CN108510493A - Boundary alignment method, storage medium and the terminal of target object in medical image - Google Patents

Abstract

The invention discloses boundary alignment method, storage medium and the terminal of target object in medical image, the method is applied to technical field of medical image processing, specifically includes：Terminal obtains the pending target object video data that clinical acquisitions arrive, and video data is input to the pre- segmentation network model for first passing through off-line training foundation；The segmentation network model is partitioned into the region where the target object in video data automatically, exports segmentation result.The present invention is by the way that pending medical image video input to be split to study is first passed through in advance with training foundation segmentation network model, the boundary of target object can be accurately positioned out in real time, eliminate the influence that bloom speck and artefact in imaging position target area upper and lower interface, the accuracy that target area boundaries judge is improved, the high-precision requirement clinically required is reached.

Description

Boundary positioning method, storage medium and terminal of target object in medical image

technical field

The invention relates to the technical field of medical image processing, in particular to a boundary positioning method, a storage medium and a terminal of a target object in a medical image.

Background technique

In the prior art, the research on the boundary location technology of a specific target object in medical images has always been an important subject. For example, in the medical field, corneal thickness is a condition that must be considered before surgery, and it will also affect the choice of surgical method, the setting of the size of the cutting area, and the evaluation of postoperative recovery. Therefore, accurate positioning of the corneal boundary is the first prerequisite for corneal pachymetry.

At present, the positioning of the corneal boundary mostly adopts the method directly related to the gray level of the image. Specifically, first determine the approximate location of the corneal region, then segment the cornea through Otsu threshold processing, then determine the corneal boundary, and then perform the subsequent thickness measurement. However, because it is difficult to eliminate the existence of instrument errors, the grayscale-based segmentation method is extremely susceptible to random high-brightness plaques and strip imaging artifacts near the cornea, which seriously affect the accurate measurement of corneal thickness. Therefore, in the prior art, the boundary positioning accuracy of the target object in the medical image is difficult to achieve the clinically required high accuracy.

Therefore, the prior art still needs to be improved and developed.

Contents of the invention

The technical problem to be solved by the present invention is to provide a boundary location method, storage medium and terminal of a target object in a medical image, aiming at solving the problem of the target object in a medical image in the prior art. The boundary location method of the current method is susceptible to interference from random artifacts in the imaging process, resulting in large errors, especially in the vertical direction, which is not accurate enough to achieve the high precision required by clinical practice.

The technical solution adopted by the present invention to solve technical problems is as follows:

A method for locating the boundary of a target object in a medical image, wherein the method is applied in the technical field of medical image processing, and specifically includes:

Step A, the terminal acquires the video data of the target object collected clinically to be processed, and inputs the video data into the segmentation network model established in advance through offline training;

Step B. The segmentation network model automatically segments the region where the target object is located in the video data, and outputs the segmentation result.

The boundary location method of the target object in the medical image, wherein, before the step A, it also includes:

In step S, a segmentation network model for automatically segmenting the region where the target object is located is established in advance through offline training.

The boundary location method of the target object in the medical image, wherein, after the step B, also includes:

Step C, performing polynomial fitting on the segmented target object area to obtain the upper and lower boundaries of the target object.

The boundary location method of the target object in the medical image, wherein the step S includes:

Step S1. Obtain the collected video of the target object, select the first frame of image and mark the area where the target object is located in the first frame of image;

Step S2, obtaining the regional features of the target object, and performing deep learning and training;

Step S3. In the training process, a sinusoidal-like displacement is introduced by piecewise affine transformation, the change of the boundary of the target object is simulated, and a segmentation network model for automatically segmenting the region where the target object is located is established.

The boundary location method of the target object in the medical image, wherein, the step S also includes:

Add conventional affine transformation, projective transformation or transmission transformation in the training process, and use image erosion technology or spike removal technology to smooth the sharp edges in the change of the target object area in the front and rear frame images.

The boundary location method of the target object in the medical image, wherein, the step B includes:

Step B1, the segmentation network model sequentially performs target object region segmentation on each frame of image in the target object video data;

Step B2, when performing target object area segmentation on the next frame image, automatically input the segmentation result of the previous frame image into the segmentation network model;

Step B3, the segmentation network model uses the segmentation result of the previous frame image as a reference, and performs target object area segmentation on the next frame image.

