CN114549425A - Medical image detection method and device, storage medium and computer equipment - Google Patents

Abstract

The application provides a medical image detection method, a medical image detection device, a storage medium and computer equipment, wherein the method comprises the following steps: acquiring a lung medical image; extracting arterial blood vessels and bronchus in the lung medical image, wherein the bronchus comprises a plurality of bronchus sections with different levels; determining a target arterial vessel section corresponding to the bronchial section in the pulmonary medical image according to the distance between the centerline of the bronchial section and the centerline of the arterial vessel; and detecting the bronchial segment and the target arterial vessel segment, and determining the detection result of the bronchial segment. The method can automatically detect the columnar signs of the bronchus, is favorable for improving the film reading efficiency of doctors, and reduces the workload.

Description

Medical image detection method, device, storage medium and computer equipment

technical field

The present application relates to the technical field of medical image processing, and in particular, to a medical image detection method, device, storage medium and computer equipment.

Background technique

Bronchiectasis (referred to as bronchiectasis) is due to chronic suppurative inflammation and fibrosis of the bronchus and its surrounding lung tissue, which destroys the muscle and elastic tissue of the bronchial wall, resulting in bronchial deformation and persistent expansion. The imaging manifestations of bronchiectasis can be divided into columnar, cystic columnar, and cystic dilatation. For the same bronchial segment, columnar bronchiectasis is mostly a sign of early disease development, while cystic is a sign of late disease, and cystic columnar bronchiectasis is between the two. However, for different lobes, the degree of imaging involvement of bronchiectasis varies greatly between individuals.

The scoring system in the related art can comprehensively evaluate bronchiectasis and its accompanying imaging signs, and the system does not treat the three signs differently in terms of procedures and methods during model training and use. However, cystic and cystic-columnar signs are easy to distinguish from normal bronchial imaging features, while columnar signs are difficult to distinguish from normal bronchus, and the degree of imaging involvement of bronchiectasis varies greatly among individuals. If the same method is used to detect three signs at the same time, the lack of consideration of a single individual factor will inevitably lead to low detection accuracy of columnar signs, so that the graded evaluation of bronchiectasis cannot be achieved, and there is bound to be a large deviation in the initial diagnosis of bronchiectasis. and blank space.

SUMMARY OF THE INVENTION

In view of this, the present application provides a medical image detection method, device, storage medium and computer equipment, which can automatically detect the columnar signs of the bronchus, which is beneficial to improve the doctor's reading efficiency and reduce the workload.

According to one aspect of the present application, a method for detecting medical images is provided, comprising:

Acquire lung medical images; extract arterial blood vessels and bronchi in lung medical images, bronchus includes multiple bronchial segments at different levels; determine lung medical images according to the distance between the centerline of the bronchial segment and the centerline of the arterial blood vessel The target arterial vessel segment corresponding to the middle bronchial segment; the bronchial segment and the target arterial vessel segment are detected and processed to determine the detection result of the bronchial segment.

Optionally, performing detection processing on the bronchial segment and the target arterial vessel segment to determine the detection result of the bronchial segment, specifically including:

According to the preset interval, at least one of the second bronchial segment, the target arterial vessel segment corresponding to the first bronchial segment, and the first bronchial segment in the pulmonary medical image are intercepted to obtain a first image, the second bronchial segment is the bronchial segment of the previous level of the first bronchial segment; the first image and/or the mask of the first image is input into the columnar sign detection model to determine the detection result of the first bronchial segment; wherein, the columnar sign detection model is based on The lung sample images and the columnar sign labels corresponding to the lung sample images are trained.

Optionally, the detection method of the medical image further includes:

Obtain a lung sample image, the lung sample image includes a plurality of different levels of bronchial segment samples and a plurality of different levels of bronchial segment samples corresponding to the target arterial blood vessel segment samples; the diameter of the first bronchial segment sample is larger than the corresponding target artery When the diameter of the blood vessel segment sample, or the difference between the diameter of the first bronchial segment sample and the diameter of the second bronchial segment sample is greater than the preset difference, a columnar sign label of the first bronchial segment sample is generated, and the columnar sign label is It is used to mark the presence of columnar signs in the bronchus; at least one of the second bronchial segment sample marked with the columnar sign label, the target arterial blood vessel segment sample corresponding to the first bronchial segment sample, and the first bronchial segment sample are intercepted to obtain the first bronchial segment sample. Two images; according to the second image and/or the mask of the second image, a neural network model is trained to obtain a columnar sign detection model.

Optionally, according to the distance between the centerline of the bronchial segment and the centerline of the arterial vessel, determine the target arterial vessel segment corresponding to the bronchial segment in the pulmonary medical image, specifically including:

Determine the first sampling point on the centerline of the bronchial segment; calculate the distances between the first sampling point and a plurality of second sampling points on the centerline of the arterial vessel; determine the second sampling point corresponding to the smallest distance as A companion point corresponding to the first sampling point; at least one arterial vessel segment corresponding to the bronchial segment is determined according to the start and end coordinates of the bronchial segment; and a target arterial vessel segment is selected from at least one arterial vessel segment according to the location of the companion point.

Optionally, extract arterial blood vessels and bronchi in lung medical images, including:

Input the lung medical image into the arterial blood vessel segmentation model to obtain the arterial blood vessels in the lung medical image; input the lung medical image into the bronchial segmentation model to obtain the bronchus in the lung medical image; wherein, the arterial blood vessel segmentation model is based on the lung samples The images and their corresponding arterial and vessel segmentation labels are trained, and the lung segmentation model is trained based on the lung sample images and their corresponding lung segmentation labels.

Optionally, the detection method of the medical image further includes:

When the detection result is that there are columnar signs, the bronchial segments with columnar signs in the lung medical images are displayed differentially.

Optionally, the detection method of the medical image further includes:

When the detection result is that there is a columnar sign, the center point and the inner diameter of the bronchial segment where the columnar sign is present are associated and stored.

Optionally, acquiring lung medical images, including:

Obtaining the medical image to be detected; preprocessing the medical image to be detected; performing lung segmentation processing on the preprocessed medical image to be detected to obtain a medical lung image; wherein the preprocessing includes at least one of the following: noise reduction processing, normalization Normalization processing, size correction processing, CT value truncation processing, resampling processing.

