Angiography image processing device based on cardiac computerized tomography

Abstract

The invention discloses an angiography image processing device based on cardiac computerized tomography, and relates to the technical field of image processing. In the processing device based on the heart electronic computed tomography angiography image, the electronic computed tomography angiography image is predicted based on a deep learning network model, and meanwhile, an aorta region and a coronary artery region in an artery region on the electronic computed tomography angiography image are distinguished; the artery region detection efficiency is improved; compared with a mode of directly distinguishing the aorta region and the coronary artery region according to a prediction result of a deep learning network model, in the processing device provided by the invention, after the initial segmentation result is output through the output module, the initial segmentation result is continuously obtained through the second acquisition module and the determination module; and the prediction result of the deep learning network model is processed again, so that the accuracy of the determined artery region on the current electronic computed tomography angiography image is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a processing device based on cardiac electronic computer tomography angiography images. Background Art

[0002] In order to detect cardiovascular diseases in users, related technologies examine the diseases based on cardiac computed tomography angiography (CTA) images.

[0003] Using traditional methods to segment coronary arteries based on cardiac CTA images requires first segmenting the aorta, then the trunk, then detail segmentation, and then removing venous false positives. This method cannot segment the aorta and coronary arteries at the same time, and the method is relatively complex and has low detection efficiency.

[0004] It can be seen that how to process CTA images to improve the efficiency of artery detection is a technical problem that people in this field need to solve urgently. Summary of the invention

[0005] The purpose of the present invention is to provide a processing device based on cardiac computed tomography angiography images to solve the technical problems that the method of segmenting coronary arteries based on cardiac CTA images using traditional methods cannot simultaneously complete the segmentation of the aorta and coronary arteries, and the method is relatively complex and has low detection efficiency.

[0006] In order to solve the above technical problems, the present invention provides a processing device based on cardiac computed tomography angiography images, comprising:

[0007] A first acquisition module, configured to acquire a current computerized tomography angiography image of the heart;

[0008] An input module, used for inputting the current CT angiography image into a pre-established deep learning network model; wherein the input during the training of the deep learning network model is the CT angiography image and the categories marked on the CT angiography image; the categories at least include the aorta, the coronary artery and the image background;

[0009] An output module, configured to output an initial segmentation result corresponding to the current CT angiography image through the deep learning network model;

[0010] A second acquisition module is used to perform connected area analysis on the artery according to the initial segmentation result; wherein the connected area analysis method is different for different blood vessel types;

[0011] The determination module is used to determine the arterial region on the current electronic computer tomography angiography image according to the connected region analysis result.

[0012] Exemplarily, the processing device based on cardiac computed tomography angiography images includes a training module, and the training module is used to train the deep learning network model;

[0013] The training module specifically includes:

[0014] a processing operation module, configured to perform a processing operation on the CT angiography image to obtain the processed CT angiography image; wherein the processing operation at least includes an image preprocessing operation, a data enhancement operation, and an operation of extracting a sub-block image of a preset size;

[0015] a labeling module, used for labeling the categories marked on the processed CT angiography image;

[0016] The training submodule is used to input the processed CT angiography images and the annotated categories into the deep learning network model for training.

[0017] Exemplarily, the processing operation module performs the image preprocessing operation including:

[0018] Changing the pixel spacing of the CT angiography image to a preset pixel spacing;

[0019] converting the computed tomography values ​​in the computed tomography angiography image with the pixel spacing changed to a preset pixel spacing into a matrix format;

[0020] Acquiring computed tomography values ​​within a preset range and performing normalization processing, and taking the computed tomography values ​​outside the preset range among all the computed tomography values ​​as endpoint values;

[0021] The data enhancement operation performed by the processing operation module is one or more of a multi-dimensional image flipping operation, an image contrast changing operation, a motion artifact adding operation, an image affine transformation operation, an image elastic deformation operation, and an Gaussian noise adding operation;

[0022] The processing operation module performs the operation of extracting the sub-block image of the preset size, including:

[0023] Acquiring hardware conditions for processing the CT angiography image; wherein the hardware conditions at least include video memory occupancy;

[0024] The computerized tomography angiography image after the data enhancement operation is divided into sub-block images of the preset size according to the hardware condition.

[0025] Exemplarily, the training submodule uses a mixed loss function and an optimizer to optimize model performance during training; wherein the mixed loss function is obtained by adding at least a dice loss and a cross entropy loss.