The boundary location method of target object in described medical image, wherein, described step B also comprises:

The segmentation result is input into the segmentation network model to update the segmentation network model.

The boundary location method of the target object in the medical image, wherein, after the step C, also includes:

Step D. Calculate the thickness of the target object according to the upper and lower boundaries of the target object.

A storage medium, on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor, so as to realize the method for locating the boundary of a target object in a medical image according to any one of the above.

A terminal, including: a processor, and a storage medium communicated with the processor, the storage medium is adapted to store a plurality of instructions; the processor is adapted to call the instructions in the storage medium to execute the above-mentioned The boundary location method of the target object in the medical image described in any one.

Beneficial effects of the present invention: the present invention can accurately locate the boundary of the target object in real time by inputting the medical image video to be processed to the segmentation network model established through learning and training in advance, and eliminate the bright spots and artifacts in the imaging The impact on the positioning of the upper and lower interfaces of the target area improves the accuracy of the boundary judgment of the target area and meets the high-precision requirements required in clinical practice.

Description of drawings

Fig. 1 is a schematic flowchart of a preferred embodiment of the method for locating the boundary of a target object in a medical image according to the present invention.

Fig. 2 is the first frame image A and the marked image B obtained when the cornea segmentation network model is established according to the present invention.

Fig. 3 is a schematic diagram of the change of the corneal boundary simulated by the piecewise affine transformation when the odd-even Amor segmentation network model is established in the present invention.

Fig. 4 is an effect diagram of the cornea segmentation and polynomial fitting by using the cornea segmentation network model of the present invention.

Fig. 5 is a schematic functional diagram of the terminal of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention more clear and definite, the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Because the boundary positioning accuracy of the target object in the medical image in the prior art is difficult to meet the clinically required high precision requirement. Especially for the positioning analysis of the corneal region, it has been a difficult problem to overcome. Therefore, the present invention provides a method for locating a boundary of a target object in a medical image, especially a method for locating a boundary of a corneal region. As shown in FIG. 1 , FIG. 1 is a schematic flowchart of a preferred embodiment of the method for locating the boundary of a target object in a medical image according to the present invention. The methods include:

Step S100 , the terminal obtains the clinically collected video data of the target object to be processed, and inputs the video data into the cornea segmentation network model previously established through offline training.

Specifically, in order to better illustrate the technical solution of the present invention, the present invention takes the analysis of the corneal region as an example. Therefore, in this embodiment, the target object is the cornea, and the segmentation network model is the cornea segmentation network model. Since corneal thickness is a condition that must be considered in many vision correction methods, it will also affect the choice of surgical method, the setting of the size of the cutting area, and the evaluation of postoperative recovery. The accurate positioning of the corneal boundary is the first prerequisite for corneal pachymetry.

Most of the existing methods for locating the corneal boundary are directly related to the gray level of the image. Specifically, first determine the approximate location of the corneal region, then segment the cornea through Otsu threshold processing, then determine the corneal boundary, and then perform the subsequent thickness measurement. However, because it is difficult to eliminate the existence of instrument errors, the grayscale-based segmentation method is extremely susceptible to the influence of random high-brightness plaques and strip imaging artifacts near the cornea. If you continue to use the Otsu method, the threshold value will be The large error is actually not only Otsu's threshold segmentation method, but also the segmentation results of other methods such as Li, Mean, Yen, etc. are also unsatisfactory, which seriously affects the accurate measurement of corneal thickness, which is far below the high clinical requirements. precision.

For the segmentation of the corneal region, since the human body is an elastic tissue, the segmentation method needs to have continuous operability for the elastic deformation and brightness changes of the target area. In the field of medical image processing, deep neural networks are usually used to judge the presence or absence of specific diseases or to segment lesions or vital organs related to the human body. With the development of deep learning, the correct rate of computer-aided image pre-diagnosis has been greatly improved, which has greatly improved the efficiency of doctors' consultations, and also contributed to the healthy development of the medical industry. Deep learning will have a broader role in the medical field. prospect.

