A method, apparatus, and storage medium for calculating coronary artery calcium scores.

Abstract

This application provides a method, apparatus, and storage medium for calculating coronary artery calcium scores, comprising: segmenting an initial three-dimensional medical image of the cardiac region from a plain coronary CT image using an image segmentation model; identifying calcified plaque regions in the initial three-dimensional medical image to determine at least one target region image and the location information of each image; identifying the coronary vessel category in each image by recognizing the input consisting of the target region image and the image location information; performing overall recognition of the initial three-dimensional medical image using a second coronary classification model to determine the calcified plaque regions included in the initial three-dimensional medical image, the location information of each region, and the category of coronary vessels; and calculating the calcium score for the calcified regions identified by both models to obtain the total coronary artery calcium score. Thus, the method provided in this application can quickly and accurately calculate the coronary artery calcium score based on plain CT scans.

Description

Technical Field

[0001] This application relates to the field of calcium score calculation technology, and in particular to a method, apparatus and storage medium for calculating coronary artery calcium score. Background Technology

[0002] Cardiovascular disease is the leading cause of death worldwide. The degree of calcification of plaques in the coronary arteries is a crucial indicator for monitoring and predicting cardiovascular disease. Plain CT scans, a commonly used screening tool for coronary artery disease, are frequently used to calculate indicators such as calcification scores. The calcification score calculated from plain CT images is an evaluation indicator reflecting the degree of calcification in the coronary arteries. However, in practice, plain CT images often lack clear visualization of the coronary arteries, and the absence of synchronized electrocardiogram (ECG) results in motion artifacts and scan dose noise, leading to incorrect labeling or inaccurate scoring of manually marked calcified areas. Furthermore, the method of manually marking calcified areas suffers from low efficiency in calculating coronary artery calcification scores. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, device and storage medium for calculating coronary artery calcium score, which can quickly and accurately calculate coronary artery calcium score based on plain CT scan.

[0004] This application provides a method for calculating coronary artery calcium scores, the calculation method including:

[0005] The acquired three-dimensional coronary artery plain CT image of the target patient is input into a pre-trained image segmentation model for image segmentation to obtain the initial three-dimensional medical image of the cardiac region of the target patient.

[0006] The regions of calcified plaques in the initial three-dimensional medical image are identified to determine at least one target region image corresponding to a calcified plaque and the location information of each target region image in the initial three-dimensional medical image.

[0007] The location information of each target region image identified from the initial three-dimensional medical image is input into a pre-constructed first coronary artery classification model to determine the category of coronary vessels in each target region image in the initial three-dimensional medical image;

[0008] The initial three-dimensional medical image is input into a pre-constructed second coronary artery classification model to determine the calcification prediction region of the calcified plaque included in the initial three-dimensional medical image, the location information of each calcification prediction region in the initial three-dimensional medical image, and the corresponding coronary artery category.

[0009] The regions of calcified plaques with the same location information and the same category of coronary arteries identified by the first coronary artery classification model and the second coronary artery classification model are labeled in the initial three-dimensional medical image to obtain the target three-dimensional medical image.

[0010] Based on the calcified plaque regions marked in the target three-dimensional medical image, the calcification integral is calculated to obtain the total coronary artery calcification integral.

[0011] Optionally, an image segmentation model can be constructed using the following steps:

[0012] Multiple 3D coronary artery plain CT images to be trained are acquired; the cardiac region in each 3D coronary artery plain CT image to be trained is segmented in advance to determine the true segmentation result of the cardiac region in each 3D coronary artery plain CT image to be trained.

[0013] The multiple three-dimensional coronary artery plain CT images to be trained are input into a segmentation neural network to determine the predicted segmentation result of each three-dimensional coronary artery plain CT image to be trained; wherein, the segmentation neural network is built based on 3D-U-net;

[0014] The predicted segmentation results of each 3D coronary artery plain CT image to be trained are compared with the actual segmentation results of the same 3D coronary artery plain CT image to determine the value of the hybrid loss function.

[0015] Based on the value of the hybrid loss function, the segmentation neural network is iteratively trained using the backpropagation algorithm to obtain the image segmentation model.

