A cardiovascular image analysis system and method

Abstract

The present invention relates to the field of medical image processing and analysis, and in particular to a cardiovascular image analysis system and method. The content includes: acquiring and preprocessing cardiovascular image data, converting the preprocessed cardiovascular image data into an initial feature space, and recursively optimizing the cardiovascular image features in the initial feature space, further constructing a dynamic feature map to obtain a dynamic feature map value; adjusting the dynamic feature map value to obtain an optimized feature map value; using block feature aggregation and joint optimal iteration strategy to obtain a segmentation result, further extracting cardiovascular geometric features, and constructing a vascular network diagram; by detecting abnormal areas in the segmentation results, determining the lesion area, and finally generating diagnostic information. The problem of lack of adaptive capture capability for dynamic features and local changes, the inability to analyze the detection of lesion areas in combination with the segmentation results, and the inability of the segmentation results to fully reflect the coherence of the cardiovascular structure is solved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing and analysis, and in particular to a cardiovascular image analysis system and method. Background Art

[0002] With the continuous development of medical imaging technology, non-invasive imaging technologies such as CT (computed tomography) and MRI (magnetic resonance imaging) have become important tools for the diagnosis of cardiovascular diseases in clinical practice. These imaging technologies can present the anatomical information of cardiovascular structures at extremely high resolution, providing strong support for the early detection, diagnosis and treatment of diseases. However, with the improvement of the performance of imaging equipment, the amount of data generated by a single scan is growing exponentially. How to efficiently process and analyze these large-scale medical imaging data has become one of the important challenges in the current field of medical imaging research.

[0003] In addition, traditional cardiovascular imaging analysis mostly relies on the doctor's naked eye observation and subjective judgment, and its results are easily affected by the doctor's experience and observation angle. It is not only time-consuming and laborious, but may also lead to instability in the analysis results. In addition, the cardiovascular anatomical structure is complex, especially the coronary arteries and peripheral blood vessels, which have many branches, complex directions and small structures. Manual analysis often finds it difficult to fully and accurately reflect these structural information. On the other hand, the early lesion characteristics of cardiovascular diseases (such as aneurysms, stenosis, plaques, etc.) are usually subtle and unevenly distributed. It is difficult for manual analysis to accurately identify these early lesions, thereby increasing the risk of missed diagnosis and misdiagnosis of the disease. Therefore, it is urgent to provide a cardiovascular imaging analysis method to make up for the limitations of manual analysis. By combining medical imaging technology and artificial intelligence, the diagnosis and treatment of cardiovascular diseases are expected to enter a precise and intelligent stage, bringing better prognosis and quality of life to patients. Summary of the invention

[0004] The present invention provides a cardiovascular image analysis system and method to solve the problems of lack of adaptive capture capability of dynamic features and local changes; inability to combine the global information of segmentation results with the local dynamic features for comprehensive analysis of lesion areas, resulting in low accuracy and sensitivity of lesion recognition; and inability of segmentation results to fully reflect the coherence of cardiovascular structures and meet the requirements for speed and efficiency in actual clinical applications.

[0005] A cardiovascular image analysis system and method of the present invention specifically includes the following technical solutions:

[0006] A cardiovascular image analysis method comprises the following steps:

[0007] S1. Obtain cardiovascular imaging data and perform preprocessing, convert the preprocessed cardiovascular imaging data into an initial feature space, gradually optimize the cardiovascular imaging features in the initial feature space through a recursive process, and obtain the recursive cardiovascular imaging features; construct a dynamic feature mapping based on the recursive feature space composed of the recursive cardiovascular imaging features, and obtain a dynamic feature mapping value; and use multi-level variational optimization to adjust the dynamic feature mapping value to obtain an optimized feature mapping value;

[0008] S2. Block feature aggregation is used to perform block operations on the optimized feature mapping values, and a joint optimal iterative strategy is used to dynamically adjust the segmentation boundaries of the cardiovascular region to obtain a three-dimensional segmentation result; based on the three-dimensional segmentation result, the geometric features of the cardiovascular are extracted, and a vascular network diagram is constructed; by detecting abnormal areas in the three-dimensional segmentation results, the lesion area is determined, and finally diagnostic information is generated.