According to another aspect of the present application, a medical image detection device is provided, the device comprising:

The acquisition module is used to acquire lung medical images; the extraction module is used to extract the arterial blood vessels and bronchi in the lung medical images, and the bronchi includes multiple bronchial segments of different levels; the determination module is used to extract the bronchial segments according to the center line and the bronchial segment. The distance between the centerlines of the arterial vessels determines the target arterial vessel segment corresponding to the bronchial segment in the lung medical image; the detection module is used to detect and process the bronchial segment and the target arterial vessel segment to determine the detection result of the bronchial segment.

Optionally, the medical image detection device further includes: an interception module, which, according to a preset interval, performs at least one of the second bronchial segment, the target arterial vessel segment corresponding to the first bronchial segment, and the first bronchial segment in the pulmonary medical image, according to a preset interval. The bronchial segment is intercepted to obtain the first image, and the second bronchial segment is the bronchial segment of the previous level of the first bronchial segment; the detection module is specifically used to input the first image and/or the mask of the first image into the columnar A sign detection model is used to determine the detection result of the first bronchial segment; wherein, the columnar sign detection model is trained based on the lung sample image and the columnar sign label corresponding to the lung sample image.

Optionally, the acquisition module is further configured to acquire lung sample images, where the lung sample images include multiple bronchial segment samples of different levels and samples of target arterial blood vessel segments corresponding to multiple bronchial segment samples of different levels; The detection device further includes: a marking module, used for when the diameter of the first bronchial segment sample is larger than the diameter of the corresponding target arterial blood vessel segment sample, or the difference between the diameter of the first bronchial segment sample and the diameter of the second bronchial segment sample When the difference is greater than the preset value, a columnar sign label of the first bronchial segment sample is generated, and the columnar sign label is used to mark the presence of columnar signs in the bronchus; the interception module is also used for the second bronchial segment sample marked with the columnar sign label, At least one of the target arterial blood vessel segment samples corresponding to the first bronchial segment sample and the first bronchial segment sample are intercepted to obtain a second image; the medical image detection device further includes: a training module for /or the mask of the second image, train the neural network model, and obtain the columnar sign detection model.

Optionally, a determining module is specifically configured to determine a first sampling point on the centerline of the bronchial segment; respectively calculate the distances between the first sampling point and a plurality of second sampling points on the centerline of the arterial blood vessel; the determining module, Specifically, the second sampling point corresponding to the smallest distance is determined as the accompanying point corresponding to the first sampling point; at least one arterial vessel segment corresponding to the bronchial segment is determined according to the coordinates of the starting point and the end point of the bronchial segment; position, and a target arterial vessel segment is selected from at least one arterial vessel segment.

Optionally, the extraction module is specifically configured to input the pulmonary medical image into the arterial blood vessel segmentation model to obtain the arterial blood vessels in the pulmonary medical image; input the pulmonary medical image into the bronchial segmentation model to obtain the bronchus in the pulmonary medical image; Among them, the arterial blood vessel segmentation model is trained based on the lung sample images and their corresponding arterial blood vessel segmentation labels, and the lung segmentation model is trained based on the lung sample images and their corresponding lung segmentation labels.

Optionally, the medical image detection device further includes:

The display module is used for differentially displaying the bronchial segments with columnar signs in the lung medical image when the detection result is that there are columnar signs.

Optionally, the medical image detection device further includes:

The storage module is configured to associate and store the center point and the inner diameter of the bronchial segment where the columnar sign exists when the detection result is that the columnar sign exists.

Optionally, an acquisition module is specifically used to acquire the medical image to be detected; the medical image detection device further includes: a preprocessing module, used to preprocess the medical image to be detected; a segmentation module, used to preprocess the to-be-detected image The medical image is subjected to lung segmentation processing to obtain a lung medical image; wherein, the preprocessing includes at least one of the following: noise reduction processing, normalization processing, size correction processing, CT value truncation processing, and resampling processing.

According to yet another aspect of the present application, a storage medium is provided on which a computer program is stored, and when the program is executed by a processor, the foregoing method for detecting a medical image is implemented.

According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, and the processor implements the above-mentioned detection method for medical images when the computer program is executed. .

With the above technical solution, considering the clinical criteria of columnar signs, such as the diameter of the bronchus is larger than the diameter of the accompanying pulmonary artery, or the diameter of the bronchus does not tend to decrease with the course of the bronchial tree. Using bronchial image information and accompanying arterial blood vessel information as the basis for judging columnar signs, it realizes automatic detection and grading evaluation of bronchial columnar signs, reduces differences in doctors' subjective judgments, improves the detection accuracy of columnar signs, and assists doctors in Image judgment and identification to improve the ability of accurate diagnosis and treatment of bronchiectasis.

The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the description, and in order to make the above-mentioned and other purposes, features and advantages of the present application more obvious and easy to understand , and the specific embodiments of the present application are listed below.

Description of drawings

The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The schematic embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation of the present application. In the attached image:

FIG. 1 shows one of the schematic flowcharts of the medical image detection method provided by the embodiment of the present application;

FIG. 2 shows the second schematic flowchart of the medical image detection method provided by the embodiment of the present application;

FIG. 3 shows a structural block diagram of an apparatus for detecting a medical image provided by an embodiment of the present application.

Detailed ways

Hereinafter, the present application will be described in detail with reference to the accompanying drawings and in conjunction with the embodiments. It should be noted that the embodiments in the present application and the features of the embodiments may be combined with each other in the case of no conflict.

The typical symptoms of bronchiectasis are chronic cough, copious purulent sputum and repeated hemoptysis. There may be no obvious signs in the early stage of bronchiectasis, and wet rales of different sizes can be heard in the affected area. After the bronchiectasis gradually increases, symptoms of hypoxia appear, clubbing of fingers, toes and poor general nutrition can be seen. As described in the background art, the currently used bronchiectasis evaluation system cannot accurately judge the columnar signs. In clinical application, an experienced doctor is often required to evaluate computed tomography (CT) images. Subjectivity, and the difference between the lesion area and the normal area and the boundary of the blood supply area on conventional CT images are difficult to directly observe with the naked eye, and cannot achieve a high diagnostic accuracy. Therefore, it is necessary to develop a system that can automatically evaluate CT images for the automatic detection and grading of bronchiectasis.

