A method and related device for quantifying coronary artery calcification score in cardiac CT images

Abstract

The application belongs to an image processing method, aiming at the technical problems that the coronary artery calcification is difficult to accurately extract on the CT image, the calcification integral quantification is difficult and the operation is complicated, a heart CT image coronary artery calcification integral quantification method and related device are provided, the obtained heart CT image is input into a coronary artery calcification segmentation model to obtain a coronary artery calcification segmentation result. In the coronary artery calcification segmentation model, the encoder and the decoder are multiple groups of structures with symmetric skip connection, each group of encoder and decoder adopts a residual U cell, in the encoder, the residual U cell is connected in parallel with a calcification region attention mechanism structure for obtaining a calcification region attention map, a bottleneck layer is based on a multi-head attention dissection structure feature fusion mechanism, and finally the coronary artery calcification segmentation result is integrated and calculated. While ensuring the extraction accuracy, the processing difficulty is relatively low.

Description

TECHNICAL FIELD

[0001] The application belongs to an image processing method, and particularly relates to a coronary artery calcification integral quantification method for a heart CT image and a related device. BACKGROUND

[0002] Coronary artery calcification is an important biomarker of cardiovascular disease. The degree of coronary artery calcification can be observed and the calcification integral can be calculated through a heart CT (Computed Tomography) image, so as to perform cardiovascular disease risk assessment.

[0003] Calcification usually appears as small size, few quantity and dispersion on a CT image, and is easily interfered by bones near the heart with similar CT values, resulting in great difficulty in accurate extraction. In addition, coronary artery calcification is easily confused with other types of calcification (such as aortic calcification), and therefore, accurate identification of coronary artery calcification often depends on CTA (Computed Tomography Angiography) image to provide clear coronary artery reference information, which also leads to high difficulty and cumbersome operation in accurate extraction and calculation of the corresponding calcification integral. SUMMARY

[0004] The application provides a coronary artery calcification integral quantification method for a heart CT image and a related device to solve the technical problems of great difficulty in accurate extraction of coronary artery calcification on a CT image, high difficulty and cumbersome operation in calcification integral quantification.

[0005] To achieve the above-mentioned purpose, the application adopts the following technical solutions:

[0006] In a first aspect, the application provides a coronary artery calcification integral quantification method for a heart CT image, comprising:

[0007] obtaining a heart CT image;

[0008] taking the heart CT image as input, obtaining a coronary artery calcification segmentation result through a coronary artery calcification segmentation model; the coronary artery calcification segmentation model comprises an encoder, a bottleneck layer and a decoder, wherein the encoder and the decoder are multiple groups of structures with symmetric skip connection, each group of the encoder and the decoder adopts a residual U unit, in the encoder, the residual U unit is connected in parallel with a calcification region attention mechanism structure for obtaining a calcification region attention map, and the bottleneck layer is based on a multi-head attention dissection structure feature fusion mechanism;

[0009] performing integral calculation on the coronary artery calcification segmentation result to obtain an integral quantification result.

[0010] Further, the coronary artery calcification segmentation model adopts a 3D U 2 -Net segmentation model.

[0011] Further, the training method of the coronary artery calcification segmentation model comprises:

[0012] The heart region and the coronary artery region are extracted from the heart CT image respectively using the heart extraction model and the coronary artery extraction model, which are used to determine the loss function during training of the coronary artery calcification segmentation model; the heart extraction model and the coronary artery extraction model both adopt nnUNetV2 model;

[0013] The coronary artery calcification segmentation label is determined.

[0014] The coronary artery calcification segmentation model is trained using the heart CT image containing the coronary artery calcification segmentation label in combination with the loss function.

[0015] Further, the processing method in the bottleneck layer comprises:

[0016] The feature maps output by the encoder in the heart extraction model, the feature maps output by the encoder in the coronary artery extraction model, and the output feature maps output by the encoder of the coronary artery calcification segmentation model are taken as inputs of the bottleneck layer, are converted into corresponding labels respectively through the linear projection layer with layer normalization, and then pass through the multi-head attention unit and the layer normalization layer-multilayer perceptron with residual connection in turn, and the output of the layer normalization layer-multilayer perceptron is reshaped into a three-dimensional feature map.

[0017] Further, the loss function during training of the coronary artery calcification segmentation model comprises:

[0018]

[0019] Wherein, is the total loss function, is the Dice loss term, is the weighting coefficient of the heart region membership constraint term, is the heart region membership constraint term, is the weighting coefficient of the coronary artery distance constraint term, is the coronary artery distance constraint term.