[0009] Preferably, the S1 specifically includes:

[0010] The recursive process of cardiovascular image characterization is defined by the following formula:

[0011] ,

[0012] in, is the pixel point in the recursive process In the Cardiovascular imaging features of the layer output; Indicates The weight matrix of the layer; is the bias term; is the activation function; is the pixel point in the recursive process In the Cardiovascular imaging features of the layer output; is the gradient of the cardiovascular imaging features in the previous layer; is the gradient regularization coefficient.

[0013] Preferably, the S1 specifically includes:

[0014] The dynamic feature mapping transforms the variation law of pixel gradients in the recursive feature space into significance weights through integral operation, forming a feature value distribution corresponding to the actual cardiovascular area. The specific formula of the dynamic feature mapping is:

[0015] ,

[0016] in, is the dynamic feature map value; is the region of interest, defined as the set of pixels containing cardiovascular structures; It is Gradients of layer cardiovascular imaging features; Used to prevent the denominator from being zero.

[0017] Preferably, the S1 specifically includes:

[0018] In the process of adjusting the dynamic feature map value, the energy function is constructed to integrate gradient smoothness, feature consistency and sparsity; the mathematical expression of the energy function is:

[0019] ,

[0020] in, is the energy function, representing the dynamic feature map value The optimization target value of is the gradient smoothing term, is the gradient magnitude of the dynamic feature map value; is the gradient of the dynamic feature map value; is the feature consistency item; is the target feature value; is the sparsity regularization term; is the weight parameter.

[0021] Preferably, the S1 specifically includes:

[0022] The variational method is used to minimize the energy function. The optimized feature mapping value is obtained by calculating the first-order variational derivative of the energy function with respect to the dynamic feature mapping value and setting the first-order variational derivative to zero.

[0023] Preferably, the S2 specifically includes:

[0024] The region of interest of the cardiovascular image is divided into two or more non-overlapping regions of interest, and the eigenvalues ​​in each region block are aggregated by weighted summation to form weighted eigenvalues ​​representing each region block.

[0025] Preferably, the S2 specifically includes:

[0026] The weighted eigenvalues ​​of each region block are analyzed and updated using the joint optimal iteration strategy. The initial segmentation result is obtained by the segmentation threshold method. In each iteration, the segmentation result is dynamically adjusted according to the segmentation status of the previous round and the weighted eigenvalues ​​of the current region block. The segmentation formula for each iteration is:

[0027] ,

[0028] in, It is The segmentation result of the iteration; It is The segmentation result of the iteration; is the step length; is a sign function; It is a regional block The weighted eigenvalue of ; is the segmentation threshold.

[0029] Preferably, the S2 specifically includes:

[0030] The final segmentation result is a three-dimensional segmentation result that identifies the cardiovascular structure. The voxel distribution of the segmentation result is analyzed to extract the geometric features of the cardiovascular system. The topological structure of the blood vessels is constructed, the connectivity of the segmented area is analyzed, and the voxel set in the segmentation result is used to establish a vascular network diagram.

[0031] A cardiovascular image analysis system, comprising the following parts:

[0032] Image data acquisition and preprocessing module, recursive feature modeling module, dynamic feature map generation module, variational optimization module, block feature aggregation module, joint optimal iterative segmentation module, geometric feature extraction and analysis module;

[0033] An image data acquisition and preprocessing module is used to acquire and preprocess cardiovascular image data, and use the preprocessed cardiovascular image data as input to the recursive feature modeling module;

[0034] The recursive feature modeling module converts the preprocessed cardiovascular image data into the initial feature space, and gradually optimizes the cardiovascular image features in the initial feature space through a recursive process to obtain the recursive cardiovascular image features; the recursive feature space is formed by the recursive cardiovascular image features; and the recursive feature space is used as the input of the dynamic feature map generation module;

[0035] A dynamic feature map generation module generates a dynamic feature map value by analyzing the recursive feature space gradient through dynamic feature map; and uses the dynamic feature map value as the input of the variational optimization module;

[0036] The variational optimization module performs global optimization on the dynamic feature mapping values, constructs and minimizes the energy function, and uses the variational method to solve the optimized feature mapping values; the optimized feature mapping values ​​are used as the input of the block feature aggregation module;

