Triangulation-based left ventricular motion prediction method, system, device, and media

Abstract

The application provides a left ventricular motion prediction method, system, device and medium based on triangulation, relates to the technical field of medical image processing, and comprises the following steps: acquiring a left ventricular endocardial contour image, and extracting a key point set and coordinate information thereof from the left ventricular endocardial contour image; processing the left ventricular endocardial contour image by using a VGG16 network model to extract feature information of the key point set; constructing a graph structure by combining a triangulation method to capture structure information of the key point set; aggregating features by using a graph convolutional neural network model to obtain embedding information of the key point set; performing feature matching by using a Sinkhorn algorithm to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour; and finally, predicting the motion of the left ventricle according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour to obtain a left ventricular motion prediction result. By combining the triangulation method and the Sinkhorn algorithm, the application constructs structure information between the key points, improves the accuracy of feature matching, and thus improves the accuracy of left ventricular motion prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a left ventricular motion prediction method, system, device and medium based on triangulation. BACKGROUND

[0002] Cardiac medical image processing technology plays a crucial role in the early diagnosis, treatment planning and disease monitoring of heart disease, and can provide key information about the structure and functional status of the heart. However, existing technologies such as generalized robust point matching, iterative closest point and robust point matching based on thin plate splines, while effective in some cases, often emphasize point features and ignore the topological relationship between key points, resulting in abnormal correspondence prediction between myocardial contours. In addition, image registration-based methods can integrate local and global information, but are computationally intensive and time-consuming when dealing with high-dimensional data; feature tracking-based methods may fall into local optimal solutions; graph matching-based methods may ignore the global geometric structure while reducing computational cost. Therefore, existing technologies face challenges in processing cardiac images, and there is an urgent need for more accurate and robust solutions to improve the accuracy of left ventricular motion prediction based on triangulation. SUMMARY

[0003] The present application provides a left ventricular motion prediction method, system, device and medium based on triangulation to solve one or more technical problems existing in the prior art and at least provide a beneficial choice or create conditions.

[0004] In one aspect, the present application provides a left ventricular motion prediction method based on triangulation, comprising the following steps:

[0005] Obtaining a left ventricular endocardial contour image;

[0006] Obtaining coordinate information of a key point set from the left ventricular endocardial contour image; the key point set includes a diastolic end key point subset and a systolic end key point subset;

[0007] Performing feature processing on the key point set using a VGG16 network model based on the left ventricular endocardial contour image and the coordinate information of the key point set to obtain feature information of the key point set;

[0008] Performing graph construction processing on the key point set using a triangulation method based on the coordinate information and feature information of the key point set to obtain structure information of the key point set;

[0009] Performing feature aggregation on the key point set using a graph convolutional neural network model based on the structure information and feature information of the key point set to obtain embedding information of the key point set;

[0010] According to the embedding information and the coordinate information of the key point set, feature matching is performed in combination with a Sinkhorn algorithm to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour.

[0011] According to the diastolic-systolic deformation relationship of the left ventricular endocardial contour, diastolic-systolic motion of the left ventricular endocardial contour is predicted to obtain a left ventricular motion prediction result.

[0012] Further, according to the left ventricular endocardial contour image and the coordinate information of the key point set, feature processing is performed by using a VGG16 network model to obtain feature information of the key point set, including:

[0013] According to the left ventricular endocardial contour image, feature processing is performed by using the VGG16 network model to obtain an initial feature tensor.

[0014] According to the initial feature tensor and the coordinate information of the key point set, the feature information of the key point set is obtained in combination with a bilinear interpolation method.

[0015] Further, according to the coordinate information and the feature information of the key point set, graph construction processing is performed by using a triangulation method to obtain structure information of the key point set, including:

[0016] According to the coordinate information and the feature information of the key point set, a distance matrix is obtained; the distance matrix is used to represent distances between the key points in the key point set;

[0017] According to the distance matrix, the weighted adjacency matrix of the key point set is obtained as structure information in combination with the triangulation method; the triangulation method is used to establish a connection between key point pairs in the key point set with a distance of not more than two units.

[0018] Further, according to the structure information and the feature information of the key point set, feature aggregation is performed by using a graph convolutional neural network model to obtain embedding information of the key point set, including:

[0019] According to the structure information and the feature information of the key point set, intra-graph aggregation features of the key point set are obtained by using an intra-graph convolution mechanism of the graph convolutional neural network model;

[0020] According to the structure information and the feature information of the key point set, cross-graph aggregation features of the key point set are obtained by using a cross-graph convolution mechanism of the graph convolutional neural network model in combination with a similarity matrix;

[0021] According to the intra-graph aggregation features and the cross-graph aggregation features of the key point set, embedding information of the key point set is obtained.

[0022] Further, the embedding information of the key point set comprises diastolic end embedding information and systolic end embedding information.

[0023] The feature matching is performed according to the embedding information and the coordinate information of the key point set in combination with a Sinkhorn algorithm to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour, comprising:

[0024] A similarity between the diastolic end embedding information and the systolic end embedding information is calculated to obtain a score matrix.

[0025] The feature matching is performed according to the score matrix in combination with the Sinkhorn algorithm to obtain a matching matrix.

[0026] The diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the matching matrix and comprehensive information of the key point set.

[0027] Further, the feature matching is performed according to the score matrix in combination with the Sinkhorn algorithm to obtain a matching matrix, comprising:

[0028] The score matrix is iteratively normalized by using the Sinkhorn algorithm to obtain a convergent double random matrix.

[0029] The matching matrix is obtained according to the convergent double random matrix; only one element in each row and each column of the matching matrix is 1, and all other elements are 0.

[0030] The matching matrix is optimized by using a minimum negative logarithmic loss.

[0031] Further, the diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the matching matrix and comprehensive information of the key point set, comprising:

[0032] A deformation function is obtained according to the matching matrix and the comprehensive information of the key point set.

[0033] The diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the deformation function.

[0034] In another aspect, the present application provides a left ventricular motion prediction system based on triangulation, comprising a data acquisition module, a coordinate acquisition module, a feature processing module, a graph construction module, a feature aggregation module, a feature matching module and a deformation prediction module.

[0035] The data acquisition module is configured to acquire left ventricular endocardial contour images.

