Coronary artery segmentation model training method and device, coronary artery segmentation method and device

Abstract

The present application provides a method and device for training a coronary segmentation model, and a method and device for coronary segmentation, which specifically include: obtaining medical image data, coronary midline marking data, and coronary segmentation marking data of the same coronary artery as a group of initial learning data; for each group of initial learning data, the group of initial learning data is cut out by random cutting out to obtain multiple groups of random cutting out results; for each group of random cutting out results, the medical image cutting out results and the coronary midline marking cutting out results in the group of random cutting out results are used as training sample data, and the coronary segmentation marking cutting out results are used as standard sample data in the same group as the training sample data, and the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary segmentation model. In this way, the present application can improve the segmentation accuracy of the coronary segmentation results in the details of the coronary edge.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a coronary artery segmentation model training method and device, and a coronary artery segmentation method and device. Background Art

[0002] The coronary artery (also called coronary artery) is the artery that supplies blood to the heart. It originates from the aortic sinus at the root of the aorta and is divided into two branches, the left and right branches, running on the surface of the heart. The coronary artery 3D reconstruction technology based on CTA (CT angiography, CT angiography) can help doctors have a more comprehensive and clear understanding of the heart, and plays an important role in reducing missed diagnoses and misdiagnoses. The accuracy of coronary artery segmentation determines the effect of 3D reconstruction.

[0003] At present, the existing solutions based on neural network models and using the coronary artery centerline to segment the coronary artery mostly directly use the entire coronary artery centerline data and input the global coronary artery information (i.e., the whole image information) at a certain size to segment the coronary artery. Although this solution can effectively segment the coronary artery, it cannot ensure that the model learns the edge detail features of the coronary artery, resulting in low segmentation accuracy of the coronary artery edge details in the coronary artery segmentation results. The coronary artery vessels in the coronary artery segmentation results are prone to adhesion of the vascular epidermis at dense locations and bifurcations. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a training method and device for a coronary segmentation model, and a coronary segmentation method and device, so as to solve the problem that coronary vessels in dense areas and bifurcations in the coronary segmentation results are prone to adhesion of the vascular epithelium, and to improve the segmentation accuracy of the coronary segmentation results in the details of the coronary edge.

[0005] In a first aspect, an embodiment of the present application provides a method for training a coronary artery segmentation model, the training method comprising:

[0006] Obtain medical imaging data of the same coronary artery, coronary artery centerline marking data, and coronary artery segmentation marking data as a set of initial learning data;

[0007] For each group of initial learning data, at the same randomly designated position, according to a preset cutout size, the medical image data, the coronary artery centerline marking data, and the coronary artery segmentation marking data in the group of initial learning data are cutout to obtain multiple groups of random cutout results; wherein each group of random cutout results includes medical image cutout results, coronary artery centerline marking cutout results, and coronary artery segmentation marking cutout results of the same size at the same position;

[0008] For each group of random cutout results, the medical image cutout results and the coronary artery midline marking cutout results in the group of random cutout results are used as a group of dual-channel training sample data, and the coronary artery segmentation marking cutout results in the group of random cutout results are used as standard sample data in the same group as the training sample data. According to the data difference between the training sample data and the standard sample data in the same group, the original segmentation model is trained to obtain a trained coronary artery segmentation model.

[0009] In an optional implementation, the training of the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain the trained coronary segmentation model includes:

[0010] For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data;

[0011] Acquire standard sample data from the same group as the group of training sample data, and determine a first model training loss of the original segmentation model for the group of training sample data based on the acquired standard sample data and the coronary artery segmentation prediction result;

[0012] The original segmentation model is repeatedly trained using the first model training loss of each group of training sample data until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

[0013] In an optional implementation, the training of the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain the trained coronary segmentation model further includes:

[0014] For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data;

[0015] According to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the coronary artery segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction result of each coronary artery;

[0016] Determining a second model training loss of the original segmentation model for each coronary artery according to the global segmentation prediction result of the same coronary artery and the coronary artery segmentation label data;

[0017] The original segmentation model is repeatedly trained using the second model training loss of each coronary artery until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

[0018] In an optional implementation, the coronary segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​according to the designated cutout positions corresponding to each group of training sample data under the same coronary artery to obtain the global segmentation prediction result of each coronary artery, and further includes:

[0019] For the coronary segmentation prediction results of each group of training sample data under the same coronary artery, according to the preset image segmentation size, from the central area of ​​the coronary segmentation prediction results, a target image area of ​​the image segmentation size is extracted as the key prediction area for coronary segmentation of the group of training sample data; wherein the central area is used to characterize the concentrated distribution area of ​​the image features in the coronary segmentation prediction results;

[0020] According to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the key prediction regions of coronary segmentation of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction results of each coronary artery.

[0021] In an optional implementation, determining the first model training loss of the original segmentation model for the group of training sample data based on the acquired standard sample data and the coronary segmentation prediction result includes:

[0022] From the acquired standard sample data, determine a background image region belonging to non-coronary arteries in the standard sample data as first standard background data, and determine a key image region belonging to coronary arteries in the standard sample data as standard key data;

[0023] From the coronary segmentation prediction result, determine a background image region belonging to a non-coronary artery in the coronary segmentation prediction result as first training background data, and determine a key image region belonging to a coronary artery in the coronary segmentation prediction result as training key data;

[0024] Determining a first background training loss of the original segmentation model for the group of training sample data according to a first background data difference between the first standard background data and the first training background data, and a preset first background loss weight;

[0025] Determine the first key training loss of the original segmentation model for the group of training sample data according to the first key data difference between the standard key data and the training key data, and a preset first key loss weight; wherein the first key loss weight is greater than the first background loss weight;

[0026] The first background training loss and the first key training loss are used as the first model training loss.

[0027] In an optional embodiment, determining the second model training loss of the original segmentation model for each coronary artery based on the global segmentation prediction result and the coronary artery segmentation label data of the same coronary artery comprises:

[0028] From the coronary segmentation mark data of each coronary artery, determine the background image area belonging to the non-coronary artery in the coronary segmentation mark data as the second standard background data, determine the image area belonging to the edge of the coronary artery in the coronary segmentation mark data as the standard edge data, and determine the image area belonging to the center of the coronary artery in the coronary segmentation mark data as the standard center data; wherein the standard edge data and the standard center data constitute the image area belonging to the coronary artery in the coronary segmentation mark data;

[0029] From the global segmentation prediction result of the same coronary artery, determine the background image area belonging to the non-coronary artery in the global segmentation prediction result as the second training background data, determine the image area belonging to the edge of the coronary artery in the global segmentation prediction result as the training edge data, and determine the image area belonging to the center of the coronary artery in the global segmentation prediction result as the training center data; wherein the training edge data and the training center data constitute the image area belonging to the coronary artery in the global segmentation prediction result;

[0030] Determining a second background training loss of the original segmentation model for the coronary artery according to a second background data difference between the second standard background data and the second training background data, and a preset second background loss weight;

[0031] Determining the edge training loss of the original segmentation model for the coronary artery according to the edge data difference between the standard edge data and the training edge data, and a preset edge loss weight;

[0032] Determining the center training loss of the original segmentation model for the coronary artery according to the center data difference between the standard center data and the training center data, and a preset center loss weight; wherein the center loss weight is less than the edge loss weight, and the center loss weight is greater than the second background loss weight;

[0033] The second background training loss, the edge training loss, and the center training loss are used as the second model training loss.

