ultrasonic diagnostic device and an analysis device.

Ultrasonic diagnostic equipment displays in real time ultrasound images shown by ultrasound image data generated in real time in response to ultrasound scans, or displays ultrasound images shown by ultrasound image data obtained from past ultrasound scans. Display sound wave images. The real-time performance of ultrasound diagnostic equipment is superior to that of other medical image diagnostic equipment, and is a major advantage. Furthermore, a CAD (Computer Aided Detection) function is known that automatically detects a characteristic part (characteristic part) in a medical image generated by a medical image diagnostic apparatus.

US Patent No. 9,674,447 US Patent Application Publication No. 2015/0294457 US Patent Application Publication No. 2018/0025112 US Patent Application Publication No. 2007/0118399 US Patent Application Publication No. 2004/0122790

The problem to be solved by the present invention is to provide an ultrasonic diagnostic apparatus and an analysis apparatus that can accurately obtain discrimination results regarding characteristic parts.

The ultrasonic diagnostic apparatus of the embodiment includes a storage section and a determination section. The storage unit stores each of the plurality of trained models corresponding to each of the plurality of stages, and each of the plurality of discrimination processes that outputs the discrimination result regarding the characteristic part of the subject depicted in the input medical image data. Stores each of a plurality of trained models used when executing . The discriminator uses at least one learned model among the plurality of learned models to perform a discrimination process corresponding to the at least one learned model, thereby deriving a discrimination result regarding the characteristic part of the subject. In each of the plurality of stages, the determination unit executes a determination process that is different from determination processes performed in other stages.

FIG. 1 is a diagram showing a configuration example of an ultrasound diagnostic system according to a first embodiment. FIG. 2 is a diagram for explaining an example of processing executed by the trained model generation function according to the first embodiment. FIG. 3 is a diagram for explaining an example of processing executed by the discrimination function according to the first embodiment. FIG. 4 is a diagram for explaining an example of the first stage processing during learning and the first stage processing during operation according to the first embodiment. FIG. 5 is a diagram for explaining an example of second-stage processing during learning and second-stage processing during operation according to the first embodiment. FIG. 6 is a diagram for explaining an example of the third stage processing during learning and the third stage processing during operation according to the first embodiment. FIG. 7 is a flowchart showing an example of the flow of the first determination process according to the first embodiment. FIG. 8 is a diagram for explaining an example of detection results of characteristic parts by the discrimination function according to the first embodiment. FIG. 9 is a diagram for explaining an example of detection results of characteristic parts by the discrimination function according to the first embodiment. FIG. 10 is a diagram illustrating an example of the results of the first detection process according to the first embodiment. FIG. 11 is a diagram illustrating an example of the results of the second detection process according to the first embodiment. FIG. 12 is a flowchart showing an example of a process executed by the ultrasound diagnostic apparatus according to the first embodiment. FIG. 13 is a diagram for explaining an example of processing executed by the marker information generation function according to the first embodiment. FIG. 14 is a diagram showing a display example according to the first embodiment. FIG. 15 is a diagram illustrating a configuration example of an ultrasonic diagnostic apparatus according to the second embodiment. FIG. 16 is a flowchart illustrating an example of a process for generating a second trained model and a second discrimination program according to the second embodiment. FIG. 17 is a flowchart illustrating an example of a process for generating a third trained model and a third discrimination program according to the second embodiment. FIG. 18 is a flowchart showing an example of a process executed by the ultrasound diagnostic apparatus according to the second modification of the first embodiment and the second embodiment. FIG. 19 is a flowchart showing an example of the flow of the first determination process according to the third modification of the first embodiment and the second embodiment. FIG. 20 is a diagram illustrating an example of the results of the third detection process according to the third modification of the first embodiment and the second embodiment. FIG. 21 is a diagram illustrating an example of the results of the fourth detection process according to the third modification of the first embodiment and the second embodiment. FIG. 22 is a diagram illustrating an example of the results of the fifth detection process according to the third modification of the first embodiment and the second embodiment. FIG. 23 is a diagram for explaining an example of various learned models and various discrimination programs according to the fourth modification of the first embodiment and the second embodiment. FIG. 24 is a diagram showing a configuration example of a medical image processing apparatus 300 according to the third embodiment.

DESCRIPTION OF THE PREFERRED EMBODIMENTS An ultrasound diagnostic apparatus and a medical image processing apparatus according to embodiments will be described below with reference to the drawings. Note that the content described in one embodiment or modification may be similarly applied to other embodiments or other modifications.

(First embodiment)
First, a configuration example of the analysis system according to the first embodiment will be described. FIG. 1 is a diagram showing a configuration example of an ultrasound diagnostic system according to a first embodiment. As illustrated in FIG. 1, the ultrasonic diagnostic system according to the first embodiment includes an ultrasonic diagnostic apparatus 1 and a learned model generating apparatus 200. The ultrasonic diagnostic apparatus 1 and the trained model generation apparatus 200 are communicably connected to each other. In the ultrasonic diagnostic system according to the first embodiment, a learned model generation device 200 generates various learned models. Furthermore, the trained model generation device 200 records learning conditions such as network structures and calculation parameters used for various types of learning. Then, the learned model generation device 200 transmits the various types of generated learned models and the recorded various learning conditions to the ultrasound diagnostic device 1. Then, the ultrasonic diagnostic apparatus 1 uses various learned models and various learning conditions transmitted from the learned model generation device 200 to execute various discrimination processes. When performing a certain discrimination process, the ultrasound diagnostic apparatus 1 uses a certain learned model and learning conditions corresponding to this learned model.

The trained model generation device 200 includes a processing circuit 201 . The processing circuit 201 is realized by, for example, a processor. The processing circuit 201 has a trained model generation function 201a.

The trained model generation function 201a performs learning processing to generate a trained model to be used when performing a discrimination process to output a discrimination result regarding a characteristic part of the subject P depicted in ultrasound image data. Execute. In addition, although the following description explains the case where a characteristic site|part is a tumor of a mammary gland, a characteristic site|part is not limited to this.

FIG. 2 is a diagram for explaining an example of processing executed by the trained model generation function 201a according to the first embodiment. As illustrated in FIG. 2, the trained model generation function 201a generates the first trained model 11a and records the first learning condition 11b in the first stage of learning. Further, the trained model generation function 201a generates the second trained model 12a and records the second learning condition 12b in the second stage following the first stage during learning. Further, the trained model generation function 201a generates a third trained model 13a and records the third learning condition 13b in the third stage following the second stage of learning. The trained model generation function 201a asynchronously generates a first trained model 11a, a second trained model 12a, and a third trained model 13a. Further, the trained model generation function 201a records the first learning condition 11b, the second learning condition 12b, and the third learning condition 13b asynchronously. The trained model generation function 201a is an example of a generation unit.

For example, the first learning condition 11b is a group of calculation parameters used when the ultrasound diagnostic apparatus 1 executes the first discrimination process. Moreover, the first learned model 11a is an adjusted parameter used when the ultrasound diagnostic apparatus 1 executes the first discrimination process. Details of the first discrimination process will be described later.

Further, the second learning condition 12b is a group of calculation parameters used when the ultrasound diagnostic apparatus 1 executes the second discrimination process. Further, the second learned model 12a is adjusted parameters used when the ultrasound diagnostic apparatus 1 executes the second discrimination process. Details of the second discrimination process will be described later.

Further, the third learning condition 13b is a group of calculation parameters used when the ultrasound diagnostic apparatus 1 executes the third discrimination process. Further, the third learned model 13a is an adjusted parameter used when the ultrasound diagnostic apparatus 1 executes the third discrimination process. Details of the third discrimination process will be described later.

In this way, each of the plurality of trained models and each of the plurality of learning conditions corresponds to each of the plurality of stages. Further, each of the plurality of trained models and each of the plurality of learning conditions are used when executing each of the plurality of discrimination processes.

Returning to the description of FIG. 1, the ultrasonic diagnostic apparatus 1 includes an apparatus main body 100, an ultrasonic probe 101, an input device 102, a display 103, and a speed detector 104.

The ultrasound probe 101 is used when collecting ultrasound image data. The ultrasound probe 101 includes, for example, a plurality of piezoelectric vibrators. The plurality of piezoelectric vibrators generate ultrasonic waves based on a drive signal supplied from a transmission circuit 110 included in the device main body 100, which will be described later. Further, the ultrasound probe 101 receives a reflected wave from the subject P, converts it into an electric signal (reflected wave signal), and transmits the reflected wave signal to the apparatus main body 100. Further, the ultrasonic probe 101 includes, for example, a matching layer provided on the piezoelectric vibrator and a backing material that prevents propagation of ultrasonic waves backward from the piezoelectric vibrator. Note that the ultrasonic probe 101 is detachably connected to the apparatus main body 100.

When ultrasonic waves are transmitted from the ultrasound probe 101 to the subject P, the transmitted ultrasound waves are successively reflected by acoustic impedance discontinuities in the body tissue of the subject P, and are reflected back to the ultrasound probe 101 as reflected waves. is received by a plurality of piezoelectric vibrators possessed by the camera. The amplitude of the received reflected wave depends on the difference in acoustic impedance at the discontinuity surface from which the ultrasound wave is reflected. Note that when a transmitted ultrasound pulse is reflected from a moving bloodstream or a surface such as a heart wall, the reflected wave depends on the velocity component of the moving object in the ultrasound transmission direction due to the Doppler effect. , subject to frequency deviation.

The ultrasonic probe 101 is provided detachably from the apparatus main body 100. When scanning a two-dimensional region within the subject P (two-dimensional scanning), the operator of the ultrasound diagnostic apparatus 1, for example, moves a 1D array probe in which a plurality of piezoelectric vibrators are arranged in a row to the ultrasound probe 101. It is connected to the device main body 100 as a. The 1D array probe is a linear type ultrasound probe, a convex type ultrasound probe, a sector type ultrasound probe, or the like. Further, when scanning a three-dimensional region within the subject P (three-dimensional scanning), the operator connects, for example, a mechanical 4D probe or a 2D array probe to the apparatus main body 100 as the ultrasound probe 101. Mechanical 4D probes are capable of two-dimensional scanning using multiple piezoelectric vibrators arranged in a line like a 1D array probe, and they also swing the multiple piezoelectric vibrators at a predetermined angle (swing angle). By doing so, three-dimensional scanning is possible. Furthermore, the 2D array probe is capable of three-dimensional scanning using a plurality of piezoelectric vibrators arranged in a matrix, and is also capable of two-dimensional scanning by focusing and transmitting ultrasonic waves.

The input device 102 is implemented by, for example, an input means such as a mouse, keyboard, button, panel switch, touch command screen, foot switch, trackball, joystick, or the like. The input device 102 receives various setting requests from the operator of the ultrasound diagnostic apparatus 1 and transfers the received various setting requests to the apparatus main body 100. For example, the input device 102 receives an instruction (execution instruction) from an operator of the ultrasound diagnostic apparatus 1 to execute CAD processing that automatically detects a characteristic region (characteristic region) in an ultrasound image. , transmits the received execution instruction to the processing circuit 180 of the apparatus main body 100. Further, the operator can also set a ROI (Region Of Interest), which is a search range for a characteristic part, in the ultrasound image in CAD processing via the input device 102.

The display 103 displays, for example, a GUI (Graphical User Interface) for the operator of the ultrasonic diagnostic apparatus 1 to input various setting requests using the input device 102, or displays an ultrasonic image generated in the apparatus main body 100. It also displays ultrasound images and the like indicated by the data. The display 103 is realized by a liquid crystal monitor, a CRT (Cathode Ray Tube) monitor, or the like. Display 103 is an example of a display section.

The speed detector 104 is attached to the ultrasound probe 101 and detects the moving speed (scan speed) of the ultrasound probe 101. Then, the speed detector 104 transmits a detection signal indicating the detected moving speed to the processing circuit 180 of the device main body 100. For example, speed detector 104 is realized by a magnetic sensor. Note that the speed detector 104 is not limited to a magnetic sensor, and may be realized by a known device other than a magnetic sensor capable of detecting the speed of the ultrasound probe 101.

The apparatus main body 100 generates ultrasound image data based on reflected wave signals transmitted from the ultrasound probe 101. The apparatus main body 100 can generate two-dimensional ultrasound image data based on a reflected wave signal corresponding to a two-dimensional region of the subject P transmitted by the ultrasound probe 101. Further, the apparatus main body 100 can generate three-dimensional ultrasound image data based on the reflected wave signal corresponding to the three-dimensional region of the subject P transmitted by the ultrasound probe 101.

As shown in FIG. 1, the apparatus main body 100 includes a transmission circuit 110, a reception circuit 120, a B-mode processing circuit 130, a Doppler processing circuit 140, an image generation circuit 150, an image memory 160, and a storage circuit 170. , and a processing circuit 180. The transmission circuit 110, the reception circuit 120, the B-mode processing circuit 130, the Doppler processing circuit 140, the image generation circuit 150, the image memory 160, the storage circuit 170, and the processing circuit 180 are communicably connected to each other.

The transmission circuit 110 causes the ultrasound probe 101 to transmit ultrasound under the control of the processing circuit 180. The transmission circuit 110 includes a rate pulser generation circuit, a transmission delay circuit, and a transmission pulser, and supplies a drive signal to the ultrasound probe 101. When scanning a two-dimensional region within the subject P, the transmitting circuit 110 causes the ultrasound probe 101 to transmit an ultrasound beam for scanning the two-dimensional region. Further, when scanning a three-dimensional region within the subject P, the transmitting circuit 110 causes the ultrasound probe 101 to transmit an ultrasound beam for scanning the three-dimensional region.

The rate pulsar generation circuit repeatedly generates rate pulses for forming transmission ultrasonic waves (transmission beams) at a predetermined rate frequency (PRF: Pulse Repetition Frequency). As the rate pulses pass through the transmission delay circuit, voltages are applied to the transmission pulsers with different transmission delay times. For example, the transmission delay circuit generates a transmission delay time for each piezoelectric vibrator, which is necessary to focus the ultrasound generated from the ultrasound probe 101 into a beam and determine the transmission directivity, using a rate pulser generation circuit. is given for each rate pulse. The transmission pulser applies a drive signal (drive pulse) to the ultrasound probe 101 at a timing based on the rate pulse. Note that the transmission delay circuit arbitrarily adjusts the transmission direction of the ultrasound from the piezoelectric vibrator surface by changing the transmission delay time given to each rate pulse.

The drive pulse is transmitted from the transmitting pulser to the piezoelectric vibrator in the ultrasound probe 101 via the cable, and then converted from an electrical signal into a mechanical vibration in the piezoelectric vibrator. The ultrasonic waves generated by this mechanical vibration are transmitted into the living body of the subject P. Here, the ultrasonic waves having different transmission delay times for each piezoelectric vibrator are focused and propagated in a predetermined direction.

Note that the transmitting circuit 110 has a function of instantaneously changing the transmitting frequency, transmitting drive voltage, etc. in order to execute a predetermined scan sequence under the control of the processing circuit 180. In particular, changing the transmission drive voltage is achieved by a linear amplifier-type oscillation circuit that can instantly switch its value, or by a mechanism that electrically switches multiple power supply units.

After the reflected waves of the ultrasound transmitted by the ultrasound probe 101 reach the piezoelectric vibrator inside the ultrasound probe 101, the mechanical vibrations are converted into electrical signals (reflected wave signals) in the piezoelectric vibrator. . Then, the receiving circuit 120 receives the reflected wave signal transmitted from the ultrasound probe 101. Under the control of the processing circuit 180, the receiving circuit 120 performs various processing on the reflected wave signal to generate reflected wave data, and outputs the generated reflected wave data to the B-mode processing circuit 130 and the Doppler processing circuit 140. do. For example, each time the receiving circuit 120 receives one frame of reflected wave signals, it generates one frame of reflected wave data from the received reflected wave signals. The receiving circuit 120 generates two-dimensional reflected wave data from the two-dimensional reflected wave signal transmitted from the ultrasound probe 101. Further, the receiving circuit 120 generates three-dimensional reflected wave data from the three-dimensional reflected wave signal transmitted from the ultrasound probe 101.

The receiving circuit 120 includes a preamplifier, an A/D (Analog to Digital) converter, a quadrature detection circuit, and the like. The preamplifier amplifies the reflected wave signal for each channel and performs gain adjustment (gain correction). The A/D converter converts the gain-corrected reflected wave signal into a digital signal by A/D converting the gain-corrected reflected wave signal. The quadrature detection circuit converts the digital signal into a baseband in-phase signal (I signal, I: In-phase) and a quadrature signal (Q signal, Q: Quadrature-phase). Then, the quadrature detection circuit outputs the I signal and the Q signal (IQ signal) to the B mode processing circuit 130 and the Doppler processing circuit 140 as reflected wave data.