[0026] Exemplarily, the output module specifically includes:

[0027] A prediction module, configured to predict the current CT angiography image using a sliding window size in the deep learning network model; wherein the window size is equal to the size of the sub-block image when the deep learning network model is trained; and the step size is half of the window size;

[0028] The output submodule is used to output the probability of each voxel in the current CT angiography image belonging to a category, and determine the predicted initial segmentation result according to the probability of each voxel belonging to a category.

[0029] Exemplarily, the prediction module also includes a weight setting module;

[0030] The weight setting module is used to increase the weight of the normalized exponential function aggregation center voxel based on the Gaussian importance weight.

[0031] Exemplarily, when predicting the aorta category and the coronary artery category, the prediction module sets a threshold for the aorta category that is greater than a threshold for the coronary artery category.

[0032] Exemplarily, the second acquisition module is specifically used to acquire, according to the initial segmentation result, a connected region with the largest number of voxels among all connected regions in the initial region of the aorta;

[0033] The determination module is specifically configured to use a connected region with the largest number of voxels among all connected regions in the initial region of the aorta as the region of the aorta on the current CT angiography image.

[0034] Exemplarily, the coronary artery category includes a left coronary artery category and a right coronary artery category;

[0035] The second acquisition module is specifically used to acquire a first candidate region in the initial region of the left coronary artery and a second candidate region in the initial region of the right coronary artery according to the initial segmentation result; wherein the first candidate region and the second candidate region are both connected regions with a number of pixels greater than a predetermined number of voxels;

[0036] The determination module is specifically used to screen out a first target area from the first candidate area as the area of ​​the left coronary artery on the current electronic computed tomography angiography image; and to screen out a second target area from the second candidate area as the area of ​​the right coronary artery on the current electronic computed tomography angiography image.

[0037] Exemplarily, the determining module screening out the first target area includes:

[0038] Obtain a first maximum connected area from all the first candidate areas;

[0039] Acquire a first target candidate region from all the first candidate regions; wherein the maximum value of the voxels in the three-dimensional direction of the first target candidate region is greater than a preset value;

[0040] Determine whether the center coordinate point of the first target candidate area is within the first maximum connected area;

[0041] If yes, retaining the first target candidate region as the first target region;

[0042] The determining module screening out the second target area includes:

[0043] Obtaining a second maximum connected area from all of the second candidate areas;

[0044] Acquire a second target candidate region from all the second candidate regions; wherein the maximum value of the voxels in the three-dimensional direction of the second target candidate region is greater than a preset value;

[0045] Determine whether the center coordinate point of the second target candidate area is within the second maximum connected area;

[0046] If so, retain the second target candidate region as the second target region.

[0047] The present invention provides a processing device based on cardiac CT angiography images, comprising: a first acquisition module, used to acquire a current CT angiography image of the heart; an input module, used to input the current CT angiography image into a pre-established deep learning network model; wherein the input during the training of the deep learning network model is the CT angiography image and the categories marked on the CT angiography image; the categories include at least the aorta, the coronary artery and the image background; and an output module, used to output an initial segmentation result corresponding to the current CT angiography image through the deep learning network model. In the processing device based on cardiac CT angiography images, the CT angiography images are predicted based on the deep learning network model, and the aortic region and the coronary region in the arterial region on the CT angiography image are distinguished, thereby improving the efficiency of arterial region detection; and compared with the method of directly completing the distinction between the aortic region and the coronary region according to the prediction result of the deep learning network model, in the processing device based on cardiac CT angiography images provided by the present invention, after the initial segmentation result is outputted through the output module, the prediction result of the deep learning network model is further processed through the second acquisition module and the determination module, thereby improving the accuracy of the determined arterial region on the current CT angiography image. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A schematic diagram of a processing device based on cardiac computed tomography angiography images provided by an embodiment of the present invention;

[0050] Figure 2 A flowchart of a method for performing artery segmentation based on a cardiac computed tomography angiography image processing device provided by an embodiment of the present invention;

[0051] Figure 3 The present invention provides a flowchart of a method for post-processing the aorta and coronary artery segmentation results. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The core of the present invention is to provide a processing device based on cardiac computed tomography angiography images to solve the technical problems that the method of segmenting coronary arteries based on cardiac CTA images using traditional methods cannot simultaneously complete the segmentation of the aorta and coronary arteries, and the method is relatively complex and has low detection efficiency.

