Coronary angiography calcification quantitative evaluation method and system based on shape topological graph network

Abstract

The invention relates to a coronary angiography calcification quantitative evaluation method and system based on a shape topological graph network. The method comprises the following steps: acquiring a coronary angiography image, performing standardization processing, performing processing through a twin network and a shared weight, and optimizing the twin network by minimizing contrast loss to obtain matched coronary images before and after development; introducing a Focal Loss loss function into the deep learning model, performing region segmentation on the matched coronary artery image before and after development, and identifying a calcified region in a segmentation result as a node; extracting geometric features and topological features of the nodes to construct a topological graph structure; performing graph convolution and global pooling technical operation through a graph isomorphic network to obtain a global feature vector; and obtaining a calcification quantitative evaluation result based on the global feature vector. A topological graph structure is constructed, and graph convolution, pooling and other operations are performed through a graph isomorphic network, so that the global relationship between the calcification areas can be captured, and the accuracy of calcification integral prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a method and system for quantitatively evaluating coronary angiography calcification based on a shape topology graph network. Background Art

[0002] Coronary artery disease (CAD) is one of the leading causes of illness and mortality worldwide. The development of CAD is closely associated with coronary atherosclerosis, with coronary artery calcification (CAC) being a key hallmark of atherosclerosis. The severity and quantification of CAC are considered important predictors of future cardiovascular event risk and a crucial tool for assessing a patient's cardiovascular health. Traditionally, CAC scores have been measured manually or semi-automatically using computed tomography (CT) images, such as the Agatston score. Specifically, the patient's pathological image is first segmented to identify calcified areas, and the calcification score is then estimated based on the segmentation results. Traditional methods are based on coronary CT images. While coronary CT can demonstrate calcification and vascular stenosis, it has limited ability to visualize subtle lesions such as soft plaques and microthrombi in the vessel wall, making it difficult to comprehensively assess vascular health. Furthermore, these methods are cumbersome, time-consuming, and operator-dependent, and are susceptible to inter- and intra-observer variability.

[0003] Coronary angiography, an X-ray imaging technique that uses contrast agents to visualize the coronary arteries, is widely used to assess the heart's blood flow. Its primary uses include diagnosing coronary artery disease (CAD), assessing the degree of vascular stenosis or obstruction, creating three-dimensional vascular modeling, and guiding interventional therapy. Coronary angiography provides real-time visualization of blood flow in the coronary arteries, helping physicians assess hemodynamic changes and determine the actual impact of stenosis on blood flow. Furthermore, it provides detailed coronary anatomy, helping physicians accurately diagnose CAD and its severity. Therefore, coronary angiography is considered the gold standard for assessing CAD stenosis and obstruction. Coronary artery stenosis refers specifically to a narrowing of the coronary artery lumen, typically caused by the accumulation of atherosclerotic plaques. This stenosis reduces blood flow to the myocardium, leading to myocardial ischemia and, in turn, serious cardiovascular events such as chest pain and myocardial infarction. Accurate identification and quantitative analysis of stenosis are crucial for the detection and treatment of cardiovascular disease.

[0004] However, both coronary CT and coronary angiography rely on imaging to observe the coronary arteries and determine the extent of coronary stenosis. However, detecting and quantifying coronary stenosis does not accurately reflect the patient's calcification status. Some stenotic locations lack calcification, and as the degree of calcification increases, the severity of stenosis flattens, and the correlation between the two gradually decreases. Therefore, traditional methods for determining coronary calcification often suffer from low accuracy in quantitative calcification assessment due to the small size and uneven distribution of calcified areas. Summary of the Invention

[0005] Based on this, in order to solve the above technical problems, a method and system for quantitative evaluation of coronary angiography calcification based on shape topology network is provided, which can improve the accuracy of calcification score prediction.

[0006] A method for quantitatively evaluating coronary angiography calcification based on a shape topology network, the method comprising:

[0007] Acquire coronary angiography images and perform standardization processing to obtain processed coronary artery images;

[0008] Inputting the coronary artery images before and after development into the twin network, processing the two coronary artery images by sharing weights, calculating the difference between the two coronary artery images, and optimizing the twin network by minimizing contrast loss to obtain matched coronary artery images before and after development;

[0009] Introducing a Focal Loss function into the deep learning model, performing regional segmentation on the matched coronary artery images before and after development, and identifying calcified areas in the segmentation results as nodes;

[0010] Extracting geometric features and topological features of the nodes, and constructing a topological graph structure based on the geometric features and topological features;

[0011] The topological graph structure is input into a graph isomorphism network, a graph convolution operation is performed on the topological graph structure through the graph isomorphism network, and a global pooling technique is used to summarize features to obtain a global feature vector; and a calcification quantitative assessment result is obtained based on the global feature vector.