[0037] The block feature aggregation module performs block operations on the optimized feature mapping values, divides the region of interest of the cardiovascular image into two or more non-overlapping regions of interest, and aggregates the feature values ​​in each region block by weighted summation to form weighted feature values ​​of each region block; the weighted feature values ​​of each region block are used as inputs of the joint optimal iterative segmentation module;

[0038] The joint optimal iterative segmentation module uses the joint optimal iterative strategy to analyze and update the weighted feature values ​​of the region block to obtain a three-dimensional segmentation result; the three-dimensional segmentation result is used as the input of the geometric feature extraction and analysis module;

[0039] The geometric feature extraction and analysis module extracts geometric features from the three-dimensional segmentation results, establishes the topological structure of the blood vessels, constructs a cardiovascular network diagram through connectivity analysis, detects diseased areas in the cardiovascular system, and generates diagnostic reports for cardiovascular images to assist clinical decision-making.

[0040] The beneficial effects of the technical solution of the present invention are:

[0041] 1. Through recursive feature extraction and dynamic feature map generation, multi-scale recursive features in cardiovascular imaging data are effectively extracted. Combined with the significance analysis of the recursive feature space gradient and multi-level variational optimization, the significance of the cardiovascular region is significantly enhanced and noise interference is suppressed. The optimized dynamic feature map value has smoothness, global consistency and local sparsity, which can accurately reflect the complex structure and dynamic changes of the cardiovascular region, providing high-quality data support for cardiovascular disease diagnosis, surgical planning and pathological research, and has a wide range of clinical and scientific research application value.

[0042] 2. Through block feature aggregation and joint optimal iteration strategy, the significance of cardiovascular regions is further enhanced, background noise interference is reduced, and segmentation boundaries are dynamically adjusted to generate high-precision 3D segmentation results. The final 3D segmentation results are combined with geometric feature extraction and topological structure construction, which not only fully reflect the structural characteristics of cardiovascular, but also accurately identify abnormal areas (such as stenosis, plaque or aneurysm), providing accurate basis for clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a structural diagram of a cardiovascular imaging analysis system according to the present invention;

[0044] Figure 2 The present invention is a flow chart of a cardiovascular imaging analysis method. DETAILED DESCRIPTION

[0045] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0047] The following is a detailed description of a cardiovascular image analysis system and method provided by the present invention in conjunction with the accompanying drawings.

[0048] Refer to the attached Figure 1 , which shows a structural diagram of a cardiovascular image analysis system provided by an embodiment of the present invention, the system includes the following parts:

[0049] Image data acquisition and preprocessing module, recursive feature modeling module, dynamic feature map generation module, variational optimization module, block feature aggregation module, joint optimal iterative segmentation module, geometric feature extraction and analysis module;

[0050] An image data acquisition and preprocessing module is used to acquire cardiovascular image data and preprocess the acquired cardiovascular image data to obtain preprocessed cardiovascular image data, and use the preprocessed cardiovascular image data as input of the recursive feature modeling module;

[0051] The recursive feature modeling module converts the preprocessed cardiovascular image data into the initial feature space, and gradually optimizes the cardiovascular image features in the initial feature space through a recursive process to obtain the recursive cardiovascular image features; in the recursive process, the expression ability of the cardiovascular image features is gradually enhanced through specific convolution operations to capture the global structure and local texture details of the cardiovascular system; the recursive cardiovascular image features form a recursive feature space containing global and local texture features; the recursive feature space is used as the input of the dynamic feature map generation module;

[0052] A dynamic feature map generation module generates a dynamic feature map value by analyzing the recursive feature space gradient through dynamic feature map; and uses the dynamic feature map value as the input of the variational optimization module;

[0053] The variational optimization module performs global optimization on the dynamic feature mapping values, constructs and minimizes the energy function, and uses the variational method to solve the optimized feature mapping values; the optimized feature mapping values ​​are used as the input of the block feature aggregation module;

[0054] The block feature aggregation module performs block operations on the optimized feature mapping values, divides the region of interest of the cardiovascular image into several small non-overlapping regions of interest, and aggregates the feature values ​​in each region block by weighted summation to form weighted feature values ​​of each region block; the weighted feature values ​​of each region block are used as inputs of the joint optimal iterative segmentation module;