[0054] With the continuous growth of clinical needs, manual extraction of coronary artery structure is a cumbersome and time-consuming process, which is not feasible in clinical practice. Therefore, the computer-assisted semi-automatic or automatic segmentation method in the present invention has become the blood vessel segmentation.

[0055] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 A schematic diagram of a processing device based on cardiac computed tomography angiography images provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the device comprises:

[0056] A first acquisition module 10, for acquiring a current CT angiography image of the heart;

[0057] An input module 11, used to input the current CT angiography image into a pre-established deep learning network model;

[0058] An output module 12, used for outputting an initial segmentation result corresponding to the current CT angiography image through a deep learning network model;

[0059] The second acquisition module 13 is used to perform connected area analysis on the artery according to the initial segmentation result; wherein the connected area analysis method is different for different blood vessel types;

[0060] The determination module 14 is used to determine the arterial region on the current CT angiography image according to the connected region analysis result.

[0061] Figure 2 A flowchart of a method for performing artery segmentation based on a cardiac computed tomography angiography image processing device provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the method includes:

[0062] S10: input cardiac computed tomography angiography images and annotations;

[0063] S11: Model training using deep learning network;

[0064] S12: Input the test data into the trained model for model prediction to obtain the segmentation results of the aorta and coronary artery;

[0065] S13: Post-process the aorta and coronary artery segmentation results.

[0066] In order to predict the arterial region on the CT angiography image, a processing device based on the cardiac CT angiography image includes a training module, which is used to train a deep learning network model. The input of the deep learning network model training is the CT angiography image and the categories marked on the CT angiography image; the categories include at least the aorta, the coronary artery and the image background.

[0067] The training modules include:

[0068] A processing operation module, used for performing a processing operation on the CT angiography image to obtain a processed CT angiography image; wherein the processing operation at least includes an image preprocessing operation, a data enhancement operation, and an operation of extracting a sub-block image of a preset size;

[0069] a labeling module, used for labeling the categories labeled on the processed CT angiography image;

[0070] The training submodule is used to input the processed CT angiography images and the annotated categories into the deep learning network model for training.

[0071] Cardiac CTA images are manually annotated, with three categories including aorta, coronary artery and background; or four categories including aorta, left coronary artery, right coronary artery and background. There is no limitation on the deep learning network used, such as nnUNet (full name is no-new-Net, a neural network model based on U-Net architecture). Cardiac CTA images are original medical digital images.

[0072] Specifically, the processing operation module performs image preprocessing operations including:

[0073] Changing the pixel spacing of computed tomography angiography images to a preset pixel spacing;

[0074] converting computed tomography values ​​in the computed tomography angiography image having a pixel pitch changed to a preset pixel pitch into a matrix format;

[0075] The computed tomography values ​​within a preset range are obtained and normalized, and the computed tomography values ​​outside the preset range among all the computed tomography values ​​are taken as endpoint values.

[0076] There is no limitation on the preset pixel spacing, such as changing the pixel spacing of the DICOM format cardiac CTA image to: 0.35 x 0.35 x 0.3 mm; obtain the computed tomography (CT) value in the DICOM format cardiac CTA image and convert it into numpy matrix format.

[0077] The image CT values ​​outside the set range are set as endpoint values, and there is no limit on the values ​​in the preset range. For example, the CT value setting range is: -48 to 887, and normalization is performed. The normalization method uses the Z-Scoring standardization method, that is, subtracting the mean and dividing by the standard deviation. The processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1. The mean and standard deviation are set to the mean and standard deviation of the entire training set foreground, and the set mean is 284 and the standard deviation is 198.

[0078] The processing operation module performs data enhancement operations such as one or more of a multi-dimensional image flipping operation, an image contrast changing operation, a motion artifact adding operation, an image affine transformation operation, an image elastic deformation operation, and a Gaussian noise adding operation.

[0079] Considering the hardware limitations such as video memory usage, it is also necessary to perform image block (patch) processing on the enhanced cardiac CTA image. The processing operation module performs the operation of extracting sub-block images of a preset size, including:

[0080] Acquiring hardware conditions for processing electronic computed tomography angiography images; wherein the hardware conditions at least include video memory occupancy;

[0081] The electronic computer tomography angiography image after the data enhancement operation is divided into sub-block images of preset sizes according to the hardware conditions.

[0082] There is no limitation on the preset size, for example, the image block size may be: 192*192*160.