[0036] The coordinate acquisition module is configured to obtain coordinate information of a key point set according to the left ventricular endocardial contour image; the key point set includes a diastolic end key point subset and a systolic end key point subset;

[0037] The feature processing module is configured to perform feature processing on the key point set by using a VGG16 network model according to the left ventricular endocardial contour image and the coordinate information of the key point set, to obtain feature information of the key point set.

[0038] The graph construction module is configured to perform graph construction processing on the key point set by using a triangulation method according to the coordinate information and the feature information of the key point set, to obtain structure information of the key point set.

[0039] The feature aggregation module is configured to perform feature aggregation on the key point set by using a graph convolutional neural network model according to the structure information and the feature information of the key point set, to obtain embedding information of the key point set.

[0040] The feature matching module is configured to perform feature matching on the key point set by using a Sinkhorn algorithm according to the embedding information and the coordinate information of the key point set, to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour.

[0041] The deformation prediction module is configured to predict diastolic-systolic motion of the left ventricular endocardial contour according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour, to obtain a left ventricular motion prediction result.

[0042] In another aspect, the present application provides a left ventricular motion prediction device based on triangulation, including a processor and a memory; the memory is configured to store a program;

[0043] When the program is executed by the processor, the processor implements the aforementioned left ventricular motion prediction method based on triangulation.

[0044] In another aspect, the present application provides a computer readable storage medium, which stores a processor executable program; the processor executable program is used to implement the aforementioned left ventricular motion prediction method based on triangulation when executed by a processor.

[0045] The beneficial effects of the present application are: the present application provides a left ventricular motion prediction method based on triangulation, which comprises: acquiring a left ventricular endocardial contour image; obtaining coordinate information of a key point set according to the left ventricular endocardial contour image; the key point set comprises a diastolic end key point subset and a systolic end key point subset; according to the left ventricular endocardial contour image and the coordinate information of the key point set, using a VGG16 network model for feature processing to obtain feature information of the key point set; according to the coordinate information and the feature information of the key point set, combining a triangulation method for graph construction processing to obtain structure information of the key point set; according to the structure information and the feature information of the key point set, using a graph convolutional neural network model for feature aggregation to obtain embedding information of the key point set; according to the embedding information and the coordinate information of the key point set, combining a Sinkhorn algorithm for feature matching to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour; according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour, predicting the diastolic-systolic motion of the left ventricular endocardial contour to obtain a left ventricular motion prediction result. The present application combines the triangulation method and the Sinkhorn algorithm to construct the structure information between the key points, improves the accuracy of feature matching, and thus improves the accuracy of left ventricular motion prediction. The present application also provides corresponding devices, systems and media, and the beneficial effects of the devices, systems and media are similar to those of the method, which will not be described here.

[0046] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and achieved by the structure particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0047] The accompanying drawings are intended to provide a further understanding of the technical solutions of the present application, and constitute a part of the specification, and are used together with the embodiments of the present application to explain the technical solutions of the present application, and do not constitute a limitation on the technical solutions of the present application.

[0048] Figure 1 is a flowchart of the left ventricular motion prediction method based on triangulation provided by the present application;

[0049] Figure 2 is a principle diagram for determining coordinate information, feature information and structure information of a key point set provided by the present application;

[0050] Figure 3 is a principle diagram for obtaining a left ventricular motion prediction result according to coordinate information, feature information and structure information of a key point set provided by the present application;

[0051] Figure 4is a structural diagram of a left ventricular motion prediction system based on triangulation provided by the present application;

[0052] Figure 5 is a structural diagram of a left ventricular motion prediction device based on triangulation provided by the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0054] The present application will be further described below in combination with the drawings and specific embodiments. The described embodiments should not be considered as limiting the present application, and all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0055] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0057] Cardiovascular diseases have a significant impact on the structure and function of the heart, especially the morphology and function of the left ventricle. The left ventricle, as the key chamber for pumping blood, plays a crucial role in maintaining the function of the heart. However, common cardiovascular diseases such as coronary heart disease and hypertension can lead to abnormalities in the structure of the left ventricle, such as myocardial hypertrophy or dilation, which in turn affects the systolic and diastolic function of the heart. Therefore, early and accurate prediction of left ventricular motion is of great significance for the diagnosis and treatment of heart disease. With the advancement of medical imaging technologies such as magnetic resonance imaging and computed tomography, computer-aided diagnosis plays an important role in medical imaging research. Traditionally, medical experts rely on personal experience to interpret medical images and provide diagnostic reports, but this method is easily affected by subjective factors, which may lead to misdiagnosis or missed diagnosis. By introducing computer-aided technology, the error rate can be significantly reduced, and medical experts can more accurately analyze the disease. In recent years, research combining computer technology with medical images has become a hot field, aiming to improve diagnostic accuracy and treatment effectiveness.

[0058] In the left ventricular motion prediction task, the prior art mostly optimizes the matching and prediction performance through multi-view feature integration, shape-aware loss function, etc. However, they all have some shortcomings. Among them, the image registration-based method has large computational cost and time consumption when processing high-dimensional data; the traditional feature tracking method such as the iterative closest point method may fall into a local optimal solution, resulting in a suboptimal correspondence relationship; in the feature tracking method based on graph matching, the deformable graph matching method reduces the computational cost, but pays too much attention to local structure and ignores global geometric information; the random depth map matching method introduces a hierarchical attention mechanism to reduce the noise influence, but increases the computational complexity and time overhead. Therefore, the above limitations affect the application effect of the prior art in left ventricular contour matching and motion prediction.

[0059] To solve the problems and defects of the related art, the embodiments of the present application provide a left ventricular motion prediction method, system, device and medium based on triangulation. First, the coordinate information of the left ventricular endocardial contour image and the key point set is obtained, and these key points are used to represent the left ventricular endocardial contour. Then, the VGG16 network model is used for feature processing to obtain the feature information of the key point set, and then the triangulation method is used to construct a graph according to the feature information and the coordinate information of the key point set to obtain the structure information of the key point set. Then, the graph convolutional neural network model is used for feature aggregation and the Sinkhorn algorithm is used for feature matching to obtain the diastolic-systolic deformation relationship of the left ventricular endocardial contour; according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour, the diastolic-systolic motion of the left ventricular endocardial contour is predicted to obtain the left ventricular motion prediction result. The embodiments of the present application not only improve the quality of feature representation, but also enhance the ability to capture local and global structure information, and achieve more accurate and smooth left ventricular motion prediction.