[0034] In a second aspect, an embodiment of the present application provides a coronary artery segmentation method, the coronary artery segmentation method comprising:

[0035] Acquire target medical image data of the coronary artery to be segmented and target coronary artery centerline marking data;

[0036] According to each blood vessel midline segment included in the target coronary artery midline marking data, specifying any position including the blood vessel midline segment from the target medical image data as a target cutout position;

[0037] At each designated target cutout position, the target medical image data and the target coronary artery centerline marking data are cutout according to a preset cutout size to obtain multiple groups of target cutout results; wherein each group of target cutout results includes a target medical image cutout result and a target coronary artery centerline marking cutout result of the same size at the same position;

[0038] The multiple groups of target cutout results are input into a pre-trained coronary artery segmentation model, and the output result of the coronary artery segmentation model is used as the coronary artery segmentation result of the coronary artery to be segmented.

[0039] In an optional embodiment, the coronary artery segmentation model is obtained after training according to any of the optional embodiments of the above-mentioned coronary artery segmentation model training method.

[0040] In an optional implementation, the step of specifying any position including the blood vessel midline segment in the target medical image data as the target cutout position further includes:

[0041] From the line segment endpoints of each of the blood vessel midline segments, specify any line segment endpoint as the first starting designated position of the target cutout position;

[0042] In the target coronary artery centerline marking data, taking the first starting designated position as the cutout center point, and predicting the region boundary where the cutout region is located according to the preset cutout size;

[0043] When it is detected that the region boundary and any blood vessel midline segment included in the target coronary artery midline mark data have an intersection, the intersection is designated as the next designated position of the target cutout position;

[0044] When it is detected that the region boundary does not intersect with any vascular centerline segment included in the target coronary artery centerline marking data, any segment endpoint is designated as the second starting designated position of the target cutout position from the segment endpoints of other vascular centerline segments except the target vascular centerline segment, and the step of determining the next designated position of the target cutout position is repeated; wherein the target vascular centerline segment is used to characterize the vascular centerline segment corresponding to the first starting designated position.

[0045] In an optional implementation, the target medical image data and the target coronary artery centerline marking data are cut out according to a preset cutout size at each designated target cutout position, including:

[0046] At each designated target cutout position, the target cutout position is used as the center point of the cutout region, and the target medical image data and the target coronary artery centerline marking data are cutout according to the cutout size.

[0047] In an optional implementation, the step of inputting the multiple groups of target cutout results into a pre-trained coronary segmentation model further includes:

[0048] From the multiple groups of target cutout results, duplicate target cutout results are deleted to obtain multiple groups of key cutout results that can represent the global image information of the coronary artery to be segmented;

[0049] The multiple groups of key cutout results are input into a pre-trained coronary artery segmentation model.

[0050] In a third aspect, an embodiment of the present application provides a training device for a coronary artery segmentation model, the training device comprising:

[0051] A first acquisition module is used to acquire medical image data of the same coronary artery, coronary artery centerline marking data and coronary artery segmentation marking data as a set of initial learning data;

[0052] A first cutout module is used to cutout the medical image data, the coronary artery centerline marking data and the coronary artery segmentation marking data in each group of initial learning data at the same randomly designated position according to a preset cutout size to obtain multiple groups of random cutout results; wherein each group of random cutout results includes medical image cutout results, coronary artery centerline marking cutout results and coronary artery segmentation marking cutout results of the same size at the same position;

[0053] A training module is used to, for each group of random cutout results, use the medical image cutout results and the coronary artery midline marking cutout results in the group of random cutout results as a group of dual-channel training sample data, use the coronary artery segmentation marking cutout results in the group of random cutout results as the standard sample data in the same group as the training sample data, and train the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary artery segmentation model.

[0054] In a fourth aspect, an embodiment of the present application provides a coronary artery segmentation device, the coronary artery segmentation device comprising:

[0055] The second acquisition module is used to acquire the target medical image data of the coronary artery to be segmented and the target coronary artery centerline marking data;

[0056] a position determination module, for specifying any position including the vascular centerline segment from the target medical image data as a target cutout position according to each vascular centerline segment included in the target coronary artery centerline marking data;

[0057] A second cutout module is used to cut out the target medical image data and the target coronary artery centerline marking data at each designated target cutout position according to a preset cutout size to obtain multiple groups of target cutout results; wherein each group of target cutout results includes a target medical image cutout result and a target coronary artery centerline marking cutout result of the same size at the same position;

[0058] The segmentation prediction module is used to input the multiple groups of target cutout results into a pre-trained coronary segmentation model, and use the output result of the coronary segmentation model as the coronary segmentation result of the coronary artery to be segmented.

[0059] In a fifth aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the steps of any of the above-mentioned methods for training a coronary artery segmentation model, or executes the steps of any of the above-mentioned methods for coronary artery segmentation.

[0060] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the computer program executes the steps of any of the above-mentioned coronary artery segmentation model training methods, or executes the steps of any of the above-mentioned coronary artery segmentation methods.

[0061] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0062] The embodiments of the present application provide a training method and device for a coronary segmentation model, and a coronary segmentation method and device, which obtain medical image data, coronary centerline marking data and coronary segmentation marking data of the same coronary artery as a group of initial learning data; for each group of initial learning data, at the same randomly specified position, according to a preset cutout size, the medical image data, coronary centerline marking data and coronary segmentation marking data in the group of initial learning data are cutout to obtain multiple groups of random cutout results; for each group of random cutout results, the medical image cutout results and coronary centerline marking cutout results in the group of random cutout results are used as a group of dual-channel training sample data, and the coronary segmentation marking cutout results in the group of random cutout results are used as standard sample data in the same group as the training sample data; according to the data difference between the training sample data and the standard sample data in the same group, the original segmentation model is trained to obtain a trained coronary segmentation model.

[0063] Through the above method, the present application can solve the problem that coronary vessels in the coronary segmentation results are prone to adhesion of the vascular epithelium at dense areas and bifurcations, and improve the segmentation accuracy of the coronary segmentation results in the details of the coronary edge.

[0064] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0066] Figure 1 A schematic diagram showing a flow chart of a method for training a coronary artery segmentation model provided in an embodiment of the present application is shown;

[0067] Figure 2 A schematic diagram showing a result of random cutout provided in an embodiment of the present application is shown;

[0068] Figure 3 A schematic flow chart of a first method for training an original segmentation model provided in an embodiment of the present application is shown;

[0069] Figure 4 A schematic flow chart of a second method for training an original segmentation model provided in an embodiment of the present application is shown;

[0070] Figure 5A schematic diagram of a process for determining a second model training loss provided in an embodiment of the present application is shown;

[0071] Figure 6 A schematic diagram showing a flow chart of a coronary artery segmentation method provided in an embodiment of the present application is shown;

[0072] Figure 7 A schematic diagram of the structure of a coronary artery segmentation model training device provided in an embodiment of the present application is shown;

[0073] Figure 8 A schematic diagram of the structure of a coronary artery segmentation device provided in an embodiment of the present application is shown;

[0074] Fig. 9 A schematic diagram of the structure of a computer device 900 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0076] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0077] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0078] At present, the existing solutions based on neural network models and using the coronary artery centerline to segment the coronary artery mostly directly use the entire coronary artery centerline data and input the global coronary artery information (i.e., the whole image information) at a certain size to segment the coronary artery. Although this solution can effectively segment the coronary artery, it cannot ensure that the model learns the edge detail features of the coronary artery, resulting in low segmentation accuracy of the coronary artery edge details in the coronary artery segmentation results. The coronary artery vessels in the coronary artery segmentation results are prone to adhesion of the vascular epidermis at dense locations and bifurcations.