[0055] The joint optimal iterative segmentation module uses the joint optimal iterative strategy to analyze and update the weighted feature values ​​of the regional blocks to dynamically adjust the segmentation boundaries of the cardiovascular region to obtain a three-dimensional segmentation result; the three-dimensional segmentation result is used as the input of the geometric feature extraction and analysis module;

[0056] The geometric feature extraction and analysis module extracts geometric features from the three-dimensional segmentation results, establishes the topological structure of the blood vessels, constructs a cardiovascular network diagram through connectivity analysis, and detects diseased areas in the cardiovascular system; through comprehensive analysis of cardiovascular images, a diagnostic report of cardiovascular images is generated to assist clinical decision-making.

[0057] Refer to the attached Figure 2 , which shows a flow chart of a cardiovascular image analysis method provided by an embodiment of the present invention, the method comprising the following steps:

[0058] S1. Obtain cardiovascular imaging data and perform preprocessing, convert the preprocessed cardiovascular imaging data into an initial feature space, gradually optimize the cardiovascular imaging features in the initial feature space through a recursive process, and obtain the recursive cardiovascular imaging features; construct a dynamic feature mapping based on the recursive feature space composed of the recursive cardiovascular imaging features, and obtain a dynamic feature mapping value; and use multi-level variational optimization to adjust the dynamic feature mapping value to obtain an optimized feature mapping value;

[0059] Obtain cardiovascular imaging data from medical imaging equipment. Cardiovascular imaging data is collected by imaging equipment such as CT or MRI. The data format is usually the internationally accepted DICOM format, so as to maintain compatibility with different medical systems in subsequent processing. Perform basic preprocessing on cardiovascular imaging data, such as cropping out the target area to remove the background part that is not related to the cardiovascular structure, so as to obtain preprocessed cardiovascular imaging data.

[0060] In order to extract multi-scale recursive feature information from cardiovascular imaging data, it is first necessary to convert the preprocessed cardiovascular imaging data into the initial feature space, and gradually optimize the cardiovascular imaging features in the initial feature space through a recursive process, so that the cardiovascular imaging features can not only capture the global structural information of the cardiovascular region, but also distinguish complex local texture changes; the recursive process generates new features through specific convolution operations, combining the advantages of feedforward networks and backward propagation, further enhancing the expression ability of the current features through the gradient information of the previous layer, and simulating the complex vascular structure in cardiovascular images; after the recursive process, the recursive cardiovascular imaging features are obtained. The recursive process of cardiovascular imaging features is defined by the following formula:

[0061] ,

[0062] ,

[0063] in, is the pixel point in the recursive process In the Cardiovascular image features output by the layer, representing pixel points In the Cardiovascular imaging feature values ​​in layer feature space; Indicates The weight matrix of the layer is learned during the training process; is a bias term used to adjust the central distribution of features; activation function ReLU is usually used to avoid the gradient vanishing problem and simulate nonlinear feature changes; is the pixel point in the recursive process In the Cardiovascular imaging features of the layer output; It is the gradient of the cardiovascular image features of the previous layer. The detailed information of the cardiovascular edge is captured by calculating the spatial variation of the cardiovascular image features of the previous layer. The gradient regularization coefficient The value of is adjusted through experimental optimization and is usually Through multi-layer recursion, the feature space contains multi-scale recursive features and dynamic change information, providing high-quality input for subsequent dynamic feature map generation.

[0064] Based on the recursive feature space composed of recursive cardiovascular image features, a dynamic feature map reflecting changes in cardiovascular regions is constructed; the goal of dynamic feature mapping is to give cardiovascular regions higher weights through significance analysis of the gradient of the recursive feature space, while suppressing background noise interference; dynamic feature mapping converts the change law of pixel gradients in the recursive feature space into significance weights through integral operations, and gradually forms a eigenvalue distribution that corresponds to the actual cardiovascular region. The specific formula of dynamic feature mapping is:

[0065] ,

[0066] in, is the dynamic feature map value, indicating the pixel point Importance in the entire recursive feature space; By weighted accumulation of all gradient amplitudes in the recursive feature space, the areas with significant changes, especially the edges, are highlighted. The influence of the size of the region of interest was eliminated by normalization, stabilizing the distribution of results; is the region of interest, defined as the set of pixels containing cardiovascular structures; It is The gradient of the cardiovascular image features of the layer; in order to prevent the denominator from being zero, the value is added , usually taken .