[0060] First, the implementation steps of the left ventricular motion prediction method based on triangulation provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0061] The left ventricular motion prediction method based on triangulation proposed in the embodiments of the present application can be applied in a terminal, can be applied in a server, and can also be software running in a terminal or a server, etc. The terminal can be a tablet computer, a notebook computer, a desktop computer, etc., but is not limited thereto. The server can be a standalone physical server, can be a server cluster or a distributed system composed of multiple physical servers, and can also be a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks, and basic cloud computing services such as big data and artificial intelligence platforms.

[0062] Reference Figure 1The implementation process of the left ventricular motion prediction method based on triangulation provided by the embodiments of the present application can include but is not limited to the following steps.

[0063] In step 101, the left ventricular endocardial contour image is obtained.

[0064] In step 101, the left ventricular endocardial contour image is obtained, which provides raw data for subsequent analysis. The endocardial contour image of the heart is a direct representation of the internal structure of the heart, especially the left ventricle. The changes in shape and position during the cardiac cycle are crucial for diagnosing cardiovascular diseases. These images obtained through medical imaging techniques such as MRI or CT scans are the basis for left ventricular motion analysis.

[0065] In step 102, the coordinate information of the key point set is obtained based on the left ventricular endocardial contour image.

[0066] It should be noted that the key point set includes a diastolic key point subset and a systolic key point subset. The contour of the left ventricular endocardium can reflect the morphological changes of the heart at different cardiac cycle stages (such as diastole and systole). By analyzing the changes of these contours, the functional status of the heart can be evaluated, including the contraction ability of the myocardium and the change of the ventricular volume. Many cardiovascular diseases can cause abnormalities in the structure and function of the heart, which are usually reflected in the left ventricular endocardial contour. For example, myocardial infarction, cardiomyopathy and other diseases can cause local or global myocardial motion to weaken or disappear, thereby changing the normal contour of the left ventricular endocardium. Therefore, accurate identification and matching of the left ventricular endocardial contour can help early detection and diagnosis of these diseases.

[0067] In step 102, the key points and the coordinate information of the key points are extracted from the left ventricular endocardial contour image. These key points are usually located at significant positions of the contour, such as vertices, edge intersection points, etc. The selection of key points is very important for accurately tracking the motion of the heart. By dividing the key points into two subsets of diastole and systole, it is helpful to compare the morphological changes of the heart at different cardiac stages, so as to better understand the function of the heart.

[0068] In step 103, the VGG16 network model is used for feature processing based on the left ventricular endocardial contour image and the coordinate information of the key point set, to obtain the feature information of the key point set.

[0069] It should be noted that the VGG16 network model is a deep convolutional neural network composed of 13 convolutional layers and 3 fully connected layers, which is known for its excellent performance in image classification tasks. The VGG16 network model can extract deep features from the left ventricular endocardial contour image, which can capture complex structural information in the image.

[0070] In step 103, a VGG16 network model is used to automatically extract deep features in the left ventricular endocardial contour image, obtaining feature information of the key point set. Compared with manually designed features, the VGG16 network model can capture more complex and more representative feature patterns, which is very beneficial to improve the matching accuracy and robustness.

[0071] In step 104, according to the coordinate information and feature information of the key point set, a graph construction process is performed by combining the triangulation method, obtaining the structure information of the key point set.

[0072] In step 104, the connection relationship between the key points is established by introducing the triangulation method, forming a graph that can reflect the topology structure of the left ventricular endocardial contour. This method not only considers the directly adjacent key points, but also considers the influence of the key points at a distance, so as to more comprehensively represent the spatial structure of the heart contour and improve the accuracy of the subsequent matching task.

[0073] In step 105, according to the structure information and feature information of the key point set, a graph convolutional neural network model is used for feature aggregation, obtaining the embedding information of the key point set.

[0074] It should be noted that the graph convolutional neural network model is a deep learning model specially used for processing graph structure data. Unlike traditional convolutional neural networks, the graph convolutional neural network model can perform feature extraction and information propagation in non-Euclidean space. It defines convolution operations through the adjacency matrix of the graph and the key point feature matrix, so that the features of each key point not only contain its own information, but also fuse the features of its neighbor key points. This mechanism can effectively capture the local and global relationships between key points, and is suitable for key point classification, graph classification, link prediction and other tasks.

[0075] In step 105, the graph convolutional neural network model is used for feature aggregation, which fuses the structure information and feature information of the key point set to generate more rich embedding information. The graph convolutional neural network model can effectively capture the relationship between key points and reflect the topology structure of the graph. By propagating and aggregating information along the edge and key point, local and global relationship information can be fully captured. The embedding information generated in this way not only contains the features of the key points, but also considers their positions in the overall structure and relationships with other key points. This provides more comprehensive and accurate feature representation for subsequent key point matching.

[0076] In step 106, according to the embedding information and coordinate information of the key point set, a feature matching is performed by combining the Sinkhorn algorithm, obtaining the diastolic-systolic deformation relationship of the left ventricular endocardial contour.

[0077] In step 106, based on the embedding information and coordinate information obtained in the previous steps, and combined with the Sinkhorn algorithm, the assignment matrix between the feature points is calculated. This step ensures one-to-one correspondence matching relationship of the feature points, avoids the problem of repeated matching or missing matching of the feature points, and improves the accuracy and consistency of the diastolic-systolic deformation relationship prediction of the left ventricular endocardial contour.

[0078] In step 107, according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour, the diastolic-systolic motion of the left ventricular endocardial contour is predicted to obtain the left ventricular motion prediction result.

[0079] In step 107, the motion of the left ventricle is predicted according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour. Through the high-precision and robust matching results obtained in the foregoing steps, the change of the left ventricle from diastole to systole can be more accurately tracked, and then the functional state thereof is evaluated. This has important significance for early diagnosis and treatment of heart disease, and can provide valuable reference information for clinicians.

[0080] In some embodiments of the present application, with reference to Figure 2 In step 103, according to the left ventricular endocardial contour image and the coordinate information of the key point set, the VGG16 network model is used for feature processing to obtain the feature information of the key point set. The implementation process includes but is not limited to the following steps.

[0081] In step 201, according to the left ventricular endocardial contour image, the VGG16 network model is used for feature processing to obtain an initial feature tensor.