[0079] Based on this, the embodiments of the present application provide a training method and device for a coronary segmentation model, and a coronary segmentation method and device, by acquiring medical image data, coronary centerline marking data and coronary segmentation marking data of the same coronary artery as a group of initial learning data; for each group of initial learning data, at the same randomly specified position, according to a preset cutout size, the medical image data, coronary centerline marking data and coronary segmentation marking data in the group of initial learning data are cutout to obtain multiple groups of random cutout results; for each group of random cutout results, the medical image cutout results and coronary centerline marking cutout results in the group of random cutout results are used as a group of dual-channel training sample data, and the coronary segmentation marking cutout results in the group of random cutout results are used as standard sample data in the same group as the training sample data; according to the data difference between the training sample data and the standard sample data in the same group, the original segmentation model is trained to obtain a trained coronary segmentation model.

[0080] Through the above method, the present application can solve the problem that coronary vessels in the coronary segmentation results are prone to adhesion of the vascular epithelium at dense areas and bifurcations, and improve the segmentation accuracy of the coronary segmentation results in the details of the coronary edge.

[0081] To facilitate understanding of this embodiment, a coronary artery segmentation model training method and device, and a coronary artery segmentation method and device provided in an embodiment of the present application are introduced in detail below.

[0082] Reference Figure 1 As shown, Figure 1 A flow chart of a coronary segmentation model training method provided in an embodiment of the present application is shown. The training method includes steps S101-S103; specifically:

[0083] S101, obtaining medical image data of the same coronary artery, coronary artery centerline marking data, and coronary artery segmentation marking data as a set of initial learning data.

[0084] Here, as an optional embodiment, the medical imaging data may be CT image data of the coronary arteries, and the coronary artery centerline marking data may be obtained by using a pre-trained coronary artery centerline extraction model. For example, the medical imaging data of the same coronary artery is input into a pre-trained coronary artery centerline extraction model, and the model output result of the coronary artery centerline extraction model is used as the above-mentioned coronary artery centerline marking data; wherein the coronary artery centerline marking data includes a plurality of line segments, and each line segment corresponds to a vascular centerline of a coronary artery in the marked coronary artery.

[0085] It should be noted that the embodiments of the present application are mainly used to solve the problem that the coronary segmentation results have low segmentation accuracy in the coronary edge details, which is caused by neglecting to learn the edge details in the coronary images during the training of the existing coronary segmentation model on the image processing side. That is, the embodiments of the present application do not focus on the accuracy of the coronary centerline marking data. Based on this, under the premise that the embodiments of the present application do not involve "improvements in the coronary centerline extraction method", the embodiments of the present application do not make any restrictions on the method of obtaining the above-mentioned coronary centerline marking data and the specific training method of the above-mentioned coronary centerline extraction model.

[0086] S102, for each group of initial learning data, at the same randomly designated position, according to a preset cutout size, the medical image data, the coronary artery centerline marking data and the coronary artery segmentation marking data in the group of initial learning data are cutout to obtain multiple groups of random cutout results.

[0087] Here, each group of random cutout results includes medical image cutout results of the same size and position, coronary artery midline marking cutout results, and coronary artery segmentation marking cutout results.

[0088] In an embodiment of the present application, the medical imaging data and the coronary artery centerline marking data in each group of initial learning data are isotropic; wherein, isotropy means that the performance values ​​measured by the three data in different directions are exactly the same, also known as homogeneity; based on the characteristic that isotropy is a physical property that does not change with the measurement direction, as an optional embodiment, when the amount of initial learning data is insufficient (such as less than a preset number threshold), in order to avoid adverse results such as overfitting, the medical imaging data and the coronary artery centerline marking data in each group of initial learning data can also be subjected to data enhancement processing, such as data rotation, data flipping and the like.

[0089] Here, the embodiments of the present application do not impose any limitation on the specific processing method of data enhancement processing.

[0090] Specifically, Figure 2 A schematic diagram of a random cutout result provided by an embodiment of the present application is shown. Figure 2As shown, the medical image data 201, the coronary artery centerline marking data 202 and the coronary artery segmentation marking data 203 are located in the same group of initial learning data, that is, the medical image data 201, the coronary artery centerline marking data 202 and the coronary artery segmentation marking data 203 correspond to the same coronary artery; at a randomly specified position coordinate a, with the position coordinate a as the cutout center, according to the preset cutout size (that is, the size of the dotted rectangular box in the figure), the medical image data 201, the coronary artery centerline marking data 202 and the coronary artery segmentation marking data 203 are cutout respectively to obtain a group of random cutout results with the position coordinate a as the cutout center, and the group of random cutout results includes: a medical image cutout result 210, a coronary artery centerline marking cutout result 220 and a coronary artery segmentation marking cutout result 230 of the same size at the same position coordinate a.

[0091] S103, for each group of random cutout results, the medical image cutout results and the coronary artery centerline marking cutout results in the group of random cutout results are used as a group of dual-channel training sample data, and the coronary artery segmentation marking cutout results in the group of random cutout results are used as standard sample data in the same group as the training sample data, and the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary artery segmentation model.

[0092] It should be noted that in the embodiments of the present application, the segmentation network model in deep learning can be used as the above-mentioned original segmentation model. The embodiments of the present application do not impose any limitation on the specific neural network type to which the original segmentation model belongs.

[0093] Here, in step S103, when the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group, based on different training ideas, the embodiment of the present application provides the following two different training methods, specifically:

[0094] In the specific implementation process of the above step S103, based on the data difference between the training sample data after cutting out and the standard sample data in each group of random cutting out results, when training the original segmentation model, refer to Figure 3 As shown, Figure 3 The flowchart of the first method for training the original segmentation model provided in the embodiment of the present application is shown. The method includes steps S301-S303; specifically:

[0095] S301, for each group of training sample data, according to the coronary artery centerline marking cutout results in the group of training sample data, train the original segmentation model to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data.

[0096] S302, obtaining standard sample data from the same group as the group of training sample data, and determining a first model training loss of the original segmentation model for the group of training sample data based on the obtained standard sample data and the coronary artery segmentation prediction result.

[0097] S303, repeatedly training the original segmentation model using the first model training loss of each group of training sample data until the original segmentation model reaches convergence, and using the original segmentation model that has reached convergence as the trained coronary artery segmentation model.

[0098] Regarding the implementation of the above steps S301-S303, in a first optional implementation, when determining the training loss of the above-mentioned first model, the training purpose of the original segmentation model and the degree of correlation between different image regions in the image data may not be considered. At this time, the commonly used cross entropy loss function or semantic segmentation loss function and other loss functions may be directly used to train the original segmentation model.

[0099] For example, a segmentation network can be used as the original segmentation model, and a cross entropy loss function or a commonly used semantic segmentation loss function, such as a focal loss loss function, can be used as the loss function of the original segmentation model to calculate the loss function value between the standard sample data and the coronary segmentation prediction result, and the calculated loss function value is used as the first model training loss of the original segmentation model for the group of training sample data; then, the back propagation algorithm is used to repeatedly train the original segmentation model based on the first model training loss of each group of training sample data until the original segmentation model converges (such as the first model training loss reaches the minimum), and the original segmentation model that has reached convergence is used as the trained coronary segmentation model to end the model training process.