[0067] In order to make the generated dynamic feature map values ​​smoother and enhance global consistency, multi-level variational optimization is used to further adjust the dynamic feature map values. By constructing an energy function, the gradient smoothness, feature consistency and sparsity are integrated into a unified goal, aiming to eliminate noise interference while enhancing the significance of the cardiovascular area. The mathematical expression of the energy function is:

[0068] ,

[0069] in, is the energy function, representing the dynamic feature map value The overall optimization target value of Used to control the smoothness of dynamic feature maps in space, is the gradient amplitude of the dynamic feature map value, and the weight parameter It is usually determined by experimental verification. ; is the gradient of the dynamic feature map value; feature consistency term Emphasizes dynamic feature mapping values ​​and target feature values The closeness, weight parameter Dynamically adjusted during the training process, directly related to the contrast of the input cardiovascular image; target feature value It is usually set to the average characteristic value of the cardiovascular region, or set according to the specific implementation scenario; the sparsity regularization term Used to reduce the impact of noise on dynamic feature mapping, weight parameters Determined by optimization method, usually .

[0070] Minimize the energy function To obtain the optimized feature map, the process of minimizing the energy function requires the use of variational methods. By calculating the energy function About Dynamic Feature Map Values The first-order variational derivative of and set it to zero to obtain the optimal solution. The first-order variational derivative of the energy function is:

[0071] ,

[0072] Let the variational derivative be zero, that is, to meet the optimal condition, and Expressed in explicit form:

[0073] ,

[0074] Taking the exponential to eliminate the logarithmic term, assuming , taking the exponents on both sides to obtain the analytical solution of the optimized feature map:

[0075] ,

[0076] Optimized feature map values It is a smooth and locally coherent feature map, and its saliency is highly consistent with the structural distribution of cardiovascular regions.

[0077] S2. Block feature aggregation is used to perform block operations on the optimized feature mapping values, and a joint optimal iterative strategy is used to dynamically adjust the segmentation boundaries of the cardiovascular region to obtain a three-dimensional segmentation result; based on the three-dimensional segmentation result, the geometric features of the cardiovascular are extracted, and a vascular network diagram is constructed; by detecting abnormal areas in the three-dimensional segmentation results, the lesion area is determined, and finally diagnostic information is generated.

[0078] In order to better enhance the significance of the cardiovascular region and reduce the interference of background noise on the segmentation results, block feature aggregation is used to perform block operations on the optimized feature mapping values. Specifically, the region of interest of the entire cardiovascular image is divided into several small non-overlapping regions of interest. The eigenvalues ​​in each region block are aggregated by weighted summation to form weighted eigenvalues ​​representing the region block; by weighting the gradient features in the region block, the dynamic significance of the cardiovascular edge is highlighted, while avoiding the computational complexity problem caused by direct operation of pixel-level data; within the region block, the eigenvalue of each pixel is dynamically weighted according to the significance of its gradient amplitude, while considering the overall normalization characteristics of the region block to ensure that the feature contrast between different region blocks can remain stable; thereby enhancing the consistency of the region and ensuring that the segmentation results can not only accurately outline the boundaries of the blood vessels, but also maintain continuity and consistency visually. The mathematical expression of aggregation is as follows:

[0079] ,

[0080] in, It is a regional block The weighted eigenvalue of represents the area block The overall characteristic strength of Weighted to emphasize the characteristics of significant edges in the region; normalization factor It is the bi-norm of the optimized feature map value, which is used to balance the gradient value.