[0020] Further, the method for integrating and calculating the coronary artery calcification segmentation result comprises:

[0021] The slice containing the calcification mask in the coronary artery calcification segmentation result is traversed to extract the calcification connected domain.

[0022] Traverse the calcification connected domain, calculate the area of the calcification connected domain and the maximum CT value of the corresponding region in the heart CT image;

[0023] According to the mapping relationship of the maximum CT value to the calcification density weight, the density weight of the calcification connected domain is determined;

[0024] The density weight of the calcification connected domain is multiplied by the area of the calcification connected domain to obtain the calcification integral of the calcification connected domain.

[0025] The calcification integrals of all calcification connected domains of all slices are added to obtain the coronary artery calcification integral of the current case, that is, the integral quantification result.

[0026] Further, the calcification region attention mechanism structure adopts a 3D Inception module.

[0027] In a second aspect, the present application provides a heart CT image coronary artery calcification integral quantification system, comprising:

[0028] A data module is configured to obtain a heart CT image.

[0029] A segmentation module is configured to take the heart CT image as input, and obtain a coronary artery calcification segmentation result by using a coronary artery calcification segmentation model. The coronary artery calcification segmentation model comprises an encoder, a bottleneck layer and a decoder. The encoder and the decoder are multiple groups of structures with symmetric skip connections. Each group of the encoder and the decoder adopts a residual U-shaped unit. In the encoder, the residual U-shaped unit is connected in parallel with a calcification region attention mechanism structure for obtaining a calcification region attention map. The bottleneck layer is based on a multi-head attention dissection structure feature fusion mechanism.

[0030] A calculation module is configured to perform integral calculation on the coronary artery calcification segmentation result to obtain an integral quantification result.

[0031] In a third aspect, the present application provides an electronic device, comprising a memory and one or more processors. The memory is coupled to the processor. The memory stores computer program code, which comprises computer instructions. When the computer instructions are executed by the processor, the electronic device performs the steps of the heart CT image coronary artery calcification integral quantification method.

[0032] In a third aspect, the present application provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the heart CT image coronary artery calcification integral quantification method are implemented.

[0033] Compared with the prior art, the present application has the following beneficial effects:

[0034] The application provides a coronary artery calcification quantification method for heart CT images. The coronary artery calcification quantification method comprises the following steps: inputting a heart CT image into a coronary artery calcification segmentation model to obtain a coronary artery calcification segmentation result. The coronary artery calcification segmentation model comprises an encoder, a bottleneck layer and a decoder. The encoder and the decoder are multiple groups of structures with symmetric skip connections. Each group of the encoder and the decoder adopts a residual U-shaped unit. In the encoder, the residual U-shaped unit is connected in parallel with a calcification region attention mechanism structure for obtaining a calcification region attention map. The bottleneck layer is based on a multi-head attention anatomical structure feature fusion mechanism. Finally, the coronary artery calcification segmentation result is integrated to obtain an integral quantification result. The application introduces the calcification region attention mechanism in the encoder of the coronary artery calcification segmentation model, thereby enhancing the attention to the calcification region and the feature extraction capability thereof. The multi-head attention anatomical structure feature fusion mechanism is introduced in the bottleneck layer, thereby enhancing the spatial semantic association representation of the coronary artery calcification, the heart and the coronary artery.

[0035] The application further provides a coronary artery calcification quantification system for heart CT images, an electronic device and a computer readable storage medium, which have all the advantages of the coronary artery calcification quantification method for heart CT images. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0037] Figure 1 FIG. 1 is a schematic diagram of the coronary artery calcification quantification method for heart CT images according to the application;

[0038] Figure 2 FIG. 2 is another schematic diagram of the coronary artery calcification quantification method for heart CT images according to the application;

[0039] Figure 3 FIG. 3 is a schematic diagram of the principle of the coronary artery calcification segmentation model according to the embodiment of the application;

[0040] Figure 4 FIG. 4 is a schematic diagram of the encoder structure of the coronary artery calcification segmentation model according to the embodiment of the application;

[0041] Figure 5 FIG. 5 is a schematic diagram of the bottleneck layer based on the multi-head attention anatomical structure feature fusion mechanism according to the embodiment of the application;

[0042] Figure 6 For an embodiment of the present application, the original CT image and the 3D segmentation result obtained by the method of the present application are shown in the following figure:

[0043] Figure 7 For an embodiment of the present application, the original CT image and the 3D segmentation result obtained by the method of the present application are shown in the following figure: DETAILED DESCRIPTION

[0044] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0045] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0046] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0047] In the description of the embodiments of the present application, it should be noted that, if the orientation or position relationship indicated by the terms “up”, “down”, “horizontal”, “inner” and the like is based on the orientation or position relationship shown in the drawings, or is the orientation or position relationship when the product of the present application is used, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, therefore, it cannot be understood as a limitation on the present application. In addition, the terms “first”, “second” and the like are only used for differentiation in description, and cannot be understood as indicating or implying relative importance.

