Strain analysis method and system based on heart CT data

Abstract

The invention discloses a strain analysis method and system based on heart CT data. The method comprises the following steps: segmenting an original heart CT image through an image segmentation network structure to obtain a segmentation result; inputting the segmentation result into a pre-trained key point detection model to obtain a plurality of key points; determining a ventricular long axis and a ventricular short axis according to the plurality of key points, and obtaining a segmentation result; registering the heart mask image by using an image registration framework to obtain a deformation field of myocardial movement from the end stage of myocardial diastole to the end stage of myocardial systole; and performing strain analysis according to the segmentation result and the deformation field, and calculating to obtain the overall radial strain, the overall longitudinal strain and the overall circumferential strain of the heart. The myocardial strain analysis is realized through the myocardial strain analysis framework, different segments of the myocardial can be analyzed, various parameter indexes are obtained, and efficient processing and accurate analysis of heart CT data are realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing, and in particular relates to a strain analysis method and system based on cardiac CT data. Background Art

[0002] Existing processing methods for cardiac CT data mainly include the following categories: (1) Harmonic Phase (HARP) analysis method: HARP is a phase-based analysis method that is obtained by using a special imaging technique called tagging MRI in cardiac magnetic resonance imaging. In this imaging technique, a series of marker lines or grids are applied to the surface or inside of the heart, and these markers deform during cardiac contraction and relaxation. HARP analysis is to evaluate the local motion and strain of the myocardium by tracking the changes of these markers. (2) Feature Tracking (FT) analysis method: Feature Tracking is an image analysis-based technology. It analyzes the motion and strain of the myocardium by automatically tracking the myocardial boundaries in cardiac MRI and CT images in continuous image frames. (3) Elastix registration framework: Elastix is ​​an open source image registration toolkit that is specifically used for image registration tasks in medical image processing. Image registration is a key step in the field of medical imaging, which involves spatially aligning two or more images to facilitate comparison or fusion of these images. Elastix provides a range of algorithms and tools for performing different types of image registration, including rigid, affine, and non-rigid.

[0003] However, the above technologies have some shortcomings in practical application. For example, the HARP analysis method takes a long time to collect and analyze in clinical applications, requires professional equipment and well-trained operators, and has not yet been applied in clinical practice. The Feature Tracking analysis method requires SAX, 2CH and 4CH views of the myocardium to work properly. This step requires doctors to mark on software such as Medis, which has high repetitiveness and poor results for local motion analysis of the myocardium. Summary of the invention

[0004] In view of the shortcomings of the prior art, the strain analysis method based on cardiac CT data proposed in the present invention includes three stages: myocardial segmentation, myocardial segmentation and registration and strain analysis. In the first stage, the cardiac CT data set is segmented by nnU-net. In the second stage, the segmented mask is input into 3D-Unet, the mitral valve center, apex and tricuspid valve center are automatically detected, the long axis and short axis of the left ventricle are determined according to the three key points, and the myocardium is divided into 17 segments based on this. In the third stage, the elastix framework is used for non-rigid registration of myocardial end-systolic and end-diastolic images. The myocardial strain tensor is calculated by the deformation field obtained by registration, and it is projected onto the radial, longitudinal and circumferential axes to obtain strain indicators such as GRS, GLS and GCS. Only the original CT image needs to be input, which realizes efficient processing and accurate analysis of cardiac CT data, and provides strong support for the diagnosis and treatment of heart diseases.

[0005] In order to achieve the aforementioned object of the invention, the present invention adopts the following scheme:

[0006] One aspect of the present invention provides a strain analysis method based on cardiac CT data, comprising the following steps:

[0007] S1. Segmenting the original cardiac CT image through an image segmentation network structure to obtain a segmentation result, wherein the segmentation result includes a segmentation mask image;

[0008] S2, inputting the segmentation result into a pre-trained key point detection model to obtain a plurality of key points; determining the ventricular long axis and the ventricular short axis according to the plurality of key points; dividing the heart into segments according to the ventricular long axis, the ventricular short axis, and the angles and proportions occupied by each segment of the heart to obtain a segmentation result;

[0009] S3. Use the image registration framework to perform registration of the cardiac mask image to obtain the deformation field of the myocardial movement from the end of myocardial diastole to the end of myocardial systole; perform strain analysis based on the segmentation results and the deformation field to obtain the strain components in the radial, longitudinal and circumferential directions, and calculate the overall radial strain, overall longitudinal strain and overall circumferential strain of the heart.