Under the control of the processing circuit 180, the B-mode processing circuit 130 performs logarithmic amplification, envelope detection processing, logarithmic compression, etc. on the reflected wave data output from the reception circuit 120, and generates a signal for each sample point. Data (B mode data) in which intensity (amplitude intensity) is expressed by brightness of luminance is generated. For example, each time the B-mode processing circuit 130 receives one frame of reflected wave data, it generates one frame of B-mode data from the received reflected wave data. B-mode processing circuit 130 outputs the generated B-mode data to image generation circuit 150. B-mode processing circuit 130 is realized by, for example, a processor.

The Doppler processing circuit 140 performs frequency analysis on the reflected wave data output from the receiving circuit 120 under the control of the processing circuit 180 to detect moving objects (blood flow, tissues, contrast agent echo components, etc.) based on the Doppler effect. The motion information is extracted, and data (Doppler data) indicating the extracted motion information is generated. For example, the Doppler processing circuit 140 extracts average velocity, dispersion, power, etc. at multiple points as motion information of the moving object, and generates Doppler data indicating the extracted motion information of the moving object. For example, each time the Doppler processing circuit 140 receives one frame of reflected wave data, it generates one frame of Doppler data from the received reflected wave data. The Doppler processing circuit 140 outputs the generated Doppler data to the image generation circuit 150. Doppler processing circuit 140 is realized by, for example, a processor.

Further, by using the function of the Doppler processing circuit 140 described above, the ultrasound diagnostic apparatus 1 according to the present embodiment can perform a color Doppler method also called a color flow mapping method (CFM). In the CFM method, ultrasonic waves are transmitted and received multiple times on multiple scanning lines. In the CFM method, a moving target indicator (MTI) filter is applied to the data string at the same position to suppress signals (clutter signals) originating from stationary or slow-moving tissues. , extracting signals originating from blood flow. In the CFM method, blood flow information such as blood flow velocity, blood flow dispersion, and blood flow power is estimated from this blood flow signal. An image generation circuit 150, which will be described later, generates ultrasound image data (color Doppler image data) in which the distribution of the estimation results is displayed in two-dimensional color, for example. Then, the display 103 displays a color Doppler image indicated by the color Doppler image data.

The B-mode processing circuit 130 and the Doppler processing circuit 140 are capable of processing both two-dimensional reflected wave data and three-dimensional reflected wave data.

The image generation circuit 150 generates ultrasound image data from various data (B-mode data and Doppler data) output by the B-mode processing circuit 130 and the Doppler processing circuit 140 under the control of the processing circuit 180. For example, every time the image generation circuit 150 receives one frame of various data output from the B-mode processing circuit 130 and the Doppler processing circuit 140, the image generation circuit 150 generates one frame of ultrasound image data from the received various data. generate. That is, the image generation circuit 150 generates a plurality of pieces of ultrasound image data in time series at a predetermined frame rate in real time with respect to ultrasound scanning. In this way, in the present embodiment, the ultrasound probe 101, the receiving circuit 120, the B-mode processing circuit 130, the Doppler processing circuit 140, and the image generation circuit 150 generate multiple ultrasound image data in real time in chronological order. is collected at a predetermined frame rate. The ultrasound image data is data based on reflected wave signals obtained by ultrasound scanning performed by the ultrasound probe 101. The ultrasound probe 101, the receiving circuit 120, the B-mode processing circuit 130, the Doppler processing circuit 140, and the image generation circuit 150 are examples of the acquisition section. Further, ultrasound image data is an example of medical image data.

Then, the image generation circuit 150 causes the image memory 160 to store a plurality of chronologically generated ultrasonic image data in real time. To explain with a specific example, the image generation circuit 150 stores the generated ultrasound image data in the image memory 160 every time it generates one frame of ultrasound image data. For example, among a plurality of chronologically generated ultrasound image data obtained by ultrasound scanning, the ultrasound image data generated first is referred to as the first frame ultrasound image data. Similarly, the Nth (N is an integer of 1 or more) generated ultrasound image data is referred to as the Nth frame ultrasound image data.

Image generation circuit 150 is realized by a processor. Here, the image generation circuit 150 converts (scan convert) the scanning line signal sequence of ultrasound scanning into a scanning line signal sequence of a video format typified by television etc., and generates ultrasound image data for display. . For example, the image generation circuit 150 generates ultrasound image data for display by performing coordinate transformation in accordance with the scanning form of ultrasound by the ultrasound probe 101. In addition to scan conversion, the image generation circuit 150 performs various types of image processing, such as image processing (smoothing processing) that regenerates an average luminance image using a plurality of image frames after scan conversion; Performs image processing (edge enhancement processing), etc. using a differential filter within the image. Further, the image generation circuit 150 synthesizes text information of various parameters, scales, body marks, etc. to the ultrasound image data.

Further, the image generation circuit 150 generates three-dimensional B-mode image data by performing coordinate transformation on the three-dimensional B-mode data generated by the B-mode processing circuit 130. Further, the image generation circuit 150 generates three-dimensional Doppler image data by performing coordinate transformation on the three-dimensional Doppler data generated by the Doppler processing circuit 140. That is, the image generation circuit 150 generates "three-dimensional B-mode image data and three-dimensional Doppler image data" as "three-dimensional ultrasound image data (volume data)." The image generation circuit 150 then performs various rendering processes on the volume data in order to generate various types of two-dimensional image data for displaying the volume data on the display 103.

The B-mode data and Doppler data are ultrasound image data before scan conversion processing, and the data generated by the image generation circuit 150 is ultrasound image data for display after scan conversion processing. Note that B-mode data and Doppler data are also called raw data.

The image memory 160 is a memory that stores various image data generated by the image generation circuit 150. Image memory 160 also stores data generated by B-mode processing circuit 130 and Doppler processing circuit 140. The B-mode data and Doppler data stored in the image memory 160 can be called up by the operator after diagnosis, for example, and become ultrasound image data for display via the image generation circuit 150. For example, the image memory 160 is implemented by a RAM, a semiconductor memory device such as a flash memory, a hard disk, or an optical disk.

The storage circuit 170 stores control programs for transmitting and receiving ultrasound waves, image processing, and display processing, diagnostic information (e.g., patient ID, doctor's findings, etc.), and various data such as diagnostic protocols and various body marks. . Furthermore, the storage circuit 170 is also used for storing data stored in the image memory 160, etc., as necessary. For example, the storage circuit 170 is realized by a semiconductor memory device such as a flash memory, a hard disk, or an optical disk. The memory circuit 170 is an example of a memory section.

The processing circuit 180 controls the entire processing of the ultrasound diagnostic apparatus 1. The processing circuit 180 is realized by, for example, a processor. The processing circuit 180 has a control function 181, a speed detection function 182, an acquisition function 183, a discrimination function 184, a marker information generation function 185, and a display control function 186.

Here, for example, each processing function that the processing circuit 180 has is stored in the storage circuit 170 in the form of a computer-executable program. The processing functions of the control function 181, speed detection function 182, acquisition function 183, discrimination function 184, marker information generation function 185, and display control function 186, which are the components of the processing circuit 180 shown in FIG. 1, can be executed by a computer. It is stored in the storage circuit 170 in the form of a program. The processing circuit 180 reads each program from the storage circuit 170 and executes each read program, thereby realizing a function corresponding to each program. In other words, the processing circuit 180 in a state where each program is read has each function shown in the processing circuit 180 of FIG.

The term "processor" used in the above description refers to, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an Application Specific Integrated Circuit (ASIC), or a programmable logic device (for example, It means a circuit such as a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), or a field programmable gate array (FPGA). The processor realizes its functions by, for example, reading and executing a program stored in the storage circuit 170. Note that instead of storing the program in the storage circuit 170, the program may be directly incorporated into the circuit of the processor. In this case, the processor realizes its functions by reading and executing a program built into the circuit. Note that each processor of this embodiment is not limited to being configured as a single circuit for each processor, but may also be configured as a single processor by combining multiple independent circuits to realize its functions. good. Furthermore, multiple components in FIG. 1 may be integrated into one processor to implement its functions. The same applies to the word "processor" used in the following explanation.

The control function 181 controls the transmission circuit 110, the reception circuit 120, the B-mode processing circuit 130, and the Doppler processing based on various setting requests input by the operator via the input device 102 and various data read from the storage circuit 170. The processing of the circuit 140 and the image generation circuit 150 is controlled.

The speed detection function 182 detects the moving speed of the ultrasound probe 101 by calculating the moving speed of the ultrasound probe 101 indicated by the detection signal based on the detection signal transmitted from the speed detector 104. For example, the speed detection function 182 detects the moving speed of the ultrasound probe 101 at predetermined time intervals. The speed detector 104 and the speed detection function 182 are examples of a detection unit.

The acquisition function 183 acquires the first trained model 11a, the second trained model 12a, and the third trained model 13a transmitted from the trained model generation device 200. Then, the acquisition function 183 causes the storage circuit 170 to store the acquired first trained model 11a, second trained model 12a, and third trained model 13a. Furthermore, the acquisition function 183 acquires the first learning condition 11b, second learning condition 12b, and third learning condition 13b transmitted from the trained model generation device 200. Then, the acquisition function 183 causes the storage circuit 170 to store the acquired first learning condition 11b, second learning condition 12b, and third learning condition 13b. The various learned models and various learning conditions stored in the storage circuit 170 are used when the discrimination function 184 performs various discrimination processes.

Further, when receiving an instruction (execution instruction) to execute CAD processing input from the operator via the input device 102, the acquisition function 183 acquires a plurality of ultrasound image data in chronological order. do. For example, after receiving an execution instruction, the acquisition function 183 acquires one newly generated frame of ultrasound image data from the image memory 160 every time one frame of newly generated ultrasound image data is stored in the image memory 160. Acquire ultrasound image data for minutes. For example, a plurality of chronologically arranged ultrasound image data is a data group obtained by ultrasound scanning. Such ultrasonic scanning is performed, for example, while the scanning area inside the subject P is being moved by the movement of the ultrasonic probe 101 that is moved along the body surface of the subject P by the operator, or while the scanning area inside the subject P is being moved. This may be performed while the probe 101 is stopped. Furthermore, in the following description, each of a plurality of pieces of ultrasound image data in time series is, for example, B-mode image data.

The determination function 184 uses various learned models and various learning conditions stored in the storage circuit 170 to execute a process during operation of deriving a determination result regarding a characteristic part. For example, every time one newly generated frame of ultrasound image data is acquired by the acquisition function 183, the determination function 184 uses the acquired one frame of ultrasound image data to perform processing during operation. Execute.

FIG. 3 is a diagram for explaining an example of a process executed by the determination function 184 according to the first embodiment. As illustrated in FIG. 3, in the first stage of operation, the discrimination function 184 uses the first trained model 11a and the first learning conditions 11b to identify characteristic parts in the ultrasound image as CAD processing. A first determination process is executed to determine whether the content is included or not, and to derive a determination result (first determination result). That is, if the determining function 184 determines that the ultrasound image includes a characteristic region, this means that a characteristic region has been detected. Furthermore, if the determining function 184 determines that the ultrasound image does not include a characteristic region, it means that no characteristic region has been detected.

In addition, in the second stage of operation, the determination function 184 uses the second learned model 12a and the second learning condition 12b to determine whether the feature region detected in the first stage of operation is benign. or malignant, and performs a second determination process to derive a determination result (second determination result). In addition, in the third stage of operation, the discrimination function 184 uses the third learned model 13a and the third learning condition 13b to determine the malignancy of the characteristic part that was determined to be malignant in the second stage of operation. A third determination process is performed to determine the determination result (third determination result). For example, the malignancy includes two malignancies: malignancy A and malignancy B, which is higher in malignancy than malignancy A. In this case, the determination function 184 determines whether the degree of malignancy of the characteristic site determined to be malignant is grade A or grade B. A case will be described below in which the discrimination function 184 discriminates between two malignancies, malignancy A and malignancy B. However, the malignancy is limited to two malignancies, malignancy A and malignancy B. I can't do it. The determination function 184 is an example of a determination unit.

FIG. 4 is a diagram for explaining an example of the first stage processing during learning and the first stage processing during operation according to the first embodiment. As illustrated in FIG. 4, in the first stage of learning, the learned model generation function 201a creates the first learned model by learning the relationship between the ultrasound image data 14 and the first label 15. A model 11a is generated and a first learning condition 11b is recorded. Here, the first label 15 is a label indicating whether or not a characteristic region is included in the ultrasound image represented by the ultrasound image data 14. That is, the learned model generation function 201a generates the first learned model 11a by learning the ultrasound image data 14 and the first label 15 in association with each other, and records the first learning condition 11b. do. The ultrasound image data 14 is, for example, B-mode image data generated from B-mode data. The ultrasound image data 14 is data generated by the ultrasound diagnostic apparatus 1 or an ultrasound diagnostic apparatus other than the ultrasound diagnostic apparatus 1.

Note that the ultrasound image data 14 may be B-mode image data that has been subjected to preprocessing such as edge enhancement and contrast enhancement. Further, the ultrasound image data 14 may be, for example, B-mode data (B-mode raw data). Further, the ultrasound image data 14 may be color Doppler image data. Further, the ultrasound image data 14 may be image data indicating FLR (Fat Lesion Ratio) obtained by strain elastosis. Further, the ultrasound image data 14 is shear velocity image data obtained by SWE (Shear Wave Elastography), in which pixel values are assigned to each sample point according to the shear velocity at each of a plurality of sample points within the scanning area. (shear wave speed image data). Further, the ultrasound image data 14 may be B-mode image data obtained by the ultrasound diagnostic apparatus 1 or an ultrasound diagnostic apparatus other than the ultrasound diagnostic apparatus 1 while performing strain elastography or SWE.

Further, the first label 15 is a label obtained by, for example, an operator determining whether or not a characteristic region is included in the ultrasound image represented by the ultrasound image data 14. Note that the first label 15 may be a label obtained by performing CAD processing on the ultrasound image data 14 using the ultrasound diagnostic apparatus 1 or other equipment.

For example, the trained model generation function 201a performs machine learning by inputting a set of ultrasound image data 14 and first label 15 to a machine learning engine as learning data (teacher data).

For example, machine learning engines include deep learning, neural networks, logistic regression analysis, nonlinear discriminant analysis, support vector machines (SVM), and random forests. Machine learning is performed using various algorithms such as , Naive Bayes, etc.

As a result of such machine learning, the learned model generation function 201a generates a first learned model used in a first discrimination process to derive a first discrimination result for input ultrasound image data. 11a. Further, as a result of the machine learning described above, the learned model generation function 201a records the first learning condition 11b used for calculation in the first discrimination process. The learned model generation function 201a then transmits the generated first learned model 11a and the recorded first learning condition 11b to the ultrasound diagnostic apparatus 1. As a result, the first learned model 11a and the first learning condition 11b are stored in the storage circuit 170 of the ultrasound diagnostic apparatus 1.

Note that the trained model generation device 200 and the ultrasound diagnostic device 1 do not need to be connected via a network. In this case, various trained models and various learning conditions generated by the trained model generation device 200 are stored in a storage medium such as a USB (Universal Serial Bus) memory. Then, the ultrasonic diagnostic apparatus 1 reads out various learned models and various learning conditions stored in the storage medium, and stores the read out various learned models and various learning conditions in the storage circuit 170.

Next, an example of the first stage processing during operation will be described. As illustrated in FIG. 4, in the first stage of operation, the discrimination function 184 uses the first learned model 11a and the first learning condition 11b to generate an ultrasound image shown in the ultrasound image data 16. A first discrimination process is performed as a CAD process in which it is determined whether a characteristic part is included or not, and a discrimination result (first discrimination result) 17 is derived. For example, the type of ultrasound image data 16 is the same as the type of ultrasound image data 14. For example, B-mode image data subjected to the above-mentioned preprocessing, B-mode image data not subjected to the above-mentioned preprocessing, B-mode data, color Doppler image data, image data indicating FLR obtained by strain elastography, Examples of the various types of ultrasound image data include shear velocity image data, B-mode image data obtained during strain elastography or SWE, and the like.

For example, when determining that the ultrasound image includes a characteristic region, the determination function 184 outputs, as the first determination result 17, information indicating that the ultrasound image contains the characteristic region. Further, when determining that the ultrasound image does not include the characteristic region, the determination function 184 outputs information indicating that the ultrasound image does not include the characteristic region as the first determination result 17.