[0048] In addition, if the term “horizontal” appears, it does not mean that the component must be absolutely horizontal, but can be slightly inclined. For example, “horizontal” only means that its direction is relatively more horizontal than “vertical”, and does not mean that the structure must be completely horizontal, but can be slightly inclined.

[0049] In the description of the embodiments of this application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0050] Cardiovascular disease refers to a class of conditions that affect the function of the heart and blood vessels. Because it is difficult to detect in its early stages, cardiovascular disease has long been one of the leading causes of death worldwide and has become a major public health problem. Early and accurate risk assessment is crucial for the prevention, diagnosis, and treatment of cardiovascular disease.

[0051] Coronary artery calcification, as a core pathological indicator of cardiovascular disease, has significant clinical value for disease risk assessment through quantitative analysis. Cardiac CT images can visually present the distribution characteristics of coronary artery calcifications, and integral models can be established based on parameters such as calcification volume and density, providing crucial evidence for clinical diagnosis and treatment. With breakthroughs in artificial intelligence technology, image analysis systems based on deep learning algorithms have effectively solved the problems of low efficiency and long processing times associated with traditional manual annotation methods, achieving rapid and automated calculation of coronary artery calcification integrals. However, several technical bottlenecks still exist in this field: First, coronary artery calcifications in CT images often exhibit small, sparse, and dispersed physical characteristics. Simultaneously, the similarity in CT values ​​between surrounding bone tissues can easily cause signal interference, significantly increasing the difficulty of accurate segmentation. Second, coronary artery calcifications and ectopic calcifications such as aortic calcifications show high similarity in imaging appearance. Accurate identification relies on three-dimensional anatomical reference information provided by coronary CT angiography, which places higher demands on the quality and integrity of the image data.

[0052] This application addresses the above-mentioned problems by proposing a method and related device for quantifying coronary artery calcification scores in cardiac CT images, thereby enabling automatic extraction and analysis of coronary artery calcifications in cardiac CT images and accurate assessment of cardiovascular disease risk.

[0053] like Figure 1 The diagram shown is a schematic representation of a method for quantifying coronary artery calcium in cardiac CT images according to this application, which may include:

[0054] S101, acquire cardiac CT images.

[0055] Cardiac CT imaging uses the rotating X-ray source and detector array of a CT scanner to generate tomographic images of the heart and coronary arteries by utilizing the differences in X-ray absorption by different tissues.

[0056] S102 uses cardiac CT images as input and obtains the coronary artery calcification segmentation results through the coronary artery calcification segmentation model.

[0057] The coronary artery calcification segmentation model includes an encoder, a bottleneck layer, and a decoder. The encoder and decoder are multiple sets of structures with symmetrical skip connections. Each set of encoders and decoders uses residual U-shaped units. In the encoder, the residual U-shaped units are connected in parallel with a calcification region attention mechanism structure for acquiring the attention map of the calcification region. The bottleneck layer is based on a multi-head attention anatomical feature fusion mechanism.

[0058] It should be noted that the coronary artery calcification segmentation model employs an encoder-decoder architecture. The encoder and decoder reuse features through skip connections, and each set of residual U-shaped units mitigates the vanishing gradient problem through residual connections, ensuring the stability of deep network training. A calcification region attention mechanism connected in parallel with the residual U-shaped units generates a calcification probability map, dynamically focusing on high-probability areas (such as near the coronary artery wall) and suppressing interfering signals such as bone. The multi-head attention anatomical feature fusion mechanism in the bottleneck layer fuses coronary artery anatomical features across channels, such as simultaneously focusing on structural information like vessel orientation and lumen diameter, improving the ability to identify ectopic calcifications. The coronary artery calcification segmentation model extracts calcification features layer by layer through the encoder, fuses anatomical information in the bottleneck layer, restores spatial details using skip connections in the decoder, and dynamically focuses on calcification regions through the calcification region attention mechanism.

[0059] S103, perform integral calculation on the segmentation results of the coronary artery calcification to obtain the integral quantification result.