[0100] With respect to the implementation of the above steps S301-S303, in a second optional implementation, considering that the training purpose of the original segmentation model is to learn to segment the coronary arteries in the image, that is, whether in the training sample data or in the standard sample data, the key learning area of ​​the original segmentation model is the image area belonging to the coronary artery, and other image areas except the coronary artery are background image areas that are irrelevant to the training purpose (i.e., learning to segment the coronary artery in the image). Based on this, in addition to the above first optional implementation (i.e., using a commonly used cross entropy loss function or other indifferent type of loss function to determine the above first model training loss), in an embodiment of the present application, the step of determining the first model training loss in the above step S302 can be performed according to the following steps, specifically:

[0101] Step 1: From the acquired standard sample data, determine the background image area belonging to the non-coronary artery in the standard sample data as the first standard background data, and determine the key image area belonging to the coronary artery in the standard sample data as the standard key data.

[0102] Step 2: From the coronary segmentation prediction results, determine the background image area belonging to the non-coronary artery in the coronary segmentation prediction results as the first training background data, and determine the key image area belonging to the coronary artery in the coronary segmentation prediction results as the training key data.

[0103] Step 3: Determine the first background training loss of the original segmentation model for the group of training sample data according to the first background data difference between the first standard background data and the first training background data, and a preset first background loss weight.

[0104] Step 4: Determine the first key training loss of the original segmentation model for the group of training sample data based on the first key data difference between the standard key data and the training key data, and the preset first key loss weight.

[0105] Step 5: Use the first background training loss and the first key training loss as the first model training loss.

[0106] It should be noted that in the above steps 1 to 5, the first key loss weight is greater than the first background loss weight; that is, the key image area belonging to the coronary artery, which is more relevant to the purpose of model training (equivalent to the key area of ​​concern in model training), accounts for a higher proportion in the model training loss, and the background image area belonging to the non-coronary artery, which is less relevant to the purpose of model training, accounts for a lower proportion in the model training loss.

[0107] Based on the implementation of steps 1 to 5 above, different weights can be assigned to the training losses obtained by the model for different image regions (i.e., image regions belonging to coronary arteries / background image regions belonging to non-coronary arteries) according to the degree of correlation between different image regions and the purpose of model training. This is conducive to further improving the training effect of the coronary artery segmentation model compared to the first optional implementation method mentioned above.

[0108] In the specific implementation process of the above step S103, based on the data difference between the training sample data after cutting out and the standard sample data in the whole coronary artery, the original segmentation model is trained with reference to Figure 4 As shown, Figure 4 A flow chart of a second method for training an original segmentation model provided in an embodiment of the present application is shown, the method comprising steps S401-S404; specifically:

[0109] S401, for each group of training sample data, according to the coronary artery centerline marking cutout results in the group of training sample data, train the original segmentation model to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data.

[0110] Here, the execution of step S401 is the same as the execution of the above step S301, and the repeated parts are not repeated here.

[0111] S402, according to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the coronary artery segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​to obtain a global segmentation prediction result of each coronary artery.

[0112] Here, in step S402, the designated cutout position corresponding to each group of training sample data is used to represent the designated cutout position corresponding to each group of random cutout results in step S102 when randomly designated.

[0113] Specifically, in step S402, since each group of random cutout results corresponds to a group of training sample data and a group of standard sample data, as an optional embodiment, the coronary segmentation prediction results of each group of training sample data under the same coronary artery can be directly spliced ​​according to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, and the splicing results are used as the global segmentation prediction results of the coronary artery, so that in the process of predicting each group of random cutout results, the original segmentation model is trained to learn the edge details of the coronary artery, so as to solve the problem that the coronary vessels in the coronary segmentation results are prone to adhesion of the vascular epidermis in dense areas and bifurcations, and improve the segmentation accuracy of the coronary segmentation results in the coronary edge details.

[0114] In step S402 of the embodiment of the present application, considering that the original segmentation model usually concentrates the learned features more on the central area of ​​the image based on the convolution processing in the process of predicting each group of random cutout results, as another optional embodiment, an efficient splicing process can also be performed according to the following steps based on the central area of ​​the coronary segmentation prediction results of each group of training sample data, specifically:

[0115] Step (1), for the coronary segmentation prediction result of each group of training sample data under the same coronary artery, according to the preset image segmentation size, extract the target image area of ​​the image segmentation size from the central area of ​​the coronary segmentation prediction result as the key prediction area of ​​the coronary segmentation of the group of training sample data.

[0116] Here, the central area is used to characterize the concentrated distribution area of ​​the image features in the coronary artery segmentation prediction result.

[0117] Step (2), according to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the key prediction regions of coronary segmentation of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction results of each coronary artery.

[0118] It should be noted that in the embodiment of the present application, for the two optional stitching processing methods in the above step S402, when actually performing the stitching processing, in order to make the connection between the stitched images smoother and more natural, Gaussian smoothing can also be preferably used as the specific stitching processing method; in addition to the Gaussian smoothing method, other stitching processing methods are also applicable to the above two optional stitching processing methods. The embodiment of the present application does not impose any limitation on the specific processing method of the stitching processing.

[0119] S403, determining the second model training loss of the original segmentation model for each coronary artery based on the global segmentation prediction result of the same coronary artery and the coronary artery segmentation label data.

[0120] S404, repeatedly training the original segmentation model using the second model training loss of each coronary artery until the original segmentation model reaches convergence, and using the original segmentation model that has reached convergence as the trained coronary artery segmentation model.

[0121] Regarding the above steps S403-S404, it should be noted that: similar to determining the first model training loss in steps S301-S303, the two optional implementation methods for determining the first model training loss given in the above steps S301-S303 are also applicable to determining the second model training loss in steps S403-S404, and the repetitions will not be repeated here.

[0122] Here, based on the second optional implementation manner (i.e., step 1 to step 5) given in the above steps S301-S303, considering that the distance between each pixel point in the image area where the coronary artery is located and the coronary artery centerline (which can be obtained from the coronary artery centerline marking data) can be further determined, the image area belonging to the edge (such as the distance between the pixel point and any coronary artery centerline is greater than a preset threshold) and the image area belonging to the center (such as the distance between the pixel point and any coronary artery centerline is less than a preset threshold) in the image area where the coronary artery is located can be further determined. Based on this, in addition to the above two optional implementation manners, the embodiment of the present application also provides the following third method for determining the model training loss, specifically:

[0123] Reference Figure 5 As shown, Figure 5 A flow chart of a method for determining a second model training loss provided in an embodiment of the present application is shown, the method comprising steps S501-S506; specifically:

[0124] S501, from the coronary segmentation labeling data of each coronary artery, determine the background image area belonging to the non-coronary artery in the coronary segmentation labeling data as the second standard background data, determine the image area belonging to the edge of the coronary artery in the coronary segmentation labeling data as the standard edge data, and determine the image area belonging to the center of the coronary artery in the coronary segmentation labeling data as the standard center data.

[0125] Here, the standard edge data and the standard center data constitute the image area belonging to the coronary artery in the coronary artery segmentation mark data, that is, the standard edge data and the standard center data are equivalent to regional division of the image area belonging to the coronary artery in the coronary artery segmentation mark data based on the relative size relationship between the distance between the pixel point and the midline of any coronary artery and the preset distance threshold.