[0012] In one embodiment, obtaining a coronary angiography image and performing normalization processing to obtain a processed coronary artery image includes:

[0013] Acquiring a coronary angiography image, adjusting the brightness and contrast of the coronary angiography image, and performing normalization processing on the coronary angiography image to obtain a preliminary processed image;

[0014] Noise in the preliminary processed image is eliminated, and the contrast of the calcified area is enhanced by an enhancement algorithm to obtain a processed coronary artery image.

[0015] In one embodiment, calculating the difference between the two coronary artery images, optimizing the twin network by minimizing contrast loss, and obtaining matched coronary artery images before and after development, comprises:

[0016] respectively calculating the Euclidean distance and cosine similarity between the two coronary artery images;

[0017] Calculating the difference between the two coronary artery images according to the Euclidean distance and cosine similarity;

[0018] Determining a contrast loss, and optimizing the twin network by minimizing the contrast loss, wherein the optimized twin network minimizes the distance between similar coronary artery image pairs and maximizes the distance between dissimilar image pairs;

[0019] Based on the difference, the optimized twin network is used to perform a matching operation on the two coronary artery images to obtain matched coronary artery images before and after development.

[0020] In one embodiment, a FocalLoss loss function is introduced into a deep learning model to perform regional segmentation on the matched coronary artery images before and after development, including:

[0021] Determine a cross entropy loss function in a deep learning model and introduce a modulation factor into the cross entropy loss function to obtain a Focal Loss loss function;

[0022] The twin network after introducing the Focal Loss loss function is used to detect and segment the regions in the matched coronary artery images before and after development.

[0023] In one embodiment, identifying the calcified region in the segmentation result as a node includes:

[0024] The calcified area in each segmentation result is marked and identified as a node, and each node represents a separate calcified area.

[0025] In one embodiment, extracting the geometric features and topological features of the nodes, and constructing a topological graph structure based on the geometric features and topological features includes:

[0026] For each of the nodes, extracting geometric features used to describe the physical characteristics of the calcified area, and calculating the topological features of each of the nodes;

[0027] Based on the geometric features and topological features, the Euclidean distance between any two node centroids is calculated. When the Euclidean distance is less than or equal to a distance threshold, an edge connection is established between the two nodes, and the connection weight is calculated to construct a topological graph structure.

[0028] In one embodiment, the topological graph structure is input into a graph isomorphism network, and a graph convolution operation is performed on the topological graph structure through the graph isomorphism network, including:

[0029] Inputting the topological graph structure into a graph isomorphic network, determining learnable parameters in the graph isomorphic network, and obtaining a set of neighbor nodes of each node in the topological graph structure;

[0030] Calculating updated features of each node based on the learnable parameters and the set of neighboring nodes;

[0031] A node feature matrix is ​​obtained according to the updated features, a graph adjacency matrix is ​​obtained according to the neighbor node set, and high-dimensional features of each node and graph in the topological graph structure are obtained based on the node feature matrix and the graph adjacency matrix.

[0032] In one embodiment, a global pooling technique is used to aggregate features to obtain a global feature vector, including:

[0033] A global pooling technique is used to aggregate the high-dimensional features of each node in the topological graph structure and the graph to obtain a global feature vector.

[0034] In one embodiment, obtaining a quantitative calcification assessment result based on the global feature vector includes:

[0035] The global feature vector is input into a fully connected layer, and regression prediction of the calcification integral is performed through the fully connected layer to output a calcification quantitative assessment result.

[0036] A coronary angiography calcification quantitative assessment system based on a shape topology network, the system comprising:

[0037] An image preprocessing module is used to obtain coronary angiography images and perform standardization processing to obtain processed coronary artery images;

[0038] an image matching module, configured to input the coronary artery images before and after development into a twin network, process the two coronary artery images by sharing weights, calculate the difference between the two coronary artery images, and optimize the twin network by minimizing contrast loss to obtain matched coronary artery images before and after development;

[0039] a calcification region segmentation module, configured to introduce a FocalLoss loss function into a deep learning model, perform region segmentation on the matched coronary artery images before and after development, and identify the calcification regions in the segmentation results as nodes;

[0040] A topology graph structure construction module is used to extract the geometric features and topological features of the nodes and construct a topology graph structure based on the geometric features and topological features;

[0041] The calcification quantitative assessment module is used to input the topological graph structure into a graph isomorphism network, perform graph convolution operation on the topological graph structure through the graph isomorphism network, and use global pooling technology to summarize features to obtain a global feature vector; and obtain a calcification quantitative assessment result based on the global feature vector.