Therefore, in this embodiment, a deep learning method is adopted to establish a corneal segmentation network model that can automatically segment the corneal region in the video image. Specifically, by acquiring the clinically collected corneal video, the first frame of image is selected and the corneal region in the first frame of image is marked. As shown in FIG. 2 , FIG. 2 is the first frame image A and the marked image B obtained when the cornea segmentation network model is established in the present invention. Preferably, the corneal area is usually marked by manual tracing, and the accuracy is ensured as much as possible, so as to establish a more accurate network model. Then obtain the marked corneal region features, and carry out deep learning and training; in the process of training, a sine-wave-like displacement is introduced through a piecewise affine transformation to simulate the change of the corneal boundary. As shown in FIG. 3 , FIG. 3 is a schematic diagram of changes in corneal boundaries simulated by segmented affine transformation when establishing a corneal segmentation network model in the present invention. The present invention performs deep learning based on a convolutional neural network. The features learned by the deep neural network in this embodiment are not sensitive to random bright spots, and can accurately and robustly locate the upper and lower boundaries of the cornea. In addition, image features are automatically obtained through deep learning, which replaces the inconvenience of traditional cumbersome manual feature acquisition, eliminates the interference of random noise, and improves the accuracy of feature acquisition.

Furthermore, during the training process, in order to more accurately simulate the segmentation result of the previous frame, the present invention also adds conventional affine transformation, projective transformation or transmission transformation, such as up and down displacement, during offline training. It is worth noting that in this step, it is necessary to use the image erosion technology with a structural element of 1 pixel or the spike removal technology to smooth the sharp edges in the corneal area changes in the front and rear frame images, so as to better simulate the changes in the corneal boundary . The affine transformation, projective transformation, transmission transformation, and de-smoothing technology added in the above steps are only used to illustrate the technical solution of the present invention, and are not used to limit the present invention. Other forms of transformation technology or de-smoothing technology belong to protection scope of the present invention.

In this embodiment, the established corneal segmentation network model is embedded into the terminal, so that the terminal has the function of automatically performing corneal segmentation. When it is necessary to perform corneal segmentation on clinically collected video data, the terminal acquires the corneal video data to be processed, and inputs the video data into the aforementioned corneal segmentation network model established through offline training.

Further, in step S200, the segmentation network model automatically segments the region where the target object is located in the video data, and outputs the segmentation result.

Preferably, the step S200 specifically includes:

Step S201, the segmentation network model sequentially performs target object region segmentation on each frame image in the target object video data;

Step S202, when performing target object region segmentation on the next frame image, automatically input the segmentation result of the previous frame image into the segmentation network model;

Step S203 , the segmentation network model uses the segmentation result of the previous frame image as a reference, and performs target object region segmentation on the next frame image.

During specific implementation, since the target object is the cornea in this embodiment, and the segmentation network model is the cornea segmentation network model, the cornea segmentation network model sequentially performs corneal region segmentation on each frame of image in the corneal video data. When traditionally processing a single image, the timing problem is not taken into consideration, but when the corneal segmentation network model of the present invention performs corneal region segmentation on the next frame image, the segmentation result of the previous frame image is automatically input into the cornea Segmentation network model; the segmentation result of the previous frame image is used as a reference, and the corneal boundary is roughly estimated, and then the corneal region is segmented for the next frame image, so as to better improve the accuracy of corneal segmentation.

Preferably, in this embodiment, the segmentation result is further input into the corneal segmentation network model, so as to update the corneal segmentation network model, thereby further improving the accuracy of corneal segmentation.

Specifically, this embodiment can perform polynomial fitting on the upper and lower boundaries of the segmentation results, and the final result is shown in Figure 4. Figure 4 is the effect of the cornea segmentation and polynomial fitting using the cornea segmentation network model of the present invention picture. It can be seen from Figure 4 that even if there are random highlight artifacts, the effect on the fitting result is very small. It is worth noting that when performing polynomial fitting, because the number of times is high, overfitting is prone to occur. Therefore, when dealing with this problem, this technology does not consider the area of about 60 pixels on the left and right borders of the image, which can be greatly improved. Improve the accuracy of the fit.

Further, in this embodiment, the corneal thickness can be calculated according to the determined upper and lower boundaries of the cornea. In order to further improve the robustness, when calculating the thickness, an area of 10 pixels around the center of the image, with a total of 20 pixels, is used to calculate the average thickness. The corneal boundary automatically measured by the corneal segmentation network model established by the present invention is more accurate, so the obtained corneal thickness is also more accurate.

Of course, although the method for locating the boundary of the target object in the medical image proposed by the present invention takes the method of locating the corneal boundary as an example, it does not limit the present invention to be used only for medical image processing, and can also be used in natural image video Splitting.