[0016] Optionally, the step of identifying the regions of calcified plaques in the initial three-dimensional medical image, and determining at least one target region image corresponding to a calcified plaque and the location information of each target region image in the initial three-dimensional medical image, includes:

[0017] The initial three-dimensional medical image is binarized according to a preset CT threshold to generate a three-dimensional binary medical image; the three-dimensional binary medical image is used to identify the region of suspected calcified plaque in the coronary artery.

[0018] Based on the three-dimensional binary medical image, connected components are constructed, and each constructed connected component is identified as a region of suspected calcified plaques included in the three-dimensional binary medical image, thereby identifying at least one initial calcified region.

[0019] Based on the threshold method, at least one initial calcification region that meets the threshold requirement is selected from the at least one initial calcification region, and the selected initial calcification region is determined as the target calcification region, thus obtaining at least one target calcification region;

[0020] For each target calcified region, based on the value of the bounding box of the target calcified region, a target region image of the corresponding size is cropped from the initial three-dimensional medical image, and the position information of each cropped target region image in the initial three-dimensional medical image is determined.

[0021] Optionally, the step of calculating the total coronary artery calcium score based on the calcified plaque regions marked in the target three-dimensional medical image includes:

[0022] Calculate the Agaston integral, volume integral, and mass integral of the calcified plaque region for each coronary artery in each slice.

[0023] The Agaston integral, volume integral, and mass integral of each coronary artery in each slice are added together to obtain the Agaston integral, volume integral, and mass integral of each coronary artery.

[0024] The Agaston integral, volume integral, and mass integral of each coronary artery are added together to obtain the total coronary artery calcification integral.

[0025] Optionally, the step of selecting at least one initial calcification region that meets the threshold requirement from the at least one initial calcification region based on the threshold method, and determining the selected initial calcification region as the target calcification region, includes:

[0026] The initial calcified regions with a volume smaller than the first volume threshold, an area greater than the first area threshold, and a number of voxels greater than the first number are defined as the target calcified regions.

[0027] Optionally, the hybrid loss function includes the cross-entropy loss function, the structural similarity loss function, and the IOU loss function.

[0028] Optionally, the types of coronary vessels include: left anterior descending artery, right coronary artery, left circumflex artery, left main coronary artery, and branch vessels.

[0029] Optionally, the second coronary artery classification model includes three parts: a backbone part, a neck part, and a prediction part.

[0030] This application embodiment also provides a device for calculating coronary artery calcium scores, the device comprising:

[0031] The segmentation module is used to input the acquired target three-dimensional coronary artery plain CT image of the target patient into a pre-trained image segmentation model for image segmentation, thereby obtaining an initial three-dimensional medical image of the target patient's cardiac region.

[0032] The identification module is used to identify the regions of calcified plaques in the initial three-dimensional medical image, and determine the target region image corresponding to at least one calcified plaque and the position information of each target region image in the initial three-dimensional medical image;

[0033] The first determining module is used to input the location information of each target region image identified from the initial three-dimensional medical image into a pre-constructed first coronary artery classification model to determine the category of coronary vessels in each target region image in the initial three-dimensional medical image;

[0034] The second determining module is used to input the initial three-dimensional medical image into a pre-constructed second coronary artery classification model to determine the calcification prediction region of the calcified plaque included in the initial three-dimensional medical image, the location information of each calcification prediction region in the initial three-dimensional medical image, and the corresponding coronary artery category.

[0035] The annotation module is used to annotate the regions of calcified plaques with the same location information and the same category of coronary vessels identified by the first coronary artery classification model and the second coronary artery classification model in the initial three-dimensional medical image to obtain the target three-dimensional medical image.

[0036] The calculation module is used to calculate the calcification integral based on the calcified plaque regions marked in the target three-dimensional medical image, and obtain the total calcification integral of the coronary artery.

[0037] Optionally, the computing device further includes a model building module, the model building module being used for:

[0038] Multiple 3D coronary artery plain CT images to be trained are acquired; the cardiac region in each 3D coronary artery plain CT image to be trained is segmented in advance to determine the true segmentation result of the cardiac region in each 3D coronary artery plain CT image to be trained.

[0039] The multiple three-dimensional coronary artery plain CT images to be trained are input into a segmentation neural network to determine the predicted segmentation result of each three-dimensional coronary artery plain CT image to be trained; wherein, the segmentation neural network is built based on 3D-U-net;

[0040] The predicted segmentation results of each 3D coronary artery plain CT image to be trained are compared with the actual segmentation results of the same 3D coronary artery plain CT image to determine the value of the hybrid loss function.