[0083] The training submodule uses a mixed loss function and an optimizer to optimize model performance during training; wherein the mixed loss function is at least obtained by adding a dice loss and a cross entropy loss.

[0084] The training adopts a mixed loss function Loss, which is obtained by adding the dice loss Dice Loss and the cross entropy loss Cross Entropy Loss. The calculation formula of the dice loss Dice Loss is as follows:

[0085] ;

[0086] The calculation formula of Cross Entropy Loss is as follows:

[0087] ;

[0088] in, Indicates The label value of each pixel, Indicates The predicted value of pixels, is the total number of pixels, which is equal to the number of pixels in a single sample multiplied by the batch size.

[0089] The AdamW optimizer is used during training, the initial learning rate is 1e-3, and the Poly learning strategy is used to adjust the learning rate. The specific formula is:

[0090] ;

[0091] in, represents the new learning rate, represents the initial learning rate, is the number of iterations this time, represents the maximum number of iterations, Represents the parameter that controls the shape of the curve and is set to 0.9.

[0092] When using the established deep learning network model for prediction, the output module specifically includes:

[0093] A prediction module, used for predicting a current CT angiography image using a sliding window size in a deep learning network model; wherein the window size is equal to the size of the sub-block image when training the deep learning network model; and the step size is half of the window size;

[0094] The output submodule is used to output the probability of each voxel in the current CT angiography image belonging to a category, and determine the predicted initial segmentation result according to the probability of each voxel belonging to the category.

[0095] The model predicts images using a sliding window method, where the window size is equal to the image patch size during training, and the step size is half the window size, so two adjacent windows will have an overlap of size patch / 2. In order to suppress splicing artifacts and reduce the influence of positions close to the boundary, Gaussian importance weights are applied, thereby increasing the weight of the center voxel of the normalized exponential function softmax aggregation.

[0096] In order to improve the segmentation sensitivity of the aorta and coronary arteries, the prediction module sets the threshold of the aorta category greater than the threshold of the coronary category when predicting the aorta category and the coronary category. The output result of the model prediction is the probability value of each voxel of each cardiac CTA image belonging to 4 categories. For example, the size of the input cardiac CTA image is: 512 x 512 x 510, and the output size predicted by the model is: 512 x 512 x 510 x 4. In order to improve the segmentation sensitivity of the aorta and coronary arteries, different thresholds are set for the aorta and coronary arteries in the processing device based on cardiac electronic computed tomography angiography images. The aorta area is relatively large, and a higher threshold can be selected, such as the threshold set for the aorta can be 0.9. The coronary area is relatively small, and a smaller threshold needs to be selected, such as the threshold set for the coronary artery can be 0.1. The segmentation result obtained is the category predicted by the model for each voxel of the cardiac CTA image.

[0097] Due to the low image resolution, there are (motion) artifacts in the data, adjacent structures with similar intensity to blood vessels, and lesions such as stenosis and calcification. At the same time, in CTA images, the coronary arteries have the characteristics of large scale changes and complex topology, which increases the difficulty of coronary artery segmentation in CTA. Therefore, in the processing device based on cardiac electronic computed tomography angiography images provided by the present invention, after prediction by the deep learning network model, the segmentation results of the aorta and coronary arteries are further post-processed. The aorta is first post-processed, and then the coronary arteries are post-processed.

[0098] The second acquisition module is specifically used to acquire the connected area with the largest number of voxels among all connected areas in the initial area of ​​the aorta according to the initial segmentation result;

[0099] The determination module is specifically used to take the connected area with the largest number of voxels among all the connected areas in the initial area of ​​the aorta as the area of ​​the aorta on the current electronic computer tomography angiography image.

[0100] When the coronary artery is post-processed, the post-processing method is connected domain analysis. Since the coronary artery is divided into a left coronary artery and a right coronary artery, the second acquisition module is specifically used to acquire a first candidate region in the initial region of the left coronary artery and a second candidate region in the initial region of the right coronary artery according to the initial segmentation result; wherein the first candidate region and the second candidate region are both connected regions with a pixel number greater than a predetermined number of voxels;

[0101] The determination module is specifically used to select a first target area from the first candidate area as the area of ​​the left coronary artery on the current electronic computed tomography angiography image; and select a second target area from the second candidate area as the area of ​​the right coronary artery on the current electronic computed tomography angiography image.