[0060] In practical applications, a modified Agatston method can be used, which introduces CT value density weighting and spatial distribution weighting on top of the traditional area-weighted method. For example, calcifications near the left main coronary artery may have a higher score because they are more likely to cause hemodynamic changes. Alternatively, a simple volume integral method can be used.

[0061] This application constructs a coronary artery calcification integral quantification system based on anatomical structure attention constraints for cardiac CT images. It can simultaneously achieve precise segmentation of the heart, coronary arteries and their calcifications, and calculate the calcification integral based on this segmentation, providing a reference for cardiovascular risk assessment.

[0062] like Figure 2 The diagram shown is another schematic representation of the coronary artery calcium integration quantification method for cardiac CT images according to this application, which may include:

[0063] S201 extracts the heart region and coronary artery region based on the nnUNetV2 model.

[0064] The nnUNetV2 model trained on the TotalSegmentator dataset is used as the heart extraction model and the coronary artery extraction model respectively to extract the heart region and the coronary artery region in the heart CT image respectively.

[0065] It should be noted that the TotalSegmentator dataset has a large data size and a high diversity, covering various pathologies, scanners, imaging sequences, and data from different medical institutions. The TotalSegmentator dataset has high annotation accuracy and reliability. In addition, the TotalSegmentator dataset not only covers common organs, but also annotates structures that are rare in other datasets.

[0066] In addition, the nnUNetV2 model used is an advanced version of nnU-Net, which is based on the original 3D U-Net architecture, adopts advanced training strategies, flexible data preprocessing, and improved upsampling methods, and can automatically configure all hyperparameters according to the characteristics of the TotalSegmentator dataset, including optimized network depth, input image local region size, and training parameters based on input data characteristics, improving the segmentation performance of the heart region and the coronary artery region.

[0067] The structure of the nnUNetV2 model mainly includes an encoder and a decoder. The encoder is usually based on a convolutional neural network structure, which converts the original image into a high-level feature representation suitable for subsequent processing. The decoder maps the high-level feature representation extracted by the encoder back to the size of the original image and generates a pixel-level prediction. The decoder gradually restores the high-level feature representation to the same size as the original image through layer-by-layer upsampling and convolution operations to obtain a segmentation result.

[0068] In actual application, the heart extraction model and the coronary artery extraction model are set respectively. During training, the heart CT image obtained is input into the heart extraction model and the coronary artery extraction model, and the heart region and the coronary artery region are extracted through the heart extraction model and the coronary artery extraction model respectively. The heart CT image containing the coronary artery calcification segmentation label can be used to train the coronary artery calcification segmentation model. The extracted heart region and coronary artery region are used to determine the loss function of the coronary artery calcification segmentation model during training.

[0069] In actual application, the decoder of the heart extraction model and the decoder of the coronary artery extraction model can also be used to obtain the corresponding heart feature map and coronary artery feature map for auxiliary viewing.

[0070] S202, the coronary artery calcification segmentation model based on anatomical structure attention constraint.

[0071] The application designs a coronary artery calcification segmentation model based on anatomical structure attention constraint, including a 3D U 2 -Net segmentation backbone, a bottleneck layer based on a multi-head attention anatomical structure feature fusion mechanism, and an optimization target based on anatomical structure constraint, for joint learning of multi-scale spatial features and anatomical semantic features of coronary artery calcification, so as to realize accurate segmentation of coronary artery calcification. As Figure 3 shown, it is a principle schematic diagram of the coronary artery calcification segmentation model of the application.

[0072] The specific execution method can include:

[0073] (1) A 3D U 2 -Net segmentation backbone based on a calcification region attention mechanism.

[0074] The embodiment adopts a 3D version of U 2 -Net model as the coronary artery calcification segmentation backbone. The backbone is a two-level nested U-shaped structure, including five encoder-decoders with symmetric skip connection structure. Specifically, the 3D version of U 2 -Net model structure includes an encoder, a bottleneck layer, and a decoder. Each encoder adopts a residual U-unit in parallel with a 3D Inception module, and each decoder adopts a residual U-unit. The output of the encoder is input to the bottleneck layer, and the output of the bottleneck layer is input to the decoder. Through the residual U-unit, multi-scale features of the coronary artery calcification can be effectively extracted. In the residual U-unit, local features are first extracted from the input feature map by a convolution layer, then the extracted local features are input to the symmetric U-shaped structure to further capture multi-scale features, and finally the local features and the multi-scale features are fused through residual connection. Through such design, the embodiment aims to efficiently extract multi-scale features inside each encoder-decoder with low storage cost, which helps to improve the segmentation performance of the entire framework for coronary artery calcifications of different sizes and spatial distributions. Considering the difference in spatial resolution of the input feature maps of different encoder-decoders, the number of down-sampling and up-sampling in the residual U-unit of the encoder-decoder from the input-output end to the bottleneck layer is set to 5-1 in turn.