[0042] The above-mentioned coronary angiography calcification quantitative assessment method and system based on shape topology network effectively solves the matching error problem caused by heart beating and morphological changes by introducing twin networks for precise image matching; through weight sharing and contrast loss functions, the consistency of image shape and position can be ensured, thereby improving the annotation accuracy of calcification areas; Focal Loss loss function can improve the focusing ability during calcification area segmentation and reduce interference from background areas; constructing a topological map structure and performing graph convolution, pooling and other operations through a graph isomorphism network can capture the global relationship between calcification areas and improve the accuracy of calcification integral prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a diagram illustrating an application environment of a method for quantitatively evaluating coronary angiography calcification based on a shape topology network in one embodiment;

[0044] Figure 2 1 is a flow chart of a method for quantitatively evaluating coronary angiography calcification based on a shape topology network in one embodiment;

[0045] Figure 3 A schematic diagram of twin network matching results in one embodiment;

[0046] Figure 4 is a schematic diagram of the calcification area segmentation result in one embodiment;

[0047] Figure 5 A comparison chart of calcification score prediction results of various network models in one embodiment;

[0048] Figure 6 FIG1 is a structural block diagram of a coronary angiography calcification quantitative assessment system based on a shape topology network in one embodiment;

[0049] Figure 7 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] The method for quantitative evaluation of coronary angiography calcification based on shape topology network provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can acquire a coronary angiography image and perform normalization processing to obtain a processed coronary image; the computer device 110 can input the coronary images before and after development into a twin network, process the two coronary images by sharing weights, calculate the difference between the two coronary images, optimize the twin network by minimizing contrast loss, and obtain matched coronary images before and after development; the computer device 110 can introduce a FocalLoss loss function into a deep learning model, perform regional segmentation on the matched coronary images before and after development, and identify the calcification area in the segmentation result as a node; the computer device 110 can extract the geometric features and topological features of the nodes, and construct a topological graph structure based on the geometric features and topological features; the computer device 110 can input the topological graph structure into a graph isomorphism network, perform a graph convolution operation on the topological graph structure through the graph isomorphism network, and use a global pooling technology to summarize features to obtain a global feature vector; and obtain a quantitative calcification assessment result based on the global feature vector. The computer device 110 may be, but is not limited to, various personal computers, laptops, smart phones, robots, tablet computers, and other devices.

[0052] In one embodiment, Figure 2 As shown, a method for quantitative evaluation of coronary angiography calcification based on a shape topology network is provided, comprising the following steps:

[0053] Step 202: Acquire a coronary angiography image and perform standardization processing to obtain a processed coronary artery image.

[0054] The computer device can acquire various coronary angiography images. In order to ensure the consistency of various coronary angiography images, the computer device can standardize the various coronary angiography images so that they have a uniform pixel range and a uniform resolution to ensure consistency in subsequent processing.

[0055] In one embodiment, a method for quantitatively assessing coronary angiography calcification based on a shape topology network is provided, which may also include a process of standardizing the image. The specific process includes: acquiring a coronary angiography image, adjusting the brightness and contrast of the coronary angiography image, and performing normalization processing to obtain a preliminary processed image; eliminating noise in the preliminary processed image, and enhancing the contrast of the calcified area through an enhancement algorithm to obtain a processed coronary image.

[0056] Computer equipment can adjust the brightness and contrast of acquired coronary angiography images and perform normalization processing to ensure consistency between coronary angiography images from different sources during subsequent processing. Because coronary images often contain noise, which can affect subsequent segmentation, further denoising can be performed. Specifically, methods such as Gaussian filtering and mean filtering can be used to remove noise from the image, retaining valid image information and helping to improve the accuracy of subsequent segmentation of calcified areas.

[0057] Next, the contrast of the calcified areas can be enhanced through an enhancement algorithm. In this embodiment, histogram equalization can be used to enhance image contrast, particularly the contrast of the calcified areas, making them more prominent in the image. This facilitates subsequent identification and segmentation of the calcified areas and aids in the training of the subsequent segmentation model. The computer device can also use data augmentation techniques (such as rotation, translation, and scaling) to increase the number of samples in the training set, reduce the risk of model overfitting, improve model generalization, and increase sample diversity, thereby enhancing the robustness of the model.

[0058] In step 204, the coronary artery images before and after development are input into the twin network, the two coronary artery images are processed by sharing weights, and the difference between the two coronary artery images is calculated. The twin network is optimized by minimizing the contrast loss to obtain matched coronary artery images before and after development.

[0059] Coronary angiography images usually have shape dislocation or deformation due to factors such as heartbeat, so matching processing is required. In order to improve the segmentation accuracy, in this embodiment, a Siamese Network can be used for image matching. Among them, the Siamese Network consists of two identical sub-networks, and the input is two coronary angiography images at different time points. The network processes the two images by sharing weights. The image matching results before and after development are as follows. Figure 3 As shown in FIG, the matching process solves the shape misalignment problem caused by heartbeat in coronary angiography. It can find the coronary artery image after development that best matches the image before development, providing an accurate reference for the subsequent segmentation of the calcified area.

[0060] In one embodiment, a method for quantitative assessment of coronary angiography calcification based on a shape topology network is provided, which may also include an image matching process, the specific process including: calculating the Euclidean distance between two coronary images respectively; calculating the difference between the two coronary images based on the Euclidean distance; determining the contrast loss, and optimizing the twin network by minimizing the contrast loss, wherein the optimized twin network minimizes the distance between similar coronary image pairs and maximizes the distance between dissimilar image pairs; and performing a matching operation on the two beating coronary images based on the difference to obtain matched coronary images before and after development.