[0082] In step 201, the VGG16 network model is used to automatically learn rich visual features in the left ventricular endocardial contour image as an initial feature tensor, including texture, edge and other complex patterns, thereby providing highly accurate and comprehensive feature description. These high-quality initial feature tensors provide a solid foundation for subsequent graph construction, feature aggregation and matching, effectively improving the accuracy and robustness of left ventricular contour matching and motion prediction.

[0083] In step 202, according to the initial feature tensor and the coordinate information of the key point set, combined with the bilinear interpolation method, the feature information of the key point set is obtained.

[0084] It should be noted that the bilinear interpolation method is a commonly used interpolation technique in image processing, which is used to predict the pixel values of non-grid points in the image. This method is based on the calculation of the four nearest neighbor pixel values around the target position. Specifically, first, linear interpolation is performed on the two pixel values on both sides of the target position in the horizontal direction to obtain two intermediate values; then, linear interpolation is performed again on the two intermediate values in the vertical direction to obtain the pixel value of the target position. The bilinear interpolation can generate smooth and continuous images by considering the influence of the four surrounding pixels, and is suitable for image scaling, rotation and deformation operations.

[0085] In step 202, the feature tensor output by the VGG16 network model is bilinearly interpolated to accurately align the feature map with the key point coordinates, thereby generating a feature vector at each key point position and obtaining more detailed and accurate key point feature information. This not only improves the detail of the feature representation, but also ensures that the feature information is highly consistent with the actual position of the key point, providing a solid foundation for subsequent feature matching and motion prediction.

[0086] In some embodiments of the present application, when performing feature processing, the output of the fourth layer second convolution and the fifth layer first convolution of the VGG16 network model is selected as the initial feature tensor. The reason is that the features of these intermediate layers can capture the transition information between local details and high-level patterns. The features of lower levels mainly capture local information such as edges and textures of the image, while the features of higher levels gradually transition to capturing more abstract object parts. The fourth layer and the fifth layer are at the middle position of the VGG16 network model, which retains sufficient spatial details and contains a certain degree of abstract information, suitable for left ventricular contour matching tasks.

[0087] In addition, the features of the fourth layer and the fifth layer of the VGG16 network model achieve a good balance between generalization and specialization, avoiding the overfitting problem that may be caused by deeper layers and the lack of generalization ability that may be caused by shallower layers. Therefore, selecting the feature representation of these two layers can provide a rich and accurate information basis for subsequent analysis.

[0088] Optionally, according to the coordinate information of the left ventricular endocardial contour image and the key point set, in combination with the VGG16 network model and the bilinear interpolation method, the implementation process of obtaining the feature information of the key point set includes but is not limited to the following steps.

[0089] First, the left ventricular endocardial contour image is input into the VGG16 network model to obtain an initial feature tensor ;

[0090] Then, using the bilinear interpolation method , according to each key point coordinate information of the key points interpolating the feature information of each key point in the initial feature tensor satisfies the following formula (1):

[0091] (1).

[0092] The above two steps work together. First, the deep features of the heart image are extracted through the VGG16 network, and then these features are accurately assigned to each key point using bilinear interpolation, thereby generating high-quality key point feature information. Such a process ensures that the feature information is both rich and accurate, providing strong support for the subsequent graph construction and feature aggregation.

[0093] In some embodiments of the present application, with reference to Figure 2 In step 104, according to the coordinate information and feature information of the key point set, the implementation process of the structure information of the key point set obtained by combining the triangulation method includes but is not limited to the following steps.

[0094] Step 301, according to the coordinate information and feature information of the key point set, a distance matrix is obtained.

[0095] It should be noted that the distance matrix is used to represent the distance between each key point in the key point set.

[0096] Step 301, the spatial relationship between each key point in the key point set is quantified by the distance between them, and a distance matrix is obtained. By calculating the distance between the key points, a preliminary spatial topology can be established, which helps to more accurately describe the relative positions between the key points in the subsequent steps.

[0097] Optionally, the Euclidean distance is used to calculate the distance between each key point in the key point set.

[0098] Step 302, according to the distance matrix, combining the triangulation method, the weighted adjacency matrix of the key point set is obtained as the structure information.

[0099] It should be noted that the triangulation method is used to establish a connection between key point pairs in the key point set whose distance does not exceed two units. The distance between each key point in the key point set includes coordinate distance and feature distance. The coordinate distance refers to the distance between two key points in the coordinate system. The feature distance refers to the distance between two key points in the feature space, and the feature distance is usually used to measure the similarity of two key points in the feature space.

[0100] ​Optionally, the coordinate distance calculation method includes Euclidean distance, Manhattan distance and Chebyshev distance, and the feature distance calculation method includes Euclidean distance and cosine similarity.

[0101] In step 302, connections are established between pairs of key points in the key point set that are within a distance of two units by using triangulation, and a weighted adjacency matrix is formed as the structure information of the key point set. The triangulation method not only considers the direct connection between adjacent key points, but also considers the indirect connection between key points that are far apart but still have relevance. This construction method can better capture the local and global topological relationships between key points. The weighted adjacency matrix represents the connection strength between key points, where the weight reflects the similarity or distance between key points. Such structure information provides important topological relationship support for subsequent feature aggregation and matching, which helps to improve the accuracy and robustness of left ventricular contour matching and motion prediction.

[0102] Through the above two steps, the structure information of the key point set can be effectively constructed, which not only enhances the ability to capture local features, but also considers global structure information, thereby providing a solid foundation for subsequent feature aggregation and matching tasks. This method overcomes the limitations of traditional methods that overemphasize point features while ignoring the topological relationships between points, and improves the overall matching accuracy.

[0103] In some embodiments of the present application, with reference to Figure 3 In step 105, according to the structure information and feature information of the key point set, the feature aggregation is performed using a graph convolutional neural network model to obtain the embedding information of the key point set. The implementation process includes but is not limited to the following steps.

[0104] In step 401, according to the structure information and feature information of the key point set, the intra-graph convolution mechanism of the graph convolutional neural network model is used to obtain the intra-graph aggregation features of the key point set.