FIG. 5 is a diagram for explaining an example of second-stage processing during learning and second-stage processing during operation according to the first embodiment. As illustrated in FIG. 5, in the second stage of learning, the learned model generation function 201a learns the relationship between the ultrasound image data 20 and the second label 21, thereby generating the second learned model. A model 12a is generated and a second learning condition 12b is recorded. Here, the second label 21 is a label indicating whether the characteristic region included in the ultrasound image represented by the ultrasound image data 20 is malignant or benign. That is, the learned model generation function 201a generates the second learned model 12a by learning the ultrasound image data 20 and the second label 21 in association with each other, and records the second learning condition 12b. do.

The ultrasound image data 20 used when generating the second trained model 12a and recording the second learning conditions 12b is image data in which characteristic parts are depicted. The ultrasound image data 20 is, for example, B-mode image data generated from B-mode data. The ultrasound image data 20 is data generated by the ultrasound diagnostic apparatus 1 or an ultrasound diagnostic apparatus other than the ultrasound diagnostic apparatus 1.

For example, the trained model generation function 201a performs machine learning by inputting a set of ultrasound image data 20 and second label 21 as learning data to a machine learning engine.

As a result of such machine learning, the learned model generation function 201a is used in a second discrimination process to derive a second discrimination result for input ultrasound image data in which characteristic parts are depicted. A second trained model 12a is generated. Further, as a result of the machine learning described above, the learned model generation function 201a records the second learning condition 12b used for calculation in the second discrimination process. The trained model generation function 201a then transmits the generated second trained model 12a and the recorded second learning conditions 12b to the ultrasound diagnostic apparatus 1. As a result, the second learned model 12a and the second learning condition 12b are stored in the storage circuit 170 of the ultrasound diagnostic apparatus 1.

Next, an example of the second stage processing during operation will be described. As illustrated in FIG. 5, in the second stage of operation, the discrimination function 184 uses the second learned model 12a and the second learning condition 12b to A second discrimination process is executed to determine whether the characteristic region depicted in the ultrasound image data 22 is benign or malignant, and to derive a discrimination result (second discrimination result) 23.

For example, when the determination function 184 determines that the characteristic region depicted in the ultrasound image data 22 is benign, the determination function 184 outputs information indicating that the characteristic region is benign as the second determination result 23. . Further, when determining that the characteristic region depicted in the ultrasound image data 22 is malignant, the determination function 184 outputs information indicating that the characteristic region is malignant as a second determination result 23. .

Here, the ultrasound image data 22 used when deriving the second discrimination result 23 is the ultrasound image data that was determined to include a characteristic part in the first discrimination process at the first stage during operation. be. That is, the ultrasound image data 22 used in the second discrimination process is the ultrasound image data that was determined to include the characteristic region in the first discrimination process. Note that, for example, the type of ultrasound image data 22 is the same as the type of ultrasound image data 20.

FIG. 6 is a diagram for explaining an example of the third stage processing during learning and the third stage processing during operation according to the first embodiment. As illustrated in FIG. 6, in the third stage of learning, the learned model generation function 201a learns the relationship between the ultrasound image data 26 and the third label 27, thereby generating the third learned model. A model 13a is generated and a third learning condition 13b is recorded. Here, the third label 27 is a label indicating whether the malignancy level of the malignant characteristic site included in the ultrasound image shown by the ultrasound image data 26 is malignancy level A or malignancy level B. . That is, the learned model generation function 201a generates the third learned model 13a by learning the ultrasound image data 26 and the third label 27 in association with each other, and records the third learning condition 13b. do.

The ultrasound image data 26 used when generating the third learned model 13a and recording the third learning condition 13b is image data in which malignant characteristic parts are depicted. The ultrasound image data 26 is, for example, B-mode image data generated from B-mode data. The ultrasound image data 26 is data generated by the ultrasound diagnostic apparatus 1 or an ultrasound diagnostic apparatus other than the ultrasound diagnostic apparatus 1.

For example, the trained model generation function 201a performs machine learning by inputting a set of ultrasound image data 26 and third label 27 to a machine learning engine as learning data.

As a result of such machine learning, the learned model generation function 201a performs a third discrimination process to derive a third discrimination result for input ultrasound image data in which malignant characteristic parts are depicted. A third trained model 13a to be used is generated. Further, as a result of the machine learning as described above, the learned model generation function 201a records the third learning condition 13b used for calculation in the third discrimination process. The trained model generation function 201a then transmits the generated third trained model 13a and the recorded third learning condition 13b to the ultrasound diagnostic apparatus 1. As a result, the third learned model 13a and the third learning condition 13b are stored in the storage circuit 170 of the ultrasound diagnostic apparatus 1.

Next, an example of the third stage processing during operation will be described. As illustrated in FIG. 6, in the third stage of operation, the discrimination function 184 uses the third learned model 13a and the third learning condition 13b to A third step that determines whether the malignancy of the malignant characteristic site depicted in the ultrasound image data 28 is malignancy level A or malignancy level B, and derives a determination result (third determination result) 29. Execute the determination process.

For example, when determining that the malignancy of the malignant characteristic site depicted in the ultrasound image data 28 is malignancy grade A, the determination function 184 determines that the malignancy of the malignant characteristic site is malignant as the third determination result 29. Information indicating that the grade is malignant grade A is output. Further, when determining that the malignancy of the malignant characteristic site depicted in the ultrasound image data 28 is malignancy grade B, the determination function 184 determines that the malignant characteristic site is malignant as a third determination result 29. Information indicating that the grade is malignant grade B is output.

Here, the ultrasonic image data 28 used when deriving the third discrimination result 29 is an ultrasonic image data 28 that depicts a characteristic region determined to be malignant in the second discrimination process at the second stage during operation. This is sound wave image data. That is, the ultrasound image data 28 used in the third discrimination process is ultrasound image data in which a characteristic region determined to be malignant in the second discrimination process is depicted. Note that, for example, the type of ultrasound image data 28 is the same as the type of ultrasound image data 26.

An example of the process during learning and the process during operation has been described above with reference to FIGS. 2 to 6. As described above, the learned model generation function 201a generates various learned models by learning the association between ultrasound image data and labels related to characteristic parts depicted in the ultrasound image data, Record various learning conditions. Various types of discrimination processing using various learned models and various learning conditions output discrimination results regarding characteristic parts. The trained model generation function 201a generates such a trained model at each of a plurality of stages during learning (the first to third stages during learning described above). Further, the trained model generation function 201a records such learning conditions at each of a plurality of stages during learning.

The trained model generation function 201a generates a trained model specialized for a function that performs a specific discrimination at each stage during learning, and records learning conditions. Therefore, the discrimination process based on the trained model generated at each stage of learning and the recorded learning conditions is a process that allows specific discrimination to be performed with high accuracy. For example, in the first stage of learning, the trained model generation function 201a generates a first trained model 11a specialized for the function of determining whether or not a characteristic region is included in an ultrasound image. and record the first learning condition 11b. Therefore, according to the first discrimination process based on the first learned model 11a and the first learning condition 11b, it is not possible to accurately discriminate whether or not a characteristic region is included in an ultrasound image. can.

Furthermore, in the second stage of learning, the trained model generation function 201a generates a second trained model 12a specialized for the function of determining whether a characteristic part is malignant or benign, Record the second learning condition 12b. Therefore, according to the second discrimination process based on the second trained model 12a and the second learning condition 12b, it is possible to accurately discriminate whether the characteristic region is malignant or benign.

In addition, in the third stage of learning, the learned model generation function 187 has a function specialized for determining whether the malignancy of a malignant characteristic part is malignancy level A or malignancy level B. The third learned model 13a is generated, and the third learning condition 13b is recorded. Therefore, according to the third discrimination process based on the third learned model 13a and the third learning condition 13b, it is determined whether the malignancy of the malignant characteristic part is malignancy level A or malignancy level B. can be determined with high accuracy.

That is, the determination function 184 executes a unique determination process that is different from determination processes performed in other stages at each of a plurality of stages during operation. Therefore, it has a first trained model 11a, a second trained model 12a, and a third trained model 13a, as well as a first learning condition 11b, a second learning condition 12b, and a third learning condition 13b. According to the ultrasonic diagnostic apparatus 1, various discrimination results regarding characteristic parts can be obtained with high accuracy.

Furthermore, as described above, the trained model generation function 201a can, for example, detect a characteristic region without using ultrasound image data in which a characteristic region to be determined as malignant or benign is depicted. A second trained model 12a is generated using the visualized ultrasound image data. Therefore, the trained model generation function 201a can accelerate the convergence of machine learning when generating the second trained model 12a.

In addition, the trained model generation function 201a can perform such a discrimination without using ultrasound image data depicting benign characteristic areas that are not targets for discrimination between malignant grade A and malignant grade B, for example. A third trained model 13a is generated using ultrasound image data in which the target malignant characteristic site is depicted. Therefore, the trained model generation function 201a can accelerate the convergence of machine learning when generating the third trained model 13a.

In addition, as described above, the determination function 184 uses various learned models and various learning conditions in each of the plurality of stages during operation (the first to third stages during operation described above). Deriving various discrimination results regarding characteristic parts. For example, in the first stage of operation, the discrimination function 184 derives the first discrimination result 17 regarding the characteristic region depicted in the input ultrasound image data 16. Further, the discrimination function 184 derives a second discrimination result 23 regarding the characteristic region depicted in the input ultrasound image data 22 in the second stage of operation. Further, in the third stage of operation, the discrimination function 184 derives a third discrimination result 29 regarding the characteristic region depicted in the input ultrasound image data 28.

Further, the discrimination function 184 inputs the ultrasound image data 22 (see FIG. 5) corresponding to the predetermined discrimination result derived by the first discrimination process to the second discrimination process. Here, the predetermined discrimination result derived by the first discrimination process refers to the first discrimination result 17 (see FIG. 4) that indicates that a characteristic region is included in the ultrasound image.

In other words, the discrimination function 184 performs the second discrimination process without inputting ultrasound image data in which the characteristic region to be determined as malignant or benign is depicted. The rendered ultrasound image data 22 is input. Then, the determination function 184 derives the second determination result 23 derived by the second determination process as a determination result corresponding to the second stage during operation.

Further, the discrimination function 184 inputs the ultrasound image data 28 (see FIG. 6) corresponding to the predetermined discrimination result derived by the second discrimination process to the third discrimination process. Here, the predetermined discrimination result derived by the second discrimination process refers to the second discrimination result 23 (see FIG. 5) that indicates that the characteristic site is malignant.

In other words, the discrimination function 184 inputs ultrasonic image data in which a characteristic malignant region that is to be determined as being grade A or grade B is not visualized for the third discrimination process. Ultrasonic image data 28 in which a characteristic malignant region is depicted is input without any modification. Then, the discrimination function 184 derives the third discrimination result 29 derived by the third discrimination process as the discrimination result corresponding to the third stage.

In this way, the discrimination function 184 selects the ultrasound image data corresponding to the predetermined discrimination result derived by the discrimination process corresponding to one of the plurality of stages to the next stage of the first stage. The input is input to the discrimination process, and the discrimination result derived by the discrimination process corresponding to the next stage of the first stage is derived as the discrimination result corresponding to the next stage of the first stage.

As described above, the discrimination function 184 uses the first trained model 11a and the first learning condition 11b that can accurately discriminate whether a characteristic region is included or not included in an ultrasound image. , the first discrimination result 17 (see FIG. 4) is derived. Therefore, the discrimination function 184 can derive the first discrimination result 17 with good discrimination accuracy.

Further, the discrimination function 184 uses the second trained model 12a and the second learning condition 12b, which can accurately discriminate whether the characteristic region is malignant or benign, to obtain a second discrimination result. 23 (see FIG. 5). Therefore, the discrimination function 184 can derive the second discrimination result 23 with good discrimination accuracy.

Further, the discrimination function 184 uses the third learned model 13a and the third learning condition 13b that can accurately discriminate whether the malignancy of the characteristic region is malignancy level A or malignancy level B. Then, a third determination result 29 (see FIG. 6) is derived. Therefore, the discrimination function 184 can derive the third discrimination result 29 with good discrimination accuracy.

For these reasons, according to the ultrasonic diagnostic apparatus 1 according to the first embodiment, it is possible to accurately obtain discrimination results regarding characteristic parts.

Next, an example of the first discrimination process executed by the discrimination function 184 using the first learned model 11a and the first learning condition 11b in the first stage of operation will be described. FIG. 7 is a flowchart showing an example of the flow of the first determination process according to the first embodiment.

As shown in FIG. 7, the determination function 184 executes a first detection process (step S101) and a second detection process (step S102).

First, the first detection process executed in step S101 will be described. The determination function 184 uses the search window, the first trained model for search, and the search algorithm to search for a characteristic region in a search range (ROI) set in the ultrasound image by the operator. For example, the determination function 184 moves the search window to multiple positions within the search range. In addition, the discrimination function 184 analyzes the image information in the search window at each position using the first search trained model and the search algorithm, so that the image information in each search window becomes a characteristic part. A corresponding probability (first probability) is calculated.

The search window is, for example, a unit of area within the search range that is compared with the first trained model for search. The first trained model for search is, for example, a model that has been trained on image information that is a sample of a characteristic part. The search algorithm is an algorithm for calculating the above-mentioned first probability and searching for image information corresponding to a characteristic region in an ultrasound image to be processed.

The first trained model for search is an example of a first pattern indicating a characteristic part.

Then, when the calculated first probability is greater than or equal to a predetermined threshold, the determination function 184 detects the range surrounded by the search window as a characteristic part (range of characteristic parts). That is, the discrimination function 184 determines that when the first probability indicating the degree of similarity between the feature of a part of the ultrasound image and the feature of the first search trained model indicating the characteristic region is greater than or equal to the threshold value. determines that a characteristic region is included in the ultrasound image. Further, the determination function 184 determines that the ultrasound image does not include the characteristic region when the first probability is less than the threshold value. Then, the determination function 184 derives a first determination result that is a result of determining whether or not the characteristic region is included in the ultrasound image. The first detection process is an example of a characteristic part detection process.

Then, the determination function 184 causes the storage circuit 170 to store the search results (detection results) for the ultrasound images for each frame of ultrasound image data. For example, if the determining function 184 is able to detect a characteristic region from one frame of an ultrasound image, the determination function 184 causes the storage circuit 170 to store position information indicating the position of the characteristic region as a search result. Note that the position of the characteristic region referred to here is the position in the image space of ultrasound image data.

Here, with reference to FIGS. 8 and 9, the characteristic parts detected by the discrimination function 184 include a correctly detected characteristic part and a characteristic part that is not actually a characteristic part but is detected incorrectly. It will be explained that there are two types of characteristic parts. FIGS. 8 and 9 are diagrams for explaining an example of detection results of characteristic parts by the discrimination function 184 according to the first embodiment.

FIG. 8 shows a case where the discrimination function 184 correctly detects the detection. As illustrated in FIG. 8, for example, the discrimination function 184 performs a first detection process on the ultrasound image, and selects the range that includes the actual characteristic region 31 surrounded by the search window 30 as the characteristic region. To detect. Then, the discrimination function 184 causes the storage circuit 170 to store information indicating the position of the characteristic region detected in the ultrasound image.

On the other hand, FIG. 9 shows a case where a characteristic part is erroneously detected by the discrimination function 184. As illustrated in FIG. 9, for example, the discrimination function 184 performs the first detection process on the ultrasound image and erroneously detects a range that does not include the actual characteristic region 31 surrounded by the search window 30. It may be detected as a characteristic part. Incorrectly detecting a feature region in this way is called overdetection (false positive). Even in such a case, the discrimination function 184 causes the storage circuit 170 to store information indicating the position of the characteristic region detected in the ultrasound image.

Further, when the characteristic region cannot be detected from one frame of the ultrasound image, the determination function 184 stores information indicating that the characteristic region could not be detected in the storage circuit 170 as a search result. to be memorized.

Note that the determination function 184 calculates the first probability as an example of an index indicating the degree of similarity between the features of the image information in the search window 30 and the features of the first trained model for search. explained. However, instead of using the first probability, the determination function 184 uses other factors such as similarity, which indicates the degree to which the features of the image information in the search window 30 and the features of the first trained model for search are similar. An index may also be calculated.

An example of the first detection process executed in step S101 has been described above. FIG. 10 is a diagram illustrating an example of the results of the first detection process according to the first embodiment. The “results determined as characteristic regions” shown in FIG. 10 are image information in the search window 30 at the position of the search window 30 determined to be a characteristic region by the first detection process from one frame of the ultrasound image. An example is shown. 10 shows an example of the image information in the search window 30 when a characteristic region is not detected in the first detection process from one frame of the ultrasound image. . Note that FIG. 10 shows images of 10 locations as an example of the results in which the search window 30 is placed at multiple locations within the search range in the first detection process and the first probability is calculated at each location. Information is shown.

As shown in FIG. 10, the discrimination function 184 detects characteristic parts at six positions in the first detection process. Specifically, as shown in FIG. 10, the determination function 184 determines that six pieces of image information 40a to 40f within the search window 30 are characteristic parts.