[0010] Another aspect of the present invention provides a strain analysis system based on cardiac CT data, comprising:

[0011] A myocardial segmentation module, used to segment the original cardiac CT image through an image segmentation network structure to obtain a segmentation result, wherein the segmentation result includes a segmentation mask image;

[0012] The myocardial segmentation module is used to input the segmentation result into a pre-trained key point detection model to obtain multiple key points; determine the ventricular long axis and the ventricular short axis according to the multiple key points; divide the heart into segments according to the ventricular long axis, the ventricular short axis, and the angles and proportions occupied by each segment of the heart to obtain a segmentation result;

[0013] The registration and strain analysis module is used to use the image registration framework to perform registration of the cardiac mask image to obtain the deformation field of the myocardial movement from the end of myocardial diastole to the end of myocardial systole; perform strain analysis based on the segmentation results and the deformation field to obtain the strain components in the radial, longitudinal and circumferential directions, and calculate the overall radial strain, overall longitudinal strain and overall circumferential strain of the heart.

[0014] Compared with the prior art, the present invention has at least the following advantages: (1) Automated analysis framework: Compared with other methods that require equipment and manual annotation, the present invention proposes an automated myocardial strain analysis framework to improve the efficiency and accuracy of the analysis. (2) Segmental analysis: Compared with the prior art, the present invention can accurately analyze different segments of the myocardium, which helps to more accurately evaluate the local function and lesions of the myocardium in combination with the blood vessels at different segments. This segmental analysis is particularly important for the early detection and location of heart disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a flowchart of a strain analysis method based on cardiac CT data provided by a typical implementation case of the present invention;

[0017] Figure 2 It is a schematic diagram of a local cardiac muscle coordinate system provided by a typical implementation case of the present invention;

[0018] Figure 3 This is a myocardial segmentation model effect diagram provided by a typical implementation case of the present invention;

[0019] Figure 4 A diagram of left ventricular volume changes during the entire myocardial motion cycle provided by a typical embodiment of the present invention;

[0020] Figure 5 is a schematic diagram of myocardial segmentation provided by a typical embodiment of the present invention;

[0021] Figure 6This is a myocardial registration effect diagram provided by a typical embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present invention clearer, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Examples of these preferred embodiments are illustrated in the accompanying drawings. The embodiments of the present invention shown in the accompanying drawings and described according to the accompanying drawings are merely exemplary, and the present invention is not limited to these embodiments.

[0023] In the report diagnosis stage, the various indicators obtained by myocardial strain analysis are very important, such as the volume of the heart, the overall longitudinal strain (the degree of deformation of the heart along its long axis during contraction, GLS), the overall radial strain (the degree of deformation of the heart along its radial direction from the center of the heart cavity to the ventricular wall during contraction, GRS) and the overall circumferential strain (the degree of deformation of the heart along the direction around the ventricular wall during contraction, GCS). These parameters can help doctors evaluate the overall and local functions of the heart, including the blood pumping capacity and the contractile capacity of the myocardium; detect heart diseases such as myocardial infarction and cardiomyopathy. By comprehensively analyzing these parameters and the original CT images, doctors can have a more comprehensive understanding of the patient's heart condition and formulate personalized treatment plans. Based on this, the present invention realizes strain analysis of the myocardium through an automated myocardial strain analysis framework, and can analyze different segments of the myocardium to obtain various parameter indicators to help doctors make judgments.