[0061] The key to the Siamese network is to achieve matching by calculating the similarity between image pairs. In this embodiment, the computer device can use Euclidean distance (Cosine Similarity) as a similarity metric. Based on this metric, the difference between the two image pairs is calculated. The network is then optimized by minimizing contrast loss, minimizing the distance between similar image pairs and maximizing the distance between dissimilar image pairs, ultimately achieving accurate matching of beating coronary artery images.

[0062] Specifically, in this embodiment, by calculating the distance, the contrast loss is minimized, so that the network can accurately match the coronary artery shape in a beating state. The Euclidean distance can be expressed as: I1 and I2 represent the two coronary artery images input respectively. The formula for cosine similarity can be:

[0063] During the training process of the Siamese network, contrastive loss can be used to optimize the model to minimize the distance between similar image pairs and maximize the distance between dissimilar image pairs. The loss function can be defined as: Where y is the label of the image pair (1 means similar, 0 means dissimilar), d eucl is the Euclidean distance, and m is the minimum separation of dissimilar image pairs.

[0064] In step 206, a Focal Loss function is introduced into the deep learning model to perform regional segmentation on the matched coronary artery image to obtain calcified areas, perform boundary detection on the calcified areas, and identify the calcified areas in the segmentation results as nodes.

[0065] In view of the sparsity and imbalance of calcification areas, in this embodiment, the Focal Loss function can be used to train the deep learning model. The calcification area segmentation results are as follows: Figure 4 As shown in the figure, Focal Loss can reduce the impact of background areas on the model and enhance the model's ability to learn calcified areas. This can significantly improve segmentation accuracy, especially when the calcified areas are small or blurred. After obtaining the matched coronary artery images, the Focal Loss loss function can be further used to accurately segment the calcified areas and construct the graph structure.

[0066] In coronary angiography images, calcified regions typically appear as small, irregularly shaped areas with relatively sparse distribution. This results in a standard binary segmentation task where background pixels (non-calcified regions) far outnumber foreground pixels (calcified regions). Under the standard cross-entropy loss, all samples contribute equally to the gradient, causing the model to favor background objects during training, making it difficult to focus on feature extraction and learning of foreground calcified regions.

[0067] In one embodiment, a method for quantitatively evaluating coronary angiography calcification based on a shape topology network is provided, which may also include a process of introducing a loss function to segment the calcified area. The specific process includes: determining the cross-entropy loss function in the deep learning model, and introducing a modulation factor into the cross-entropy loss function to obtain a Focal Loss loss function; using the twin network after introducing the Focal Loss loss function to detect and segment the areas in the matched coronary artery images before and after development.

[0068] The Focal Loss loss function introduces a modulation factor based on the cross-entropy loss to reduce the contribution of easy-to-classify samples to the loss calculation and highlight the weight of hard-to-classify samples. The standard binary cross-entropy loss function (BCE) can be expressed as: CE(p,y) = -[ylog(p) + (1-y)log(1-p)]; where y is the true label and p is the predicted probability. Focal Loss adds a modulation factor (1-p) on this basis. γ And add a balance factor α to adapt to category imbalance. The Focal Loss loss function can be expressed as: FL(p,y)=-αy(1-p) γ log(p)-(1-α)(1-y)p γ log(1-p); where γ≥0 is the modulation factor exponent. When γ>0, the gradient contribution of samples that the model can easily and correctly classify is greatly reduced; while for difficult samples, the loss will be amplified, increasing the model's attention to these samples. The parameter α is used to balance the ratio of positive and negative samples.

[0069] To address the technical deficiencies of existing solutions, such as insufficient segmentation accuracy for calcified areas and greater difficulty in matching images of beating coronary arteries, this application introduces a twin network to accurately match images before and after imaging, effectively resolving the matching error caused by heart beating and morphological changes. By using a shared weight and contrast loss function, the consistency of image shape and position is ensured, thereby improving the accuracy of labeling calcified areas. Furthermore, the combination of the FocalLoss loss function improves the model's focusing ability during calcified area segmentation, reduces background interference on the model, and maintains a high level of segmentation accuracy, especially when the calcified areas are sparse or inconspicuous.

[0070] In one embodiment, a method for quantitatively evaluating coronary angiography calcification based on a shape topology network may further include a node identification process, specifically including: marking each calcification region in the segmentation result and identifying it as a node, where each node represents a separate calcification region.

[0071] After calcification segmentation, each segmentation result image is first subjected to boundary detection using morphological operations (such as erosion and dilation). Each calcification region in the segmentation result is identified as a separate node, each representing a small calcification area. Features such as the area, shape, and location of the calcification region are used in subsequent graph neural network processing.