[0126] For example, taking the preset distance threshold of 1 mm as an example, if the distance between the pixel point x and each coronary artery centerline in the coronary artery centerline marking data is greater than 1 mm, it can be determined that the pixel point x belongs to the image area at the edge of the coronary artery; if the distance between the pixel point x and the coronary artery centerline L1 in the coronary artery centerline marking data is less than 1 mm, it can be determined that the pixel point x belongs to the image area at the center of the coronary artery.

[0127] S502, from the global segmentation prediction result of the same coronary artery, determine the background image area belonging to the non-coronary artery in the global segmentation prediction result as the second training background data, determine the image area belonging to the edge of the coronary artery in the global segmentation prediction result as the training edge data, and determine the image area belonging to the center of the coronary artery in the global segmentation prediction result as the training center data.

[0128] Here, the training edge data and the training center data constitute the image region belonging to the coronary artery in the global segmentation prediction result.

[0129] Specifically, the implementation of step S502 is the same as that of step S501, and the repeated parts are not repeated here.

[0130] S503: Determine a second background training loss of the original segmentation model for the coronary artery according to a second background data difference between the second standard background data and the second training background data, and a preset second background loss weight.

[0131] S504, determining the edge training loss of the original segmentation model for the coronary artery according to the edge data difference between the standard edge data and the training edge data, and a preset edge loss weight.

[0132] S505, determining the center training loss of the original segmentation model for the coronary artery according to the center data difference between the standard center data and the training center data, and a preset center loss weight.

[0133] Here, the center loss weight is smaller than the edge loss weight, and the center loss weight is larger than the second background loss weight, that is, since the correlation between the image area belonging to the edge of the coronary artery and the training purpose of the model is the highest, the correlation between the image area belonging to the center of the coronary artery and the training purpose of the model is second, and the correlation between the background image area belonging to the non-coronary artery and the training purpose of the model is the lowest, therefore, similar to the weight allocation method of the above steps 1 to 5, different weights are allocated to the training losses obtained by the model for different image areas (i.e., the image area belonging to the edge of the coronary artery / the image area belonging to the center of the coronary artery / the background image area belonging to the non-coronary artery) according to the correlation between different image areas and the training purpose of the model. Therefore, compared with the second optional implementation shown in the above steps 1 to 5, it is beneficial to further improve the training effect of the coronary artery segmentation model.

[0134] S506: Use the second background training loss, the edge training loss, and the center training loss as the second model training loss.

[0135] Regarding the above steps S501-S506, it should be noted that the implementation principles of the above steps S501-S506 are the same as the second optional implementation method (ie, steps 1-5) given in the above steps S301-S303, and the repeated parts will not be repeated here.

[0136] The training method for the above-mentioned coronary segmentation model provided in an embodiment of the present application obtains medical image data, coronary centerline marking data and coronary segmentation marking data of the same coronary artery as a group of initial learning data; for each group of initial learning data, at the same randomly specified position, according to a preset cutout size, the medical image data, coronary centerline marking data and coronary segmentation marking data in the group of initial learning data are cutout to obtain multiple groups of random cutout results; for each group of random cutout results, the medical image cutout results and coronary centerline marking cutout results in the group of random cutout results are used as a group of dual-channel training sample data, and the coronary segmentation marking cutout results in the group of random cutout results are used as standard sample data in the same group as the training sample data; according to the data difference between the training sample data and the standard sample data in the same group, the original segmentation model is trained to obtain a trained coronary segmentation model.

[0137] Through the above method, the present application can solve the problem that coronary vessels in the coronary segmentation results are prone to adhesion of the vascular epithelium at dense areas and bifurcations, and improve the segmentation accuracy of the coronary segmentation results in the details of the coronary edge.

[0138] In the embodiment of the present application, after the coronary segmentation model is trained, in the application stage of the coronary segmentation model, the embodiment of the present application also provides the following coronary segmentation method, specifically:

[0139] Reference Figure 6 As shown, Figure 6 A schematic flow chart of a coronary artery segmentation method provided in an embodiment of the present application is shown, the method comprising steps S601-S604; specifically:

[0140] S601, obtaining target medical image data of the coronary artery to be segmented and target coronary artery centerline marking data.

[0141] Here, the execution method of step S601 is the same as that of step S101, and the repeated parts are not repeated here.

[0142] It should be noted that, unlike step S101 which also requires obtaining coronary segmentation mark data, since step S601 belongs to the application step of the coronary segmentation model, it is not necessary to obtain coronary segmentation mark data in step S601.

[0143] S602 , according to each blood vessel centerline segment included in the target coronary artery centerline marking data, specify any position including the blood vessel centerline segment from the target medical image data as a target cutout position.

[0144] Here, the target coronary artery centerline marking data includes multiple vascular centerline segments, each vascular centerline segment corresponds to the vascular centerline of a coronary artery in the coronary artery to be segmented; wherein each vascular centerline segment has at least two endpoints (i.e., the first endpoint and the tail endpoint of the segment).

[0145] It should be noted that, in step S602, in addition to the optional implementation of randomly specifying the target cutout position, in the embodiment of the present application, as another optional implementation, specifying any position including the blood vessel midline segment in the target medical image data as the target cutout position further includes:

[0146] From the line segment endpoints of each of the blood vessel midline segments, specify any line segment endpoint as the first starting designated position of the target cutout position;

[0147] In the target coronary artery centerline marking data, taking the first starting designated position as the cutout center point, and predicting the region boundary where the cutout region is located according to the preset cutout size;

[0148] When it is detected that the region boundary and any blood vessel midline segment included in the target coronary artery midline mark data have an intersection, the intersection is designated as the next designated position of the target cutout position;

[0149] When it is detected that the region boundary does not intersect with any vascular centerline segment included in the target coronary artery centerline marking data, any segment endpoint is designated as the second starting designated position of the target cutout position from the segment endpoints of other vascular centerline segments except the target vascular centerline segment, and the step of determining the next designated position of the target cutout position is repeated; wherein the target vascular centerline segment is used to characterize the vascular centerline segment corresponding to the first starting designated position.

[0150] S603, at each designated target cutout position, the target medical image data and the target coronary artery midline marking data are cutout according to a preset cutout size to obtain multiple groups of target cutout results.

[0151] Here, each group of target cutout results includes target medical image cutout results of the same size and position and target coronary artery midline marking cutout results.

[0152] Specifically, as an optional embodiment, at each designated target cutout position, the target cutout position is used as the center point of the cutout area, and the target medical image data and the target coronary artery centerline marking data are cutout according to the cutout size (e.g., Figure 2 cutout method shown).

[0153] S604, inputting the multiple groups of target cutout results into a pre-trained coronary artery segmentation model, and using the output result of the coronary artery segmentation model as the coronary artery segmentation result of the coronary artery to be segmented.

[0154] Here, the coronary artery segmentation model is obtained after training according to any of the optional implementations of the above-mentioned coronary artery segmentation model training method, and the specific training method is not repeated here.

[0155] Specifically, considering the influence of repeated prediction on the accuracy of the model output result, as an optional embodiment, the step of inputting the multiple sets of target cutout results into a pre-trained coronary artery segmentation model further includes:

[0156] From the multiple groups of target cutout results, duplicate target cutout results are deleted to obtain multiple groups of key cutout results that can represent the global image information of the coronary artery to be segmented;

[0157] The multiple groups of key cutout results are input into a pre-trained coronary artery segmentation model.