[0024] One aspect of the present invention provides a strain analysis method based on cardiac CT data, comprising the following steps:

[0025] S1. Segmenting the original cardiac CT image through an image segmentation network structure to obtain a segmentation result, wherein the segmentation result includes a segmentation mask image;

[0026] S2, inputting the segmentation result into a pre-trained key point detection model to obtain a plurality of key points; determining the ventricular long axis and the ventricular short axis according to the plurality of key points; dividing the heart into segments according to the ventricular long axis, the ventricular short axis, and the angles and proportions occupied by each segment of the heart to obtain a segmentation result;

[0027] S3. Use the image registration framework to perform registration of the cardiac mask image to obtain the deformation field of the myocardial movement from the end of myocardial diastole to the end of myocardial systole; perform strain analysis based on the segmentation results and the deformation field to obtain the strain components in the radial, longitudinal and circumferential directions, and calculate the overall radial strain, overall longitudinal strain and overall circumferential strain of the heart.

[0028] In one embodiment, step S2 specifically includes:

[0029] Input the segmented mask image into a pre-trained key point detection model to obtain the coordinates of multiple key points; the multiple key points include: the center point of the mitral valve, the apex, and the center point of the tricuspid valve; take the line connecting the center point of the mitral valve and the apex as the long axis of the ventricle, and take the line connecting the center point of the mitral valve and the center point of the tricuspid valve as the short axis of the ventricle;

[0030] A three-dimensional Cartesian coordinate system is established to transform the coordinates of the heart. First, a transformation matrix is ​​used to align the axis of the myocardial mesh with the z-axis, so that the apex is at 0 on the z-axis, the center of the mitral valve is above the z-axis, and the long axis is the z-axis. Another transformation matrix is ​​used to rotate the myocardial mesh so that the x-axis is aligned parallel to the line connecting the center of the mitral valve and the center of the tricuspid valve. The y-axis is perpendicular to the x- and z-axes. Finally, the heart is segmented according to the range and angle range of the z-axis occupied by each segment.

[0031] Specifically, the Z-axis range and angle range of each segment are as shown in Table 1: Table 1

[0032]

[0033]

[0034] In one embodiment, the step S3 uses an image registration framework to perform registration of the heart mask image, including:

[0035] The segmented mask image at the end of myocardial contraction is used as a fixed image in the registration process, and the mask image at the end of myocardial diastole is used as a moving image. B-spline non-rigid registration is used to obtain a strain field; the strain field defines the transformation of each point in the fixed image to the corresponding position in the moving image.

[0036] In one embodiment, the strain analysis in step S3 includes:

[0037] The myocardial strain tensor is estimated directly from the spatial derivative of the strain field as follows:

[0038]

[0039] in, represents the spatial derivative, I represents the unit tensor;

[0040] For each point in the myocardium, a three-dimensional Cartesian coordinate system is established, and the three axes correspond to the radial, longitudinal and circumferential axes respectively. The strain components are projected onto the three axes, and the calculation formula is as follows:

[0041] εp(x)=p T ε(x)p

[0042] Where p represents the radial, longitudinal or circumferential direction.

[0043] In one embodiment, the calculation process of the global radial strain, the global longitudinal strain and the global circumferential strain of the heart is:

[0044]

[0045] Among them, S represents the strain in three directions, L ed and L es They represent the projection lengths of the myocardial strain components at the end of diastole and end of systole on the three axes respectively; for each segment, points are sampled at equal intervals to calculate the average strain, and finally the overall radial strain, overall longitudinal strain and overall circumferential strain are obtained.

[0046] Another aspect of the present invention provides a strain analysis system based on cardiac CT data, characterized in that it includes:

[0047] A myocardial segmentation module, used to segment the original cardiac CT image through an image segmentation network structure to obtain a segmentation result, wherein the segmentation result includes a segmentation mask image;

[0048] The myocardial segmentation module is used to input the segmentation result into a pre-trained key point detection model to obtain multiple key points; determine the ventricular long axis and the ventricular short axis according to the multiple key points; divide the heart into segments according to the ventricular long axis, the ventricular short axis, and the angles and proportions occupied by each segment of the heart to obtain a segmentation result;

[0049] The registration and strain analysis module is used to use the image registration framework to perform registration of the cardiac mask image to obtain the deformation field of the myocardial movement from the end of myocardial diastole to the end of myocardial systole; perform strain analysis based on the segmentation results and the deformation field to obtain the strain components in the radial, longitudinal and circumferential directions, and calculate the overall radial strain, overall longitudinal strain and overall circumferential strain of the heart.