[0041] Based on the value of the hybrid loss function, the segmentation neural network is iteratively trained using the backpropagation algorithm to obtain the image segmentation model.

[0042] Optionally, when the recognition module identifies regions of calcified plaques in the initial three-dimensional medical image and determines at least one target region image corresponding to a calcified plaque and the location information of each target region image in the initial three-dimensional medical image, the recognition module is used to:

[0043] The initial three-dimensional medical image is binarized according to a preset CT threshold to generate a three-dimensional binary medical image; the three-dimensional binary medical image is used to identify the region of suspected calcified plaque in the coronary artery.

[0044] Based on the three-dimensional binary medical image, connected components are constructed, and each constructed connected component is identified as a region of suspected calcified plaques included in the three-dimensional binary medical image, thereby identifying at least one initial calcified region.

[0045] Based on the threshold method, at least one initial calcification region that meets the threshold requirement is selected from the at least one initial calcification region, and the selected initial calcification region is determined as the target calcification region, thus obtaining at least one target calcification region;

[0046] For each target calcified region, based on the value of the bounding box of the target calcified region, a target region image of the corresponding size is cropped from the initial three-dimensional medical image, and the position information of each cropped target region image in the initial three-dimensional medical image is determined.

[0047] Optionally, when the calculation module is used to calculate the total coronary artery calcification score based on the calcified plaque regions marked in the target three-dimensional medical image, the calculation module is used to:

[0048] Calculate the Agaston integral, volume integral, and mass integral of the calcified plaque region for each coronary artery in each slice.

[0049] The Agaston integral, volume integral, and mass integral of each coronary artery in each slice are added together to obtain the Agaston integral, volume integral, and mass integral of each coronary artery.

[0050] The Agaston integral, volume integral, and mass integral of each coronary artery are added together to obtain the total coronary artery calcification integral.

[0051] Optionally, when the identification module is used to select at least one initial calcification region that meets the threshold requirement from the at least one initial calcification region based on the threshold method, and to determine the selected initial calcification region as the target calcification region, the identification module is used to:

[0052] The initial calcified regions with a volume smaller than the first volume threshold, an area greater than the first area threshold, and a number of voxels greater than the first number are defined as the target calcified regions.

[0053] Optionally, the hybrid loss function includes the cross-entropy loss function, the structural similarity loss function, and the IOU loss function.

[0054] Optionally, the types of coronary vessels include: left anterior descending artery, right coronary artery, left circumflex artery, left main coronary artery, and branch vessels.

[0055] Optionally, the second coronary artery classification model includes three parts: a backbone part, a neck part, and a prediction part.

[0056] This application also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the calculation method described above are performed.

[0057] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the calculation method described above.

[0058] This application provides a method, apparatus, and storage medium for calculating coronary artery calcium scores. The calculation method includes: inputting a target three-dimensional coronary artery plain CT image of a target patient into a pre-trained image segmentation model for image segmentation to obtain an initial three-dimensional medical image of the cardiac region of the target patient; identifying the regions of calcified plaques in the initial three-dimensional medical image to determine at least one target region image corresponding to a calcified plaque and the position information of each target region image in the initial three-dimensional medical image; inputting each target region image identified from the initial three-dimensional medical image and the position information of each target region image into a pre-constructed first coronary artery classification model to determine each target region image in the initial three-dimensional medical image. The initial three-dimensional medical image is input into a pre-constructed second coronary artery classification model to determine the calcification prediction regions of calcified plaques included in the initial three-dimensional medical image, the location information of each calcification prediction region in the initial three-dimensional medical image, and the corresponding coronary artery category. Regions of calcified plaques with the same location information and category of coronary arteries identified by the first and second coronary artery classification models are marked in the initial three-dimensional medical image to obtain a target three-dimensional medical image. Based on the marked calcified plaque regions in the target three-dimensional medical image, a calcification score is calculated to obtain the total coronary artery calcification score.

[0059] Thus, this application improves the efficiency and accuracy of cardiac region image segmentation by using an image segmentation model trained based on a hybrid loss function. Furthermore, it achieves comprehensive recognition of local and global features of the three-dimensional image of the cardiac region through two coronary artery classification models, thereby improving the accuracy of calcified region and coronary artery category recognition results, and consequently improving the accuracy of calcification integral calculation results.