[0158] It should be noted that, in the actual application of the coronary artery segmentation model, the "method of performing cutout based on target coronary artery centerline marking data" described in another optional implementation manner in the above steps S602-S603 is only used to improve the model output efficiency of the coronary artery prediction model for the model output results (i.e., the coronary artery segmentation results of the coronary arteries to be segmented) on the basis of randomly specifying the cutout position and obtaining the model input data (i.e., steps S102-S103). That is, in addition to the method for obtaining the target cutout results described in the above steps S602-S603, the method of obtaining the model input data (i.e., the medical image cutout results of the same size at the same position in multiple groups of random cutout results and the coronary artery centerline marking cutout results) by random cutout in the above steps S102-S103 is still applicable to the application process of the coronary artery segmentation model.

[0159] Through the above-mentioned coronary artery segmentation method provided in the embodiment of the present application, after training the coronary artery segmentation model, the present application can use the coronary artery segmentation model to perform coronary artery segmentation on the target medical image data of the coronary artery to be segmented with the help of the target coronary artery midline marking data of the coronary artery to be segmented, which is beneficial to improve the segmentation accuracy of the coronary artery segmentation result in the coronary artery edge details learned by the coronary artery segmentation model on the image processing side.

[0160] Based on the same inventive concept, the embodiments of the present application also provide a training device for a coronary artery segmentation model corresponding to the training method for the coronary artery segmentation model in the above embodiments. Since the principle of solving the problem by the training device in the embodiments of the present application is similar to the training method in the above embodiments of the present application, the implementation of the training device can refer to the implementation of the above-mentioned training method, and the repeated parts will not be repeated.

[0161] Reference Figure 7 As shown, Figure 7 A schematic diagram of the structure of a training device for a coronary artery segmentation model provided in an embodiment of the present application is shown; the training device comprises:

[0162] A first acquisition module 701 is used to acquire medical image data, coronary artery centerline marking data and coronary artery segmentation marking data of the same coronary artery as a set of initial learning data;

[0163] The first cutout module 702 is used to cutout the medical image data, the coronary artery centerline marking data and the coronary artery segmentation marking data in each group of initial learning data at the same randomly designated position according to a preset cutout size to obtain multiple groups of random cutout results; wherein each group of random cutout results includes medical image cutout results, coronary artery centerline marking cutout results and coronary artery segmentation marking cutout results of the same size at the same position;

[0164] The training module 703 is used to, for each group of random cutout results, use the medical image cutout results and the coronary artery centerline marking cutout results in the group of random cutout results as a group of dual-channel training sample data, and use the coronary artery segmentation marking cutout results in the group of random cutout results as standard sample data in the same group as the training sample data, and train the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary artery segmentation model.

[0165] In an optional implementation, when the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group to obtain the trained coronary segmentation model, the training module 703 is used to:

[0166] For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data;

[0167] Acquire standard sample data from the same group as the group of training sample data, and determine a first model training loss of the original segmentation model for the group of training sample data based on the acquired standard sample data and the coronary artery segmentation prediction result;

[0168] The original segmentation model is repeatedly trained using the first model training loss of each group of training sample data until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

[0169] In an optional implementation, when the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group to obtain the trained coronary segmentation model, the training module 703 is further used to:

[0170] For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data;

[0171] According to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the coronary artery segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction result of each coronary artery;

[0172] Determining a second model training loss of the original segmentation model for each coronary artery according to the global segmentation prediction result of the same coronary artery and the coronary artery segmentation label data;

[0173] The original segmentation model is repeatedly trained using the second model training loss of each coronary artery until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

[0174] In an optional implementation, when the coronary segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​according to the designated cutout positions corresponding to each group of training sample data under the same coronary artery to obtain the global segmentation prediction result of each coronary artery, the training module 703 is used to:

[0175] For the coronary segmentation prediction results of each group of training sample data under the same coronary artery, according to the preset image segmentation size, from the central area of ​​the coronary segmentation prediction results, a target image area of ​​the image segmentation size is extracted as the key prediction area for coronary segmentation of the group of training sample data; wherein the central area is used to characterize the concentrated distribution area of ​​the image features in the coronary segmentation prediction results;

[0176] According to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the key prediction regions of coronary segmentation of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction results of each coronary artery.

[0177] In an optional implementation, when determining the first model training loss of the original segmentation model for the group of training sample data based on the acquired standard sample data and the coronary segmentation prediction result, the training module 703 is used to:

[0178] From the acquired standard sample data, determine a background image region belonging to non-coronary arteries in the standard sample data as first standard background data, and determine a key image region belonging to coronary arteries in the standard sample data as standard key data;

[0179] From the coronary segmentation prediction result, determine a background image region belonging to a non-coronary artery in the coronary segmentation prediction result as first training background data, and determine a key image region belonging to a coronary artery in the coronary segmentation prediction result as training key data;

[0180] Determining a first background training loss of the original segmentation model for the group of training sample data according to a first background data difference between the first standard background data and the first training background data, and a preset first background loss weight;

[0181] Determine the first key training loss of the original segmentation model for the group of training sample data according to the first key data difference between the standard key data and the training key data, and a preset first key loss weight; wherein the first key loss weight is greater than the first background loss weight;

[0182] The first background training loss and the first key training loss are used as the first model training loss.

[0183] In an optional implementation, when determining the second model training loss of the original segmentation model for each coronary artery based on the global segmentation prediction result and the coronary artery segmentation label data of the same coronary artery, the training module 703 is used to:

[0184] From the coronary segmentation mark data of each coronary artery, determine the background image area belonging to the non-coronary artery in the coronary segmentation mark data as the second standard background data, determine the image area belonging to the edge of the coronary artery in the coronary segmentation mark data as the standard edge data, and determine the image area belonging to the center of the coronary artery in the coronary segmentation mark data as the standard center data; wherein the standard edge data and the standard center data constitute the image area belonging to the coronary artery in the coronary segmentation mark data;

[0185] From the global segmentation prediction result of the same coronary artery, determine the background image area belonging to the non-coronary artery in the global segmentation prediction result as the second training background data, determine the image area belonging to the edge of the coronary artery in the global segmentation prediction result as the training edge data, and determine the image area belonging to the center of the coronary artery in the global segmentation prediction result as the training center data; wherein the training edge data and the training center data constitute the image area belonging to the coronary artery in the global segmentation prediction result;

[0186] Determining a second background training loss of the original segmentation model for the coronary artery according to a second background data difference between the second standard background data and the second training background data, and a preset second background loss weight;

[0187] Determining the edge training loss of the original segmentation model for the coronary artery according to the edge data difference between the standard edge data and the training edge data, and a preset edge loss weight;

[0188] Determining the center training loss of the original segmentation model for the coronary artery according to the center data difference between the standard center data and the training center data, and a preset center loss weight; wherein the center loss weight is less than the edge loss weight, and the center loss weight is greater than the second background loss weight;

[0189] The second background training loss, the edge training loss, and the center training loss are used as the second model training loss.

[0190] Based on the same inventive concept, the present application also provides a coronary artery segmentation device corresponding to the above-mentioned coronary artery segmentation method. Since the principle of solving the problem by the coronary artery segmentation device in the embodiment of the present application is similar to the above-mentioned coronary artery segmentation method in the embodiment of the present application, the implementation of the coronary artery segmentation device can refer to the implementation of the coronary artery segmentation method, and the repeated parts will not be repeated.