In order to reduce the differences in the subjectivity of doctors, improve the efficiency and accuracy of the diagnosis of columnar bronchiectasis, reduce the waiting time of patients, and improve the treatment effect, the present application provides a medical image detection method, device, storage medium and computer equipment. The following is a detailed description through examples.

As shown in FIG. 1 , a method for detecting medical images is provided in this embodiment, and the method includes:

Step 101, obtaining a lung medical image;

Wherein, the chest of the patient can be scanned by the chest photographing equipment, so as to obtain a medical image of the patient's lungs, and the medical image of the lungs can be a CT image.

Specifically, the step of acquiring the pulmonary medical image specifically includes: acquiring the medical image to be detected; preprocessing the medical image to be detected; and performing lung segmentation processing on the preprocessed medical image to be detected to obtain the pulmonary medical image.

The preprocessing includes at least one of the following: noise reduction processing, normalization processing, size correction processing, CT value truncation processing, and resampling processing.

In this embodiment, the chest region of the patient is scanned by the CT equipment to obtain the medical image to be detected. After obtaining the medical image to be detected, the medical image to be detected is preprocessed, so that the subsequent processing can obtain image input with uniform specifications, reduce the interference of non-critical information, thus ensure the imaging quality, and improve the detection accuracy, stability and consistency . After preprocessing, lung segmentation processing is performed on the medical image to be detected based on the lung segmentation model to obtain at least one area of interest in the lung corresponding to the medical image to be detected, that is, the medical image of the lung, so that the medical image to be detected can be segmented. The parenchymal area of the lung is extracted to remove the image of the sternum, so as to avoid the interference of the sternum image in the subsequent image registration, which is convenient to accurately identify the bronchial columnar signs of the lung.

In practical application scenarios, taking the 3D plain CT image as the original input image as an example, during preprocessing, the size of the plain CT image is adjusted to a preset size through size correction processing, so as to unify all medical images and facilitate the The extracted features are then compared with the disease features. Then use linear or nonlinear smoothing and filtering techniques to denoise the plain CT image. For example, noise (eg, salt and pepper noise) in the image is removed using at least one of Gaussian smoothing filtering, median filtering, and the like. Taking median filtering as an example, the gray value of each pixel in the plain CT image is set to the median value of all pixel gray values in a certain neighborhood window of the point, so that the surrounding pixel values are close to the true value. , thereby eliminating isolated noise points. After denoising, the CT value of the plain CT image is truncated and normalized between -1000 and 1000. The obtained image value is distributed between 0 and 1, and the pixel size is 0.6 mm. The direction is re-interpolated to obtain a normalized CT image with a width, height and depth of W×H×D. Finally, lung ROIs are extracted using a lung segmentation model or a region growing method. Among them, the lung segmentation model can use the UNet network architecture or related network architecture. Since the neural network model has the advantages of fast processing speed and high processing accuracy, lung segmentation through the lung segmentation model can effectively improve the lung segmentation. Therefore, the rate and accuracy of subsequent bronchial columnar sign detection are improved, and the robustness is high.

Step 102, extracting the arterial blood vessels and bronchi in the lung medical image;

Among them, the bronchus includes multiple bronchial segments of different levels. There are about 24 branches from human bronchi to alveoli, forming a bronchial tree. The bronchial segments of different levels can be recorded as L ₁ , L ₂ ......L _n . It can be understood that a bronchial segment at the same level can include multiple bronchial branches. Taking a first-level bronchial segment as an example, multiple bronchial branches at the same level can be denoted as L _1-1 , L _1-2 ......L _{1- m} . Using the centerline extraction method, the centerlines of bronchial segments of different levels are extracted, which can be recorded as line0 _nm , and each centerline includes the starting point P ₀ and the ending point P _w .

In practical application scenarios, three bronchial grading algorithms, namely traditional algorithms, deep learning models, template matching methods and semantic segmentation algorithms, can be used to classify bronchial tubes.

Specifically, the lung medical image is input into a bronchial segmentation model to obtain the bronchi in the lung medical image, and the lung segmentation model is trained based on the lung sample image and its corresponding lung segmentation label. The pulmonary medical images are input into the arterial blood vessel segmentation model to obtain the arterial blood vessels in the pulmonary medical images. The arterial blood vessel segmentation model is trained based on the lung sample images and their corresponding arterial blood vessel segmentation labels.

It is worth mentioning that since the arteries and veins cross each other on the image, the two are adjacent and the grayscale is very close, especially in the non-contrast-enhanced CT images, the difference between the arteriovenous bifurcation and the intersection is very small. Therefore, the CT value corresponding to each pixel in the lung medical image is obtained first; all the target pixels corresponding to the CT value within the threshold range are screened out, and the pulmonary vascular area and impurity area of the lung medical image are roughly segmented, so as to combine the lung. The CT value corresponding to each pixel in the partial CT image ensures that all the pulmonary blood vessel regions in the lung CT image can be extracted and avoids omissions during segmentation and extraction. Further, according to the boundary composed of any pixel point in the pulmonary blood vessel area and the impurity area and the pixel point in its neighborhood, the pulmonary blood vessel area is finely segmented. The image of the pulmonary vascular area is input into the arterial vascular segmentation model to effectively segment the arterial area and the venous area, and improve the accuracy and efficiency of pulmonary arterial and venous separation.

Step 103, according to the distance between the centerline of the bronchial segment and the centerline of the arterial vessel, determine the target arterial vessel segment corresponding to the bronchial segment in the pulmonary medical image;

Among them, the target arterial vessel segment is the arterial vessel with the shortest distance from the bronchial segment, and is also called the accompanying arterial vessel.

Step 104 , perform detection processing on the bronchial segment and the target arterial vessel segment, and determine the detection result of the bronchial segment.