[0060] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0061] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart illustrating a method for calculating coronary artery calcium scores provided in an embodiment of this application;

[0063] Figure 2 A schematic diagram of the SPP structure provided in this application;

[0064] Figure 3 A schematic diagram of the technical architecture of the method for calculating coronary artery calcium scores provided in this application;

[0065] Figure 4 One of the schematic diagrams of a coronary artery calcium integration calculation device provided in the embodiments of this application;

[0066] Figure 5 A second schematic diagram of a coronary artery calcification integral calculation device provided in an embodiment of this application;

[0067] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0069] Cardiovascular disease is the leading cause of death worldwide. The degree of calcification of plaques in the coronary arteries is a crucial indicator for monitoring and predicting cardiovascular disease. Plain CT scans, a commonly used screening tool for coronary artery disease, are frequently used to calculate indicators such as calcification scores. The calcification score calculated from plain CT images is an evaluation indicator reflecting the degree of calcification in the coronary arteries. However, in practice, plain CT images often lack clear visualization of the coronary arteries, and the absence of synchronized electrocardiogram (ECG) results in motion artifacts and scan dose noise, leading to incorrect labeling or inaccurate scoring of manually marked calcified areas. Furthermore, the method of manually marking calcified areas suffers from low efficiency in calculating coronary artery calcification scores.

[0070] Based on this, the embodiments of this application provide a method for calculating coronary artery calcium scores, which can quickly and accurately calculate coronary artery calcium scores based on plain CT scans.

[0071] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for calculating coronary artery calcium scores, provided as an embodiment of this application. Figure 1 As shown in the embodiments of this application, the calculation method includes:

[0072] S101. Input the acquired target three-dimensional coronary artery plain CT image of the target patient into a pre-trained image segmentation model for image segmentation to obtain the initial three-dimensional medical image of the target patient's heart region.

[0073] Here, the image segmentation model is used to segment the heart region from a coronary artery plain CT image to obtain a three-dimensional medical image containing only the heart region of the target patient.

[0074] The reason for using an image segmentation model to segment the three-dimensional medical image of the heart region from the coronary artery plain CT image is to exclude areas with high false positive noise such as bones, thereby improving the accuracy of subsequent calcification integral calculation results.

[0075] Plain CT images are three-dimensional medical images acquired during the patient's imaging process without contrast agents. Plain CT images are often used for initial screening of coronary artery disease and for calculating calcification scores. While high-density structures such as calcifications and stents can be observed on plain CT scans, blood vessels are difficult to visualize.

[0076] In one embodiment provided in this application, an image segmentation model is constructed through the following steps: acquiring multiple three-dimensional coronary artery plain CT images to be trained; wherein, the heart region in each of the three-dimensional coronary artery plain CT images to be trained is pre-segmented to determine the true segmentation result of the heart region in each of the three-dimensional coronary artery plain CT images to be trained; inputting the multiple three-dimensional coronary artery plain CT images to be trained into a segmentation neural network to determine the predicted segmentation result of each three-dimensional coronary artery plain CT image to be trained; wherein, the segmentation neural network is constructed based on 3D-U-net; comparing and calculating the predicted segmentation result and the true segmentation result of each three-dimensional coronary artery plain CT image to be trained to determine the value of the hybrid loss function; and iteratively training the segmentation neural network using the backpropagation algorithm based on the value of the hybrid loss function to obtain the image segmentation model.

[0077] Here, for each 3D coronary artery plain CT image to be trained, the heart region in the image is segmented in advance to determine the true segmentation result of the heart region in each 3D coronary artery plain CT image to be trained. This is to add label information to the 3D coronary artery plain CT image to be trained that can be used for model recognition and training.

[0078] Here, the hybrid loss function includes the cross-entropy loss function (BCE), the structural similarity loss function (SSIM), and the IOU loss function. The hybrid loss function is composed of these three loss functions.

[0079] Among them, BCE does not consider neighborhood labels and is calculated per pixel; SSIM is a patch-level loss, which helps optimize focusing on boundary and foreground regions; IOU loss is a block-level metric, which places more emphasis on large foreground regions, thus generating relatively uniform and more accurate probabilities for these regions. The mixture of the three losses retains the characteristics of BCE in preserving the smooth gradient of all pixels, the property of IOU in focusing more on the foreground, and the advantage of SSIM in respecting the structure of the original image by using a larger loss near the boundary and further pushing the background prediction to zero.