[0191] Reference Figure 8 As shown, Figure 8 A schematic diagram of the structure of a coronary artery segmentation device provided in an embodiment of the present application is shown, wherein the coronary artery segmentation device comprises:

[0192] The second acquisition module 801 is used to acquire the target medical image data of the coronary artery to be segmented and the target coronary artery centerline marking data;

[0193] The position determination module 802 is used to designate any position including the vascular centerline segment from the target medical image data as a target cutout position according to each vascular centerline segment included in the target coronary artery centerline mark data;

[0194] The second cutout module 803 is used to cut out the target medical image data and the target coronary artery centerline marking data at each designated target cutout position according to a preset cutout size to obtain multiple groups of target cutout results; wherein each group of target cutout results includes a target medical image cutout result and a target coronary artery centerline marking cutout result of the same size at the same position;

[0195] The segmentation prediction module 804 is used to input the multiple groups of target cutout results into a pre-trained coronary segmentation model, and use the output result of the coronary segmentation model as the coronary segmentation result of the coronary artery to be segmented.

[0196] In an optional implementation, in the segmentation prediction module 804, the coronary segmentation model is obtained after training according to any of the optional implementations of the above-mentioned coronary segmentation model training method.

[0197] In an optional implementation, when any position including the blood vessel midline segment is designated as the target cutout position in the target medical image data, the position determination module 802 is further configured to:

[0198] From the line segment endpoints of each of the blood vessel midline segments, specify any line segment endpoint as the first starting designated position of the target cutout position;

[0199] In the target coronary artery centerline marking data, taking the first starting designated position as the cutout center point, and predicting the region boundary where the cutout region is located according to the preset cutout size;

[0200] When it is detected that the region boundary and any blood vessel midline segment included in the target coronary artery midline mark data have an intersection, the intersection is designated as the next designated position of the target cutout position;

[0201] When it is detected that the region boundary does not intersect with any vascular centerline segment included in the target coronary artery centerline marking data, any segment endpoint is designated as the second starting designated position of the target cutout position from the segment endpoints of other vascular centerline segments except the target vascular centerline segment, and the step of determining the next designated position of the target cutout position is repeated; wherein the target vascular centerline segment is used to characterize the vascular centerline segment corresponding to the first starting designated position.

[0202] In an optional implementation, when the target medical image data and the target coronary artery centerline marking data are cut out at each designated target cutout position according to a preset cutout size, the second cutout module 803 is used to:

[0203] At each designated target cutout position, the target cutout position is used as the center point of the cutout region, and the target medical image data and the target coronary artery centerline marking data are cutout according to the cutout size.

[0204] In an optional implementation, when the plurality of target cutout results are input into a pre-trained coronary segmentation model, the segmentation prediction module 804 is used to:

[0205] From the multiple groups of target cutout results, duplicate target cutout results are deleted to obtain multiple groups of key cutout results that can represent the global image information of the coronary artery to be segmented;

[0206] The multiple groups of key cutout results are input into a pre-trained coronary artery segmentation model.

[0207] like Fig. 9 As shown, an embodiment of the present application provides a computer device 900 for executing the steps of any method for training a coronary artery segmentation model described in the present application, or executing the steps of any of the above-mentioned coronary artery segmentation methods. The device includes a memory 901, a processor 902, and a computer program stored in the memory 901 and executable on the processor 902, wherein the processor 902 implements the steps of any of the above-mentioned methods for training a coronary artery segmentation model, or executes the steps of any of the above-mentioned methods for coronary artery segmentation when executing the computer program.

[0208] Specifically, the above-mentioned memory 901 and processor 902 can be general-purpose memory and processor, which are not specifically limited here. When the processor 902 runs the computer program stored in the memory 901, it can execute the steps of any of the above-mentioned coronary artery segmentation model training methods, or execute the steps of any of the above-mentioned coronary artery segmentation methods.

[0209] Corresponding to the training method of the coronary artery segmentation model and the coronary artery segmentation method in the present application, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of any of the above-mentioned training methods for the coronary artery segmentation model, or executes the steps of any of the above-mentioned coronary artery segmentation methods.

[0210] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, it can execute the steps of any of the above-mentioned coronary artery segmentation model training methods, or execute the steps of any of the above-mentioned coronary artery segmentation methods.

[0211] In the embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0212] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0213] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0214] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0215] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0216] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for training a coronary artery segmentation model, characterized in that: The training method comprises: Obtain medical imaging data of the same coronary artery, coronary artery centerline marking data, and coronary artery segmentation marking data as a set of initial learning data; For each group of initial learning data, at the same randomly designated position, according to a preset cutout size, the medical image data, the coronary artery centerline marking data, and the coronary artery segmentation marking data in the group of initial learning data are cutout to obtain multiple groups of random cutout results; wherein each group of random cutout results includes medical image cutout results, coronary artery centerline marking cutout results, and coronary artery segmentation marking cutout results of the same size at the same position; For each group of random cutout results, the medical image cutout results and the coronary artery midline mark cutout results in the group of random cutout results are used as a group of dual-channel training sample data, and the coronary artery segmentation mark cutout results in the group of random cutout results are used as standard sample data in the same group as the training sample data, and the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary artery segmentation model; The method of training the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain the trained coronary segmentation model includes: For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data; Acquire standard sample data from the same group as the group of training sample data, and determine a first model training loss of the original segmentation model for the group of training sample data based on the acquired standard sample data and the coronary artery segmentation prediction result; The original segmentation model is repeatedly trained using the first model training loss of each group of training sample data until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

2. The training method according to claim 1, characterized in that: The method further comprises: training the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary segmentation model; For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data; According to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the coronary artery segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction result of each coronary artery; Determining a second model training loss of the original segmentation model for each coronary artery according to the global segmentation prediction result of the same coronary artery and the coronary artery segmentation label data; The original segmentation model is repeatedly trained using the second model training loss of each coronary artery until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

3. The training method according to claim 2, characterized in that: The coronary segmentation prediction results of each group of training sample data under the same coronary artery are spliced ​​according to the designated cutout positions corresponding to each group of training sample data under the same coronary artery to obtain the global segmentation prediction results of each coronary artery, and further includes: For the coronary segmentation prediction results of each group of training sample data under the same coronary artery, according to the preset image segmentation size, from the central area of ​​the coronary segmentation prediction results, a target image area of ​​the image segmentation size is extracted as the key prediction area for coronary segmentation of the group of training sample data; wherein the central area is used to characterize the concentrated distribution area of ​​the image features in the coronary segmentation prediction results; According to the designated cutout positions corresponding to each group of training sample data under the same coronary artery, the key prediction regions of coronary segmentation of each group of training sample data under the same coronary artery are spliced ​​to obtain the global segmentation prediction results of each coronary artery.

4. The training method according to claim 1, characterized in that: Determining the first model training loss of the original segmentation model for the group of training sample data according to the acquired standard sample data and the coronary artery segmentation prediction result includes: From the acquired standard sample data, determine a background image region belonging to non-coronary arteries in the standard sample data as first standard background data, and determine a key image region belonging to coronary arteries in the standard sample data as standard key data; From the coronary segmentation prediction result, determine a background image region belonging to a non-coronary artery in the coronary segmentation prediction result as first training background data, and determine a key image region belonging to a coronary artery in the coronary segmentation prediction result as training key data; Determining a first background training loss of the original segmentation model for the group of training sample data according to a first background data difference between the first standard background data and the first training background data, and a preset first background loss weight; Determine the first key training loss of the original segmentation model for the group of training sample data according to the first key data difference between the standard key data and the training key data, and a preset first key loss weight; wherein the first key loss weight is greater than the first background loss weight; The first background training loss and the first key training loss are used as the first model training loss.