[0075] (2) Encoder.

[0076] As Figure 4 shown, it is an encoder schematic diagram of the coronary artery calcification segmentation model. It includes a residual U-unit and a 3D Inception module in parallel. In the 3D U 2On the basis of the Net backbone model, the application further proposes a calcification region attention mechanism and introduces it to each encoder. The calcification region attention mechanism adopts a 3D Inception module in parallel with the corresponding residual U unit to predict a calcification region attention map, and makes the calcification region attention map and the output feature map of the residual U unit element-level multiplication to obtain a segmentation backbone output feature map. Specifically, the 3D Inception module includes four parallel branches of 1⨉1⨉1 convolution, 3⨉3⨉3 convolution, 5⨉5⨉5 convolution and 3⨉3⨉3 maximum pooling. After the feature maps output by the four parallel branches are fused in the channel dimension, a 1⨉1⨉1 convolution layer with a Sigmoid activation function is used to obtain the calcification region attention map.

[0077] During the training process, in order to effectively optimize the 3D Inception module corresponding to each encoder and constrain it to accurately predict the calcification region attention map, the embodiment constructs a calcification region heat map based on the calcification region attention map, and down-samples it to the size of the output feature map of each encoder, as the calcification region attention map label for optimizing the 3D Inception module corresponding to the corresponding encoder. The construction method of the calcification region heat map is as follows:

[0078] 1) Perform connected component labeling on the coronary artery calcification label in the calcification region attention map, and calculate the geometric center of each connected component;

[0079] 2) Traverse the geometric center of each connected component, and use a three-dimensional Gaussian function to construct the corresponding region heat map:

[0080]

[0081] wherein, is the activation value of the constructed calcification region heat map at the three-dimensional coordinates , is the geometric center coordinate of the th connected component, n and and are the number of connected components and the standard deviation, respectively, n is the serial number of the connected component.

[0082] (3) Bottleneck layer based on multi-head attention anatomical structure feature fusion mechanism.

[0083] The application proposes an anatomical structure feature fusion mechanism based on multi-head attention, aiming to utilize the spatial-semantic association between calcification and the two anatomical structures of heart and coronary artery to achieve accurate segmentation of coronary artery calcification. The mechanism realizes the spatial-semantic association representation of coronary artery calcification by introducing a multi-head attention unit in the bottleneck layer of the segmentation backbone to fuse the anatomical features of heart and coronary artery, thereby improving the performance of coronary artery calcification segmentation.

[0084] As shown in Figure 5 , it is a bottleneck layer schematic diagram of the multi-head attention anatomical structure feature fusion mechanism. When training or actually applying the coronary artery calcification segmentation model, the feature map output by the encoder in the heart extraction model (heart feature map), the feature map output by the encoder in the coronary artery model (coronary artery feature map), and the output feature map output by the encoder of the coronary artery calcification segmentation model (calcification feature map) are taken as the inputs of the bottleneck layer. In the bottleneck layer, the linear projection layer with layer normalization is used to convert the three inputs into corresponding labels, which are taken as the three inputs of the multi-head attention unit . The calculation formula of the multi-head attention unit is as follows:

[0085]

[0086] Among them, represents the multi-head attention calculation, are the key vector, query vector and value vector, respectively corresponding to the heart region, coronary artery region and calcification label, represents the Softmax activation function, is the dimension of the key vector. Through the multi-head attention unit, the adaptive weighting of the features of heart, coronary artery and calcification is realized, which can effectively enhance the spatial-semantic association representation of coronary artery calcification.

[0087] Finally, the output of the multi-head attention unit is input into a layer normalization layer-multilayer perceptron with residual connection, and the finally obtained vector is reshaped into a three-dimensional feature map, which is input into the decoder of the coronary artery calcification segmentation model to obtain the final segmentation result. As shown in Figure 6 , it is a schematic diagram of the original CT image and the 3D segmentation result obtained by using the method of the application in one embodiment of the application. Among them, the left side is the original CT image, and the right side is the 3D segmentation result obtained by using the method of the application.