[0102] The first target area selected by the determination module includes:

[0103] Obtain a first maximum connected area from all first candidate areas;

[0104] Acquire a first target candidate region from all first candidate regions; wherein the maximum value of voxels in the three-dimensional direction of the first target candidate region is greater than a preset value;

[0105] Determine whether the center coordinate point of the first target candidate area is within the first maximum connected area;

[0106] If yes, retain the first target candidate region as the first target region;

[0107] The second target area selected by the determination module includes:

[0108] Obtain the second largest connected area from all second candidate areas;

[0109] Acquire a second target candidate region from all the second candidate regions; wherein the maximum value of voxels in the three-dimensional direction of the second target candidate region is greater than a preset value;

[0110] Determine whether the center coordinate point of the second target candidate area is within the second largest connected area;

[0111] If yes, the second target candidate region is retained as the second target region.

[0112] Figure 3 A flowchart of a method for post-processing the aorta and coronary artery segmentation results provided by the present invention is shown in FIG. Figure 3 As shown, the method includes:

[0113] S14: Obtain the segmentation results of the aorta and coronary arteries predicted by the deep learning network;

[0114] S15: Perform maximum connected domain analysis on the aorta region and retain the connected region with the largest number of voxels;

[0115] S16: Perform connectivity analysis on the coronary artery domain and retain connectivity areas that meet preset conditions.

[0116] The aorta and coronary artery segmentation results are post-processed. First, the aorta is post-processed using the maximum connected domain analysis method to screen out the connected regions with the largest number of voxels.

[0117] Secondly, the coronary artery is post-processed, and the post-processing method is connected domain analysis. Since the coronary artery is divided into the left coronary artery and the right coronary artery, the preset conditions include: for all the segmentation results of the left coronary artery and the right coronary artery, the connected domain analysis is performed, and the connected areas with the number of pixels greater than the predetermined number of voxels are screened out, and the predetermined number of voxels can be 200, and the screened connected areas are used as candidate areas. Then the candidate areas are processed, and the maximum connected domain belonging to the left coronary artery and the right coronary artery is first taken for all candidate areas; secondly, all candidate areas are traversed, and if the maximum value of the candidate area in the three directions of space xyz is greater than the preset value, the next processing flow is entered, and the preset value can be 30; the subsequent processing flow needs to first determine the category to which the candidate area belongs. If the candidate area belongs to the left coronary artery, calculate whether the central coordinate point of this area is within the maximum connected domain of the left coronary artery, and if so, retain this candidate area as the left coronary artery area; if the candidate area belongs to the right coronary artery, calculate whether the central coordinate point of this area is within the maximum connected domain of the right coronary artery, and if so, retain this candidate area as the right coronary artery area.

[0118] The present invention provides a processing device based on cardiac electronic computer tomography angiography images. After prediction is performed through a deep learning network model, the aorta and coronary artery segmentation results are further post-processed, which can retain the aorta, especially the coronary artery as a whole, to the greatest extent, reduce false positives such as veins, reduce coronary artery ruptures caused by lesions such as stenosis and calcification, and reduce partial coronary artery loss caused by poor quality, noise, and insufficient vascular development of the CTA image itself, thereby improving the accuracy and robustness of aorta and coronary artery segmentation based on CTA images.

[0119] The above is a detailed introduction to a processing device based on cardiac computed tomography angiography images provided by the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified, and these improvements and modifications also fall within the scope of protection of the present invention.

[0120] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. A processing device based on cardiac computed tomography angiography images, characterized in that: include: A first acquisition module, configured to acquire a current computerized tomography angiography image of the heart; An input module, used for inputting the current CT angiography image into a pre-established deep learning network model; wherein the input during the training of the deep learning network model is the CT angiography image and the categories marked on the CT angiography image; the categories at least include the aorta, the coronary artery and the image background; An output module, configured to output an initial segmentation result corresponding to the current CT angiography image through the deep learning network model; A second acquisition module is used to perform connected area analysis on the artery according to the initial segmentation result; wherein the connected area analysis method is different for different blood vessel types; The determination module is used to determine the arterial region on the current electronic computer tomography angiography image according to the connected region analysis result.

2. The processing device based on cardiac computed tomography angiography images according to claim 1, characterized in that: The processing device based on cardiac computed tomography angiography images includes a training module, and the training module is used to train the deep learning network model; The training module specifically includes: a processing operation module, configured to perform a processing operation on the CT angiography image to obtain the processed CT angiography image; wherein the processing operation at least includes an image preprocessing operation, a data enhancement operation, and an operation of extracting a sub-block image of a preset size; a labeling module, used for labeling the categories marked on the processed CT angiography image; The training submodule is used to input the processed CT angiography images and the annotated categories into the deep learning network model for training.