5. The training method according to claim 2, characterized in that: The step of determining the second model training loss of the original segmentation model for each coronary artery based on the global segmentation prediction result of the same coronary artery and the coronary artery segmentation label data comprises: From the coronary segmentation mark data of each coronary artery, determine the background image area belonging to the non-coronary artery in the coronary segmentation mark data as the second standard background data, determine the image area belonging to the edge of the coronary artery in the coronary segmentation mark data as the standard edge data, and determine the image area belonging to the center of the coronary artery in the coronary segmentation mark data as the standard center data; wherein the standard edge data and the standard center data constitute the image area belonging to the coronary artery in the coronary segmentation mark data; From the global segmentation prediction result of the same coronary artery, determine the background image area belonging to the non-coronary artery in the global segmentation prediction result as the second training background data, determine the image area belonging to the edge of the coronary artery in the global segmentation prediction result as the training edge data, and determine the image area belonging to the center of the coronary artery in the global segmentation prediction result as the training center data; wherein the training edge data and the training center data constitute the image area belonging to the coronary artery in the global segmentation prediction result; Determining a second background training loss of the original segmentation model for the coronary artery according to a second background data difference between the second standard background data and the second training background data, and a preset second background loss weight; Determining the edge training loss of the original segmentation model for the coronary artery according to the edge data difference between the standard edge data and the training edge data, and a preset edge loss weight; Determining the center training loss of the original segmentation model for the coronary artery according to the center data difference between the standard center data and the training center data, and a preset center loss weight; wherein the center loss weight is less than the edge loss weight, and the center loss weight is greater than the second background loss weight; The second background training loss, the edge training loss, and the center training loss are used as the second model training loss.

6. A coronary artery segmentation method, characterized in that: The coronary artery segmentation method comprises: Acquire target medical image data of the coronary artery to be segmented and target coronary artery centerline marking data; According to each blood vessel midline segment included in the target coronary artery midline marking data, specifying any position including the blood vessel midline segment from the target medical image data as a target cutout position; At each designated target cutout position, the target medical image data and the target coronary artery centerline marking data are cutout according to a preset cutout size to obtain multiple groups of target cutout results; wherein each group of target cutout results includes a target medical image cutout result and a target coronary artery centerline marking cutout result of the same size at the same position; The multiple groups of target cutout results are input into a pre-trained coronary artery segmentation model, and the output result of the coronary artery segmentation model is used as the coronary artery segmentation result of the coronary artery to be segmented; wherein the coronary artery segmentation model is obtained after training according to any one of the training methods described in claims 1 to 5.

7. The coronary artery segmentation method according to claim 6, characterized in that: The step of specifying any position including the blood vessel midline segment in the target medical image data as a target cutout position further includes: From the line segment endpoints of each of the blood vessel midline segments, specify any line segment endpoint as the first starting designated position of the target cutout position; In the target coronary artery centerline marking data, taking the first starting designated position as the cutout center point, and predicting the region boundary where the cutout region is located according to the preset cutout size; When it is detected that the region boundary and any blood vessel midline segment included in the target coronary artery midline mark data have an intersection, the intersection is designated as the next designated position of the target cutout position; When it is detected that the region boundary does not intersect with any vascular centerline segment included in the target coronary artery centerline marking data, any segment endpoint is designated as the second starting designated position of the target cutout position from the segment endpoints of other vascular centerline segments except the target vascular centerline segment, and the step of determining the next designated position of the target cutout position is repeated; wherein the target vascular centerline segment is used to characterize the vascular centerline segment corresponding to the first starting designated position.

8. The coronary artery segmentation method according to claim 7, characterized in that: The step of performing cutout processing on the target medical image data and the target coronary artery centerline marking data at each designated target cutout position according to a preset cutout size includes: At each designated target cutout position, the target cutout position is used as the center point of the cutout region, and the target medical image data and the target coronary artery centerline marking data are cutout according to the cutout size.

9. The coronary artery segmentation method according to claim 6, characterized in that: The step of inputting the plurality of target image cutout results into a pre-trained coronary segmentation model further includes: From the multiple groups of target cutout results, duplicate target cutout results are deleted to obtain multiple groups of key cutout results that can represent the global image information of the coronary artery to be segmented; The multiple groups of key cutout results are input into a pre-trained coronary artery segmentation model.

10. A training device for a coronary artery segmentation model, characterized in that: The training device comprises: A first acquisition module is used to acquire medical image data of the same coronary artery, coronary artery centerline marking data and coronary artery segmentation marking data as a set of initial learning data; A first cutout module is used to cutout the medical image data, the coronary artery centerline marking data and the coronary artery segmentation marking data in each group of initial learning data at the same randomly designated position according to a preset cutout size to obtain multiple groups of random cutout results; wherein each group of random cutout results includes medical image cutout results, coronary artery centerline marking cutout results and coronary artery segmentation marking cutout results of the same size at the same position; a training module, for taking, for each group of random cutout results, the medical image cutout results and the coronary artery midline mark cutout results in the group of random cutout results as a group of dual-channel training sample data, taking the coronary artery segmentation mark cutout results in the group of random cutout results as standard sample data in the same group as the training sample data, and training the original segmentation model according to the data difference between the training sample data and the standard sample data in the same group to obtain a trained coronary artery segmentation model; Wherein, when the original segmentation model is trained according to the data difference between the training sample data and the standard sample data in the same group to obtain the trained coronary segmentation model, the training module is used to: For each group of training sample data, according to the coronary artery midline marking cutout results in the group of training sample data, the original segmentation model is trained to perform coronary artery segmentation prediction on the medical image cutout results in the group of training sample data to obtain the coronary artery segmentation prediction results of the group of training sample data; Acquire standard sample data from the same group as the group of training sample data, and determine a first model training loss of the original segmentation model for the group of training sample data based on the acquired standard sample data and the coronary artery segmentation prediction result; The original segmentation model is repeatedly trained using the first model training loss of each group of training sample data until the original segmentation model reaches convergence, and the original segmentation model that reaches convergence is used as the trained coronary artery segmentation model.

11. A coronary artery segmentation device, characterized in that: The coronary artery segmentation device comprises: The second acquisition module is used to acquire the target medical image data of the coronary artery to be segmented and the target coronary artery centerline marking data; a position determination module, for specifying any position including the vascular centerline segment from the target medical image data as a target cutout position according to each vascular centerline segment included in the target coronary artery centerline marking data; A second cutout module is used to cut out the target medical image data and the target coronary artery centerline marking data at each designated target cutout position according to a preset cutout size to obtain multiple groups of target cutout results; wherein each group of target cutout results includes a target medical image cutout result and a target coronary artery centerline marking cutout result of the same size at the same position; A segmentation prediction module is used to input the multiple groups of target cutout results into a pre-trained coronary artery segmentation model, and use the output result of the coronary artery segmentation model as the coronary artery segmentation result of the coronary artery to be segmented; wherein the coronary artery segmentation model is obtained after training according to any one of the training methods described in claims 1 to 5.

12. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the training method for the coronary artery segmentation model as described in any one of claims 1 to 5 are executed, or the steps of the coronary artery segmentation method as described in any one of claims 6 to 9 are executed.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the coronary artery segmentation model training method as described in any one of claims 1 to 5, or executes the steps of the coronary artery segmentation method as described in any one of claims 6 to 9.

Images