[0088] (4) Optimization target based on anatomical structure constraint

[0089] This application also proposes an optimization objective based on anatomical constraints. Its design aims to utilize the spatial relationship between the segmentation results of the heart, coronary arteries, and their calcifications to ensure the accuracy of coronary artery calcification segmentation. By introducing anatomical constraints of the heart and coronary arteries into the loss function, interference from similar targets such as pericardial bones and other types of calcifications can be effectively eliminated, thereby further achieving precise segmentation of coronary artery calcifications.

[0090] Specifically, the optimization objective based on anatomical constraints includes a Dice loss term, a cardiac region membership constraint term, and a coronary artery distance constraint term. The Dice loss term measures the region overlap between the coronary artery calcification segmentation result and the label, and can be expressed as:

[0091]

[0092] in, For Dice loss terms, The confidence map of coronary artery calcification segmentation. Individual phenotypic value The first segmentation label for coronary artery calcification Individual phenotypic value The total number of voxels. is the Laplacian smoothing factor in the Dice loss term, used to avoid numerical instability and accelerate the convergence of the training process.

[0093] Based on the prior knowledge that coronary artery calcifications are located within the cardiac region, this application constructs a cardiac region membership constraint term, defined as the overlap rate between the predicted coronary artery calcification region and the non-cardiac region, expressed as:

[0094]

[0095] in, For the heart region, there are membership constraints. The first confidence plot of the heart segment Individual phenotypic value The confidence map of coronary artery calcification segmentation. Individual phenotypic value This is the Laplace smoothing factor in the membership constraint term for the heart region, used to avoid numerical instability and accelerate the convergence of the training process.

[0096] Further, the application also designs a coronary artery distance constraint term, aiming to guide the model to identify the coronary artery calcification according to the distance between the calcification and the coronary artery. The closer the calcification is to the coronary artery, the greater the probability of belonging to the coronary artery calcification; otherwise, the farther the calcification is from the coronary artery, the smaller the probability of belonging to the coronary artery calcification. The coronary artery distance constraint term is used to measure the confidence weighted distance between the calcification and the coronary artery, which can be expressed as:

[0097]

[0098] wherein, is the coronary artery distance constraint term, is the Euclidean distance between the th voxel coordinate and the th voxel coordinate , and are the th and the th voxel value of the coronary artery segmentation confidence map, respectively, is a small normal number to prevent the denominator from being zero.

[0099] Finally, the loss function of the coronary artery calcification segmentation model established by the application can be expressed as:

[0100]

[0101] wherein, is the total loss function, and are the weighting coefficients of the heart region membership constraint term and the coronary artery distance constraint term, respectively.

[0102] The loss function set for the coronary artery calcification segmentation model by the application uses the prior constraint of the heart and coronary artery anatomical structure on the region where the coronary artery calcification is located, guiding the model to accurately identify the coronary artery calcification from the similar structures such as the bone around the heart and different types of calcification.

[0103] S203, coronary artery calcification score calculation based on the Agatston scoring method.

[0104] On the basis of the coronary artery calcification segmentation result, the application adopts the Agatston scoring method to calculate the calcification score. In this embodiment, taking the CT scan at 12 keV as the reference benchmark, the density weight and area of the coronary artery calcification in the cardiac CT image are comprehensively evaluated to assess the degree of calcification lesion. In this embodiment, the density weight of a single coronary artery calcification is determined by the maximum CT value in the calcification region, and the determination standard can be as shown in Table 1.

[0105] Table 1. Mapping relationship between the maximum CT value and the density weight of a single coronary artery calcification region.

[0106]

[0107] The higher the maximum CT value, the greater the density weight assigned, indicating that the calcification density is higher, the degree of calcification is greater, and the risk of cardiovascular disease is also higher.

[0108] The specific calculation method is as follows: 1) Traverse the slices containing calcification masks in the coronary artery calcification segmentation results and extract the calcification connected regions; 2) Traverse the calcification connected regions, calculate the area of ​​the calcification connected region and the maximum CT value of the corresponding area in the original cardiac CT image, determine the density weight of the calcification connected region according to Table 1, and multiply it by the area of ​​the calcification connected region to obtain the calcification integral of the calcification connected region; 3) Add up the calcification integrals of all calcification connected regions in all slices to obtain the coronary artery calcification integral of the current case, which is used as the integral quantification result.