The image information 40a is image information in which the actual characteristic region 31a is depicted. The image information 40b is over-detected image information. The image information 40c is image information in which the actual characteristic region 31b is depicted. The image information 40d is image information in which the actual characteristic region 31c is depicted. The image information 40e is image information in which the actual characteristic region 31d is depicted. The image information 40f is over-detected image information. That is, in the case shown in FIG. 10, two of the six pieces of image information determined to be characteristic parts are overdetected.

Further, as shown in FIG. 10, the determination function 184 did not detect characteristic parts at four positions in the first detection process. Specifically, as shown in FIG. 10, the determination function 184 determines that the four pieces of image information 40g to 40j within the search window 30 are not characteristic parts.

Next, the second detection process executed in step S102 will be explained. The discrimination function 184 uses a trained model for overdetection discrimination and a search algorithm to determine whether the feature region (image information within the search window 30) detected in the first detection process is overdetected in the first detection process. A probability (second probability) corresponding to the image information estimated to be generated is calculated.

The trained model for overdetection discrimination is, for example, a model that has learned to discriminate between image information estimated to be overdetected in the first detection process and image information of a characteristic part. The search algorithm is an algorithm that calculates the second probability described above and searches for image information corresponding to image information that is estimated to be over-detected in the first detection process in the ultrasound image to be processed. It is.

In the first detection process, the trained model for overdetection discrimination extracts image information that is estimated to be incorrectly determined if a characteristic part is included even though the ultrasound image does not contain a characteristic part. This is a model that was trained for the purpose of The trained model for overdetection discrimination is an example of the second pattern.

Here, as illustrated in FIG. 10, a case will be described in which six pieces of image information 40a to 40f are detected in the first detection process. In this case, the determination function 184 calculates a second probability that each of the six pieces of image information 40a to 40f corresponds to image information that is estimated to be over-detected in the first detection process in the second detection process. calculate. In this way, the determination function 184 calculates the second probability for each of the six pieces of image information 40a to 40f.

Then, the determination function 184 detects, among the six image information 40a to 40f, image information for which the calculated second probability is greater than or equal to a predetermined threshold value as image information that was overdetected in the first detection process. . Then, the determination function 184 causes the storage circuit 170 to store the search results (detection results) for the ultrasound images for each frame of ultrasound image data. For example, if over-detected image information is detected in one frame of ultrasound image, the determination function 184 stores position information indicating the position of the detected image information in the storage circuit 170 as a search result. Make me remember.

Note that in the second detection process, the determination function 184 may be able to correctly detect the image information that was overdetected in the first detection process, and may not be able to correctly detect it. If the detection cannot be performed correctly, this includes erroneously detecting image information that is not image information that was over-detected in the first detection process as image information that was over-detected in the first detection process.

Further, for example, if over-detected image information is not detected for one frame of ultrasound image, the determination function 184 uses position information indicating the position of the undetected image information as a search result. The data is stored in the memory circuit 170.

Note that a case has been described in which the discrimination function 184 calculates the second probability as an example of an index indicating the degree of similarity between the features of the image information and the features of the trained model for overdetection discrimination. However, instead of the second probability, the discrimination function 184 may calculate other indicators such as similarity, which indicates the degree to which the features of the image information and the features of the trained model for overdetection discrimination are similar. good.

An example of the second detection process executed in step S102 has been described above. FIG. 11 is a diagram illustrating an example of the results of the second detection process according to the first embodiment. FIG. 11 shows the results of performing the second detection process on each of the six pieces of image information 40a to 40f shown in FIG. 10 above. The "result of over-detection" shown in FIG. 11 shows an example of image information detected by the second detection process. The "result determined not to be overdetection" shown in FIG. 11 indicates an example of image information that was not detected by the second detection process.

As shown in FIG. 11, in the second detection process, the determination function 184 detects two pieces of image information 40b and 40f as image information that was overdetected in the first detection process. Furthermore, the determination function 184 does not detect the four pieces of image information 40a, 40c, 40d, and 40e in the second detection process. Then, in the second detection process, the determination function 184 selects four image information 40a, 40c, 40d, and 40e other than the detected two image information 40b and 40f out of the six image information 40a to 40f. It is detected as image information that includes the characteristic parts of the image. As a result, in the second detection process, four pieces of image information 40a, 40c, 40d, and 40e are finally detected as characteristic parts.

As described above, the discrimination function 184 finally detects the feature parts detected in the first detection process, excluding over-detected feature parts, as feature parts in the second detection process. .

That is, the discrimination function 184 uses a second detection function that indicates the degree of similarity between the feature of the feature region determined to be included in the ultrasound image by the first detection process and the feature of the trained model for overdetection discrimination. If the probability is greater than or equal to the threshold, it is corrected that the characteristic region determined to be included in the ultrasound image in the first detection process is not included in the ultrasound image. Therefore, the discrimination function 184 can detect the characteristic region with high accuracy. The second detection process is an example of a correction process.

Here, if the operator wants to grasp whether or not there is a characteristic region rather than the details of the characteristic region, it is considered that the operator moves the ultrasound probe 101 relatively quickly over a relatively wide range. On the other hand, if the operator already knows the existence of a characteristic region and wants to grasp the details of the characteristic region, the operator can move the ultrasound probe 101 relatively slowly in a relatively narrow range around the characteristic region. Conceivable.

Therefore, the ultrasound diagnostic apparatus 1 according to the first embodiment executes appropriate processing depending on the moving speed of the ultrasound probe 101, as described below.

FIG. 12 is a flowchart showing an example of a process executed by the ultrasound diagnostic apparatus 1 according to the first embodiment. For example, the ultrasound diagnostic apparatus 1 executes the process shown in FIG. 12 at predetermined time intervals.

As shown in FIG. 12, the determination function 184 determines whether the moving speed of the ultrasound probe 101 is greater than a predetermined threshold Th1 (step S201). Here, if the moving speed of the ultrasound probe 101 is greater than the predetermined threshold Th1, for example, the operator may want to grasp whether or not there is a characteristic region, rather than the details of the characteristic region. In step S201, the speed detection function 182 detects the moving speed of the ultrasound probe 101. Then, in step S201, the determination function 184 determines whether the moving speed of the ultrasound probe 101 detected by the speed detection function 182 is greater than the threshold Th1. The threshold Th1 is an example of a first threshold.

If it is determined that the moving speed of the ultrasound probe 101 is greater than the predetermined threshold Th1 (step S201: Yes), the determination function 184 performs the process described below in step S202.

For example, in step S202, the determination function 184 selects the first frame rate (for example, 30 fps) and the second frame rate (for example, 15 fps) as the frame rate when deriving the determination result. Select frame rate 1. Note that the frame rate here is, for example, the number of ultrasound image data processed per unit time. Also, the second frame rate is lower than the first frame rate.

Further, for example, in step S202, the determination function 184 selects the first derivation process from the first derivation process and the second derivation process as the derivation process when deriving the determination result. Here, the first derivation process is the first determination process at the first stage during operation shown in FIG. 4 above. The second derivation process includes the first determination process, the second determination process at the second stage of operation shown in FIG. 5, and the third determination process at the third stage of operation shown in FIG. This is the determination process No. 3.

The first derivation process has a lower processing load than the second derivation process. Therefore, the determination function 184 can increase the frame rate when executing the first derivation process. Therefore, as described above, the determination function 184 selects a relatively high first frame rate from a plurality of types of frame rates (the first frame rate and the second frame rate) in step S202. .

Furthermore, if an affirmative determination is made in step S201 (step S201: Yes), as described above, the operator may want to grasp whether or not there is a characteristic region, rather than the details of the characteristic region. Therefore, in step S202, the determination function 184 does not perform the second derivation process in which multiple types of derivation processes are determined from among the multiple types of derivation processes (the first derivation process and the second derivation process). A first derivation process (first discrimination process) that performs one type of discrimination as to whether or not a characteristic region is included in an ultrasound image is selected.

Then, in step S203, the determination function 184 executes the selected first derivation process. In step S203, the control function 181 controls the transmitting circuit 110, the receiving circuit 120, the B-mode processing circuit 130, the Doppler processing circuit 140, and the image generation circuit so as to perform various processes at the selected first frame rate. 150, an acquisition function 183, a discrimination function 184, a marker information generation function 185, and a display control function 186 are controlled.

Further, in step S203, the marker information generation function 185 generates marker information indicating the detection result of the ultrasound image based on the first discrimination result 17 (see FIG. 4) derived by the first derivation process. . For example, in step S203, the marker information generation function 185 generates marker information according to the derived first determination result 17 every time the first determination result 17 is derived by the first derivation process.

For example, when the first discrimination result 17 indicates that a characteristic region is included in the ultrasound image, the marker information generation function 185 generates a rectangle representing the range detected as the characteristic region. Generate marker information indicating the marker.

FIG. 13 is a diagram for explaining an example of a process executed by the marker information generation function 185 according to the first embodiment. For example, when the first discrimination result 17 indicates that the ultrasound image includes a characteristic region, the marker information generation function 185 generates a rectangle representing the range detected as the characteristic region, as shown in FIG. The marker information indicating the marker 32 is generated. In addition, in FIG. 13, the code "31" refers to an actual characteristic part.

Each time the marker information generation function 185 generates marker information, it superimposes the marker 32 indicated by the marker information on the ultrasound image, as shown in FIG. 13. Then, the marker information generation function 185 causes the image memory 160 to store ultrasound image data indicating the ultrasound image on which the marker 32 is superimposed.

Further, in step S203, the display control function 186 causes the display 103 to display an ultrasound image indicated by the ultrasound image data stored in the image memory 160. Ultrasound images are an example of medical images. The display control function 186 is an example of a display control section.

For example, the display control function 186 causes the display 103 to display the ultrasound image on which the marker 32 is superimposed in real time. For example, each time the marker information generation function 185 stores ultrasound image data indicating an ultrasound image on which the marker 32 is superimposed in the image memory 160, the display control function 186 extracts the ultrasound image data from the image memory 160. get. Then, the display control function 186 causes the display 103 to display the ultrasound image on which the marker 32 is superimposed, every time the ultrasound image data indicating the ultrasound image on which the marker 32 is superimposed is obtained.

On the other hand, when the first determination result 17 indicates that the ultrasound image does not include a characteristic region, the marker information generation function 185 does not generate marker information.

However, the display control function 186 also causes the display 103 to display in real time an ultrasound image on which the marker 32 is not superimposed. For example, when the first determination result 17 indicates that the ultrasound image does not include a characteristic region, the display control function 186 causes the display 103 to display the ultrasound image illustrated in FIG. 14 . Note that FIG. 14 shows an ultrasound image that actually includes the characteristic region 31, but in which the characteristic region was not detected in the first derivation process. FIG. 14 is a diagram showing a display example according to the first embodiment.

When the process of step S203 is completed, the ultrasound diagnostic apparatus 1 ends the process shown in FIG. 12.

On the other hand, if it is determined that the moving speed of the ultrasound probe 101 is less than or equal to the predetermined threshold Th1 (step S201: No), the determination function 184 determines that the moving speed of the ultrasound probe 101 is lower than the predetermined threshold Th2. It is determined whether or not the value is also smaller (step S204). Here, the threshold Th2 is smaller than the threshold Th1. Further, if the moving speed of the ultrasound probe 101 is smaller than the threshold value Th2, for example, it is considered that the operator already knows the existence of the characteristic region and wants to understand the details of the characteristic region. The threshold Th2 is an example of a second threshold.

If it is determined that the moving speed of the ultrasound probe 101 is smaller than the threshold Th2 (step S204: Yes), the determination function 184 performs the process described below in step S205.

For example, in step S205, the determination function 184 selects the second frame rate from the first frame rate and the second frame rate as the frame rate for deriving the determination result.

Further, for example, in step S205, the determination function 184 selects the second derivation process from the first derivation process and the second derivation process as the derivation process when deriving the determination result.

Here, the second derivation process derives the first determination result 17, the second determination result 23 (see FIG. 5), and the third determination result 29 (see FIG. 6). Therefore, the second derivation process derives more determination results than the first derivation process. In this way, the second derivation process has a higher processing load than the first derivation process. Therefore, when executing the second derivation process, it is preferable to lower the frame rate. Therefore, as described above, the determination function 184 selects the relatively low second frame rate from the first frame rate and the second frame rate in step S205.

Furthermore, since the operator is likely to want to grasp the details of the characteristic parts, in step S205, the discrimination function 184 derives more discrimination results than the first derivation process from among the plurality of types of derivation processes. Select the second derivation process.

Then, in step S206, the determination function 184 executes the first detection process (the first stage process during operation) included in the selected second derivation process, and derives the first determination result 17. .

Furthermore, in step S206, the marker information generation function 185 performs the same process as the process in step S203. Furthermore, in step S206, the display control function 186 performs the same process as the process in step S203.

Note that in step S206 and steps S207 and S208 described below, the control function 181 controls the transmitting circuit 110, receiving circuit 120, and B-mode processing so as to perform various processes at the selected second frame rate. It controls the circuit 130, the Doppler processing circuit 140, the image generation circuit 150, the acquisition function 183, the discrimination function 184, the marker information generation function 185, and the display control function 186. That is, the control function 181 controls each process in steps S206 to S208 to be performed at the second frame rate.

Then, in step S207, the determination function 184 executes the second determination process (second-stage process during operation) included in the selected second derivation process, and derives the second determination result 23. .

In step S207, the marker information generation function 185 generates the character string "malignant" or the character string "benign" according to the derived second determination result 23. For example, the marker information generation function 185 generates the character string "malignant" when the second determination result 23 indicates that the characteristic site is malignant. Further, the marker information generation function 185 generates the character string "benign" when the second determination result 23 indicates that the characteristic site is benign. The marker information generation function 185 generates a character string "malignant" or a character string "benign" each time the second discrimination result 23 is derived.

Then, in step S207, the display control function 186 superimposes the generated character string "malignant" or the character string "benign" on the ultrasound image every time the character string "malignant" or the character string "benign" is generated. and display it on the display 103.

Note that each time the marker information generation function 185 generates the character string "malignant" or the character string "benign," it adds the generated character string "malignant" to the report information of the subject P stored in the storage circuit 170. Alternatively, the character string "benign" may be automatically registered. For example, the report information is information such as the history of the results of ultrasound image diagnosis of the subject P that is registered by an operator such as a doctor via the input interface 190. Then, the operator displays the report screen indicated by the report information on the display 103 as needed, and confirms the history of the ultrasound image diagnosis results indicated by the report screen. In this embodiment, the marker information generation function 185 automatically registers the determination result of whether a characteristic region is malignant or benign in the report information of the subject P, so that the operator can manually determine whether the characteristic region is malignant or benign. There is no need to register the results in the report information. Therefore, according to the present embodiment, it is possible to suppress the burden on the operator when registering the determination result in report information. Note that the marker information generation function 185 may automatically register the second determination result 23 in the report information instead of the character string "malignant" and the character string "benign."

Then, in step S208, the determination function 184 executes the third determination process (the third stage process during operation) included in the selected second derivation process, and derives the third determination result 29. .

In step S208, the marker information generation function 185 generates a character string "malignancy level A" or a character string "malignancy level B" according to the derived third determination result 29. For example, the marker information generation function 185 generates the character string "malignancy level A" when the third determination result 29 indicates that the malignancy level of the characteristic site is malignancy level A. Further, the marker information generation function 185 generates a character string "malignancy level B" when the third determination result 29 indicates that the malignancy level of the characteristic site is malignancy level B. The marker information generation function 185 generates a character string "malignancy level A" or a character string "malignancy level B" each time the third determination result 29 is derived.

Then, in step S208, the display control function 186 controls the generated character string "malignancy level A" or the character string "malignancy level B" each time the character string "malignancy level A" or the character string "malignancy level B" is generated. B'' is superimposed on the ultrasound image and displayed on the display 103.

Note that each time the marker information generation function 185 generates the character string "malignancy grade A" or the character string "malignancy grade B", the marker information generation function 185 adds the generated characters to the report information of the subject P stored in the storage circuit 170. The column "malignancy level A" or the character string "malignancy level B" may be automatically registered. In this embodiment, the marker information generation function 185 automatically registers the determination result of the malignancy of the characteristic part in the report information of the subject P, so that the operator can manually input the determination result into the report information. There is no need to register. Therefore, according to the present embodiment, it is possible to suppress the burden on the operator when registering the determination result in report information. Note that the marker information generation function 185 may automatically register the third determination result 29 in the report information instead of the character string "malignancy level A" and the character string "malignancy level B".

When the process of step S208 is completed, the ultrasonic diagnostic apparatus 1 ends the process shown in FIG. 12.