[0105] In step 401, the intra-graph convolution mechanism of the graph convolutional neural network model is used to aggregate features of the key points within the diastolic key point subset and the systolic key point subset. The intra-graph convolution mechanism can effectively capture the local relationships between key points and integrate these relationship information into the feature representation of the key points. The intra-graph aggregation features generated in this way not only contain the features of the key points themselves, but also consider the influence of their neighbor key points, thereby enhancing the discriminability of the features. This step ensures that the feature information within the key point subset is fully fused and enhanced.

[0106] Optionally, the intra-graph convolution mechanism of the graph convolutional neural network model satisfies the following formula (2):

[0107] (2);

[0108] In formula (2), the edge set is obtained from the weighted adjacency matrix of the key point set. , Indicate key points With key points There are connections between them. Key point The Layer embedded information; Indicate key points The Layer embedded information, Indicate key points The Layer embedded information, , ; This represents the feature aggregation function along the key points. Represents the feature aggregation function along the edge; and Together with key points There are key points of connection. The Layer Embedded Information and key points The Layer Embedded Information Aggregation yields key points The Layer Embedded Information .

[0109] Step 402: Based on the structural and feature information of the keypoint set, the cross-graph convolution mechanism of the graph convolutional neural network model is used, combined with the similarity matrix, to obtain the cross-graph aggregation features of the keypoint set.

[0110] In this step, the cross-graph convolution mechanism of the graph convolutional neural network model, combined with a similarity matrix, is used to aggregate features from the end-diastolic and end-systolic keypoint subsets. The cross-graph convolution mechanism can propagate and aggregate information across different graph structures, thereby capturing the global relationships between different keypoint subsets. The similarity matrix measures the similarity between keypoints in different graphs, helping to selectively transmit and integrate cross-graph information. The resulting cross-graph aggregated features not only enhance the feature representation within a single graph but also consider the correlations between different graphs, further improving the richness and robustness of the features.

[0111] Optionally, the cross-graph convolution mechanism of the graph convolutional neural network model satisfies the following formula (3):

[0112] (3):

[0113] In formula (3), This represents a cross-graph convolution function used to aggregate information across graphs. It first aggregates information from the previous layer, and then uses a similarity matrix. This emphasizes the features of similar keypoints and weights and aggregates the features of neighboring keypoints to better capture the local and global relationships between keypoints, then performs another round of information aggregation. Similarity matrix Represents a subset of end-diastolic keypoints in the keypoint set. Key points With the key point subset at the end of the contraction Key points The similarity between keypoints is updated using a weight matrix and implemented as an exponential function to ensure all member values ​​are non-negative, thus better capturing local and global relationships between keypoints. Similarity Matrix Satisfy the following formula (4):

[0114] (4);

[0115] In formula (4), Represents the similarity function. It can be a similarity function based on Euclidean distance or a cosine similarity function.

[0116] Step 403: Obtain the embedding information of the keypoint set based on the intra-graph aggregation features and cross-graph aggregation features of the keypoint set.

[0117] In step 403, intra-graph aggregated features and cross-graph aggregated features are combined to generate the final embedding information. By combining intra-graph and cross-graph aggregated features, the embedding information not only includes detailed features within the end-diastolic keypoint subset and the end-systolic keypoint subset, but also considers the global relationships between different subsets. This comprehensive embedding information provides a more comprehensive and accurate foundation for subsequent keypoint matching, helping to improve matching accuracy and robustness.

[0118] The three steps described above work together: first, they enhance local features through intra-graph convolution; then, they incorporate global information through cross-graph convolution; and finally, they synthesize and generate high-quality embedding information. This method ensures the comprehensiveness and accuracy of feature information, providing a solid foundation for left ventricular motion prediction based on triangulation.

[0119] In some embodiments of this application, the embedding information of the keypoint set includes end-diastolic embedding information and end-systolic embedding information. (Refer to...) Figure 3, and the implementation process of the diastolic-systolic deformation relationship of the left ventricular endocardial contour includes but is not limited to the following steps.

[0120] In step 501, the similarity between the end-diastolic embedding information and the end-systolic embedding information is calculated to obtain a score matrix.

[0121] In step 501, the similarity between the end-diastolic embedding information and the end-systolic embedding information is calculated using methods such as inner product or cosine similarity to obtain a score matrix, where each element represents the similarity between the key point subsets at the corresponding positions of the two time points. This score matrix reflects the similarity between key point pairs at different time points.

[0122] In step 502, feature matching is performed based on the score matrix and Sinkhorn algorithm to obtain a matching matrix.

[0123] It should be noted that Sinkhorn algorithm is an iterative method based on optimal transport theory, which is used to generate a doubly stochastic matrix to ensure that each key point has a unique and optimal matching relationship between different time points. This algorithm normalizes the rows and columns of the matrix alternately, so that the sum of each row and column is 1, and gradually converges to a stable doubly stochastic matrix. The reason for choosing Sinkhorn algorithm for feature matching is that it can ensure the uniqueness and consistency of matching, avoiding repeated or missed matching. In addition, Sinkhorn algorithm considers global information, improving the accuracy and robustness of matching, especially when dealing with low-quality images or abnormal contours. Although it involves an iterative process, it has high computational efficiency, suitable for large-scale data processing, and can be combined with various similarity measures to provide a flexible matching scheme. These characteristics make Sinkhorn algorithm suitable for feature matching tasks in left ventricular motion estimation based on deep learning.

[0124] In step 502, the Sinkhorn algorithm is used to process the score matrix to generate a matching matrix, effectively solving the uniqueness and consistency of key points, avoiding repeated matching or missed matching, thereby improving the accuracy and reliability of matching. The matching matrix reflects the mapping relationship between key points at different time points, which is crucial for understanding cardiac motion, especially in assessing myocardial displacement and deformation that needs to be tracked when evaluating cardiac function.

[0125] In step 503, the diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained based on the matching matrix and the coordinate information of the key point set.

[0126] It should be noted that the diastolic-systolic deformation relationship describes how the endocardial contour of the left ventricle changes shape in a complete cardiac cycle, which is very important for diagnosing heart diseases. For example, abnormal deformation patterns may be a manifestation of certain pathological conditions, such as myocardial infarction or ventricular wall motion disorders. Therefore, accurately estimating this deformation relationship is of great significance for clinical applications.

[0127] In step 503, the diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the matching matrix and the coordinate information of the key point set. The relative movement of each key point in the diastolic and systolic states can be determined, and the change trend of the entire left ventricular endocardial contour can be described accordingly.