[0072] Step 208: extract the geometric features and topological features of the nodes, and construct a topological graph structure based on the geometric features and topological features.

[0073] Based on the segmentation of the calcified area, the computer equipment can extract the features of the nodes and construct the graph to obtain the topological graph structure.

[0074] In one embodiment, a method for quantitative assessment of coronary angiography calcification based on a shape topology graph network is provided, which may also include a process of feature extraction and graph construction. The specific process includes: for each node, extracting geometric features used to describe the physical characteristics of the calcified area, and calculating the topological features of each node; based on the geometric features and topological features, calculating the Euclidean distance between the centroids of any two nodes; when the Euclidean distance is less than or equal to the distance threshold, establishing an edge connection between the two nodes, calculating the connection weight, and constructing a topological graph structure.

[0075] For each node, the computer device can extract its geometric features, including the area A i , perimeter P i , Compactness, Spindle Ratio SR i , fractal dimension FD i , Euler number χ iEtc. Among them, the compactness can be expressed as These geometric features can describe the physical properties of each calcification area and help further analyze its distribution in the coronary artery.

[0076] Computer equipment can also construct a relationship model between nodes, namely a topological graph structure, by calculating topological features such as the distance, relative position, and adjacency between nodes. It can be expressed as: i =[A i ,P i ,Compactness i ,SR i ,FD i ,χ i ] Τ The computer device can also construct a series of nested sub-complexes of the filter complex K based on the spatial distribution of the calcified area. Each layer K t Corresponding to different scale parameters t, it is defined as: Among them, B(v i ,t) represents node v i For each topological dimension d (such as 0-dimensional, 1-dimensional), record the generation and extinction moments of the topological features:

[0077] in, and are the generation and extinction moments of the i-th d-dimensional topological feature respectively; B d is a persistent strip; then the persistent strip is converted into a feature vector Used for subsequent graph construction tasks.

[0078] In this embodiment, the topological features that can be selected include zero persistence, including the mean μ 0,i , maximum value Standard deviation σ 0,i ; 1 Maintain durability, including mean μ 1,i , maximum value Standard deviation σ 1,i , so the topological feature vector of the node can be expressed as: Node feature vector h i =(s i ,t i ) is expressed as: For any two nodes v i and v j , calculate the Euclidean distance d between their centroids ij : Where (x i ,y i ) and (x j ,y j ) are nodes v i and v j The centroid coordinates of ; set a distance threshold τ, when d ij ≤τ, node v i and v j Edge ij The weight of the edge is set to 1, and the connectivity of the graph is ensured while maintaining the sparsity of the graph by adjusting the distance threshold τ. The constructed calcification segmentation graph G = (V, E) satisfies: V = {v1, v2, ..., v N},E={e ij ∣d ij ≤τ}. That is, by setting a distance threshold, if the distance between two nodes is less than the threshold, a connection is established between the two nodes. The established connection is the edge in the graph, and the weight of the edge can be calculated based on the distance between the nodes.

[0079] In this example, each node corresponds to a calcified area, and edges are established using distance weights between nodes. Edge weights between nodes are calculated using factors such as Euclidean distance and relative position to define the topological relationships between calcified areas. The resulting topological structure serves as input data for further processing in the subsequent Graph Isomorphism Network (GIN).

[0080] The calcification segmentation results are converted into a topological graph structure containing nodes, and feature extraction and quantitative evaluation are performed through a graph isomorphic network. Each calcification area is constructed as a node in the graph through its geometric and topological features, and the relationship between nodes is connected through weights to form edges. This not only considers the local characteristics of a single calcification area, but also effectively captures the global relationship between calcification areas, providing a more accurate calcification score prediction. Compared with traditional methods, it significantly improves the accuracy and comprehensiveness of calcification quantitative evaluation.

[0081] In step 210 , the topological graph structure is input into a graph isomorphism network, graph convolution is performed on the topological graph structure through the graph isomorphism network, and global pooling technology is used to aggregate features to obtain a global feature vector; a quantitative calcification assessment result is obtained based on the global feature vector.

[0082] Computers can use the constructed topological graph structure as input to a graph isomorphism network (GIN). GIN effectively extracts complex relationships between nodes and learns global graph structural features. The network gradually extracts high-dimensional features for each node and the entire graph through multi-layer convolution operations, gradually aggregating node features.

[0083] In one embodiment, a method for quantitative assessment of coronary angiography calcification based on a shape topological graph network is provided, which may also include a process of performing a convolution operation. The specific process includes: inputting the topological graph structure into a graph isomorphism network, determining the learnable parameters in the graph isomorphism network, and obtaining a set of neighbor nodes for each node in the topological graph structure; calculating updated features of each node based on the learnable parameters and the set of neighbor nodes; obtaining a node feature matrix based on the updated features, obtaining a graph adjacency matrix based on the set of neighbor nodes, and obtaining high-dimensional features of each node in the topological graph structure and the graph based on the node feature matrix and the graph adjacency matrix.