[0050] In one embodiment, the myocardial segmentation module is specifically used to:

[0051] Input the segmented mask image into a pre-trained key point detection model to obtain the coordinates of multiple key points; the multiple key points include: the center point of the mitral valve, the apex, and the center point of the tricuspid valve; take the line connecting the center point of the mitral valve and the apex as the long axis of the ventricle, and take the line connecting the center point of the mitral valve and the center point of the tricuspid valve as the short axis of the ventricle;

[0052] A three-dimensional Cartesian coordinate system is established to transform the coordinates of the heart. First, a transformation matrix is ​​used to align the axis of the myocardial mesh with the Z axis, so that the apex point is at 0 on the Z axis, the mitral valve center is above the Z axis, and the long axis is the Z axis; another transformation matrix is ​​used to rotate the myocardial mesh so that the X axis is parallel to the line connecting the mitral valve center and the tricuspid valve center; the Y axis is perpendicular to the X and Z axes; finally, the heart segments are divided according to the range and angle range of the Z axis occupied by each segment.

[0053] In one embodiment, the registration and strain analysis module uses an image registration framework to perform registration of a cardiac mask image, including:

[0054] The segmented mask image at the end of myocardial contraction is used as a fixed image in the registration process, and the mask image at the end of myocardial diastole is used as a moving image. B-spline non-rigid registration is used to obtain a strain field; the strain field defines the transformation of each point in the fixed image to the corresponding position in the moving image.

[0055] In one embodiment, the strain analysis in the registration and strain analysis module includes:

[0056] The myocardial strain tensor is estimated directly from the spatial derivative of the strain field as follows:

[0057]

[0058] in, represents the spatial derivative, I represents the unit tensor;

[0059] For each point in the myocardium, a three-dimensional Cartesian coordinate system is established, and the three axes correspond to the radial, longitudinal and circumferential axes respectively. The strain components are projected onto the three axes, and the calculation formula is as follows:

[0060] ε p (x) = p T ε(x)p

[0061] Where p represents the radial, longitudinal or circumferential direction.

[0062] In one embodiment, the calculation process of the global radial strain, the global longitudinal strain and the global circumferential strain of the heart is:

[0063]

[0064] Among them, S represents the strain in three directions, L ed and L esThey represent the projection lengths of the myocardial strain components at the end of diastole and end of systole on the three axes respectively; for each segment, points are sampled at equal intervals to calculate the average strain, and finally the overall radial strain, overall longitudinal strain and overall circumferential strain are obtained.

[0065] like Figure 1 As shown, the present invention consists of three stages: myocardial segmentation, myocardial segmentation and myocardial registration. In the first stage, the myocardium is segmented by nnU-net. In the second stage, the segmented myocardium is input into the key point detection network Unet, and three key points (mitral valve center, apex, and tricuspid valve center) are found. The long axis is determined by the mitral valve center and the apex, and the short axis is determined by the mitral valve center and the tricuspid valve center, thereby realizing the division of the myocardium into 17 segments. In the third stage, the elastix registration framework is used to register the myocardial left ventricular mask, and the deformation field of the myocardial movement from the end of myocardial diastole (ED) to the end of myocardial systole (ES) is obtained. By analyzing its Lagrangian deformation, the deformation components in the radial, longitudinal and circumferential directions are obtained, and then the three indicators GRS, GLS and GCS are calculated. Figure 1 In the figure, the orange box illustrates the process of myocardial segmentation; the green box illustrates the process of myocardial segmentation; and the blue box illustrates the process of registration to obtain deformation field and strain analysis.

[0066] In a specific embodiment, the myocardial segmentation process specifically includes: first pre-training on the public dataset MM-WHS, MM-WHS provides 20 labeled and 40 unlabeled CT images, which are input into nnU-net as training sample pairs, and then the predicted label (3D) of the left ventricular myocardium is output, and then the pre-trained model is applied to its own dataset, and a smaller learning rate is used to update the model weights to optimize the model performance.