[0081] In order to further dynamically adjust the segmentation boundary of the cardiovascular region, the weighted eigenvalues ​​of the regional block are analyzed and updated using the joint optimal iteration strategy; in each iteration, the segmentation result is dynamically adjusted according to the segmentation status of the previous round and the weighted eigenvalues ​​of the current regional block; the initial segmentation result is obtained by the segmentation threshold method, and the Otsu method is usually used to calculate the segmentation threshold. The specific formula is as follows:

[0082] ,

[0083] in, is the segmentation threshold; is the initial threshold; Indicates the initial threshold The corresponding between-class variance is used to describe the contrast between foreground and background; Represents the intra-class variance, reflecting the tightness of the current segmentation result. The optimized feature map value is binarized through the initial threshold to generate the initial segmentation result ; In each subsequent iteration, the segmentation result is dynamically adjusted by analyzing the weighted eigenvalues ​​of the current region block. The segmentation formula for each iteration is:

[0084] ,

[0085] in, It is The segmentation result of the iteration; It is The segmentation result of the iteration; is the step length, usually ; is a symbolic function defined as ; Segmentation threshold It is a value that is dynamically adjusted based on the weighted feature value of the region block to ensure that the segmentation result can better fit the actual vascular structure in each round of iteration. Through round-by-round iterative optimization, accurate and coherent segmentation results are finally obtained. , which combines the global segmentation capability of initialization and the flexibility of dynamic adjustment of local areas, can significantly improve the accuracy and adaptability of segmentation.

[0086] The final segmentation result is a high-precision three-dimensional segmentation result that identifies the cardiovascular structure. The voxel distribution of the segmentation result is analyzed to extract the geometric features of the cardiovascular, such as the vessel diameter, length, branch angle, etc., to obtain a voxel set. The method of extracting geometric features can be used with existing technology. The topological structure of the blood vessel is constructed, and the vascular branches, nodes and connection relationships are determined by analyzing the connectivity of the segmented area. Using the voxel set in the segmentation result, a vascular network diagram is established.

[0087] Abnormal areas (such as stenosis, plaque or aneurysm) are detected in the segmentation results, and the lesion area is determined by comparing the local changes of geometric features with the size of the threshold, such as the sharp reduction of the blood vessel diameter or the abnormal increase of the curvature. The threshold is set according to the expert experience method. The segmentation results are used to achieve a comprehensive analysis of cardiovascular images, covering geometric structure and lesion identification, and finally generate diagnostic information for clinical use.

[0088] In summary, a cardiovascular image analysis system and method are completed.

[0089] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0090] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0091] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A cardiovascular image analysis method, characterized in that: The following steps are involved: S1, obtaining cardiovascular imaging data and preprocessing it, converting the preprocessed cardiovascular imaging data into an initial feature space, and gradually optimizing the cardiovascular imaging features in the initial feature space through a recursive process to obtain the recursive cardiovascular imaging features; Based on the recursive feature space composed of cardiovascular image features after recursion, a dynamic feature map is constructed. The dynamic feature map converts the variation law of pixel gradients in the recursive feature space into a significance weight through an integral operation, forms a feature value distribution corresponding to the actual cardiovascular region, and obtains a dynamic feature map value. Multi-level variational optimization is used to adjust the dynamic feature mapping value. In the process of adjusting the dynamic feature mapping value, an energy function is constructed to integrate gradient smoothness, feature consistency and sparsity, and the energy function is minimized by using a variational method. The optimized feature mapping value is obtained by calculating the first-order variational derivative of the energy function with respect to the dynamic feature mapping value and setting the first-order variational derivative to zero. S2. Block feature aggregation is used to perform block operations on the optimized feature mapping values, and a joint optimal iterative strategy is used to dynamically adjust the segmentation boundaries of the cardiovascular region to obtain a three-dimensional segmentation result; based on the three-dimensional segmentation result, the geometric features of the cardiovascular are extracted, and a vascular network diagram is constructed; by detecting abnormal areas in the three-dimensional segmentation results, the lesion area is determined, and finally diagnostic information is generated.

2. The cardiovascular image analysis method according to claim 1, characterized in that: The S1 specifically includes: The recursive process of cardiovascular image characterization is defined by the following formula: ， in, is the pixel point in the recursive process In the Cardiovascular imaging features of the layer output; Indicates The weight matrix of the layer; is the bias term; is the activation function; is the pixel point in the recursive process In the Cardiovascular imaging features of the layer output; is the gradient of the cardiovascular imaging features in the previous layer; is the gradient regularization coefficient.