[0109] The cardiac extraction module and coronary artery extraction model module in this application both use the nnUNetV2 model trained on the TotalSegmentator dataset to extract the cardiac region and coronary artery region from cardiac CT images, respectively. The coronary artery calcification segmentation model uses 3D U... 2 The -Net model serves as the backbone for calcification segmentation. Each encoder incorporates a calcification region attention mechanism to extract features from calcifications of varying sizes and spatial distributions. Furthermore, a multi-head attention mechanism is used at the bottleneck layer to fuse features from the encoders of the heart and coronary artery segmentation modules. The Dice loss function is combined with constraints related to the anatomical structures of the heart and coronary arteries to guide the coronary artery calcification segmentation model in eliminating interference from the surrounding bone structure and accurately identifying coronary artery calcifications from cardiac calcifications. Finally, calcification integral quantization is performed using the Agatston scoring method to calculate the calcification integral value of the identified coronary artery calcifications.

[0110] The method described in this application has been rigorously quantitatively evaluated on cardiac CT imaging data of a group of high-risk individuals for cardiovascular disease. The results show that the method can effectively improve the segmentation accuracy of coronary artery calcifications in cardiac CT images, and ultimately provide more accurate and reliable calcification score calculation results, which has potential clinical application value.

[0111] like Figure 7 The diagram shown is a schematic representation of a coronary artery calcification scoring and quantification system for cardiac CT imaging according to this application, which may include:

[0112] The data module is used to acquire cardiac CT images;

[0113] The segmentation module is configured to take the cardiac CT image as input, and obtain a coronary artery calcification segmentation result by a coronary artery calcification segmentation model. The coronary artery calcification segmentation model comprises an encoder, a bottleneck layer, and a decoder. The encoder and the decoder are of a structure with symmetric skip connection in multiple groups. Each group of the encoder and the decoder adopts a residual U-shaped unit. In the encoder, the residual U-shaped unit is connected in parallel with a calcification region attention mechanism structure for obtaining a calcification region attention map. The bottleneck layer is based on a multi-head attention dissection structure feature fusion mechanism.

[0114] The calculation module is configured to perform integral calculation on the coronary artery calcification segmentation result to obtain an integral quantization result.

[0115] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, another division manner can be used, for example, a plurality of modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components can be or can not be physically separated. The components displayed as modules can be one physical unit or multiple physical units, that is, can be located in one place or can be distributed to multiple different places. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiments.

[0116] In addition, each module in the embodiments of the present application can be integrated in one processing unit, or each module can exist physically, or two or more modules can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0117] The embodiments of the present application also provide an electronic device. The electronic device can include one or more processors, memories, and communication interfaces.

[0118] The memory, the communication interface, and the processor are coupled. For example, the memory, the communication interface, and the processor can be coupled together through a bus.

[0119] The communication interface is configured to perform data transmission with other devices. The memory stores computer program codes. The computer program codes include computer instructions. When the computer instructions are executed by the processor, the electronic device performs the steps of the cardiac CT image coronary artery calcification integral quantization method.

[0120] The processor can be a processor or a controller, for example, a Central Processing Unit (CPU), a general-purpose processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. It can implement or execute the various exemplary logical blocks, modules, and circuits described in connection with the present disclosure. The processor can also be a combination of implementing computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The processor can be used to support the electronic device to perform the method steps provided in the above embodiments.

[0121] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like.

[0122] The computer readable storage medium provided by the embodiments of the present application stores a computer program, and the computer program is executed by the processor to implement the steps of the above-mentioned heart CT image coronary artery calcification score quantification method.

[0123] The computer readable storage medium involved in the present application includes a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable magnetic disk, a CD ROM, or any other form of storage medium known in the technical field.

[0124] The above is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for quantifying coronary artery calcification scores in cardiac CT images, characterized in that, include: Obtain cardiac CT images; Using cardiac CT images as input, the coronary artery calcification segmentation results are obtained through a coronary artery calcification segmentation model. The coronary artery calcification segmentation model includes an encoder, a bottleneck layer, and a decoder. The encoder and decoder are multiple sets of structures with symmetrical skip connections. Each set of encoders and decoders uses residual U-shaped units. In the encoder, the residual U-shaped units are connected in parallel with a calcification region attention mechanism structure for acquiring the attention map of the calcification region. The bottleneck layer is based on a multi-head attention anatomical structure feature fusion mechanism to fuse the anatomical features of the heart and coronary arteries. The loss function used during training of the coronary artery calcification segmentation model includes: in, For the overall loss function, For Dice loss terms, These are the weighting coefficients for the membership constraint terms of the heart region. The heart region is a member of the constraint terms. These are the weighting coefficients for the coronary artery distance constraint term. For coronary artery distance constraints; Dice Loss Items Used to measure the regional overlap between coronary artery calcification segmentation results and labels, cardiac region membership constraint. Represented as: Coronary artery distance constraint Used to measure the confidence-weighted distance between calcifications and coronary arteries; in, The total number of voxels. The confidence map of coronary artery calcification segmentation. Individual phenotypic value The first confidence plot of the heart segment Individual phenotypic value The Laplace smoothing factor in the membership constraint term for the cardiac region; The segmentation results of the coronary artery calcifications are integrated to obtain the integrated quantification results.