[0084] Graph isomorphism networks can extract complex relationships between nodes and learn global graph structural features. The learned features include node geometry and topological relationships. Each node updates its own feature representation by aggregating information from its neighboring nodes. The formula is: in, is the updated node v at layer k i The eigenvector of is a learnable parameter used to adjust the weight of the node’s own features. N(i) is the weight of the node v i The neighbor node set represents the feature splicing operation, MLP (k) is the k-th layer of the multi-layer perceptron, which includes several fully connected layers and activation functions. The model uses a multi-layer GIN stacking method. Each layer of GIN not only aggregates the information of neighboring nodes, but also performs nonlinear transformation through MLP, which is specifically expressed as: H (k) =GINLayer (k) (H (k-1) ,A); among them, is the node feature matrix of the k-th layer, is the adjacency matrix of the graph, defined as: GINLayer (k) It is the GIN module of the kth layer, which includes feature aggregation and MLP transformation.

[0085] In this embodiment, the feature representation of each node is updated based on the information of its neighboring nodes. In each convolution layer, each node aggregates the features of its neighboring nodes. By stacking multiple layers of GIN, the complex relationships between nodes are extracted. The node features are fused through graph convolution operations to form a global feature representation of the graph.

[0086] Specifically, in one embodiment, a method for quantitatively evaluating coronary angiography calcification based on a shape topology graph network is provided, which may also include a feature aggregation process. The specific process includes: using global pooling technology to aggregate the high-dimensional features of each node in the topology graph structure and the graph to obtain a global feature vector.

[0087] After the graph convolution operation is completed, the features of all nodes are summarized using global pooling techniques (such as maximum pooling or average pooling) to obtain the global feature vector of the graph. By expanding the stack of multiple layers of GIN, we can get: Commonly used global pooling methods include sum pooling, average pooling and max pooling: G =Pooling(H (k) ); where K is the number of layers of GIN, is the global eigenvector.

[0088] In one embodiment, a method for quantitative assessment of coronary angiography calcification based on a shape topology network may further include a process for performing quantitative calcification assessment, the specific process including: inputting a global feature vector into a fully connected layer, performing regression prediction of the calcification integral through the fully connected layer, and outputting a quantitative calcification assessment result.

[0089] By processing the global feature vector through the fully connected layer, a quantitative assessment result of the calcification area can be obtained. The result is a calcification degree score that can be used for clinical diagnosis and treatment decision support.

[0090] Specifically, the computer device can use the global feature h G Input the fully connected layer to perform regression prediction of calcification score: Among them, FC represents the fully connected layer, and the output is the predicted calcification integral value Indicates the degree of calcification in the coronary artery. A higher score indicates a more severe degree of calcification. Figure 5 As shown, the network used in this application is TDA+GIN, and the prediction integral correlation is the correlation between the predicted value and the true value.

[0091] The present application discloses a method for quantitatively evaluating coronary angiography calcification based on a shape topology network. By adopting a twin network to match the beating coronary artery images before and after development, the method successfully solves the matching error problem of coronary angiography images caused by heart beating and morphological changes. By optimizing the contrast loss function, the shape and position of the coronary artery images can be accurately matched, thereby ensuring the accurate alignment of the calcified areas in the images at different time points and improving the annotation accuracy. In addition, the introduction of the FocalLoss loss function significantly improves the model's ability to focus on the calcified areas, especially when the calcified areas are sparse, blurred or unbalanced. By reducing the weight of the background area, the segmentation accuracy of the calcified area is effectively improved. This effectively avoids the over-learning of the background area in the traditional method, thereby greatly improving the accurate recognition and segmentation capabilities of the calcified areas.

[0092] The segmentation results of the calcified area are converted into a topological graph structure, and quantitative calcification assessment is performed with the help of a graph convolutional network (GIN). By constructing a calcified area graph structure based on geometric and topological features, each calcified area is represented as a node in the graph, and the relationship between nodes is weighted by distance and topological features to establish edges; the graph convolutional network extracts the local features and global graph features of each node through layer-by-layer convolution operations, and aggregates the information of the entire graph into the final feature vector through global pooling; it not only effectively captures the individual characteristics of the calcified area, but also considers the relationship between the calcified areas, thereby providing a more comprehensive and accurate quantitative assessment of calcification. Compared with traditional quantitative assessment methods, the introduction of the graph structure enables this method to maintain efficient and accurate calcification score prediction in more complex and variable coronary images.