In this embodiment, considering the clinical criteria for columnar signs, such as the diameter of the bronchus is larger than the diameter of the accompanying pulmonary artery, or the diameter of the bronchus does not tend to decrease along the course of the bronchial tree. Extract arterial vessels and bronchi in lung medical images to obtain accurate bronchial and arterial information. The target arterial vessel segment closest to the bronchial segment is determined by the distance between the centerline of the bronchial segment and the centerline of the arterial vessel. Then, the bronchial image information and target arterial blood vessel information are used as the basis for judging the columnar signs, which realizes the automatic detection and grading evaluation of the bronchial columnar signs, reduces the differences in the subjective judgment of doctors, improves the detection accuracy of the columnar signs, and then assists the doctors. Perform image judgment and identification to improve the ability of accurate diagnosis and treatment of bronchiectasis.

In the embodiment of the present application, further, the steps of performing detection processing on the bronchial segment and the target arterial vessel segment, and determining the detection result of the bronchial segment, specifically include:

Step 104-1, according to a preset interval, perform interception processing on at least one of the second bronchial segment, the target arterial vessel segment corresponding to the first bronchial segment, and the first bronchial segment in the pulmonary medical image to obtain a first image ;

Wherein, the first bronchial segment is the bronchus of the level that the system user needs to detect the columnar signs, and the second bronchial segment is the bronchial segment of the previous level of the first bronchial segment. For example, if the first bronchial segment is the fourth-level bronchial segment, then The second bronchial segment is a third-order bronchial segment, in other words, the first bronchial segment is a branch of the second bronchial segment.

Further, the first image includes at least one three-dimensional image block of the first bronchial segment, and at least one three-dimensional image block of the second bronchial segment and/or the target arterial vessel segment, for example, the first image includes two of the first bronchial segment. A partial image and 2 partial images of the target arterial vessel segment, or 2 partial images of the first bronchial segment and 2 partial images of the second bronchial segment, or each of the first bronchial segment, the second bronchial segment, and the target arterial vessel segment. 2 partial images. Moreover, on the bronchial segment or the target arterial vessel segment, the distance between the center points of two adjacent three-dimensional image blocks is a preset distance.

Step 104-2: Input the first image and/or the mask of the first image into the columnar sign detection model to determine the detection result of the first bronchial segment.

The columnar sign detection model is trained based on the lung sample images and the columnar sign labels corresponding to the lung sample images. The columnar sign label is used to distinguish the presence or absence of columnar signs in the bronchial segment samples in the lung sample images.

In this embodiment, the first image and/or the mask of the first image is used as input data, and the columnar sign detection model is used to determine the diameter difference of the first bronchial segment and its accompanying pulmonary artery, or the first bronchus The change trend of the diameter of the bronchial segment relative to the bronchial segment of the previous level is analyzed, so as to determine whether the columnar sign occurs in the first bronchial segment. The classification model constructed by deep learning technology can automatically detect bronchial columnar signs. At the same time, because the bronchus is graded in advance, it can detect the bronchus of the required level in a targeted manner, and realize the bronchial classification evaluation. This reduces differences in doctors' subjective judgments, improves the diagnostic accuracy of columnar signs, and assists doctors in image diagnosis and identification, supports subsequent treatment and rehabilitation, and improves the ability to accurately diagnose and treat bronchiectasis.

It should be noted that since there is no second bronchial segment in the first-level bronchial segment, it is impossible to analyze whether the columnar sign occurs in the first-level bronchial segment through the change trend of the tube diameter.

It can be understood that, in order to avoid the detection error of the model, the first image related to the first bronchial segment of a level can be input according to the bronchial level, and the detection result of the first bronchial segment of this level can be obtained. The first images related to the first bronchial segment of another level are then input. This cycle is repeated until all the first bronchial segments of the system user demand level are detected, and the detection results of all the first bronchial segments are aggregated.

Further, the step of obtaining the columnar sign detection model includes: obtaining a lung sample image, where the lung sample image includes a plurality of bronchial segment samples of different levels and a plurality of bronchial segment samples of different levels and corresponding target arterial blood vessel segment samples; When the diameter of the first bronchial segment sample is greater than the diameter of the corresponding target artery segment sample, or the difference between the diameter of the first bronchial segment sample and the diameter of the second bronchial segment sample is greater than the preset difference, the first bronchial segment sample is generated. A columnar sign label of the bronchial segment sample, the columnar sign label is used to mark the presence of columnar signs in the bronchus; for at least one of the second bronchial segment sample marked with the columnar sign label and the target artery segment sample corresponding to the first bronchial segment sample, And the first bronchial segment sample is intercepted to obtain a second image; according to the second image and/or the mask of the second image, a neural network model is trained to obtain a columnar sign detection model.

Wherein, the second bronchial segment sample is the bronchial segment of the previous level of the first bronchial segment sample. As the human body is healthy, the diameter of the bronchus tends to decrease as the bronchial tree travels, that is, the diameter of the first bronchial segment sample is smaller than that of the second bronchial segment sample. Then, in order to determine whether the first bronchial segment sample has columnar signs, the preset difference can be set to a value less than 0, and considering that the higher the level, the thinner the bronchial diameter, the difference between the bronchial diameters of two adjacent levels The smaller the value, the higher the level of the first bronchial segment sample can be set, and the larger the preset difference value is.

In this embodiment, the diameters of the first bronchial segment sample, the second bronchial segment sample, and the target arterial vessel segment sample in the lung sample image are identified. Compare the diameters of the first bronchial segment sample and the second bronchial segment sample in the lung sample image, if the difference between the diameter of the first bronchial segment sample and the second bronchial segment sample is greater than the preset difference, it means that the bronchial The change trend of the diameter of the segment satisfies the columnar sign; or compare the diameters of the first bronchial segment sample and the corresponding target arterial blood vessel segment sample in the lung sample image, if the diameter of the first bronchial segment sample is larger than the corresponding target arterial blood vessel The diameter of the segment sample, indicating the presence of columnar signs. After the columnar sign is detected, a columnar sign label is generated to label the first bronchial segment sample. The neural network model is trained using the first bronchial segment sample marked with the columnar sign label and the second image of the second bronchial segment sample and/or the target arterial vessel segment sample corresponding to the first bronchial segment sample, or using the second image The mask is used to train the neural network model to obtain a columnar feature detection model. Therefore, a feature model (columnar sign detection model) for detecting normal bronchi and columnar sign bronchi is established by using big data training through deep learning, and then the model is used to distinguish normal bronchi and columnar dilated bronchi in clinical lung image data. Reduce the workload of doctors.