On the other hand, if it is determined that the moving speed of the ultrasound probe 101 is equal to or higher than the threshold Th2 (step S204: No), the ultrasound diagnostic apparatus 1 performs the process described below in step S209.

For example, in step S209, the ultrasound diagnostic apparatus 1 performs the same process as the process in step S203. However, in step S209, the control function 181 controls the transmission circuit 110, the reception circuit 120, the B-mode processing circuit 130, the Doppler processing circuit 140, the image generation circuit 150, and the acquisition function so as to perform various processes at a predetermined frame rate. 183, the discrimination function 184, the marker information generation function 185, and the display control function 186 are controlled. The predetermined frame rate may be a frame rate other than the first frame rate and the second frame rate, or may be the first frame rate or the second frame rate.

When the process of step S209 is completed, the ultrasonic diagnostic apparatus 1 ends the process shown in FIG. 12.

In the process shown in FIG. 12, the discrimination function 184 selects a frame rate for deriving a discrimination result from among a plurality of types of frame rates of a plurality of ultrasound image data according to the moving speed of the ultrasound probe 101. Execute the selected process. The process of selecting such a frame rate is an example of the first selection process.

In addition, in the process shown in FIG. 12, the discrimination function 184 performs a derivation process for deriving a discrimination result from among a plurality of types of derivation processes for deriving a discrimination result according to the moving speed of the ultrasound probe 101. Execute the process that selects the process. The process of selecting such a derivation process is an example of the second selection process. For example, in the process shown in FIG. 12, the determination function 184 selects at least one learned model and at least one learning condition according to the moving speed of the ultrasound probe 101, out of the plurality of learned models and the plurality of learning conditions. By using the learned model and performing a discrimination process corresponding to the learning conditions, a discrimination result regarding the characteristic part of the subject P is derived.

Therefore, according to the ultrasound diagnostic apparatus 1 according to the first embodiment, appropriate processing can be performed depending on the moving speed of the ultrasound probe 101.

The ultrasonic diagnostic apparatus 1 according to the first embodiment has been described above. According to the ultrasonic diagnostic apparatus 1 according to the first embodiment, as described above, it is possible to obtain discrimination results regarding characteristic parts with high accuracy. Further, according to the ultrasound diagnostic apparatus 1 according to the first embodiment, as described above, appropriate processing can be performed depending on the moving speed of the ultrasound probe 101.

(Second embodiment)
In the first embodiment described above, the learned model generation function 201a of the trained model generation device 200 generates various learned models and records various learning conditions. However, the ultrasound diagnostic apparatus 1 may generate various learned models and record various learning conditions. Therefore, such an embodiment will be described as a second embodiment. Note that in the description of the second embodiment, differences from the first embodiment will be mainly described, and descriptions of the same configuration and processing as the first embodiment may be omitted.

FIG. 15 is a diagram showing a configuration example of an ultrasound diagnostic apparatus 1 according to the second embodiment. The second embodiment differs from the first embodiment in that the processing circuit 180 includes a trained model generation function 187.

In the first embodiment described above, the learned model generation function 201a generates various learned models using ultrasound image data generated by the ultrasound diagnostic apparatus 1 or other ultrasound diagnostic equipment, Record various learning conditions.

On the other hand, the learned model generation function 187 uses the ultrasound image data (for example, B-mode image data) generated by the image generation circuit 150 to generate various learned models, similar to the learned model generation function 201a. and record various learning conditions. The trained model generation function 187 is an example of a generation unit.

Specifically, the trained model generation function 187 generates a first trained model 11a, a second trained model 12a, and a third trained model 13a. Further, the trained model generation function 187 records the first learning condition 11b, the second learning condition 12b, and the third learning condition 13b in the storage circuit 170. That is, the trained model generation function 187 learns by associating the ultrasound image data with the discrimination results (labels) related to the characteristic parts depicted in the ultrasound image data, thereby applying a model to the input ultrasound image data. A trained model is generated to be used when performing a discrimination process that outputs discrimination results regarding the characteristic parts of the object to be visualized, and learning conditions are recorded. The trained model generation function 187 generates a trained model and records learning conditions at each of a plurality of stages during learning (the first to third stages during learning described above).

Here, similarly to the trained model generation function 201a, the trained model generation function 187 asynchronously generates the first trained model 11a, the second trained model 12a, and the third trained model 13a. generate. Further, the learned model generation function 187 asynchronously records the first learning condition 11b, the second learning condition 12b, and the third learning condition 13b.

The trained model generation function 187 then stores the first trained model 11a, the second trained model 12a, and the third trained model 13a in the storage circuit 170. Further, the trained model generation function 187 causes the storage circuit 170 to store the first learning condition 11b, the second learning condition 12b, and the third learning condition 13b.

In the second embodiment, the determination function 184 executes the same process as in the first embodiment using various learned models and various learning conditions stored in the storage circuit 170.

The second embodiment has been described above. According to the second embodiment, similarly to the first embodiment, it is possible to obtain the determination results regarding the characteristic parts with high accuracy. Furthermore, according to the second embodiment, similar to the first embodiment, appropriate processing can be performed depending on the moving speed of the ultrasound probe 101.

Note that in the second embodiment, the trained model generation function 187 generates the first trained model 11a, the second trained model 12a, and the third trained model synchronously rather than asynchronously. 13a may be generated. Similarly, the trained model generation function 187 may record the first learning condition 11b, the second learning condition 12b, and the third learning condition 13b synchronously instead of asynchronously.

For example, first, the trained model generation function 187 generates the first trained model 11a and records the first learning condition 11b. After generating the first trained model 11a and recording the first learning condition 11b, the determination function 184 uses the first trained model 11a and the first learning condition 11b to generate the first trained model 11a and the first learning condition 11b. Execute determination processing. Then, the trained model generation function 187 generates the second trained model 12a by using the ultrasound image data determined to include the characteristic part in the first determination process as part of the learning data. , the second learning condition 12b is recorded. This makes it possible to automatically prepare part of the learning data used when generating the second learned model 12a without the operator having to prepare it.

An example of a method for generating the second trained model 12a in such a case will be described with reference to FIG. 16. FIG. 16 is a flowchart illustrating an example of a process for generating the second trained model 12a according to the second embodiment.

As illustrated in FIG. 16, the trained model generation function 187 determines whether or not the ultrasound image data in which the characteristic region is depicted and the second label 21 have been input (step S301). Here, an example of ultrasound image data to be determined in step S301 as to whether or not it has been input will be described. For example, the ultrasound image data to be determined is the ultrasound image data that was determined to include a characteristic region in the first discrimination process, and it has been confirmed by the operator's visual inspection that the characteristic region is included. This is ultrasound image data. The operator inputs such ultrasound image data together with the second label 21 into the apparatus main body 100 via the input device 102.

If it is determined that the ultrasound image data and the second label 21 have not been input (step S301: No), the learned model generation function 187 performs the determination in step S301 again. On the other hand, if it is determined that the ultrasound image data and the second label 21 have been input (step S301: Yes), the trained model generation function 187 input the ultrasound image data and the second label 21. The second learned model 12a is generated using the second learned model 12a, and the second learning condition 12b is recorded (step S302). The trained model generation function 187 then ends the process shown in FIG. 16.

Then, after the second trained model 12a is generated and the second learning condition 12b is recorded, the determination function 184 uses the second trained model 12a and the second learning condition 12b to A second determination process is executed. The trained model generation function 187 then generates a third trained model by using the ultrasound image data depicting the characteristic region determined to be malignant in the second discrimination process as part of the learning data. 13a is generated. This makes it possible to automatically prepare part of the learning data used when generating the third learned model 13a without the operator having to prepare it.

An example of a method for generating the third learned model 13a in such a case will be described with reference to FIG. 17. FIG. 17 is a flowchart illustrating an example of a process for generating the third learned model 13a according to the second embodiment.

As illustrated in FIG. 17, the trained model generation function 187 receives ultrasound image data depicting a characteristic region determined to be malignant in the second discrimination process and a third label 27. It is determined whether or not (step S401). Here, an example of ultrasound image data to be determined in step S401 as to whether or not it has been input will be described. For example, the ultrasound image data to be determined is ultrasound image data that depicts a characteristic region that was determined to be malignant in the second discrimination process, and the operator visually observes that the characteristic region is determined to be malignant. This is ultrasound image data that has been confirmed to be true. The operator inputs such ultrasound image data together with the third label 27 into the apparatus main body 100 via the input device 102.

If it is determined that the ultrasound image data and the third label 27 have not been input (step S401: No), the learned model generation function 187 performs the determination in step S401 again. On the other hand, if it is determined that the ultrasound image data and the third label 27 have been input (step S401: Yes), the trained model generation function 187 input the ultrasound image data and the third label 27. The third trained model 13a is generated using the third learned model 13a, and the third learning condition 13b is recorded (step S402). The trained model generation function 187 then ends the process shown in FIG. 17.

Other methods of generating trained models have been described above. In another method for generating a trained model, the trained model generation function 187 combines ultrasonic image data corresponding to a predetermined discrimination result output from a discrimination process corresponding to one stage among a plurality of stages; A trained model corresponding to the next step after the first step is generated by learning in association with the discrimination result (label) corresponding to the next step after the first step. Therefore, according to another method of generating a trained model, training data can be easily prepared.

(Modification 1 of the first embodiment and the second embodiment)
In the first and second embodiments described above, the determination function 184 performs a derivation process when deriving a determination result from among a plurality of types of derivation processes according to the moving speed of the ultrasound probe 101. We have explained the case of selecting . However, the determination function 184 may select the derivation process when deriving the determination result depending on whether the determination result is derived in real time or as a post-process. Therefore, such an embodiment will be described as a first modification of the first embodiment and the second embodiment. In addition, in the explanation of Modification 1, the points that are different from the first embodiment and the second embodiment will be mainly explained, and the same configuration and processing as the first embodiment and the second embodiment will not be explained. May be omitted.

For example, in Modification 1, an instruction (real-time processing instruction) for deriving a discrimination result in real time for ultrasound scanning is input from the operator to the ultrasound diagnostic apparatus 1 via the input device 102. Ru. Further, in Modification 1, an instruction (post-processing instruction) for deriving a discrimination result is input by the operator via the input device 102 as post-processing (post-processing) to be executed after ultrasonic scanning. Ru.

When the real-time processing instruction is input, the ultrasound diagnostic apparatus 1 according to the first modification executes the same processing as the processing in step S202 and step S203 shown in FIG. 12 above.

Further, when a post-processing instruction is input, the ultrasound diagnostic apparatus 1 performs post-processing on the plurality of ultrasound image data stored in the storage circuit 170, not in real time, but in the process shown in FIG. Processing similar to that shown in steps S205 to S208 is executed.

Therefore, according to the ultrasonic diagnostic apparatus 1 according to the first modification of the first embodiment and the second embodiment, depending on whether the discrimination result is derived in real time or as post-processing. appropriate processing can be carried out. Moreover, according to the ultrasonic diagnostic apparatus 1 according to the first modification, similarly to the first embodiment and the second embodiment, it is possible to obtain the determination result regarding the characteristic region with high accuracy.

(Modification 2 of the first embodiment and the second embodiment)
In the first and second embodiments described above, the determination function 184 determines whether to perform the first stage of operation or the first stage of operation depending on the moving speed of the ultrasound probe 101. The case where the user selects whether to perform the processing in steps 3 to 3 has been explained.

However, the determination function 184 performs the first detection process shown in FIG. 7 or the first detection process and the second detection process shown in FIG. 7 depending on the moving speed of the ultrasound probe 101. You may choose either. Therefore, such an embodiment will be described as a second modification of the first embodiment and the second embodiment. In addition, in the description of the second modification, the points that are different from the first embodiment and the second embodiment will be mainly explained, and the same configuration and processing as the first embodiment and the second embodiment will not be explained. May be omitted.

FIG. 18 is a flowchart showing an example of a process executed by the ultrasound diagnostic apparatus 1 according to the second modification of the first embodiment and the second embodiment. For example, the ultrasound diagnostic apparatus 1 executes the process shown in FIG. 18 at predetermined time intervals.

As shown in FIG. 18, the determination function 184 determines whether the moving speed of the ultrasound probe 101 is greater than a predetermined threshold Th1 (step S501). In step S501, the speed detection function 182 detects the moving speed of the ultrasound probe 101. Then, in step S501, the determination function 184 determines whether the moving speed of the ultrasound probe 101 detected by the speed detection function 182 is greater than the threshold Th1.

If it is determined that the moving speed of the ultrasound probe 101 is greater than the predetermined threshold Th1 (step S501: Yes), the determination function 184 performs the process described below in step S502.

For example, in step S502, the determination function 184 selects the first frame rate from the first frame rate and the second frame rate as the frame rate for deriving the determination result.

Further, for example, in step S502, the determination function 184 selects the first derivation process from the first derivation process and the third derivation process as the derivation process when deriving the determination result. Here, the first derivation process according to the second modification is the first detection process shown in FIG. 7 above. Further, the third derivation process according to the second modification is the first detection process and the second detection process shown in FIG. 7 above.

As described above, in the second detection process, from among the characteristic parts detected in the first detection process, characteristic parts excluding overdetections in the first detection process are finally detected as characteristic parts. . Therefore, in Modification 2, the third derivation process derives the first discrimination result, which is the result of determining whether or not the ultrasound image includes a characteristic region, with higher accuracy than the first derivation process. It is possible to do so. On the other hand, the first derivation process has a lower processing load than the third derivation process.

If an affirmative determination is made in step S501 (step S501: Yes), it is considered that a high frame rate is prioritized over a high accuracy of the first determination result. Therefore, in step S502, the determination function 184 selects a first derivation process from among the plurality of types of derivation processes that determines whether or not the ultrasound image includes a characteristic region with a relatively small processing load. select.

Further, the determination function 184 can increase the frame rate when executing the first derivation process. Therefore, as described above, the determination function 184 selects a relatively high first frame rate from among the plurality of types of frame rates in step S502.

Then, in step S503, the determination function 184 executes the selected first derivation process (first detection process). Further, in step S503, the control function 181 controls the transmitting circuit 110, the receiving circuit 120, the B-mode processing circuit 130, the Doppler processing circuit 140, and the image generation circuit to perform various processing at the selected first frame rate. 150, an acquisition function 183, a discrimination function 184, a marker information generation function 185, and a display control function 186 are controlled.

Further, in step S503, the marker information generation function 185 generates marker information indicating the detection result of the ultrasound image based on the characteristic region detected by the first detection process. For example, in step S503, the marker information generation function 185 generates a rectangular marker representing the range detected as a characteristic region, as in the first embodiment, every time a characteristic region is detected by the first detection process. Generate marker information to indicate.

Then, similarly to the first embodiment, the marker information generation function 185 superimposes the marker indicated by the marker information on the ultrasound image every time marker information is generated. Then, the marker information generation function 185 causes the image memory 160 to store ultrasound image data indicating the ultrasound image on which the marker is superimposed, as in the first embodiment.

Further, in step S503, the display control function 186 causes the display 103 to display an ultrasound image indicated by the display ultrasound image data stored in the image memory 160, as in the first embodiment. That is, in step S503, the ultrasound image with the marker superimposed and the ultrasound image without the marker superimposed are displayed on the display 103 in real time.

When the process of step S503 is completed, the ultrasound diagnostic apparatus 1 ends the process shown in FIG. 18.

On the other hand, if it is determined that the moving speed of the ultrasound probe 101 is less than or equal to the predetermined threshold Th1 (step S501: No), the determination function 184 determines that the moving speed of the ultrasound probe 101 is lower than the predetermined threshold Th2. It is determined whether or not the value is also smaller (step S504).

If it is determined that the moving speed of the ultrasound probe 101 is smaller than the threshold Th2 (step S504: Yes), in step S505, the determination function 184 performs the process described below.

If an affirmative determination is made in step S504 (step S504: Yes), it is considered that the accuracy of the first determination result is prioritized over a high frame rate. Here, the third derivation process is able to more accurately derive the first discrimination result, which is the result of determining whether or not the ultrasound image includes a characteristic region, than the first derivation process. be. Therefore, in step S505, the determination function 184 selects the third derivation process from the first derivation process and the third derivation process as the derivation process when deriving the determination result.

Furthermore, the third derivation process has a higher processing load than the first derivation process. Therefore, when executing the third derivation process, it is preferable to lower the frame rate. Therefore, in step S505, the determination function 184 selects the second frame rate from the first frame rate and the second frame rate as the frame rate for deriving the determination result.

Then, in step S506, the determination function 184 executes the first detection process included in the selected third derivation process.

Note that in step S506 and S507 described below, the control function 181 controls the transmitting circuit 110, receiving circuit 120, B-mode processing circuit 130, It controls the Doppler processing circuit 140, the image generation circuit 150, the acquisition function 183, the discrimination function 184, the marker information generation function 185, and the display control function 186. That is, the control function 181 controls each process in steps S506 and S507 to be performed at the second frame rate.