[0093] It should be understood that, although the various steps in the above flow chart are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the above flow chart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0094] In one embodiment, Figure 6 As shown, a coronary angiography calcification quantitative assessment system based on a shape topology network is provided, comprising: an image preprocessing module 610, an image matching module 620, a calcification region segmentation module 630, a topology structure construction module 640, and a calcification quantitative assessment module 650, wherein:

[0095] An image preprocessing module 610 is used to acquire a coronary angiography image and perform standardization processing to obtain a processed coronary artery image;

[0096] Image matching module 620 is used to input the pre-development and post-development coronary artery images into the Siamese network, process the two coronary artery images by sharing weights, calculate the difference between the two coronary artery images, and optimize the Siamese network by minimizing contrast loss to obtain matched pre-development and post-development coronary artery images;

[0097] The calcification region segmentation module 630 is configured to introduce a Focal Loss function into the deep learning model, perform region segmentation on the matched pre- and post-development coronary artery images, and identify the calcification regions in the segmentation results as nodes;

[0098] A topology graph structure construction module 640 is used to extract the geometric features and topological features of the nodes and construct a topology graph structure based on the geometric features and topological features;

[0099] The calcification quantitative assessment module 650 is used to input the topological graph structure into the graph isomorphism network, perform graph convolution operations on the topological graph structure through the graph isomorphism network, and use global pooling technology to summarize features to obtain a global feature vector; and obtain the calcification quantitative assessment result based on the global feature vector.

[0100] In one embodiment, the image preprocessing module 610 is also used to obtain a coronary angiography image, adjust the brightness and contrast of the coronary angiography image, and perform normalization processing to obtain a preliminary processed image; eliminate noise in the preliminary processed image, and enhance the contrast of the calcification area through an enhancement algorithm to obtain a processed coronary image.

[0101] In one embodiment, the image matching module 620 is further used to respectively calculate the Euclidean distance between the two coronary images; calculate the difference between the two coronary images based on the Euclidean distance; determine the contrast loss, and optimize the twin network by minimizing the contrast loss, so that the optimized twin network minimizes the distance between similar coronary image pairs and maximizes the distance between dissimilar image pairs; and perform matching operations on the two coronary images based on the difference to obtain matched coronary images before and after development.

[0102] In one embodiment, the calcification area segmentation module 630 is also used to determine the cross entropy loss function in the deep learning model, and introduce a modulation factor into the cross entropy loss function to obtain the Focal Loss loss function; the twin network after introducing the Focal Loss loss function is used to detect the areas in the matched coronary artery images before and after development and perform segmentation.

[0103] In one embodiment, the calcified region segmentation module 630 is further configured to mark each calcified region in the segmentation result and identify it as a node, where each node represents a separate calcified region.

[0104] In one embodiment, the topology graph structure construction module 640 is also used to extract geometric features for each node that are used to describe the physical characteristics of the calcified area, and calculate the topological features of each node; based on the geometric features and topological features, the Euclidean distance between the centroids of any two nodes is calculated. When the Euclidean distance is less than or equal to the distance threshold, an edge connection is established between the two nodes, and the connection weight is calculated to construct a topology graph structure.

[0105] In one embodiment, the calcification quantitative assessment module 650 is further used to input the topological graph structure into a graph isomorphic network, determine the learnable parameters in the graph isomorphic network, and obtain the neighbor node set of each node in the topological graph structure; calculate the updated features of each node based on the learnable parameters and the neighbor node set; obtain the node feature matrix based on the updated features, obtain the graph adjacency matrix based on the neighbor node set, and obtain the high-dimensional features of each node in the topological graph structure and the graph based on the node feature matrix and the graph adjacency matrix.

[0106] In one embodiment, the calcification quantitative assessment module 650 is further configured to use a global pooling technique to aggregate the high-dimensional features of each node in the topological graph structure and the graph to obtain a global feature vector.

[0107] In one embodiment, the calcification quantitative assessment module 650 is further configured to input the global feature vector into a fully connected layer, perform regression prediction of the calcification integral through the fully connected layer, and output a calcification quantitative assessment result.

[0108] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for quantitatively evaluating coronary angiography calcification based on a shape topology network is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0109] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0110] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of a method for quantitatively evaluating coronary angiography calcification based on a shape topology network are implemented.

[0111] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for quantitatively evaluating coronary angiography calcification based on a shape topology network are implemented.

[0112] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0113] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0114] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for quantitative assessment of coronary angiography calcification based on shape topology network, characterized in that: The method comprises: Acquire coronary angiography images and perform standardization processing to obtain processed coronary artery images; Inputting the coronary artery images before and after development into the twin network, processing the two coronary artery images by sharing weights, calculating the difference between the two coronary artery images, and optimizing the twin network by minimizing contrast loss to obtain matched coronary artery images before and after development; Introducing a Focal Loss function into the deep learning model, performing regional segmentation on the matched coronary artery images before and after development, and identifying calcified areas in the segmentation results as nodes; Extracting geometric features and topological features of the nodes, and constructing a topological graph structure based on the geometric features and topological features; The topological graph structure is input into a graph isomorphism network, a graph convolution operation is performed on the topological graph structure through the graph isomorphism network, and a global pooling technique is used to summarize features to obtain a global feature vector; and a calcification quantitative assessment result is obtained based on the global feature vector.