2. The method for quantifying coronary artery calcification in cardiac CT images according to claim 1, characterized in that, The coronary artery calcification segmentation model uses 3D U 2 -Net segmentation model.

3. The method for quantifying coronary artery calcification in cardiac CT images according to claim 1, characterized in that, The training method for the coronary artery calcification segmentation model includes: The cardiac region and coronary artery region were extracted from cardiac CT images using a cardiac extraction model and a coronary artery extraction model, respectively, to determine the loss function during the training of the coronary artery calcification segmentation model; both the cardiac extraction model and the coronary artery extraction model adopted the nnUNetV2 model. Identify coronary artery calcification segmentation labels; The coronary artery calcification segmentation model was trained using cardiac CT images containing coronary artery calcification segmentation labels and a loss function.

4. The method for quantifying coronary artery calcification in cardiac CT images according to claim 3, characterized in that, The processing method within the bottleneck layer includes: The feature maps output by the encoder in the heart extraction model, the coronary artery extraction model, and the coronary artery calcification segmentation model are used as inputs to the bottleneck layer. They are then converted into corresponding labels by a linear projection layer with layer normalization, and then passed through a multi-head attention unit and a layer normalization layer-multilayer perceptron with residual connections. Finally, the output of the layer normalization layer-multilayer perceptron is reshaped into a three-dimensional feature map.

5. The method for quantifying coronary artery calcification in cardiac CT images according to claim 1, characterized in that, A method for integrating the segmentation results of the coronary artery calcifications includes: Traverse the slices containing calcification masks in the coronary artery calcification segmentation results and extract the calcification connected components; Traverse the calcification connected regions and calculate the area of ​​the calcification connected regions and the maximum CT value of the corresponding region in the cardiac CT image. The density weight of the calcification connected region is determined based on the mapping relationship between the maximum CT value and the calcification density weight. Multiply the density weight of the calcified connected region by the area of ​​the calcified connected region to obtain the calcification integral of the calcified connected region. The coronary artery calcification score for the current case is obtained by summing the calcification integrals of all calcification connected regions in all slices, which is the quantified result.

6. The method for quantifying coronary artery calcification in cardiac CT images according to claim 1, characterized in that, The attention mechanism structure for the calcification region adopts a 3D Inception module.

7. A system for quantifying coronary artery calcification in cardiac CT imaging, characterized in that, include: The data module is used to acquire cardiac CT images; The segmentation module is used to obtain coronary artery calcification segmentation results by taking cardiac CT images as input and using a coronary artery calcification segmentation model. The coronary artery calcification segmentation model includes an encoder, a bottleneck layer, and a decoder. The encoder and decoder are multiple sets of structures with symmetrical skip connections. Each set of encoders and decoders uses residual U-shaped units. In the encoder, the residual U-shaped units are connected in parallel with a calcification region attention mechanism structure for acquiring the attention map of the calcification region. The bottleneck layer is based on a multi-head attention anatomical structure feature fusion mechanism to fuse the anatomical features of the heart and coronary arteries. The loss function used during training of the coronary artery calcification segmentation model includes: in, For the overall loss function, For Dice loss terms, These are the weighting coefficients for the membership constraint terms of the heart region. The heart region is a member of the constraint terms. These are the weighting coefficients for the coronary artery distance constraint term. For coronary artery distance constraints; Dice Loss Items Used to measure the regional overlap between coronary artery calcification segmentation results and labels, cardiac region membership constraint. Represented as: Coronary artery distance constraint Used to measure the confidence-weighted distance between calcifications and coronary arteries; The calculation module is used to perform integral calculations on the segmentation results of the coronary artery calcifications to obtain the integral quantification results.

8. An electronic device, characterized in that, include: A memory, one or more processors; the memory is coupled to the processors; wherein the memory stores computer program code, the computer program code including computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the steps of the method for quantifying coronary artery calcium in cardiac CT images as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for quantifying coronary artery calcium in cardiac CT images as described in any one of claims 1-6.

Images