Then, in step S507, the determination function 184 executes the second detection process included in the selected third derivation process, and finally detects the characteristic part.

Further, in step S507, the marker information generation function 185 generates marker information indicating the detection result of the ultrasound image based on the characteristic region finally detected by the second detection process. For example, in step S507, the marker information generation function 185 generates a rectangle representing the range detected as the characteristic region, as in the first embodiment, each time a characteristic region is finally detected by the second detection process. Generate marker information indicating the marker.

Then, similarly to the first embodiment, the marker information generation function 185 superimposes the marker indicated by the marker information on the ultrasound image every time marker information is generated. Then, the marker information generation function 185 causes the image memory 160 to store ultrasound image data indicating the ultrasound image on which the marker is superimposed, as in the first embodiment.

Further, in step S507, the display control function 186 causes the display 103 to display an ultrasound image indicated by the display ultrasound image data stored in the image memory 160, as in the first embodiment. That is, in step S507, the ultrasound image with the marker superimposed and the ultrasound image without the marker superimposed are displayed on the display 103 in real time.

When the process of step S507 is completed, the ultrasonic diagnostic apparatus 1 ends the process shown in FIG. 18.

On the other hand, if it is determined that the moving speed of the ultrasound probe 101 is equal to or higher than the threshold Th2 (step S504: No), the ultrasound diagnostic apparatus 1 performs the process described below in step S508.

For example, in step S508, the ultrasound diagnostic apparatus 1 performs the same process as the process in step S503. However, the control function 181 includes a transmission circuit 110, a reception circuit 120, a B-mode processing circuit 130, a Doppler processing circuit 140, an image generation circuit 150, an acquisition function 183, and a discrimination function so as to perform various processes at a predetermined frame rate. 184, controls the marker information generation function 185 and the display control function 186. The predetermined frame rate may be a frame rate other than the first frame rate and the second frame rate, or may be the first frame rate or the second frame rate.

When the process of step S508 is completed, the ultrasonic diagnostic apparatus 1 ends the process shown in FIG. 18.

In the process shown in FIG. 18, the discrimination function 184 selects a frame rate for deriving a discrimination result from among a plurality of types of frame rates of a plurality of ultrasound image data according to the moving speed of the ultrasound probe 101. Execute the selected process.

In addition, in the process shown in FIG. 18, the discrimination function 184 performs a derivation process for deriving a discrimination result from among a plurality of types of derivation processes for deriving a discrimination result according to the moving speed of the ultrasound probe 101. Execute the process that selects the process.

Therefore, according to the ultrasound diagnostic apparatus 1 according to the second modification of the first embodiment and the second embodiment, similarly to the first embodiment and the second embodiment, the moving speed of the ultrasound probe 101 is Appropriate processing can be performed depending on the situation. Further, according to the ultrasonic diagnostic apparatus 1 according to the second modification, similarly to the first embodiment and the second embodiment, it is possible to accurately obtain a discrimination result regarding a characteristic region.

(Variation 3 of the first embodiment and the second embodiment)
In the first embodiment and second embodiment described above, the case where the determination function 184 executes the first detection process and the second detection process has been described. However, the determination function 184 may perform another process instead of the first detection process and the second detection process. Therefore, such an embodiment will be described as a third modification of the first embodiment and the second embodiment.

In addition, in the explanation of Modification 3, the points that are different from the first embodiment and the second embodiment will be mainly explained, and the same configuration and processing as the first embodiment and the second embodiment will not be explained. May be omitted.

In Modification 3, an example of the first discrimination process executed by the discrimination function 184 using the first trained model 11a in the first stage of operation will be described. FIG. 19 is a flowchart showing an example of the flow of the first determination process according to the third modification of the first embodiment and the second embodiment.

As shown in FIG. 19, the discrimination function 184 executes the third detection process in the first discrimination process (step S601), and also executes the fourth detection process in parallel with the third detection process. (Step S602), and a fifth detection process is executed (Step S603).

First, the third detection process executed in step S601 will be described. The determination function 184 uses the search window, the second trained model for search, and the search algorithm to search for a characteristic part in the search range set in the ultrasound image by the operator. For example, the determination function 184 moves the search window to multiple positions within the search range. In addition, the discrimination function 184 analyzes the image information in the search window at each position using the second trained model for search and the search algorithm, so that the image information in each search window becomes a characteristic part. The corresponding probability (third probability) is calculated.

The search window is, for example, a unit of area within the search range that is compared with the features of the second trained model for search. The second trained model for search is, for example, a model that has learned image information that is a sample of a characteristic part. The search algorithm is an algorithm for calculating the third probability described above and searching for image information corresponding to a characteristic region in the ultrasound image to be processed.

The second trained model for search is an example of the first pattern indicating a characteristic part.

Then, when the calculated third probability is greater than or equal to a predetermined threshold, the determination function 184 detects the range surrounded by the search window as a characteristic part (range of characteristic parts). That is, the discrimination function 184 determines that when the third probability indicating the degree of similarity between the feature of a portion of the ultrasound image and the feature of the second search trained model indicating the characteristic region is greater than or equal to the threshold value. determines that a characteristic region is included in the ultrasound image. Furthermore, when the third probability is less than the threshold value, the determination function 184 determines that the ultrasound image does not include the characteristic region.

Then, the determination function 184 causes the storage circuit 170 to store the search result (detection result) in the third detection process for the ultrasound image for each frame of ultrasound image data. For example, if the discrimination function 184 is able to detect a characteristic region from one frame of ultrasound image, the determination function 184 stores position information indicating the position of the characteristic region in the storage circuit 170 as a search result in the third detection process. let Note that the position of the characteristic region referred to here is the position in the image space of ultrasound image data.

Further, when a characteristic region is not detected from one frame of the ultrasound image, the discrimination function 184 stores information indicating that a characteristic region has not been detected in the storage circuit 170 as a search result in the third detection process. Make me remember.

Note that the determination function 184 will explain a case where the third probability is calculated as an example of an index indicating the degree of similarity between the features of the image information in the search window and the features of the second trained model for search. did. However, instead of using the third probability, the discrimination function 184 uses other indicators such as similarity, which indicates the degree to which the features of the image information in the search window and the features of the second trained model for search are similar. may be calculated.

An example of the third detection process executed in step S601 has been described above. FIG. 20 is a diagram illustrating an example of the results of the third detection process according to the third modification of the first embodiment and the second embodiment. “Result of determination as a characteristic region” shown in FIG. 20 shows an example of image information in the search window 30 when a characteristic region is detected by the third detection process. The “result of determining that it is not a characteristic region” shown in FIG. 20 shows an example of image information in the search window 30 when a characteristic region is not detected by the third detection process. In addition, FIG. 20 shows images of eight locations as an example of the results in which the search window 30 is placed at a plurality of positions within the search range in the third detection process and the third probability is calculated at each position. Information is shown.

As shown in FIG. 20, the discrimination function 184 detects characteristic parts at four positions in the third detection process. Specifically, as shown in FIG. 20, the determination function 184 detects four pieces of image information 50a to 50d within the search window 30.

Image information 50a is over-detected image information. The image information 50b is image information in which the actual characteristic region 31e is depicted. The image information 50c is image information in which the actual characteristic region 31f is depicted. The image information 50d is over-detected image information. That is, in the case shown in FIG. 20, two of the four pieces of image information determined to be characteristic parts are overdetected.

Further, as shown in FIG. 20, the determination function 184 does not detect characteristic parts at four positions in the third detection process. Specifically, as shown in FIG. 20, the determination function 184 does not detect the four pieces of image information 50e to 50h within the search window 30 as characteristic parts. Note that the determination function 184 does not detect the image information 50e even though the image information 50e includes the characteristic region 31g.

Next, the fourth detection process executed in step S602 will be explained. The determination function 184 uses the search window, the third trained model for search, and the search algorithm to search for a characteristic part in the search range set in the ultrasound image by the operator. For example, the determination function 184 moves the search window to multiple positions within the search range. In addition, the discrimination function 184 analyzes the image information in the search window at each position using the third trained model for search and the search algorithm, so that the image information in each search window becomes a characteristic part. A corresponding probability (fourth probability) is calculated.

The search window is, for example, a unit of area within the search range that is compared with the features of the third trained model for search. The third learned model for search is, for example, a model that has learned image information that is a sample of a characteristic part. Here, the third trained model for search is a different model from the second trained model for search. The search algorithm is an algorithm for calculating the above-mentioned fourth probability and searching for image information corresponding to a characteristic region in an ultrasound image to be processed.

For example, the second trained model for search described above is trained using a first group of image information in which characteristic parts are depicted and a second group of image information in which characteristic parts are not depicted. This is the model obtained by On the other hand, the third trained model for search is obtained through learning performed using the third image information group in which characteristic parts are depicted and the second image information group in which characteristic parts are not depicted. This is a model that can be used. That is, the second image information group is commonly used in the second search trained model and the third search trained model during learning.

Therefore, the second search trained model and the third search trained model perform learning using different image information, even if the same feature region (for example, a tumor) is depicted. Therefore, the characteristic parts that can be detected are different. On the other hand, the second trained model for search and the third trained model for search use the same image information group (second image information group) as the image information group in which the characteristic part is not depicted. learning is being carried out. Therefore, the second trained model for search and the third trained model for search have the same images that are correctly determined not to be characteristic parts, and also have the same images that are overdetected.

The third learned model for search is an example of a second pattern that is different from the first pattern indicating a characteristic part.

Then, when the calculated fourth probability is greater than or equal to a predetermined threshold, the determination function 184 detects the range surrounded by the search window as a characteristic part (range of characteristic parts). That is, the discrimination function 184 determines that when the fourth probability indicating the degree of similarity between the feature of a portion of the ultrasound image and the feature of the third search trained model indicating the characteristic region is greater than or equal to the threshold value. determines that a characteristic region is included in the ultrasound image. Furthermore, when the fourth probability is less than the threshold value, the determination function 184 determines that the ultrasound image does not include the characteristic region.

Then, the determination function 184 causes the storage circuit 170 to store the search result (detection result) in the fourth detection process for the ultrasound image for each frame of ultrasound image data. For example, when a characteristic region is detected from one frame of an ultrasound image, the determination function 184 causes the storage circuit 170 to store position information indicating the position of the characteristic region as a search result in the fourth detection process. Note that the position of the characteristic region referred to here is the position in the image space of ultrasound image data.

Further, when a characteristic region is not detected from one frame of the ultrasound image, the determination function 184 stores information indicating that a characteristic region has not been detected in the storage circuit 170 as a search result in the fourth detection process. Make me remember.

Note that the determination function 184 will explain a case where the fourth probability is calculated as an example of an index indicating the degree of similarity between the features of the image information in the search window and the features of the third trained model for search. did. However, instead of using the fourth probability, the discrimination function 184 uses other indicators such as similarity, which indicates the degree to which the features of the image information in the search window and the features of the third trained model for search are similar. may be calculated.

An example of the fourth detection process executed in step S602 has been described above. FIG. 21 is a diagram illustrating an example of the results of the fourth detection process according to the third modification of the first embodiment and the second embodiment. “Result of determination as a characteristic region” shown in FIG. 21 shows an example of image information in the search window 30 when a characteristic region is detected by the fourth detection process. “Result of determination that it is not a characteristic region” shown in FIG. 21 shows an example of image information in the search window 30 when a characteristic region is not detected by the fourth detection process. Note that FIG. 21 shows images of eight locations as an example of the results of the fourth detection process in which the search window 30 is placed at multiple positions within the search range and the fourth probability is calculated at each position. Information is shown.

As shown in FIG. 21, the discrimination function 184 detects characteristic parts at four positions in the fourth detection process. Specifically, as shown in FIG. 21, the determination function 184 detects four pieces of image information 50i to 50l within the search window 30.

The image information 50i is image information in which the actual characteristic region 31h is depicted. Image information 50j is over-detected image information. The image information 50k is image information in which the actual characteristic region 31i is depicted. The image information 50l is over-detected image information. That is, in the case shown in FIG. 21, two of the four pieces of image information determined to be characteristic parts are overdetected.

Further, as shown in FIG. 21, the determination function 184 does not detect characteristic parts at four positions in the fourth detection process. Specifically, as shown in FIG. 21, the determination function 184 does not detect the four pieces of image information 50m to 50p within the search window 30 as characteristic parts. Note that the determination function 184 does not detect the image information 50o even though the image information 50o includes the characteristic region 31j. Similarly, the determination function 184 does not detect the image information 50p even though the characteristic region 31k is included in the image information 50p.

Next, the fifth detection process executed in step S603 will be explained. In the fifth detection process, the discrimination function 184 uses the detection results of the third detection process and the fourth detection process to finally detect the characteristic region.

Here, as described above, the image information in which the actual feature part is depicted out of the image information detected using the second search trained model in the third detection process, and the image information in which the actual feature part is depicted, and the fourth detection This differs from the image information in which the actual characteristic parts are depicted among the image information detected using the third trained search model in the process.

For example, the image information 50b and 50c shown in FIG. 20 are different from the image information 50i and 50k shown in FIG.

However, as described above, among the image information detected using the second search trained model in the third detection process, over-detected image information and the third detection process in the fourth detection process The over-detected image information among the image information detected using the trained search model matches (common). That is, the over-detected portions on the ultrasound image are the same in the third detection process and the fourth detection process.

For example, the image information 50a shown in FIG. 20 and the image information 50j shown in FIG. 21 match. Furthermore, the image information 50d shown in FIG. 20 and the image information 50l shown in FIG. 21 match.

Therefore, in the fifth detection process, the discrimination function 184 selects the characteristic part detected in the third detection process and the characteristic part detected in the fourth detection process. A characteristic region other than the characteristic region commonly detected in the two detection processes of the detection process is detected.

For example, in the fifth detection process, the determination function 184 identifies the image information 50a to 50d detected in the third detection process and the image information 50i to 50l detected in the fourth detection process. In the fifth detection process, the discrimination function 184 identifies image information 50a, 50d, 50j, and 50l that is commonly detected in the two detection processes among the identified image information 50a to 50d, and 50i to 50l. do.

FIG. 22 is a diagram illustrating an example of the results of the fifth detection process according to the third modification of the first embodiment and the second embodiment. In the fifth detection process, the determination function 184 selects the image information 50b, 50c, 50i shown in FIG. , 50k are finally detected as characteristic parts.

In this way, the discrimination function 184 uses the image information detected by the third detection process using the second trained model for search, and the image information detected by the fourth detection process using the third trained model for search. Among the image information detected, image information excluding image information commonly detected in the third detection process and the fourth detection process is finally detected as a characteristic part, and the characteristic part is included in the ultrasound image. It is determined that the The third detection process is an example of the first characteristic part detection process. Further, the fourth detection process is an example of the second characteristic part detection process.

In this way, the discrimination function 184 according to the third modification removes the over-detected feature parts so that the over-detected feature parts are not included in the finally detected feature parts. Detect characteristic parts. Therefore, according to the ultrasonic diagnostic apparatus 1 according to the third modification, it is possible to obtain the determination result regarding the characteristic region with higher accuracy.

Then, in the fifth detection process, the determination function 184 causes the storage circuit 170 to store position information indicating the position of the finally detected characteristic part as a search result.

Further, for the characteristic parts not detected in the fifth detection process, the determination function 184 causes the storage circuit 170 to store information indicating that the characteristic parts were not detected as a search result.

Then, the marker information generation function 185 generates marker information indicating the detection result of the ultrasound image based on the characteristic region finally detected by the fifth detection process. For example, every time a characteristic part is finally detected by the fifth detection process, the marker information generation function 185 indicates a rectangular marker representing the range detected as a characteristic part, similarly to the first embodiment. Generate marker information.

Then, similarly to the first embodiment, the marker information generation function 185 superimposes the marker indicated by the marker information on the ultrasound image every time marker information is generated. Then, the marker information generation function 185 causes the image memory 160 to store ultrasound image data indicating the ultrasound image on which the marker is superimposed, as in the first embodiment.

Then, the display control function 186 causes the display 103 to display an ultrasound image indicated by the display ultrasound image data stored in the image memory 160, as in the first embodiment. That is, the display control function 186 causes the display 103 to display the ultrasound image with the marker superimposed and the ultrasound image without the marker superimposed on the display 103 in real time.

The ultrasound diagnostic apparatus 1 according to the third modification of the first embodiment and the second embodiment has been described above. According to the ultrasonic diagnostic apparatus 1 according to the third modification, as described above, it is possible to obtain determination results regarding characteristic parts with higher accuracy.