3. The processing device based on cardiac computed tomography angiography images according to claim 2, characterized in that: The processing operation module performs the image preprocessing operation including: Changing the pixel spacing of the CT angiography image to a preset pixel spacing; converting the computed tomography values ​​in the computed tomography angiography image with the pixel spacing changed to a preset pixel spacing into a matrix format; Acquiring computed tomography values ​​within a preset range and performing normalization processing, and taking the computed tomography values ​​outside the preset range among all the computed tomography values ​​as endpoint values; The data enhancement operation performed by the processing operation module is one or more of a multi-dimensional image flipping operation, an image contrast changing operation, a motion artifact adding operation, an image affine transformation operation, an image elastic deformation operation, and an Gaussian noise adding operation; The processing operation module performs the operation of extracting the sub-block image of the preset size, including: Acquiring hardware conditions for processing the CT angiography image; wherein the hardware conditions at least include video memory occupancy; The computerized tomography angiography image after the data enhancement operation is divided into sub-block images of the preset size according to the hardware condition.

4. The processing device based on cardiac computed tomography angiography images according to claim 2, characterized in that: The training submodule uses a mixed loss function and an optimizer to optimize model performance during training; wherein the mixed loss function is obtained by adding at least a dice loss and a cross entropy loss.

5. The processing device based on cardiac computed tomography angiography images according to any one of claims 2 to 4, characterized in that: The output module specifically includes: A prediction module, configured to predict the current CT angiography image using a sliding window size in the deep learning network model; wherein the window size is equal to the size of the sub-block image when the deep learning network model is trained; and the step size is half of the window size; The output submodule is used to output the probability of each voxel in the current CT angiography image belonging to a category, and determine the predicted initial segmentation result according to the probability of each voxel belonging to a category.

6. The processing device based on cardiac computed tomography angiography images according to claim 5, characterized in that: The prediction module also includes a weight setting module; The weight setting module is used to increase the weight of the normalized exponential function aggregation center voxel based on the Gaussian importance weight.

7. The processing device based on cardiac computed tomography angiography images according to claim 6, characterized in that: When predicting the aorta category and the coronary artery category, the prediction module sets a threshold for the aorta category that is greater than a threshold for the coronary artery category.

8. The processing device based on cardiac computed tomography angiography images according to claim 7, characterized in that: The second acquisition module is specifically used to acquire the connected area with the largest number of voxels among all connected areas in the initial area of ​​the aorta according to the initial segmentation result; The determination module is specifically configured to use a connected region with the largest number of voxels among all connected regions in the initial region of the aorta as the region of the aorta on the current CT angiography image.

9. The processing device based on cardiac computed tomography angiography images according to claim 7, characterized in that: The coronary artery category includes a left coronary artery category and a right coronary artery category; The second acquisition module is specifically used to acquire a first candidate region in the initial region of the left coronary artery and a second candidate region in the initial region of the right coronary artery according to the initial segmentation result; wherein the first candidate region and the second candidate region are both connected regions with a number of pixels greater than a predetermined number of voxels; The determination module is specifically used to screen out a first target area from the first candidate area as the area of ​​the left coronary artery on the current electronic computed tomography angiography image; and to screen out a second target area from the second candidate area as the area of ​​the right coronary artery on the current electronic computed tomography angiography image.

10. The processing device based on cardiac computed tomography angiography images according to claim 9, characterized in that: The determining module screening out the first target area includes: Obtain a first maximum connected area from all the first candidate areas; Acquire a first target candidate region from all the first candidate regions; wherein the maximum value of the voxels in the three-dimensional direction of the first target candidate region is greater than a preset value; Determine whether the center coordinate point of the first target candidate area is within the first maximum connected area; If yes, retaining the first target candidate region as the first target region; The determining module screening out the second target area includes: Obtaining a second maximum connected area from all of the second candidate areas; Acquire a second target candidate region from all the second candidate regions; wherein the maximum value of the voxels in the three-dimensional direction of the second target candidate region is greater than a preset value; Determine whether the center coordinate point of the second target candidate area is within the second maximum connected area; If so, retain the second target candidate region as the second target region.

Images