Specifically, the neural network (Neural Networks, NN) model may be a convolutional neural network (Convolutional Neural Networks, CNN) model, a residual shrinkage network (Deep Residual Shrinkage Network, DRSN) model, a fully connected neural network (Fully Connected Neural Network, FCNN) model, Recurrent Neural Network (RNN) model or Long Short-Term Memory (LSTM) model. The embodiments of the present application are not specifically limited.

It can be understood that, if a columnar sign label has been added to the lung sample image, it is not necessary to perform the step of diameter comparison again, and the second image marked with the columnar sign label and/or the mask of the second image may be Train a neural network model. For example, the patient's historical lung CT images are used as lung sample images, and during the patient's medical treatment, the results marked by doctors are used as columnar sign labels.

In practical application scenarios, in order to ensure the detection accuracy of columnar signs, the training samples required for training the model are as consistent as possible with the data input during detection. For example, when training the columnar sign detection model, the second image of the first bronchial segment sample and the second bronchial segment sample is used as the training sample, then in the subsequent detection by the columnar sign detection model, it is necessary to include the first bronchial segment and the second bronchial segment sample. The first image of the bronchial segment is input to the columnar sign detection model. For another example, when training the columnar sign detection model, the second image mask of the first bronchial segment sample and its corresponding target arterial vessel segment sample is used as the training sample, then in the subsequent detection by the columnar sign detection model, it is necessary to include the first A first image mask of a bronchial segment and a target arterial vessel segment is input to the columnar feature detection model.

In the embodiment of the present application, further, according to the distance between the centerline of the bronchial segment and the centerline of the arterial vessel, the step of determining the target arterial vessel segment corresponding to the bronchial segment in the pulmonary medical image specifically includes:

Step 103-1, determining the first sampling point on the centerline of the bronchial segment;

Wherein, the position of the first sampling point on the centerline of the bronchial segment can be set as required, the number of the first sampling point on the centerline of a tube segment can be multiple, and the spacing between the multiple first sampling points can be the same or can be different. For example, it is preset to select 6 first sampling points on the center line of the bronchial segment, and the selection rule is that the starting point of the center line is the first first sampling point, the end point is the sixth first sampling point, and the first sampling point is the first sampling point. The points are equally spaced.

Step 103-2, respectively calculating the distances between the first sampling point and a plurality of second sampling points on the centerline of the arterial blood vessel;

Step 103-3, determining the second sampling point corresponding to the smallest distance as the accompanying point corresponding to the first sampling point;

In this embodiment, for a first sampling point, the distances between the first sampling point and a plurality of second sampling points on the centerline of the arterial blood vessel are calculated respectively, to obtain a set of distances related to the first sampling point. The second sampling point corresponding to the minimum value in the distance set, that is, the second sampling point on the arterial blood vessel that is closest to the first sampling point, is taken as the accompanying point corresponding to the first sampling point, so as to facilitate the subsequent determination of each sampling point. The bronchial segment corresponds to the target arterial vessel segment.

Step 103-4: Determine at least one arterial vessel segment corresponding to the bronchial segment according to the coordinates of the starting point and the ending point of the bronchial segment;

Step 103-5: Select a target arterial vessel segment from at least one arterial vessel segment according to the position of the accompanying point.

In this embodiment, the arterial blood vessel is intercepted according to the coordinates of the start point and the end point of the bronchial segment, so as to determine at least one arterial blood vessel segment that conforms to the length range of the bronchial segment. Considering that the multiple concomitant points corresponding to the multiple first sampling points of each bronchial segment may not be on the centerline of the same arterial vessel. By means of statistics, the arterial vessel segment with the most concomitant points is selected as the target arterial vessel segment closest to the bronchial segment. In order to facilitate the detection of columnar signs with the relevant information of the target arterial vessel segment and bronchial segment.

In the embodiment of the present application, further, the method for detecting a medical image further includes: when the detection result is that there is a columnar sign, differentially displaying the bronchial segments with the columnar sign in the lung medical image.

In this embodiment, if it is determined that the bronchial segment in the pulmonary medical image has the columnar sign, the bronchial segment with the columnar sign and the bronchial segment without the columnar sign in the pulmonary medical image are displayed differently. In order to facilitate the system user (doctor) to intuitively know the expansion of the bronchial segment, it is beneficial to improve the identification efficiency of bronchiectasis.

Specifically, the way of displaying the difference may be to generate a columnar sign label at a specified position of the bronchial segment in the lung medical image; it may also be to highlight the label information of the bronchial segment, for example, the name of the bronchial segment in the lung medical image. Highlighting is performed; text prompting information of the presence of columnar signs in the bronchial segment can also be added to the pulmonary medical image. The embodiments of the present application are not specifically limited.

Further, the step of differentially displaying bronchial segments with columnar signs in the pulmonary medical image specifically includes: extracting sign feature information of the bronchial segments with columnar signs; determining a display level according to the sign feature information; Differential display of bronchial segments with columnar signs. For example, the severity of columnar bronchiectasis is determined by the sign feature information, and the display mode of the bronchial segment is determined by the corresponding relationship between the severity and the display level. For example, if the diameter difference between the first bronchial segment and the second bronchial segment is greater than the preset difference associated with the first display level, the first bronchial segment is determined as the first display level, and the display mode corresponding to the first display level is as follows: The marking information of the bronchial segment is highlighted in red; if the diameter difference between the first bronchial segment and the second bronchial segment is greater than the preset difference associated with the second display level, the first bronchial segment is determined as the second display level, and the first bronchial segment is determined as the second display level. The display mode corresponding to a display level is to highlight the marking information of the bronchial segment in yellow, wherein the preset difference value associated with the first display level is greater than the preset difference value associated with the second display level, that is, the first display level Bronchial dilatation was more severe in grade bronchial segments than in bronchial segments in the second displayed grade. Therefore, through different display methods, the system users can intuitively distinguish bronchial segments with different bronchiectasis through lung medical images, which greatly improves the identification efficiency of bronchiectasis.