Based on the above embodiments, the present invention also discloses a terminal, as shown in FIG. 5 , including: a processor (processor) 10, and a storage medium (memory) 20 connected to the processor 10; Invoking the program instructions in the storage medium 20 to execute the method provided by the above embodiment, for example:

Step S100, the terminal obtains the clinically collected video data of the target object to be processed, and inputs the video data into the segmentation network model established in advance through offline training;

Step S200 , the segmentation network model automatically segments the region where the target object is located in the video data, and outputs the segmentation result.

An embodiment of the present invention also provides a storage medium, where computer instructions are stored on the storage medium, and the computer instructions cause a computer to execute the methods provided in the foregoing embodiments.

In summary, the present invention provides a method for locating the boundary of a target object in a medical image, a storage medium, and a terminal. The method is applied in the technical field of medical image processing, and specifically includes: the terminal acquires the video of the target object to be processed collected clinically data, and input the video data to the segmentation network model established in advance through offline training; the segmentation network model automatically segments the area where the target object in the video data is located, and outputs the segmentation result. In the present invention, by inputting the medical image video to be processed into the segmentation network model established through learning and training in advance for segmentation, the boundary of the target object can be accurately located in real time, and the high-light spots and artifacts in the imaging can be eliminated to locate the upper and lower interfaces of the target area. To improve the accuracy of boundary judgment of the target area and meet the clinically required high-precision requirements.

It should be understood that the application of the present invention is not limited to the above examples, and those skilled in the art can make improvements or transformations according to the above descriptions, and all these improvements and transformations should belong to the protection scope of the appended claims of the present invention.

Claims (10)

1. a boundary location method of target object in a medical image, it is characterized in that, described method is applied to medical image processing technical field, specifically comprises: Step A, the terminal acquires the video data of the target object collected clinically to be processed, and inputs the video data into the segmentation network model established in advance through offline training; Step B. The segmentation network model automatically segments the region where the target object is located in the video data, and outputs the segmentation result. 2. according to the boundary localization method of target object in medical image described in claim 1, it is characterized in that, also comprise before described step A: In step S, a segmentation network model for automatically segmenting the region where the target object is located is established in advance through offline training. 3. according to the boundary localization method of target object in medical image described in claim 1, it is characterized in that, also comprise after described step B: Step C, performing polynomial fitting on the segmented region of the target object to obtain the upper and lower boundaries of the target object. 4. according to the boundary localization method of target object in medical image described in claim 2, it is characterized in that, described step S comprises: Step S1. Obtain the collected video of the target object, select the first frame of image and mark the area where the target object is located in the first frame of image; Step S2, obtaining the regional features of the target object, and performing deep learning and training; Step S3. In the training process, a sinusoidal-like displacement is introduced by piecewise affine transformation, the change of the boundary of the target object is simulated, and a segmentation network model for automatically segmenting the region where the target object is located is established. 5. according to the boundary localization method of target object in medical image described in claim 2, it is characterized in that, described step S also comprises: Add conventional affine transformation, projective transformation or transmission transformation in the training process, and use image erosion technology or spike removal technology to smooth the sharp edges in the change of the target object area in the front and rear frame images. 6. according to the boundary localization method of target object in medical image described in claim 1, it is characterized in that, described step B comprises: Step B1, the segmentation network model sequentially performs target object region segmentation on each frame of image in the target object video data; Step B2, when performing target object area segmentation on the next frame image, automatically input the segmentation result of the previous frame image into the segmentation network model; Step B3, the segmentation network model uses the segmentation result of the previous frame image as a reference, and performs target object area segmentation on the next frame image. 7. according to the boundary localization method of target object in medical image described in claim 1, it is characterized in that, described step B also comprises: The segmentation result is input into the segmentation network model to update the segmentation network model. 8. according to the boundary localization method of target object in medical image described in claim 3, it is characterized in that, also comprise after described step C: Step D. Calculate the thickness of the target object according to the upper and lower boundaries of the target object. 9. A storage medium on which a plurality of instructions are stored, wherein the instructions are adapted to be loaded and executed by a processor, so as to realize the target object in a medical image according to any one of claims 1-8 The boundary location method. 10. A terminal, characterized by comprising: a processor, a storage medium communicatively connected to the processor, the storage medium is suitable for storing multiple instructions; the processor is suitable for calling the instructions in the storage medium, In order to implement the method for locating the boundary of the target object in the medical image described in any one of claims 1-8 above.