[0067] In a specific embodiment, the myocardial segmentation process specifically includes: using a unified standard left ventricular section to divide the left ventricular myocardium into 17 segments. The 17-segment division method is more in line with the anatomical structure of the heart. According to the definition, the line connecting the center of the mitral valve and the apex is taken as the long axis of the left ventricle, the mitral valve to the starting point of the chordae tendineae attachment is called the base, the area between the starting point and the end point of the chordae tendineae attachment is called the middle, the end point of the chordae tendineae attachment to the apex is called the apex, and the apex without a heart cavity is called the apex. The above three parts account for approximately 35%, 35%, and 30% of the total volume of the left ventricle, respectively, and the line connecting the center of the mitral valve and the center of the tricuspid valve is taken as the short axis.

[0068] In order to achieve automatic segmentation, a key point detection model based on 3D-Unet was trained. The model input is the mask image obtained after segmentation, and the output is the coordinates of 3 key points. After the long axis and short axis are obtained, a three-dimensional Cartesian coordinate system is established to transform the coordinates of the myocardium. The long axis is the z-axis. Then, the 17 segments are divided according to the angles occupied by each segment and the range of the z-axis.

[0069] In a specific embodiment, the registration and strain analysis process specifically includes: registration is to spatially align the images of the end-systolic phase of the myocardium and the end-diastolic phase of the myocardium. For the data obtained in the first step of segmentation, only the left ventricular myocardium is considered, and the other parts are set to 0, which can speed up the registration speed. The mask image at the end of systole is used as a fixed image in the elastix registration process, and the mask image at the end of diastole is used as a moving image, and B-spline non-rigid registration is used. After registration, a deformation field can be obtained, which defines the transformation of each point in the fixed image to the corresponding position in the moving image. After the image registration with the original input size of (512, 512, 224), a deformation field of size (512, 512, 224, 3) can be obtained, and 3 represents the displacement of each point in the x, y, and z directions.

[0070] During myocardial strain analysis, the myocardial strain tensor is directly estimated from the spatial derivative of the deformation field as follows:

[0071]

[0072] in represents the spatial derivative and I represents the unit tensor.

[0073] Strain measurements are dependent on the chosen coordinate system, as their values ​​are related to the coordinates of each myocardial point. Therefore, it is important to use a local myocardial coordinate system, such as Figure 2 shown. Figure 2 (a) shows the entire coordinate system. Figure 2 (b) and (c) represent the circumferential and longitudinal components, respectively. For each point in the myocardium, a three-dimensional Cartesian coordinate system is established, with the three axes corresponding to the radial, longitudinal, and circumferential axes, respectively. Therefore, the previously calculated myocardial strain components can be projected onto the three axes, as defined below:

[0074]

[0075] Where p represents the radial, longitudinal or circumferential direction.

[0076] After obtaining the projected length, the myocardial strain can be calculated:

[0077]

[0078] Among them, S represents strain in three directions, L ed and L es They represent the projection lengths of the myocardial strain components at the end of diastole and end of systole on the three axes respectively. Finally, for each segment, points are sampled at equal intervals to calculate the average strain, and finally the overall GLS, GRS and GCS can be obtained.

[0079] In the specific implementation process of the strain analysis system based on cardiac CT data in this application, the data set uses cardiac CT data provided by the Fuwai Hospital of the Chinese Academy of Medical Sciences, with a total of 33 records.

[0080] For the segmentation part, 20 of them are used as training sets and the other 13 as test sets, and nnU-net is used for fine-tuning training.

[0081] For the myocardial segmentation, 20 cardiac CT data and three manually calibrated key points are also used as training sets to input into 3D-Unet, and the other 13 are used as test sets. The mean square error is used as the loss function. Because the original CT image has 512*512*224 points, and the output has only three points, this will cause the loss to be very small at the beginning and difficult to converge. Therefore, the output of the model is changed to three 5*5*5 matrices, and the points in the matrix are the candidate points of the three key points. When calculating the loss, the loss of the three matrices is calculated, and finally the one with the largest probability is selected in the matrix as the result. A total of 200 epochs are trained, the batch size is 4, and the Adam optimizer is used. The initial value of the learning rate is 0.001, and the learning rate decays at an exponential rate of 0.99 in each epoch.