In the embodiment of the present application, further, the method for detecting a medical image further includes: in the case that the detection result is that there is a columnar sign, the center point and the inner diameter of the bronchial segment where the columnar sign is present are associated and stored.

In this embodiment, if it is determined that there is a columnar sign in the bronchial segment of the pulmonary medical image, the center point and inner diameter of the bronchial segment with the columnar sign are stored in association, so as to realize the information traceability of the columnar sign, so that the system user can intuitively Knowing the dilation of the bronchial segments reduces the workload of the system users in reading images, which is beneficial to improve the identification efficiency of bronchiectasis.

As shown in FIG. 2, in a specific embodiment of the present application, a method for detecting medical images is proposed, including:

Step 201, image acquisition and image preprocessing;

Specifically, the chest of the patient may be scanned by a CT device, so as to obtain an image of the chest to be detected of the patient. The chest image to be detected is normalized, and the CT value of the chest image to be detected is truncated and normalized between -1000 and 1000. The obtained image values are distributed between 0 and 1, and the pixel size is 0.6 mm. It is re-interpolated in the three-dimensional direction to obtain the normalized chest image to be detected whose width, height and depth are W×H×D.

Step 202, lung region segmentation;

Specifically, the traditional area growth method or the model-based lung parenchyma segmentation method can be used to obtain the lung area segmentation results (pulmonary medical images). The training obtained the lung parenchyma segmentation model.

Step 203, bronchial segmentation;

Specifically, bronchi segmentation is performed within the lung segmentation region, and all bronchi are labeled as mask ₀ . Bronchial segmentation can use traditional regional growth methods, such as selecting seed points from thicker parts of the trachea to start growing according to the threshold area; or using a model-based method to segment the bronchus. The neural network model is trained to obtain a bronchial segmentation model.

Step 204, pulmonary artery blood vessel segmentation;

Specifically, arterial vessel segmentation is performed within the lung segmentation region, and all pulmonary arterial vessels are labeled as mask ₁ . Since there is no difference between arterial and venous vessels in plain CT scan, the traditional pulmonary artery segmentation method is not effective. Model-based methods can be used to segment arterial vessels. Arterial vessels need to be manually marked according to requirements, and then an arterial vessel segmentation model is obtained through neural network model training. However, since there is little difference between arteries and veins, it is enough to separate the blood vessels.

Step 205, bronchial classification;

Specifically, according to the bronchial segmentation result mask ₀ , use the centerline extraction method to extract the bronchial centerline line0 _n , and calculate the connected domain according to the center line to divide the bronchus into different levels on the mask ₀ to obtain the bronchial classification L ₁ , L ₂ ...... L _n .

Step 206, searching for bronchial accompanying arteries;

Specifically, calculate the distance field from all points in the lung parenchyma area to the bronchial centerline to obtain a map; extract the arterial blood vessel centerline line1 on the pulmonary artery segmentation result mask ₁ ; extract different levels of bronchial segments line1 _n from the bronchial classification, each The center line of a bronchial segment includes the starting point P ₀ , the end point P _w and the first sampling points P ₁ , P ₂ ...... P _(w-1) at equal intervals. The accompanying points on line1 with the shortest distance from line1 are denoted as A ₀ , A ₁ ......A _(w-1) ; sometimes points A ₀ , A ₁ ...... A _(w-1) may not be in the centerline of the same segment of arterial vessels In the above, the centerline of the most voted segment can be selected as the centerline line1 _n of the accompanying arterial vessel segment (target arterial vessel segment) of the bronchial segment according to the multi-point voting method, so that the mask ₁ can be searched according to different bronchial segments on mask ₀ . The nearest concomitant arterial vessel segment.

Step 207, bronchial classification;

Specifically, the model-based bronchial classification method obtains the following information according to the bronchial segment line0 _n and the accompanying arterial vessel segment line1 _n in the lung sample image:

(1) Straighten the original image of the bronchus segment line0 _n according to the centerline of the segment, and sample and intercept the three-dimensional small block image I0 _n at equal intervals;

(2) the same operation intercepts the three-dimensional small block mask M0 _n of the bronchial segment line0 _n on the mask ₀ ;

(3) extract corresponding three-dimensional image I0 _(n-1) and three-dimensional mask M0 _(n-1 ) on the superior bronchial segment line0 _(n-1) of this bronchial segment line0 _n by the same operation;

(4) Extract the corresponding three-dimensional image I1 _n and three-dimensional mask M1 _n by the same operation on the arterial vessel segment line1 _n accompanying the bronchial segment line0 _n ;

The collected image blocks I0 _n , M0 _n , I0 _(n-1) , M0 _(n-1) , I1 _n , M1 _n are merged together as the input of the neural network, according to the results marked by the doctor (0 is normal bronchus, 1 is dilated bronchus) as a label to train a three-dimensional neural network model (columnar sign detection model), the model structure and parameter design have no special requirements;

During the detection of columnar signs, I0 _n , M0 _n , I0 _(n-1) , M0 _(n-1 related to the bronchial segment line0 _n and the accompanying arterial vessel segment line1 _n in the lung medical images were obtained in the same way. ₎ , I1 _n , M1 _n , and input the columnar sign detection model. According to the bronchial segment information in mask ₀ and the accompanying arterial blood vessel information in mask ₁ , the deep learning model is used to distinguish whether there is columnar dilation of the bronchus.

Step 208, bronchial information output;

Specifically, the bronchial segment marked as 1 after the classification in step 207 is marked with different values on the mask ₀ , and the center point of the bronchial segment and the calculated bronchial inner diameter are recorded as the output, so as to calculate the dilated bronchus and its position information. Measurements and statistics for each expansion are output.

In this embodiment, the bronchial feature model is established by using big data training through deep learning, and the bronchial image information, accompanying arterial blood vessel information, and superior bronchial image information are all used as the basis for the network to discriminate, not only the clinical lung CT image The normal bronchi in the data can be distinguished from the columnar dilated bronchi, which can improve the diagnostic accuracy of columnar signs and reduce the workload of doctors.