3. The cardiovascular image analysis method according to claim 2, characterized in that: The S1 specifically includes: The specific formula of dynamic feature mapping is: ， in, is the dynamic feature map value; The area of ​​interest All pixels within Perform integration; is the region of interest, defined as the set of pixels containing cardiovascular structures; It is Gradients of layer cardiovascular imaging features; Used to prevent the denominator from being zero.

4. The cardiovascular image analysis method according to claim 3, characterized in that: The S1 specifically includes: The mathematical expression of the energy function is: ， in, is the energy function, representing the dynamic feature map value The optimization target value of is the gradient smoothing term, is the gradient magnitude of the dynamic feature map value; is the gradient of the dynamic feature map value; is the feature consistency item; is the target feature value; is the sparsity regularization term; is the weight parameter.

5. The cardiovascular image analysis method according to claim 1, characterized in that: The S2 specifically includes: The region of interest of the cardiovascular image is divided into two or more non-overlapping regions of interest, and the eigenvalues ​​in each region block are aggregated by weighted summation to form weighted eigenvalues ​​representing each region block.

6. The cardiovascular image analysis method according to claim 5, characterized in that: The S2 specifically includes: The weighted eigenvalues ​​of each region block are analyzed and updated using the joint optimal iteration strategy. The initial segmentation result is obtained by the segmentation threshold method. In each iteration, the segmentation result is dynamically adjusted according to the segmentation status of the previous round and the weighted eigenvalues ​​of the current region block. The segmentation formula for each iteration is: ， in, It is The segmentation result of the iteration; It is The segmentation result of the iteration; is the step length; is a sign function; It is a regional block The weighted eigenvalue of ; is the segmentation threshold.

7. The cardiovascular image analysis method according to claim 1, characterized in that: The S2 specifically includes: The final segmentation result is a three-dimensional segmentation result that identifies the cardiovascular structure. The voxel distribution of the segmentation result is analyzed to extract the geometric features of the cardiovascular system. The topological structure of the blood vessels is constructed, the connectivity of the segmented area is analyzed, and the voxel set in the segmentation result is used to establish a vascular network diagram.

8. A cardiovascular image analysis system, applied to the cardiovascular image analysis method according to any one of claims 1 to 7, characterized in that: Includes the following parts: Image data acquisition and preprocessing module, recursive feature modeling module, dynamic feature map generation module, variational optimization module, block feature aggregation module, joint optimal iterative segmentation module, geometric feature extraction and analysis module; An image data acquisition and preprocessing module is used to acquire and preprocess cardiovascular image data, and use the preprocessed cardiovascular image data as input to the recursive feature modeling module; The recursive feature modeling module converts the preprocessed cardiovascular image data into the initial feature space, and gradually optimizes the cardiovascular image features in the initial feature space through a recursive process to obtain the recursive cardiovascular image features; the recursive feature space is formed by the recursive cardiovascular image features; and the recursive feature space is used as the input of the dynamic feature map generation module; A dynamic feature map generation module generates a dynamic feature map value by analyzing the recursive feature space gradient through dynamic feature map; and uses the dynamic feature map value as the input of the variational optimization module; The variational optimization module performs global optimization on the dynamic feature mapping values, constructs and minimizes the energy function, and uses the variational method to solve the optimized feature mapping values; the optimized feature mapping values ​​are used as the input of the block feature aggregation module; The block feature aggregation module performs block operations on the optimized feature mapping values, divides the region of interest of the cardiovascular image into two or more non-overlapping regions of interest, and aggregates the feature values ​​in each region block by weighted summation to form weighted feature values ​​of each region block; the weighted feature values ​​of each region block are used as inputs of the joint optimal iterative segmentation module; The joint optimal iterative segmentation module uses the joint optimal iterative strategy to analyze and update the weighted feature values ​​of the region block to obtain a three-dimensional segmentation result; the three-dimensional segmentation result is used as the input of the geometric feature extraction and analysis module; The geometric feature extraction and analysis module extracts geometric features from the three-dimensional segmentation results, establishes the topological structure of the blood vessels, constructs a cardiovascular network diagram through connectivity analysis, detects diseased areas in the cardiovascular system, and generates diagnostic reports for cardiovascular images to assist clinical decision-making.

Images