[0082] For the myocardial registration part, elastix was used for non-rigid registration, using a recursive pyramid structure and combining multiple transformations, including rigid transformation, affine transformation and B-spline transformation. The maximum number of iterations was 500 and the B-spline interpolation order was 1.

[0083] The visualization results of the above implementation process are shown as follows: Figure 3 As shown, the left ventricle and the right ventricle can be segmented. With the segmented mask image, combined with the voxel size information of the original CT image, the volume of the left ventricle can be obtained. Figure 4 The left ventricular volume change diagram of the whole myocardial motion cycle is shown. The key points obtained by the key point model establish the coordinate axis and divide it into 17 segments, such as Figure 5 As shown. The myocardium is non-rigidly registered to obtain the deformation field, as shown in Figure 6 shown.

[0084] The comparison table of strain analysis results is shown in Table 2: Table 2

[0085]

[0086] It should be understood that although this specification is described according to implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions of each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A strain analysis method based on cardiac CT data, It is characterized in that The following steps are involved: S1. Segmenting the original cardiac CT image through an image segmentation network structure to obtain a segmentation result, wherein the segmentation result includes a segmentation mask image; S2, inputting the segmentation result into a pre-trained key point detection model to obtain a plurality of key points; Determine the ventricular long axis and the ventricular short axis according to the multiple key points; Divide the heart into segments according to the ventricular long axis, ventricular short axis, angles and proportions occupied by each segment of the heart to obtain a segmentation result; S3. Use the image registration framework to perform registration of the cardiac mask image to obtain the deformation field of the myocardial movement from the end of myocardial diastole to the end of myocardial systole; perform strain analysis based on the segmentation results and the deformation field to obtain the strain components in the radial, longitudinal and circumferential directions, and calculate the overall radial strain, overall longitudinal strain and overall circumferential strain of the heart.

2. The strain analysis method based on cardiac CT data according to claim 1, It is characterized in that The step S2 specifically includes: Input the segmented mask image into a pre-trained key point detection model to obtain the coordinates of multiple key points; the multiple key points include: the center point of the mitral valve, the apex, and the center point of the tricuspid valve; take the line connecting the center point of the mitral valve and the apex as the long axis of the ventricle, and take the line connecting the center point of the mitral valve and the center point of the tricuspid valve as the short axis of the ventricle; A three-dimensional Cartesian coordinate system is established to transform the coordinates of the heart. First, a transformation matrix is ​​used to align the axis of the myocardial mesh with the Z axis, so that the apex is at 0 on the Z axis, the center of the mitral valve is above the Z axis, and the long axis is the Z axis. Another transformation matrix is ​​used to rotate the myocardial mesh so that the X axis is parallel to the line connecting the center of the mitral valve and the center of the tricuspid valve; the Y axis is perpendicular to the X and Z axes; finally, the heart segments are divided according to the range and angle range of the Z axis occupied by each segment.

3. The strain analysis method based on cardiac CT data according to claim 1, It is characterized in that In step S3, the heart mask image is registered using an image registration framework, including: The segmented mask image at the end of myocardial contraction is used as a fixed image in the registration process, and the mask image at the end of myocardial diastole is used as a moving image. B-spline non-rigid registration is used to obtain a strain field; the strain field defines the transformation of each point in the fixed image to the corresponding position in the moving image.

4. The strain analysis method based on cardiac CT data according to claim 3, It is characterized in that The strain analysis in step S3 includes: The myocardial strain tensor is estimated directly from the spatial derivative of the strain field as follows: in, represents the spatial derivative, I represents the unit tensor; For each point in the myocardium, a three-dimensional Cartesian coordinate system is established, and the three axes correspond to the radial, longitudinal and circumferential axes respectively. The strain components are projected onto the three axes, and the calculation formula is as follows: e p (x)=p T e(x)p Where p represents the radial, longitudinal or circumferential direction.