Further, as a specific implementation of the above method for detecting medical images, an embodiment of the present application provides a device for detecting medical images. As shown in FIG. 3 , the device for detecting medical images includes: an acquisition module, an extraction module, and a determination module and detection modules.

Among them, the acquisition module is used to acquire lung medical images; the extraction module is used to extract the arterial blood vessels and bronchi in the lung medical images, and the bronchi includes multiple bronchial segments of different levels; the determination module is used to determine the center of the bronchial segments according to the The distance between the line and the center line of the arterial vessel determines the target arterial vessel segment corresponding to the bronchial segment in the lung medical image; the detection module is used to detect and process the bronchial segment and the target arterial vessel segment to determine the detection result of the bronchial segment .

Further, the medical image detection device further includes: an interception module, which, according to a preset interval, analyzes at least one of the second bronchial segment, the target arterial vessel segment corresponding to the first bronchial segment, and the first bronchial segment in the lung medical image. The interception process is performed to obtain a first image, and the second bronchial segment is the bronchial segment of the previous level of the first bronchial segment; the detection module is specifically used to input the first image and/or the mask of the first image into the columnar sign detection The model is used to determine the detection result of the first bronchial segment; wherein, the columnar sign detection model is obtained by training based on the lung sample image and the columnar sign label corresponding to the lung sample image.

Further, the acquisition module is also used to acquire lung sample images, and the lung sample images include multiple bronchial segment samples of different levels and samples of target arterial blood vessel segments corresponding to multiple bronchial segment samples of different levels; detection of medical images; The apparatus further includes: a marking module for when the diameter of the first bronchial segment sample is greater than the diameter of the corresponding target arterial vessel segment sample, or the difference between the diameter of the first bronchial segment sample and the diameter of the second bronchial segment sample is greater than In the case of a preset difference, a columnar sign label of the first bronchial segment sample is generated, and the columnar sign label is used to mark the presence of a columnar sign in the bronchus; the interception module is also used for the second bronchial segment sample marked with the columnar sign label, the first At least one of the target arterial blood vessel segment samples corresponding to a bronchial segment sample, and the first bronchial segment sample are intercepted to obtain a second image; the medical image detection device further includes: a training module, configured according to the second image and/or Or the mask of the second image, train the neural network model, and obtain the columnar sign detection model.

Further, the determination module is specifically used to determine the first sampling point on the centerline of the bronchial segment; the distances between the first sampling point and a plurality of second sampling points on the centerline of the arterial blood vessel are calculated respectively; the determination module, specifically It is used to determine the second sampling point corresponding to the smallest distance as the accompanying point corresponding to the first sampling point; according to the starting point coordinates and end point coordinates of the bronchial segment, determine at least one arterial vessel segment corresponding to the bronchial segment; according to the position of the accompanying point , and select a target arterial vessel segment from at least one arterial vessel segment.

Further, the extraction module is specifically used for inputting the pulmonary medical image into the arterial blood vessel segmentation model to obtain the arterial blood vessels in the pulmonary medical image; inputting the pulmonary medical image into the bronchial segmentation model to obtain the bronchus in the pulmonary medical image; wherein , the arterial vessel segmentation model is trained based on the lung sample images and their corresponding arterial vessel segmentation labels, and the lung segmentation model is trained based on the lung sample images and their corresponding lung segmentation labels.

Further, the medical image detection device further includes: a display module, configured to display the bronchial segments with columnar signs in the lung medical image differentially when the detection result is that there are columnar signs.

Further, the medical image detection apparatus further includes: a storage module, configured to associate and store the center point and the inner diameter of the bronchial segment where the columnar sign exists when the detection result is that the columnar sign exists.

Further, the acquisition module is specifically used to acquire the medical image to be detected; the medical image detection device further includes: a preprocessing module, used to preprocess the medical image to be detected; a segmentation module, used to preprocess the medical image to be detected. The image is subjected to lung segmentation processing to obtain a lung medical image; wherein, the preprocessing includes at least one of the following: noise reduction processing, normalization processing, size correction processing, CT value truncation processing, and resampling processing.

In this embodiment, each module of the medical image detection apparatus implements the steps of the medical image detection method in any embodiment of the first aspect when each module performs its own function. Therefore, the medical image detection apparatus also includes the first aspect. All the beneficial effects of the medical image detection method in any of the embodiments will not be repeated here.

It should be noted that, for other corresponding descriptions of the functional modules involved in the medical image detection apparatus provided in the embodiments of the present application, reference may be made to the corresponding descriptions in FIG. 1 and FIG. 2 , which will not be repeated here.

In a specific application scenario, the medical image detection device in the embodiment of the present application may be a device, or may be a component, an integrated circuit, or a chip in a terminal. The apparatus may be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device may be a mobile phone, a tablet computer, a notebook computer, a palmtop computer, an in-vehicle electronic device, a smart camera device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant. (personal digital assistant, PDA), etc., the non-mobile electronic device may be a server, a network attached storage (Network Attached Storage, NAS), or a personal computer (personal computer, PC), etc., which are not specifically limited in the embodiments of the present application.

According to yet another aspect of the present application, a storage medium is provided on which a computer program is stored, and when the program is executed by a processor, the foregoing method for detecting a medical image is implemented.

According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, and the processor implements the above-mentioned detection method for medical images when the computer program is executed. .

Based on the above methods shown in FIGS. 1 and 2 , correspondingly, an embodiment of the present application further provides a storage medium on which a computer program is stored, and when the program is executed by a processor, implements the above method for detecting medical images.

Based on this understanding, the technical solution of the present application can be embodied in the form of a software product or a hardware product or a combination of software and hardware, and the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, A mobile hard disk, etc.), including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method of each implementation scenario of the present application.

Based on the medical image detection method shown in FIG. 1 and the embodiment of the medical image detection apparatus shown in FIG. 3 , in order to achieve the above purpose, the embodiment of the present application also provides a computer device, which may be a personal computer. , a server, a network device, etc., the computer device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the above method for detecting medical images.

Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (Radio Frequency, RF) circuit, a sensor, an audio circuit, a WI-FI module, and the like. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like. Optional network interfaces may include standard wired interfaces, wireless interfaces (such as Bluetooth interfaces, WI-FI interfaces), and the like.