[0128] In some embodiments of the present application, with reference to Figure 3 In step 502, feature matching is performed according to the score matrix combined with the Sinkhorn algorithm, and the implementation process of the matching matrix includes but is not limited to the following steps.

[0129] In step 601, the Sinkhorn algorithm is used to iteratively normalize the score matrix to obtain a convergent doubly stochastic matrix.

[0130] In step 601, the score matrix is converted into a doubly stochastic matrix (i.e., the sum of each row and each column is 1). The Sinkhorn algorithm adjusts the element values in the score matrix iteratively, so that the sum of the elements in each row and each column of the final matrix is close to 1. This process ensures that each end-diastolic key point is only matched with one end-systolic key point, and vice versa, thereby avoiding the problems of repeated matching or missing matching.

[0131] Specifically, the Sinkhorn algorithm gradually approaches the doubly stochastic condition by alternately normalizing the rows and columns of the score matrix. Each iteration makes the matrix closer to satisfying the doubly stochastic constraint. When the iteration reaches a certain number of times or meets the preset convergence criteria, the score matrix becomes a stable doubly stochastic matrix, which directly reflects the optimal matching possibility between key points.

[0132] In step 602, a matching matrix is obtained according to the convergent doubly stochastic matrix. In the matching matrix, only one element in each row and each column is 1, and all other elements are 0.

[0133] In step 602, the doubly stochastic matrix is converted to obtain the matching matrix, and the probability distribution in the doubly stochastic matrix is interpreted as the basis for actual matching decisions. Since the elements of the doubly stochastic matrix represent the possibility of matching, the most likely matching relationships need to be selected to construct the final matching matrix. The matching matrix is a binary matrix, in which only one element in each row and each column is 1, and all other elements are 0.

[0134] Alternatively, a matching matrix can be constructed by determining which elements should be set to 1 by selecting the maximum value in each row or column of a double random matrix.

[0135] Step 603: The matching matrix is ​​optimized by minimizing the negative logarithm loss.

[0136] In step 603, to improve the quality of the matching matrix, the matching matrix is ​​further optimized. Minimizing the negative logarithmic loss is a common optimization strategy aimed at adjusting the matching matrix to better reflect actual cardiac motion patterns.

[0137] Minimizing the negative logarithmic loss function makes the matching results more consistent with the expected myocardial motion model, reducing unreasonable matches. The optimization process may involve adjusting certain matching decisions in the matching matrix to ensure that the overall matching scheme minimizes the total loss function. This helps improve the accuracy and reliability of keypoint matching, especially in the presence of noisy or outlier data.

[0138] The above three steps together ensure that the transformation from the scoring matrix to the matching matrix not only satisfies the theoretical optimal transfer problem solution, but also provides high-quality and reliable key point matching results in practice. This is crucial for subsequent analysis of the diastolic-contraction deformation relationship of the left ventricular endocardial contour.

[0139] In some embodiments of this application, reference is made to Figure 3 In step 503, the process of obtaining the diastolic-contraction deformation relationship of the left ventricular endocardium contour based on the coordinate information of the matching matrix and the key point set includes, but is not limited to, the following steps.

[0140] Step 701: Obtain the deformation function based on the coordinate information of the matching matrix and the key point set.

[0141] In step 701, a mathematical model (i.e., a deformation function) is constructed based on the matching results between key points. This model quantifies the shape changes of the heart at different time points. The deformation function is a key tool connecting the left ventricular endothelial contour at end-diastole and end-systole. It reflects not only the spatial displacement of key points but may also include local deformations (such as stretching and compression). Specifically, the matching matrix provides information on the correspondence between each key point from end-diastole to end-systole. These one-to-one matching pairs form the basis for constructing the deformation function. The coordinate information includes the location, attribute, and confidence information of the key points. Using this information, the deformation function can be defined more accurately, ensuring that it truly reflects the actual motion patterns of the endothelial membrane.

[0142] Step 702: Based on the deformation function, obtain the diastolic-contraction deformation relationship of the left ventricular endocardium profile.

[0143] In step 702, the positions of the key points on the left ventricular endocardial contour are reconstructed using a deformation function. By applying the deformation function to all key points, the diastolic-systolic deformation relationship of the left ventricular endocardial contour can be obtained. Based on the diastolic-systolic deformation relationship of the left ventricular endocardial contour, a series of quantitative analyses can be performed, such as calculating the thickness change of the myocardial wall, the volume change rate, the strain and the strain rate, etc. These parameters are crucial for evaluating the efficiency of heart pumping and detecting potential cardiac lesions.

[0144] Specifically, first, based on the matching matrix and the coordinate information of the key points, a deformation function is constructed, which describes the spatial displacement of each key point from end diastole to end systole. Commonly used deformation functions include affine transformation, thin plate splines or radial basis functions, etc., to accurately quantify the movement path of each key point on the left ventricular endocardial contour.

[0145] Next, the deformation function is applied to all points of the entire left ventricular endocardial contour, not just the key points, to achieve modeling of the continuous deformation of the endocardial surface. For points not directly involved in matching, an interpolation method is used to predict their movement path, ensuring the integrity of the deformation model. This step generates a detailed pattern of changes in the left ventricular endocardial contour during the cardiac cycle, i.e., the diastolic-systolic deformation relationship.

[0146] Using these deformation relationships, a series of quantitative motion parameters are calculated, such as displacement field (specific movement direction and distance of each point), strain rate (local stretching or compression of the myocardium), volume change (estimating the volume change of the left ventricle in the cardiac cycle to calculate the ejection fraction EF), and wall thickness change (observing the change of myocardial wall thickness over time). Integrating the above parameters, a comprehensive left ventricular motion estimation report is generated. Finally, by comparing with clinical standard measurements and cross-validation, the deformation function and its related parameters are continuously optimized to improve the accuracy of the model.

[0147] Alternatively, according to the coordinate information and embedding information of the key point set, feature matching is performed using the Sinkhorn algorithm to obtain the diastolic-systolic deformation relationship of the left ventricular endocardial contour. The implementation process includes but is not limited to the following steps.

[0148] First, calculate the similarity between the end diastolic embedding information and the end systolic embedding information to obtain the score matrix . The score matrix satisfies the following formula (5):

[0149] (5);

[0150] In formula (5), denotes the similarity function, It can be a similarity function based on Euclidean distance or a cosine similarity function.