(Modification 4 of the first embodiment and the second embodiment)
In the first and second embodiments described above, the determination function 184 performs a derivation process when deriving a determination result from among a plurality of types of derivation processes according to the moving speed of the ultrasound probe 101. We have explained the case of selecting . However, the discrimination function 184 may select a learned model to be used when deriving the discrimination result from among a plurality of types of learned models, depending on the moving speed of the ultrasound probe 101. Therefore, such an embodiment will be described as a fourth modification of the first embodiment and the second embodiment.

In addition, in the explanation of the fourth modification of the first embodiment and the second embodiment, the points different from the first embodiment and the second embodiment will be mainly explained, and the differences between the first embodiment and the second embodiment will be explained. Descriptions of configurations and processes similar to those in the embodiment may be omitted.

FIG. 23 is a diagram for explaining examples of various learned models and various learning conditions according to Modification 4 of the first embodiment and the second embodiment. As illustrated in FIG. 23, the storage circuit 170 stores two first trained models 11a_1 and 11a_2 used for the first stage processing during operation. Furthermore, the storage circuit 170 stores two second trained models 12a_1 and 12a_2 used for second-stage processing during operation. Furthermore, the storage circuit 170 stores two third learned models 13a_1 and 13a_2 used for the third stage of processing during operation. In this way, the storage circuit 170 stores a plurality of types of trained models corresponding to each of the plurality of stages.

Furthermore, the storage circuit 170 stores two first learning conditions 11b_1 and 11b_2 used for the first stage of processing during operation. Furthermore, the storage circuit 170 stores two second learning conditions 12b_1 and 12b_2 used for second-stage processing during operation. Furthermore, the storage circuit 170 stores two third learning conditions 13b_1 and 13b_2 used in the third stage of processing during operation. In this way, the storage circuit 170 stores a plurality of types of learning conditions corresponding to each of a plurality of stages.

The first learning condition 11b_1 is used in the first discrimination process using the first learned model 11a_1. The first learning condition 11b_2 is used in the first discrimination process using the first learned model 11a_2.

The second learning condition 12b_1 is used in the second discrimination process using the second trained model 12a_1. The second learning condition 12b_2 is used in the second discrimination process using the second trained model 12a_2.

The third learning condition 13b_1 is used in the third discrimination process using the third learned model 13a_1. The third learning condition 13b_2 is used in the third discrimination process using the third trained model 13a_2.

In this way, one learned model corresponds to one learning condition. Here, the processing load of the discrimination process using the trained model is a load according to the learning conditions corresponding to this trained model.

The two first learned models 11a_1 and 11a_2 are adjusted parameters used in the first discrimination process. However, the discrimination accuracy in the first discrimination process using the first trained model 11a_1 is higher than the discrimination accuracy in the first discrimination process using the first learned model 11a_2. On the other hand, the processing load of the first discrimination process using the first trained model 11a_2 is lower than the processing load of the first discrimination process using the first learned model 11a_1.

Furthermore, the two second trained models 12a_1 and 12a_2 are adjusted parameters used in the second discrimination process. However, the discrimination accuracy in the second discrimination process using the second trained model 12a_1 is higher than the discrimination accuracy in the second discrimination process using the second learned model 12a_2. On the other hand, the processing load of the second discrimination process using the second trained model 12a_2 is lower than the processing load of the second discrimination process using the second learned model 12a_1.

Furthermore, the two third learned models 13a_1 and 13a_2 are adjusted parameters used in the third discrimination process. However, the discrimination accuracy in the third discrimination process using the third trained model 13a_1 is higher than the discrimination accuracy in the third discrimination process using the third learned model 13a_2. On the other hand, the processing load of the third discrimination process using the third learned model 13a_2 is lower than the processing load of the third discrimination process using the third learned model 13a_1.

The discrimination function 184 according to the fourth modification executes the first discrimination process when the moving speed of the ultrasound probe 101 is greater than the threshold Th1 when performing the first discrimination process in the first stage during operation. The first trained model 11a_2 is selected as the first trained model to be used. Then, the determination function 184 executes the first determination process using the selected first learned model 11a_2 and the corresponding first learning condition 11b_2.

Further, the discrimination function 184 is used when performing the first discrimination process in the first stage during operation, when the moving speed of the ultrasound probe 101 is smaller than the threshold value Th2. The first trained model 11a_1 is selected as the first trained model. In this way, when the moving speed of the ultrasound probe 101 is smaller than the threshold Th2, the discrimination function 184 selects the first model that can derive the first discrimination result with higher accuracy than the first trained model 11a_2. Select trained model 11a_1. Then, the determination function 184 executes the first determination process using the selected first learned model 11a_1 and the corresponding first learning condition 11b_1.

Further, the discrimination function 184 is used when performing the second discrimination process at the second stage during operation, when the moving speed of the ultrasound probe 101 is greater than the threshold Th1. The second trained model 12a_2 is selected as the second trained model. The determination function 184 then executes the second determination process using the selected second learned model 12a_2 and the corresponding second learning condition 12b_2.

In addition, the discrimination function 184 is used when performing the second discrimination process at the second stage during operation, when the moving speed of the ultrasound probe 101 is smaller than the threshold Th2. The second trained model 12a_1 is selected as the second trained model. In this way, when the moving speed of the ultrasound probe 101 is smaller than the threshold Th2, the discrimination function 184 uses a second model that can derive the second discrimination result more accurately than the second trained model 12a_2. Select trained model 12a_1. Then, the determination function 184 executes a second determination process using the selected second learned model 12a_1 and the corresponding second learning condition 12b_1.

Further, the discrimination function 184 is used when performing the third discrimination process at the third stage during operation, when the moving speed of the ultrasound probe 101 is greater than the threshold Th1. The third trained model 13a_2 is selected as the third trained model. Then, the determination function 184 executes a third determination process using the selected third learned model 13a_2 and the corresponding third learning condition 13b_2.

Further, the discrimination function 184 is used when performing the third discrimination process at the third stage during operation, when the moving speed of the ultrasound probe 101 is smaller than the threshold Th2. The third trained model 13a_1 is selected as the third trained model. In this way, when the moving speed of the ultrasound probe 101 is smaller than the threshold Th2, the discrimination function 184 uses the third model that can derive the third discrimination result with higher accuracy than the third learned model 13a_2. Select trained model 13a_1. Then, the determination function 184 executes a third determination process using the selected third learned model 13a_1 and the corresponding third learning condition 13b_1.

As described above, the discrimination function 184 selects a trained model and learning conditions to be used when deriving a discrimination result from among a plurality of types of trained models, depending on the moving speed of the ultrasound probe 101. Execute processing. Therefore, the ultrasound diagnostic apparatus 1 according to the fourth modification can perform appropriate processing depending on the moving speed of the ultrasound probe 101. Note that the process of selecting the trained model and learning conditions described above is an example of the third selection process.

(Modification 5 of the first embodiment and the second embodiment)
In the above-mentioned modification 4, the discrimination function 184 selects a learned model and learning conditions to be used when deriving a discrimination result from among a plurality of types of learned models, according to the moving speed of the ultrasound probe 101. The selection case has been explained. However, the discrimination function 184 is used to derive discrimination results from among multiple types of trained models, depending on whether the discrimination results are derived in real time or as post-processing. You may also select a trained model. Therefore, such an embodiment will be described as a fifth modification of the first embodiment and the second embodiment. Note that in the description of Modification Example 5, differences from Modification Example 1 and Modification Example 4 described above will be mainly explained, and descriptions of the same configurations and processes as Modification Example 1 and Modification Example 4 may be omitted.

In modification 5, when a real-time processing instruction is input, the discrimination function 184 performs the first discrimination process used when performing the first discrimination process in the first stage during operation. The first trained model 11a_2 is selected as the first trained model. Then, the determination function 184 executes the first determination process using the selected first learned model 11a_2 and the corresponding first learning condition 11b_2.

In addition, when a real-time processing instruction is input, the discrimination function 184 performs a second learning process to be used when performing the second discrimination process at the second stage during operation. The second trained model 12a_2 is selected as the trained model. The determination function 184 then executes the second determination process using the selected second learned model 12a_2 and the corresponding second learning condition 12b_2.

In addition, when a real-time processing instruction is input, the discrimination function 184 performs a third learning process to be used when performing the third discrimination process at the third stage during operation. The third trained model 13a_2 is selected as the trained model. Then, the determination function 184 executes a third determination process using the selected third learned model 13a_2 and the corresponding third learning condition 13b_2.

In addition, when a post-processing instruction is input, the discrimination function 184 performs a first learning process to be used when performing the first discrimination process in the first stage during operation. The first trained model 11a_1 is selected as the trained model. Then, the determination function 184 executes the first determination process using the selected first learned model 11a_1 and the corresponding first learning condition 11b_1.

In addition, when a post-processing instruction is input, the discrimination function 184 performs a second learning process to be used when performing the second discrimination process at the second stage during operation. The second trained model 12a_1 is selected as the trained model. Then, the determination function 184 executes a second determination process using the selected second learned model 12a_1 and the corresponding second learning condition 12b_1.

In addition, when a post-processing instruction is input, the determination function 184 performs a third learning process to be used when performing the third determination process at the third stage during operation. The third trained model 13a_1 is selected as the trained model. Then, the determination function 184 executes a third determination process using the selected third learned model 13a_1 and the corresponding third learning condition 13b_1.

Therefore, according to the ultrasonic diagnostic apparatus 1 according to the fifth modification of the first embodiment and the second embodiment, it depends on whether the discrimination result is derived in real time or as post-processing. appropriate processing can be carried out. Further, according to the ultrasonic diagnostic apparatus 1 according to the fifth modification, similarly to the first embodiment and the second embodiment, it is possible to obtain the discrimination result regarding the characteristic region with high accuracy.

(Variation 6 of the first embodiment and the second embodiment)
Next, a modification example for increasing the discrimination accuracy of the first discrimination process using the first trained model 11a and the second discrimination process using the second learned model 12a will be described in accordance with the first embodiment. This will be described as a sixth modification of the second embodiment. In addition, in the explanation of the modification 6, the points that are different from the first embodiment and the second embodiment will be mainly explained, and the same configuration and processing as the first embodiment and the second embodiment will not be explained. May be omitted.

For example, the second discrimination process based on the second learned model 12a and the second learning condition 12b according to Modification 6 is different from the second discrimination process according to the first embodiment and the second embodiment. Differently, it is determined whether a characteristic region included in an ultrasound image is malignant, benign, or neither malignant nor benign.

In modification 6, in the second stage of learning, the learned model generation function 201a or the learned model generation function 187 learns the relationship between the ultrasound image data 20 and the label, thereby generating the second learned model. A model 12a is generated and a second learning condition 12b is recorded. In modification 6, the label used when generating the second trained model 12a is based on whether the characteristic region included in the ultrasound image shown by the ultrasound image data is malignant or benign, or Indicates whether it is malignant or benign.

Here, it is assumed that the ultrasound image represented by the ultrasound image data input to the second discrimination process includes a characteristic region. However, there may be cases where over-detection occurs in the first discrimination process, and the over-detected ultrasound image data is input to the second discrimination process.

Therefore, in the sixth modification, the second discrimination process discriminates whether the characteristic region included in the ultrasound image is malignant, benign, or neither malignant nor benign.

If the second discrimination process determines that the characteristic site included in the ultrasound image is neither malignant nor benign, it is possible that the characteristic site is not included in the ultrasound image. That is, if it is determined that the object is neither malignant nor benign, it is considered that over-detected ultrasound image data has been input to the second determination process. That is, ultrasound image data determined to be neither malignant nor benign is considered to be overdetected ultrasound image data.

Therefore, the learned model generation function 201a or the learned model generation function 187 updates the first learned model 11a and the first learning condition 11b using the ultrasound image data determined to be neither malignant nor benign. do. In the following description, a case will be described in which the trained model generation function 187 updates the first trained model 11a and the first learning condition 11b. However, the trained model generation function 201a updates the first trained model 11a and the first learning condition 11b by executing the same process as the trained model generation function 187 described below. It's okay.

The trained model generation function 187 learns by associating ultrasound image data determined to be neither malignant nor benign with the first label 15 indicating that the ultrasound image does not include a characteristic region. , updates the first learned model 11a and the first learning condition 11b. Thereby, the discrimination accuracy of the first discrimination process based on the first learned model 11a and the first learning condition 11b can be increased.

Further, for example, the third discrimination process based on the third learned model 13a and the third learning condition 13b according to the sixth modification is the third discrimination process according to the first embodiment and the second embodiment. In contrast, it is determined whether the malignancy of the malignant characteristic site included in the ultrasound image is malignancy level A, malignancy level B, or neither malignancy level A nor malignancy level B.

In modification example 6, in the third stage of learning, the learned model generation function 201a or the learned model generation function 187 learns the relationship between the ultrasound image data 25 and the label, thereby generating the third learned model. A model 13a is generated and a third learning condition 13b is recorded. In modification 6, the label used when generating the third learned model 13a is whether the malignancy level of the malignant feature site included in the ultrasound image indicated by the ultrasound image data is malignancy level A. , indicates whether the grade is B, or whether it is neither A nor B.

Here, it is assumed that the characteristic site included in the ultrasound image indicated by the ultrasound image data input to the third discrimination process is malignant. However, in the second discrimination process, a benign characteristic region is mistakenly judged as malignant, and the ultrasound image data incorrectly judged as malignant is input to the third discrimination process. It is also possible.

Therefore, in Modified Example 6, the third discrimination process determines whether the malignancy of the malignant characteristic part included in the ultrasound image is malignant grade A, malignant grade B, or even if malignant grade A is malignant. Determine whether it is B or not.

If the third discrimination process determines that the malignancy of the characteristic part included in the ultrasound image is neither grade A nor grade B, the characteristic part included in the ultrasound image is considered to be benign. It will be done. That is, if it is determined that the malignancy level is neither A nor B, it is considered that the ultrasound image data in which a benign characteristic region was erroneously determined to be malignant was input to the third determination process. That is, the ultrasound image data determined to be neither malignant grade A nor malignant grade B is considered to be ultrasound image data in which a benign characteristic region was erroneously determined to be malignant.

Therefore, the learned model generation function 201a or the learned model generation function 187 uses the ultrasound image data determined to be neither malignant grade A nor malignant grade B to generate the second learned model 12a and the second learned model 12a. Update condition 12b. In the following description, a case will be described in which the trained model generation function 187 updates the second trained model 12a and the second learning condition 12b. However, the trained model generation function 201a updates the second trained model 12a and the second learning condition 12b by executing the same process as the trained model generation function 187 described below. It's okay.

The trained model generation function 187 learns by associating ultrasound image data that is determined to be neither malignant grade A nor malignant grade B with a label indicating that a characteristic part included in the ultrasound image is benign. As a result, the second learned model 12a and the second learning condition 12b are updated. Thereby, the discrimination accuracy of the second discrimination process based on the second learned model 12a and the second learning condition 12b can be increased.

Modification 6 has been explained. As described above, the trained model generation function 187 according to the sixth modification uses ultrasound image data corresponding to a predetermined discrimination result outputted from the discrimination process corresponding to the next stage of one stage, and this predetermined By learning in association with the discrimination results, a new trained model corresponding to the first stage is generated, and learning conditions are newly recorded. Thereby, the discrimination accuracy of the discrimination process corresponding to the first stage can be increased.

(Modification 7 of the first embodiment and the second embodiment)
Next, another modification example for improving the discrimination accuracy of the first discrimination process using the first trained model 11a and the second discrimination process using the second learned model 12a will be described. This will be described as a seventh modification of the embodiment and the second embodiment. In addition, in the explanation of Modification 7, the points that are different from the first embodiment and the second embodiment will be mainly explained, and the same configuration and processing as the first embodiment and the second embodiment will not be explained. May be omitted.

In the sixth modification described above, a case has been described in which the second trained model 12a and the third trained model 13a are generated by a learning method different from the first embodiment and the second embodiment. On the other hand, in Modification 7, a second trained model 12a and a third trained model 13a generated by the same learning method as in the first embodiment and the second embodiment are used.

However, in Modification 7, a fourth trained model, a fourth learning condition, a fifth trained model, and a fifth learning condition are used. In Modification 7, in the second stage of operation, the fourth discrimination process using the fourth learned model and the fourth learning condition is executed before the second discrimination process. Furthermore, in the third stage of operation, a fifth discrimination process using a fifth trained model and a fifth learning condition is executed before the third discrimination process.

In the second stage of operation, the discrimination function 184 executes a fourth discrimination process as a CAD process using the fourth learned model and the fourth learning condition. The fourth determination process is a process of determining whether a characteristic region is included or not included in the ultrasound image shown in the ultrasound image data 22, and deriving a determination result (fourth determination result). In this way, in the seventh modification, the determination function 184 executes the fourth determination process, which is similar to the first determination process executed in the first stage of operation, in the second stage of operation. Further, the fourth trained model is generated in the same manner as the first trained model 11a. Further, the fourth learning condition is recorded in the same manner as the first learning condition 11b.