Those skilled in the art can understand that the structure of a computer device provided in this embodiment does not constitute a limitation on the computer device, and may include more or less components, or combine some components, or arrange different components.

The storage medium may also include an operating system and a network communication module. An operating system is a program that manages and saves the hardware and software resources of computer equipment, supports the operation of information processing programs and other software and/or programs. The network communication module is used to realize the communication between various controls in the storage medium, as well as the communication with other hardware and software in the physical device.

From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, and can also be implemented by hardware.

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred implementation scenario, and the units or processes in the accompanying drawing are not necessarily necessary to implement the present application. Those skilled in the art can understand that the units in the device in the implementation scenario may be distributed in the device in the implementation scenario according to the description of the implementation scenario, or may be located in one or more devices different from the implementation scenario with corresponding changes. The units of the above implementation scenarios may be combined into one unit, or may be further split into multiple subunits.

The above serial numbers in this application are only for description, and do not represent the pros and cons of the implementation scenarios. The above disclosures are only a few specific implementation scenarios of the present application, however, the present application is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present application.

Claims (10)

1. a detection method of medical image, is characterized in that, described method comprises: Obtaining medical images of the lungs; extracting arterial blood vessels and bronchi in the lung medical image, the bronchi includes a plurality of bronchial segments of different levels; determining the target arterial vessel segment corresponding to the bronchial segment in the lung medical image according to the distance between the centerline of the bronchial segment and the centerline of the arterial vessel; The detection process is performed on the bronchial segment and the target arterial vessel segment to determine the detection result of the bronchial segment. 2 . The method for detecting medical images according to claim 1 , wherein the detecting processing of the bronchial segment and the target arterial vessel segment to determine the detection result of the bronchial segment specifically comprises: 2 . According to a preset interval, at least one of the second bronchial segment, the target arterial vessel segment corresponding to the first bronchial segment, and the first bronchial segment in the pulmonary medical image are intercepted to obtain a first image , the second bronchial segment is the bronchial segment of the previous level of the first bronchial segment; Inputting the first image and/or the mask of the first image into a columnar sign detection model to determine the detection result of the first bronchial segment; Wherein, the columnar sign detection model is obtained by training based on the lung sample image and the columnar sign label corresponding to the lung sample image. 3. The method for detecting medical images according to claim 2, wherein the method further comprises: acquiring the lung sample image, where the lung sample image includes a plurality of bronchial segment samples of different levels and a target arterial vessel segment sample corresponding to the plurality of bronchial segment samples of different levels; When the diameter of the first bronchial segment sample is greater than the diameter of the corresponding target arterial vessel segment sample, or the difference between the diameter of the first bronchial segment sample and the diameter of the second bronchial segment sample is greater than a preset difference , generating a columnar sign label of the first bronchial segment sample, where the columnar sign label is used to mark the presence of a columnar sign in the bronchus; Perform interception processing on at least one of the second bronchial segment sample marked with the columnar sign label, the target arterial blood vessel segment sample corresponding to the first bronchial segment sample, and the first bronchial segment sample to obtain the first bronchial segment sample. two images; According to the second image and/or the mask of the second image, a neural network model is trained to obtain the columnar sign detection model. 4 . The method for detecting medical images according to claim 1 , wherein, according to the distance between the center line of the bronchial segment and the center line of the arterial blood vessel, determining the pulmonary medical image in the lung medical image. 5 . The target arterial vessel segment corresponding to the bronchial segment specifically includes: determining a first sampling point on the centerline of the bronchial segment; respectively calculating the distances between the first sampling point and a plurality of second sampling points on the centerline of the arterial vessel; Determining the second sampling point corresponding to the smallest distance as the accompanying point corresponding to the first sampling point; Determine at least one arterial vessel segment corresponding to the bronchial segment according to the starting point coordinate and the ending point coordinate of the bronchial segment; A target arterial vessel segment is selected from the at least one arterial vessel segment based on the location of the companion point. 5. The method for detecting a medical image according to any one of claims 1 to 4, wherein the extracting the arterial blood vessels and bronchi in the pulmonary medical image specifically comprises: Inputting the pulmonary medical image into an arterial blood vessel segmentation model to obtain the arterial blood vessels in the pulmonary medical image; Inputting the pulmonary medical image into a bronchial segmentation model to obtain the bronchus in the pulmonary medical image; Wherein, the arterial blood vessel segmentation model is obtained by training based on the lung sample images and their corresponding arterial blood vessel segmentation labels, and the lung segmentation model is obtained by training based on the lung sample images and their corresponding lung segmentation labels. 6. The method for detecting medical images according to any one of claims 1 to 4, wherein the method further comprises: In the case that the detection result is that there are columnar signs, differentially display the bronchial segments with the columnar signs in the lung medical image, and/or associate and store the center point and the center point of the bronchial segments with the columnar signs. the inside diameter of. 7. The method for detecting medical images according to any one of claims 1 to 4, wherein the acquiring a medical image of the lungs specifically comprises: Obtain medical images to be tested; Preprocessing the medical image to be detected; performing lung segmentation processing on the preprocessed medical image to be detected to obtain the lung medical image; The preprocessing includes at least one of the following: noise reduction processing, normalization processing, size correction processing, CT value truncation processing, and resampling processing. 8. A device for detecting medical images, wherein the device comprises: an acquisition module for acquiring lung medical images; an extraction module for extracting arterial vessels and bronchi in the pulmonary medical image, the bronchi includes a plurality of bronchial segments of different levels; a determining module, configured to determine the target arterial vessel segment corresponding to the bronchial segment in the pulmonary medical image according to the distance between the centerline of the bronchial segment and the centerline of the arterial vessel; The detection module is configured to perform detection processing on the bronchial segment and the target arterial vessel segment, and determine the detection result of the bronchial segment. 9 . A storage medium on which a computer program is stored, characterized in that, when the computer program is executed by a processor, the method for detecting a medical image according to any one of claims 1 to 7 is implemented. 10 . 10. A computer device, comprising a storage medium, a processor and a computer program stored on the storage medium and running on the processor, wherein the processor implements claims 1 to 7 when executing the computer program The detection method of any one of the medical images.

Images