[0151] Secondly, the score matrix is iteratively normalized by Sinkhorn algorithm, and the iteration is performed until convergence. The converged double stochastic matrix is obtained, and each row and column of the double stochastic matrix sums to 1.

[0152] The iteration process of the Sinkhorn algorithm satisfies the following formula (6):

[0153] In formula (6), the initial matrix , represents the number of iterations; and are two normalization factors, respectively adjusting the rows and columns of ; is the normalization factor The result of the iteration times is , and represents a diagonal matrix composed of ; The normalization factor is , and the result of the iteration times is , and represents a diagonal matrix composed of

[0154] ; The iteration process of satisfies the following formula (7):

[0155] The iteration process of satisfies the following formula (8):

[0156] (8).

[0157] Then, the matching matrix is optimized by minimizing the negative log loss, and the negative log loss satisfies the following formula (9):

[0158] (9).

[0159] In an ideal case, the matching matrix should be a permutation matrix, meaning that each key point in the diastolic key point subset corresponds to only one key point in the systolic key point subset, and the element of the matching matrix represents the key point in the diastolic key point subset Key points of the subset of end-systole key points match, otherwise 0. The matching matrix obtained by Sinkhorn algorithm After optimization, it is close to a permutation matrix, but not exactly a strict permutation matrix, because the result of Sinkhorn algorithm can contain some non-zero non-integer values. However, in practical applications, this approximation is usually sufficient to determine the one-to-one correspondence between key points.

[0160] Then, according to the matching matrix and the coordinate information of the key point set, the matched key point pairs in the key point set are determined, and the deformation function is obtained. The change of the deformation function is the result of the superposition of the changes along the x-axis and y-axis directions. Taking the x-axis direction as an example, the deformation function satisfies the following formula (10):

[0161] (10);

[0162] In formula (10), the current position refers to a specific position in the image that needs to be deformed; represents the position after deformation from the current position to the deformed position, represents the number of key points; is the position of the key point , is the weight coefficient of ; is the effective radius, which determines the influence range of the key point, and is calculated based on the Euclidean distance; is the radial basis function, which calculates the contribution of each key point pair to deformation, satisfies the following formula (11):

[0163] (15);

[0164] In formula (11), represents the normalized distance, which represents the ratio of the distance from the current position to to the effective radius , when , is forcibly set to 0, which means that the key points beyond the effective radius of the control point will not be affected by this key point.

[0165] Optionally, the radial basis function is a C2 function based on Wendland.

[0166] In summary, the method can provide the following technical effects.

[0167] The method for predicting left ventricular motion based on triangulation provided by the embodiments of the present application effectively improves the accuracy and efficiency of left ventricular motion prediction. First, the VGG16 network is used to extract features from the end-diastolic and end-systolic heart images, with particular attention to the feature representation of deep convolutional layers, and the bilinear interpolation method is used to enhance the geometric structure information, providing accurate and detailed feature descriptions for subsequent steps. Then, the triangulation method is used to construct the topological structure of the key point set of the left ventricular endocardial contour, which not only captures the feature information of adjacent nodes, but also emphasizes the importance of distant nodes, optimizing the accuracy and robustness of key point matching. In addition, the application of graph convolutional neural network model enhances the ability of cross-graph information fusion, effectively reflects the topological structure of the graph by propagating embedding information along the edges of the graph, and realizes efficient cross-graph information propagation by using the similarity matrix, improving the overall adaptability and performance of the model. These technical means work together to ensure that each link from image input to feature extraction to graph construction can operate efficiently, providing a solid foundation for heart motion analysis.

[0168] On the basis of efficient feature processing and key point matching, the method further optimizes the similarity matrix through the Sinkhorn algorithm to generate accurate node correspondence, and then constructs the deformation function to accurately map the diastolic-systolic deformation relationship of the left ventricular endocardial contour, providing a reliable basis for the motion estimation of the left ventricular endocardial contour. Finally, according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour, the diastolic-systolic motion of the left ventricular endocardial contour is predicted, and the left ventricular motion prediction result is obtained. Therefore, the method for predicting left ventricular motion based on triangulation provided by the embodiments of the present application not only improves the accuracy, robustness and efficiency of left ventricular motion prediction.

[0169] Secondly, with reference to Figure 4 The system for predicting left ventricular motion based on triangulation provided by the embodiments of the present application includes a data acquisition module 801, a coordinate acquisition module 802, a feature processing module 803, a graph construction module 804, a feature aggregation module 805, a feature matching module 806, and a deformation prediction module 807.

[0170] The data acquisition module 801 is used to acquire the left ventricular endocardial contour image.

[0171] The coordinate acquisition module 802 is used to obtain the coordinate information of the key point set according to the left ventricular endocardial contour image. The key point set includes an end-diastolic key point subset and an end-systolic key point subset.

[0172] The feature processing module 803 is configured to perform feature processing on the VGG16 network model based on the left ventricular endocardial contour image and the coordinate information of the key point set, to obtain feature information of the key point set.

[0173] The graph construction module 804 is configured to perform graph construction processing based on the coordinate information and the feature information of the key point set in combination with a triangulation method, to obtain structure information of the key point set.

[0174] The feature aggregation module 805 is configured to perform feature aggregation on the graph convolutional neural network model based on the structure information and the feature information of the key point set, to obtain embedding information of the key point set.

[0175] The feature matching module 806 is configured to perform feature matching based on the embedding information and the coordinate information of the key point set in combination with the Sinkhorn algorithm, to obtain the diastolic-systolic deformation relationship of the left ventricular endocardial contour.

[0176] The deformation prediction module 807 is configured to perform prediction on the diastolic-systolic motion of the left ventricular endocardial contour based on the diastolic-systolic deformation relationship of the left ventricular endocardial contour, to obtain a left ventricular motion prediction result.

[0177] In another aspect, referring to Figure 5 The left ventricular motion prediction device based on triangulation provided by the embodiment of the present application comprises a processor and a memory.

[0178] When the program is executed by the processor, the processor implements the aforementioned left ventricular motion prediction method based on triangulation.

[0179] In addition, the embodiment of the present application provides a computer readable storage medium, which stores a program executable by a processor, and the program executable by the processor is used to implement the aforementioned left ventricular motion prediction method based on triangulation when executed by the processor.