However, when generating the fourth trained model, more ultrasound image data overdetected in the first discrimination process is used than when generating the first trained model 11a.

Therefore, in the fourth discrimination process using the fourth trained model and the fourth learning condition, over-detection of ultrasound image data that was over-detected in the first discrimination process is suppressed.

Here, a case will be described in which the determination function 184 that has executed the fourth determination process determines that the ultrasound image includes a characteristic region. In this case, in the second stage of operation, the discrimination function 184 uses the second learned model 12a and the second learning condition 12b to generate an ultrasound image that is determined to contain a characteristic region. A second discrimination process is performed on the data.

Further, a case will be described in which the determination function 184 that has executed the fourth determination process determines that the ultrasound image does not include a characteristic region. In this case, the learned model generation function 201a or the learned model generation function 187 generates the first learned model 11a and the first The learning condition 11b of is updated. In the following description, a case will be described in which the trained model generation function 187 updates the first trained model 11a and the first learning condition 11b. However, the trained model generation function 201a updates the first trained model 11a and the first learning condition 11b by executing the same process as the trained model generation function 187 described below. It's okay.

The trained model generation function 187 associates the ultrasound image data determined that the ultrasound image does not contain a characteristic region with the first label 15 indicating that the ultrasound image does not contain the characteristic region. By learning, the first learned model 11a and the first learning condition 11b are updated. Thereby, the discrimination accuracy of the first discrimination process based on the first learned model 11a and the first learning condition 11b can be increased.

Then, in the third stage of operation, the discrimination function 184 executes the fifth discrimination process using the fifth learned model and the fifth learning condition. In the fifth discrimination process, it is determined again whether the characteristic region that was determined to be benign by the second discrimination process in the second stage of operation is benign or malignant, and the discrimination result (the fifth This is the process of deriving the determination result). In this way, in the seventh modification, the determination function 184 executes the fifth determination process, which is similar to the second determination process executed in the second stage of operation, in the third stage of operation. Further, the fifth trained model is generated in the same manner as the second trained model 12a. Further, the fifth learning condition is recorded in the same manner as the second learning condition 12b.

However, in the case of generating the fifth trained model, a benign feature part was incorrectly determined to be malignant in the second discrimination process, rather than in the case of generating the second trained model 12a. Ultrasound image data is often used.

Therefore, in the fifth discrimination process using the fifth trained model and the fifth learning condition, a benign feature region is incorrectly determined to be malignant in the ultrasound image data in the second discrimination process. The included characteristic parts can be correctly determined to be benign.

Here, a case will be described in which the characteristic region included in the ultrasound image is determined to be malignant by the determination function 184 that has executed the fifth determination process. In this case, in the third stage of operation, the discrimination function 184 uses the third trained model 13a and the third learning condition 13b to apply the ultrasound image data whose characteristic region is determined to be malignant. Then, the third determination process is executed.

On the other hand, a case will be described in which the characteristic region included in the ultrasound image is determined to be benign by the determination function 184 that has executed the fifth determination process. In this case, the learned model generation function 201a or the learned model generation function 187 generates the second learned model 12a and the second learning condition 12b using the ultrasound image data in which the feature region is determined to be benign. Update. In the following description, a case will be described in which the trained model generation function 187 updates the second trained model 12a and the second learning condition 12b. However, the trained model generation function 201a updates the second trained model 12a and the second learning condition 12b by executing the same process as the trained model generation function 187 described below. It's okay.

The trained model generation function 187 learns by associating and learning the ultrasound image data in which the characteristic region is determined to be benign and the second label 21 indicating that the characteristic region is benign. The learned model 12a and the second learning condition 12b are updated. Thereby, the discrimination accuracy of the second discrimination process based on the second learned model 12a and the second learning condition 12b can be increased.

Modification 7 has been explained. As described above, the trained model generation function 187 according to Modification Example 7 uses ultrasonic image data corresponding to a predetermined discrimination result outputted from the discrimination process corresponding to the next stage after the first stage, and this predetermined By learning in association with the discrimination results, a new trained model corresponding to the first stage is generated, and learning conditions are newly recorded. Thereby, the discrimination accuracy of the discrimination process corresponding to the first stage can be increased.

(Third embodiment)
In addition, in the embodiment and modification example mentioned above, the case where the ultrasonic diagnostic apparatus 1 derives a discrimination result using a learned model and learning conditions was explained. However, the medical image processing apparatus may perform similar processing as post-processing rather than in real time. Therefore, such an embodiment will be described as a third embodiment.

FIG. 24 is a diagram showing a configuration example of a medical image processing apparatus 300 according to the third embodiment. As illustrated in FIG. 24, the medical image processing apparatus 300 includes an input device 301, a display 302, a storage circuit 310, and a processing circuit 320. Input device 301, display 302, storage circuit 310, and processing circuit 320 are connected to be able to communicate with each other. Medical image processing device 300 is an example of an analysis device.

The input device 301 is implemented by a mouse, keyboard, button, panel switch, touch command screen, foot switch, trackball, joystick, or the like. The input device 301 receives various setting requests from the operator of the medical image processing apparatus 300. The input device 301 outputs the received various setting requests to the processing circuit 320. For example, the input device 301 receives an instruction (execution instruction) for executing CAD processing from an operator of the medical image processing apparatus 300 and outputs the received execution instruction to the processing circuit 320 . Further, the operator can also set an ROI, which is a search range for a characteristic region, in the ultrasound image in CAD processing via the input device 301.

The display 302 displays, for example, a medical image or a GUI for an operator to input various setting requests using the input device 301.

The storage circuit 310 stores various programs for displaying a GUI and information used by the programs. Furthermore, the storage circuit 310 stores a plurality of pieces of ultrasound image data generated by the ultrasound diagnostic apparatus 1 in time series. Further, the storage circuit 310 stores various learned models and various learning conditions of any of the first embodiment, second embodiment, or modified examples 1 to 7 described above. These various learned models and various learning conditions are used for processing executed by the processing circuit 320.

The processing circuit 320 controls the entire processing of the medical image processing apparatus 300. The processing circuit 320 is realized by, for example, a processor. For example, as shown in FIG. 24, the processing circuit 320 includes a control function 321, a speed detection function 322, an acquisition function 323, a trained model generation function 324, a discrimination function 325, a marker information generation function 326, and a display control function 327. It has a processing function. Here, for example, the components of the processing circuit 320 shown in FIG. 24 are a control function 321, a speed detection function 322, an acquisition function 323, a trained model generation function 324, a discrimination function 325, a marker information generation function 326, and a display control function. Each processing function of 327 is recorded in the storage circuit 310 in the form of a program executable by a computer. The processing circuit 320 is a processor that reads each program from the storage circuit 310 and executes it to realize a function corresponding to each program. In other words, the processing circuit 320 in a state where each program is read has each function shown in the processing circuit 320 of FIG. 24.

The medical image processing device 300 acquires, for example, a plurality of time-series ultrasound image data generated by the ultrasound diagnostic device 1. However, a detection signal indicating the moving speed of the ultrasound probe 101 detected by the speed detector 104 is added to this ultrasound image data. Then, the medical image processing apparatus 300 performs the same processing on the ultrasound image data as in the first embodiment, the second embodiment, or any of the modified examples 1 to 7 described above, and displays the data on the display 302. The ultrasound image and information such as detected characteristic parts are displayed.

The control function 321 controls the entire processing of the medical image processing apparatus 300. The speed detection function 322 has the same function as the speed detection function 182 described above. For example, the speed detection function 322 detects the moving speed of the ultrasound probe 101 based on the detection signal added to the ultrasound image data. The acquisition function 323 has the same function as the acquisition function 183 described above. The learned model generation function 324 has the same function as the learned model generation function 187 described above. The discrimination function 325 has the same function as the discrimination function 184 described above. The marker information generation function 326 performs the same function as the marker information generation function 185 described above. The display control function 327 performs the same function as the display control function 186 described above.

The medical image processing apparatus 300 according to the third embodiment has been described above. According to the medical image processing apparatus 300 according to the third embodiment, similarly to the first embodiment and the like, it is possible to accurately obtain a discrimination result regarding a characteristic region.

In addition, in at least one embodiment or modification described above, the ultrasound diagnostic apparatus 1 may handle each function provided in the processing circuit 180, various learned models, and various learning conditions as one software package. Similarly, the medical image processing apparatus 300 may handle each function provided in the processing circuit 320, various learned models, and various learning conditions as one software package.

According to at least one embodiment or modification described above, it is possible to obtain a determination result regarding a characteristic part with high accuracy.

Although several embodiments of the invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included within the scope and gist of the invention as well as within the scope of the invention described in the claims and its equivalents.

1 Ultrasonic diagnostic equipment 184 Discrimination function 187 Learned model generation function

Claims (17)

When each of the plurality of trained models corresponding to each of the plurality of stages executes each of the plurality of discrimination processes that outputs discrimination results regarding the characteristic parts of the subject depicted in the input medical image data. a storage unit that stores each of the plurality of trained models used for the
A discriminator that derives a discrimination result regarding a characteristic part of the subject by executing a discrimination process corresponding to the at least one learned model using at least one learned model among the plurality of learned models. and,
Equipped with
The discrimination unit executes a discrimination process in each of the plurality of stages that is different from discrimination processes executed in other stages.
Ultrasound diagnostic equipment. The discrimination unit is configured to process medical image data corresponding to a predetermined discrimination result outputted from a discrimination process corresponding to one stage among the plurality of stages to a discrimination process corresponding to a next stage after the first stage. input, and derive a discrimination result output from a discrimination process corresponding to the next stage of the first stage as a discrimination result corresponding to the next stage of the first stage;
The ultrasonic diagnostic apparatus according to claim 1. a collection unit that collects a plurality of the medical image data;
a detection unit that detects a moving speed of an ultrasound probe used when collecting the plurality of medical image data;
further comprising;
The discrimination unit performs a first selection process of selecting a frame rate for deriving a discrimination result from among a plurality of types of frame rates of the plurality of medical image data, according to a moving speed of the ultrasound probe. , and executing a second selection process for selecting a derivation process for deriving the discrimination result from among a plurality of types of derivation processes for deriving the discrimination result, according to the movement speed,
The ultrasonic diagnostic apparatus according to claim 1 or 2. a collection unit that collects a plurality of the medical image data;
a detection unit that detects a moving speed of an ultrasound probe used when collecting the plurality of medical image data;
further comprising;
The storage unit stores a plurality of types of trained models corresponding to each of the plurality of stages,
The discrimination unit performs a first selection process of selecting a frame rate for deriving a discrimination result from among a plurality of types of frame rates of the plurality of medical image data, according to a moving speed of the ultrasound probe. , and executing a third selection process of selecting a trained model to be used in deriving a discrimination result from among the plurality of types of trained models according to the movement speed;
The ultrasonic diagnostic apparatus according to claim 1 or 2. In the first selection process, if the moving speed is larger than a first threshold, the determining unit selects a first frame rate as a frame rate for deriving the determination result, and selects the first frame rate as the frame rate when deriving the determination result, and is smaller than a second threshold that is smaller than the first threshold, selecting a second frame rate lower than the first frame rate as the frame rate for deriving the discrimination result;
The ultrasonic diagnostic apparatus according to claim 3 or 4. In the second selection process, if the moving speed is larger than the first threshold, the discrimination unit selects the first derivation process as the derivation process when deriving the discrimination result, and selects the first derivation process as the derivation process when deriving the discrimination result. is smaller than the second threshold value, which is smaller than the first threshold value, a second derivation process that derives more discrimination results than the first derivation process, or a second derivation process when deriving the discrimination results. selecting as the derivation process a third derivation process that is capable of deriving a discrimination result with higher accuracy than the first derivation process and has a higher processing load;
The ultrasonic diagnostic apparatus according to claim 3. In the third selection process, the discrimination unit selects a first trained model as a trained model to be used when deriving a discrimination result if the moving speed is greater than a first threshold; If the moving speed is smaller than a second threshold value, which is smaller than the first threshold value, the learned model used to derive the discrimination result is used to derive the discrimination result with higher accuracy than the first trained model. selecting a second trained model that is possible and has a high processing load;
The ultrasonic diagnostic apparatus according to claim 4. When the moving speed is greater than the first threshold, the determining unit performs the first derivation process by determining a part of the medical image indicated by the input medical image data and a first When the index indicating the degree of similarity between the two patterns is equal to or higher than a threshold value, it is determined that the medical image includes the characteristic region, and when the index is less than the threshold value, it is determined that the characteristic region is included in the medical image. Execute feature part detection processing to derive a discrimination result that is a result of determining that there is no
The ultrasonic diagnostic apparatus according to claim 6. When the moving speed is smaller than the second threshold, the determining unit determines, as the third derivation process, that the characteristic part detection process and the characteristic part detection process are included in the medical image. degree of similarity between the detected characteristic region and a second pattern indicating a characteristic region estimated to be erroneously determined to be included in the medical image even though it is not included in the medical image in the characteristic region detection process; When the index indicating is equal to or greater than a threshold, a correction process is executed to correct the determination result that the characteristic region determined to be included in the medical image is not included in the medical image.
The ultrasonic diagnostic apparatus according to claim 8. The discriminating unit is configured to detect a medical image indicated by the input medical image data, which is detected by a first characteristic part detection process using a first pattern indicating a characteristic part, for the input medical image data. and a portion of the medical image detected by a second characteristic region detection process using a second pattern different from the first pattern indicating the characteristic region with respect to the input medical image data. , a portion of the medical image excluding a portion of the medical image commonly detected in the first characteristic region detection process and the second characteristic region detection process is a characteristic region; is determined to be included,
The ultrasonic diagnostic apparatus according to claim 1. further comprising a collection unit that collects a plurality of the medical image data by performing ultrasound scanning on the subject;
The discrimination unit derives a discrimination result depending on whether the discrimination result is derived in real time with respect to the ultrasonic scanning or whether the discrimination result is derived as a post-processing performed after the ultrasonic scanning. Select the derivation process for deriving the discrimination result from among multiple types of derivation processes for
The ultrasonic diagnostic apparatus according to claim 1. further comprising a collection unit that collects a plurality of the medical image data by performing ultrasound scanning on the subject;
The storage unit stores a plurality of types of trained models corresponding to each of the plurality of stages,
The discrimination unit determines the plurality of types depending on whether the discrimination result is derived in real time with respect to the ultrasonic scan or whether the discrimination result is derived as post-processing performed after the ultrasonic scan. Select a trained model to use when deriving the discrimination result from among the trained models of
The ultrasonic diagnostic apparatus according to claim 1. a collection unit that collects medical image data depicting characteristic parts of the subject;
By associating and learning the medical image data and the discrimination results regarding the characteristic parts depicted in the medical image data, the discrimination results regarding the characteristic parts of the subject depicted in the input medical image data are output. a generation unit that generates a trained model used in executing the discrimination process in each of the plurality of stages;
Equipped with
The determination processing performed in each of the plurality of stages is different from the determination processing performed in other stages,
Ultrasound diagnostic equipment. The generation unit generates medical image data corresponding to a predetermined discrimination result output from a discrimination process corresponding to one stage among the plurality of stages, and a discrimination result corresponding to a stage next to the first stage. Generate a trained model corresponding to the next stage of the first stage by learning in association with
The ultrasonic diagnostic apparatus according to claim 13. The generation unit learns by associating and learning medical image data corresponding to a predetermined discrimination result output from the discrimination process corresponding to the next stage of the first stage, and the predetermined discrimination result. Generate a trained model corresponding to the first stage,
The ultrasonic diagnostic apparatus according to claim 14. When each of the plurality of trained models corresponding to each of the plurality of stages executes each of the plurality of discrimination processes that outputs discrimination results regarding the characteristic parts of the subject depicted in the input medical image data. a storage unit that stores each of the plurality of trained models used for the
A discriminator that derives a discrimination result regarding a characteristic part of the subject by executing a discrimination process corresponding to the at least one learned model using at least one learned model among the plurality of learned models. and,
Equipped with
The discrimination unit executes a discrimination process in each of the plurality of stages that is different from discrimination processes executed in other stages.
Analysis device. an acquisition unit that acquires medical image data depicting characteristic parts of the subject;
By associating and learning the medical image data and the discrimination results regarding the characteristic parts depicted in the medical image data, the discrimination results regarding the characteristic parts of the subject depicted in the input medical image data are output. a generation unit that generates a trained model used in executing the discrimination process in each of the plurality of stages;
Equipped with
The determination processing performed in each of the plurality of stages is different from the determination processing performed in other stages,
Analysis device.

Images