2. The method for quantitative assessment of coronary angiography calcification based on shape topology graph network according to claim 1, characterized in that: Obtain coronary angiography images and perform standardization processing to obtain processed coronary artery images, including: Acquiring a coronary angiography image, adjusting the brightness and contrast of the coronary angiography image, and performing normalization processing on the coronary angiography image to obtain a preliminary processed image; Noise in the preliminarily processed image is eliminated, and the contrast of the calcified area is enhanced by an enhancement algorithm to obtain a processed coronary artery image.

3. The method for quantitative assessment of coronary angiography calcification based on shape topology graph network according to claim 1, characterized in that: Calculating the difference between the two coronary artery images, optimizing the twin network by minimizing the contrast loss, and obtaining matched coronary artery images before and after development, including: respectively calculating the Euclidean distance and cosine similarity between the two coronary artery images; Calculating the difference between the two coronary artery images according to the Euclidean distance and cosine similarity; Determining a contrast loss, and optimizing the twin network by minimizing the contrast loss, wherein the optimized twin network minimizes the distance between similar coronary artery image pairs and maximizes the distance between dissimilar image pairs; Based on the difference, the optimized twin network is used to perform a matching operation on the two coronary artery images to obtain matched coronary artery images before and after development.

4. The method for quantitative assessment of coronary angiography calcification based on shape topology network according to claim 1, characterized in that: The FocalLoss loss function is introduced into the deep learning model to perform regional segmentation on the matched coronary artery images before and after development, including: Determine a cross entropy loss function in a deep learning model and introduce a modulation factor into the cross entropy loss function to obtain a Focal Loss loss function; The twin network after introducing the Focal Loss loss function is used to detect and segment the regions in the matched coronary artery images before and after development.

5. The method for quantitative assessment of coronary angiography calcification based on shape topology network according to claim 1, characterized in that: Identify calcified areas in the segmentation results as nodes, including: The calcified area in each segmentation result is marked and identified as a node, and each node represents a separate calcified area.

6. The method for quantitative assessment of coronary angiography calcification based on shape topology network according to claim 1, characterized in that: Extracting the geometric features and topological features of the nodes, and constructing a topological graph structure based on the geometric features and topological features, including: For each of the nodes, extracting geometric features used to describe the physical characteristics of the calcified area, and calculating the topological features of each of the nodes; Based on the geometric features and topological features, the Euclidean distance between any two node centroids is calculated. When the Euclidean distance is less than or equal to a distance threshold, an edge connection is established between the two nodes, and the connection weight is calculated to construct a topological graph structure.

7. The method for quantitative assessment of coronary angiography calcification based on shape topology network according to claim 1, characterized in that: Inputting the topological graph structure into a graph isomorphism network, and performing a graph convolution operation on the topological graph structure through the graph isomorphism network, including: Inputting the topological graph structure into a graph isomorphic network, determining learnable parameters in the graph isomorphic network, and obtaining a set of neighbor nodes of each node in the topological graph structure; Calculating updated features of each node based on the learnable parameters and the set of neighboring nodes; A node feature matrix is ​​obtained according to the updated features, a graph adjacency matrix is ​​obtained according to the neighbor node set, and high-dimensional features of each node and graph in the topological graph structure are obtained based on the node feature matrix and the graph adjacency matrix.

8. The method for quantitative assessment of coronary angiography calcification based on shape topology network according to claim 7, characterized in that: Use global pooling technology to aggregate features and obtain a global feature vector, including: A global pooling technique is used to aggregate the high-dimensional features of each node in the topological graph structure and the graph to obtain a global feature vector.

9. The method for quantitative assessment of coronary angiography calcification based on shape topology network according to claim 1, characterized in that: Obtaining a quantitative calcification assessment result based on the global eigenvector includes: The global feature vector is input into a fully connected layer, and regression prediction of the calcification integral is performed through the fully connected layer to output a calcification quantitative assessment result.

10. A coronary angiography calcification quantitative assessment system based on shape topology network, characterized in that: The system comprises: An image preprocessing module is used to obtain coronary angiography images and perform standardization processing to obtain processed coronary artery images; an image matching module, configured to input the coronary artery images before and after development into a twin network, process the two coronary artery images by sharing weights, calculate the difference between the two coronary artery images, and optimize the twin network by minimizing contrast loss to obtain matched coronary artery images before and after development; a calcification region segmentation module, configured to introduce a FocalLoss loss function into a deep learning model, perform region segmentation on the matched coronary artery images before and after development, and identify the calcification regions in the segmentation results as nodes; A topology graph structure construction module is used to extract the geometric features and topological features of the nodes and construct a topology graph structure based on the geometric features and topological features; The calcification quantitative assessment module is used to input the topological graph structure into a graph isomorphism network, perform graph convolution operation on the topological graph structure through the graph isomorphism network, and use global pooling technology to summarize features to obtain a global feature vector; and obtain a calcification quantitative assessment result based on the global feature vector.

Images