[0180] Similarly, the contents in the above method embodiments are applicable to the system embodiments, the device embodiments and the medium embodiments, the system embodiments, the device embodiments and the medium embodiments specifically implement the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0181] Although the embodiments of the present application have been shown and described, those skilled in the art can understand that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the claims and their equivalents.

[0182] The above is a specific description of the preferred embodiment of the present application, but the present application is not limited to the described embodiments, and those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the present application, and these equivalent modifications or replacements are all included in the scope defined by the claims of the present application.

Claims

1. A method for left ventricular motion prediction based on triangulation, characterized in that, The method comprises the following steps: obtaining a left ventricular endocardial contour image; obtaining coordinate information of a key point set according to the left ventricular endocardial contour image; the key point set comprises a diastolic end key point subset and a systolic end key point subset; performing feature processing on the left ventricular endocardial contour image and the coordinate information of the key point set by using a VGG16 network model to obtain feature information of the key point set; performing graph construction processing on the coordinate information and the feature information of the key point set by using a triangulation method to obtain structure information of the key point set; performing feature aggregation on the structure information and the feature information of the key point set by using a graph convolutional neural network model to obtain embedding information of the key point set; performing feature matching on the embedding information and the coordinate information of the key point set by using a Sinkhorn algorithm to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour; predicting diastolic-systolic motion of the left ventricular endocardial contour according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour to obtain a left ventricular motion prediction result.

2. The triangulation-based left ventricular motion prediction method of claim 1, wherein, The feature processing on the left ventricular endocardial contour image and the coordinate information of the key point set by using the VGG16 network model to obtain the feature information of the key point set comprises: performing feature processing on the left ventricular endocardial contour image by using the VGG16 network model to obtain an initial feature tensor; obtaining the feature information of the key point set by using a bilinear interpolation method according to the initial feature tensor and the coordinate information of the key point set.

3. The triangulation-based left ventricular motion prediction method of claim 1, wherein, The graph construction processing on the coordinate information and the feature information of the key point set by using the triangulation method to obtain the structure information of the key point set comprises: obtaining a distance matrix according to the coordinate information and the feature information of the key point set; the distance matrix is used to represent distances between key points in the key point set; obtaining a weighted adjacency matrix of the key point set as the structure information according to the distance matrix and the triangulation method; the triangulation method is used to establish a connection between key point pairs in the key point set with a distance of not more than two units.

4. The triangulation-based left ventricular motion prediction method of claim 1, wherein, The feature aggregation on the structure information and the feature information of the key point set by using the graph convolutional neural network model to obtain the embedding information of the key point set comprises: obtaining intra-graph aggregation features of the key point set by using an intra-graph convolution mechanism of the graph convolutional neural network model according to the structure information and the feature information of the key point set; obtaining cross-graph aggregation features of the key point set by using a cross-graph convolution mechanism of the graph convolutional neural network model according to the structure information and the feature information of the key point set and in combination with a similarity matrix; obtaining the embedding information of the key point set according to the intra-graph aggregation features and the cross-graph aggregation features of the key point set.

5. The triangulation-based left ventricular motion prediction method of claim 1, wherein, The embedding information of the key point set comprises diastolic end embedding information and systolic end embedding information. The embedding information and the coordinate information of the key point set are combined with the Sinkhorn algorithm to perform feature matching to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour, including: Similarity between the end-diastolic embedding information and the end-systolic embedding information is calculated to obtain a score matrix; Feature matching is performed according to the score matrix and the Sinkhorn algorithm to obtain a matching matrix; The diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the matching matrix and comprehensive information of the key point set.

6. The triangulation-based left ventricular motion prediction method of claim 5, wherein, The feature matching is performed according to the score matrix and the Sinkhorn algorithm to obtain a matching matrix, including: The Sinkhorn algorithm is used to iteratively normalize the score matrix to obtain a convergent double random matrix; The matching matrix is obtained according to the convergent double random matrix; only one element in each row and each column of the matching matrix is 1, and all other elements are 0; The matching matrix is optimized by minimizing the negative logarithmic loss.

7. The triangulation-based left ventricular motion prediction method of claim 5, wherein, The diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the matching matrix and comprehensive information of the key point set, including: A deformation function is obtained according to the matching matrix and comprehensive information of the key point set; The diastolic-systolic deformation relationship of the left ventricular endocardial contour is obtained according to the deformation function.

8. A left ventricular motion prediction system based on triangulation, characterized in that, It includes: a data acquisition module, a coordinate acquisition module, a feature processing module, a graph construction module, a feature aggregation module, a feature matching module, and a deformation prediction module; The data acquisition module is configured to acquire a left ventricular endocardial contour image; The coordinate acquisition module is configured to obtain coordinate information of a key point set according to the left ventricular endocardial contour image; the key point set includes an end-diastolic key point subset and an end-systolic key point subset; The feature processing module is configured to perform feature processing on the left ventricular endocardial contour image and the coordinate information of the key point set using a VGG16 network model to obtain feature information of the key point set; The graph construction module is configured to perform graph construction processing on the coordinate information and the feature information of the key point set using a triangulation method to obtain structure information of the key point set; The feature aggregation module is configured to perform feature aggregation on the structure information and the feature information of the key point set using a graph convolutional neural network model to obtain embedding information of the key point set; The feature matching module is configured to perform feature matching on the embedding information and the coordinate information of the key point set using the Sinkhorn algorithm to obtain a diastolic-systolic deformation relationship of the left ventricular endocardial contour; The deformation prediction module is configured to predict diastolic-systolic motion of the left ventricular endocardial contour according to the diastolic-systolic deformation relationship of the left ventricular endocardial contour to obtain a left ventricular motion prediction result.

9. A left ventricular motion prediction apparatus based on triangulation, characterized by It includes: a processor and a memory; The memory is configured to store a program; When the program is executed by the processor, the processor implements the triangulation-based left ventricular motion prediction method according to any one of claims 1 to 7.

10. A computer readable storage medium having stored thereon a program executable by a processor, the program comprising instructions for causing the processor to perform the method of any one of claims 1 to 9. The program executable by the processor, when executed by the processor, is used to implement the left ventricular motion prediction method based on triangulation as claimed in any one of claims 1 to 7. The program executable by the processor, when executed by the processor, is used to implement the left ventricular motion prediction method based on triangulation as claimed in any one of claims 1 to 7.

Images