5. The strain analysis method based on cardiac CT data according to claim 4, It is characterized in that The calculation process of the global radial strain, global longitudinal strain and global circumferential strain of the heart is: Among them, S represents the strain in three directions, L ed and L es They represent the projection lengths of the myocardial strain components at the end of diastole and end of systole on the three axes respectively; for each segment, points are sampled at equal intervals to calculate the average strain, and finally the overall radial strain, overall longitudinal strain and overall circumferential strain are obtained.

6. A strain analysis system based on cardiac CT data, It is characterized in that include: A myocardial segmentation module, used to segment the original cardiac CT image through an image segmentation network structure to obtain a segmentation result, wherein the segmentation result includes a segmentation mask image; A myocardial segmentation module, used for inputting the segmentation result into a pre-trained key point detection model to obtain a plurality of key points; Determine the ventricular long axis and the ventricular short axis according to the multiple key points; Divide the heart into segments according to the ventricular long axis, ventricular short axis, angles and proportions occupied by each segment of the heart to obtain a segmentation result; The registration and strain analysis module is used to use the image registration framework to perform registration of the cardiac mask image to obtain the deformation field of the myocardial movement from the end of myocardial diastole to the end of myocardial systole; perform strain analysis based on the segmentation results and the deformation field to obtain the strain components in the radial, longitudinal and circumferential directions, and calculate the overall radial strain, overall longitudinal strain and overall circumferential strain of the heart.

7. The strain analysis system based on cardiac CT data according to claim 6, It is characterized in that The myocardial segmentation module is specifically used for: Input the segmented mask image into a pre-trained key point detection model to obtain the coordinates of multiple key points; the multiple key points include: the center point of the mitral valve, the apex, and the center point of the tricuspid valve; take the line connecting the center point of the mitral valve and the apex as the long axis of the ventricle, and take the line connecting the center point of the mitral valve and the center point of the tricuspid valve as the short axis of the ventricle; A three-dimensional Cartesian coordinate system is established to transform the coordinates of the heart. First, a transformation matrix is ​​used to align the axis of the myocardial mesh with the z-axis, so that the apex is at 0 on the z-axis, the center of the mitral valve is above the z-axis, and the long axis is the z-axis. Another transformation matrix is ​​used to rotate the myocardial mesh so that the x-axis is aligned parallel to the line connecting the center of the mitral valve and the center of the tricuspid valve. The y-axis is perpendicular to the x- and z-axes. Finally, the heart is segmented according to the range and angle range of the z-axis occupied by each segment.

8. The strain analysis system based on cardiac CT data according to claim 6, It is characterized in that The registration and strain analysis module uses an image registration framework to perform registration of a cardiac mask image, including: The segmented mask image at the end of myocardial contraction is used as a fixed image in the registration process, and the mask image at the end of myocardial diastole is used as a moving image. B-spline non-rigid registration is used to obtain a strain field; the strain field defines the transformation of each point in the fixed image to the corresponding position in the moving image.

9. The strain analysis system based on cardiac CT data according to claim 8, It is characterized in that The strain analysis in the registration and strain analysis module includes: The myocardial strain tensor is estimated directly from the spatial derivative of the strain field as follows: in, represents the spatial derivative, I represents the unit tensor; For each point in the myocardium, a three-dimensional Cartesian coordinate system is established, and the three axes correspond to the radial, longitudinal and circumferential axes respectively. The strain components are projected onto the three axes, and the calculation formula is as follows: e p (x)=p T e(x)p Where p represents the radial, longitudinal or circumferential direction.

10. The strain analysis system based on cardiac CT data according to claim 9, It is characterized in that The calculation process of the global radial strain, global longitudinal strain and global circumferential strain of the heart is: Among them, S represents the strain in three directions, L ed and L es They represent the projection lengths of the myocardial strain components at the end of diastole and end of systole on the three axes respectively; for each segment, points are sampled at equal intervals to calculate the average strain, and finally the overall radial strain, overall longitudinal strain and overall circumferential strain are obtained.

Images