[0080] Here, the backpropagation algorithm is used to iteratively train the segmentation neural network, which updates the weight coefficients of the segmentation neural network so that the image segmentation model corresponding to the updated weights can obtain a segmentation result that is closer to the real situation when performing image segmentation.

[0081] S102. Identify the regions of calcified plaques in the initial three-dimensional medical image, and determine at least one target region image corresponding to a calcified plaque and the position information of each target region image in the initial three-dimensional medical image.

[0082] Here, identifying areas with calcified plaques in the initial 3D medical image of the heart region is to remove abnormal areas that are not calcified plaques from the initial 3D medical image before coronary artery classification. For example, suspected areas that are too large or too small are removed.

[0083] In one embodiment provided in this application, the step of identifying the region of calcified plaque in the initial three-dimensional medical image and determining at least one target region image corresponding to a calcified plaque and the position information of each target region image in the initial three-dimensional medical image includes: performing binarization processing on the initial three-dimensional medical image according to a preset CT threshold to generate a three-dimensional binary medical image; the three-dimensional binary medical image is used to determine the region of suspected calcified plaque in the coronary artery; constructing connected components based on the three-dimensional binary medical image, and determining each constructed connected component as the region of suspected calcified plaque included in the three-dimensional binary medical image, thereby determining at least one initial calcified region; based on a threshold method, selecting at least one initial calcified region that meets the threshold requirement from the at least one initial calcified region, and determining the selected initial calcified region as the target calcified region, thereby obtaining at least one target calcified region; for each target calcified region, cropping a target region image of a corresponding size from the initial three-dimensional medical image according to the value of the bounding box of the target calcified region, and determining the position information of each cropped target region image in the initial three-dimensional medical image.

[0084] Here, the initial three-dimensional medical image is transformed into a three-dimensional binary medical image based on a preset CT threshold, which involves converting a grayscale image into a black and white image. The CT threshold can be selected according to the actual situation.

[0085] For example, based on a preset CT threshold, the initial three-dimensional medical image is binarized to generate a three-dimensional binary medical image. This can be done as follows: Assuming the preset CT threshold is 130, first determine the CT value of each voxel in the initial three-dimensional medical image; then, change the CT values ​​of voxels with CT values ​​greater than or equal to 130 to the CT values ​​corresponding to pure white, and change the CT values ​​of voxels with CT values ​​less than 130 to the CT values ​​corresponding to pure black. This yields the three-dimensional binary medical image. It should be noted that the CT values ​​of calcified plaques in plain CT images are generally greater than 130.

[0086] Here, when constructing connected components based on the aforementioned three-dimensional binary medical image, the connected component function of skimage can be used to construct multiple connected components. This allows for the construction of at least one connected component, and generally, multiple connected components can be constructed. Each constructed connected component represents a region where a suspected plaque may exist.

[0087] Here, the threshold method is used to screen out target calcified regions from the initial calcified regions. In fact, some connected regions that are suspected to be calcified regions but are not calcified regions are deleted, so as to retain only the regions where calcified plaques exist.

[0088] In another embodiment provided in this application, the step of selecting at least one initial calcification region that meets the threshold requirements from the at least one initial calcification region based on the threshold method, and determining the selected initial calcification region as the target calcification region, includes: determining the initial calcification region with a volume smaller than a first volume threshold, an area larger than a first area threshold, and a voxel count greater than a first number as the target calcification region.

[0089] The first volume threshold, the first area threshold, and the first quantity are predetermined based on actual conditions. Typically, a volume greater than 1000 mm² is considered. 3 or area less than 1mm 2 Alternatively, areas with fewer than 3 voxels are not considered as areas containing calcified plaques. Thus, area and volume thresholds are used to filter out non-calcified plaque areas that are too large or too small.

[0090] Here, the target calcified region obtained through filtering is also a connected region, and the bounding box of the target calcified region is the bounding box of the corresponding connected region.

[0091] S103. Input the image of each target region identified from the initial three-dimensional medical image and the location information of each target region image into the pre-constructed first coronary artery classification model to determine the category of coronary vessels in each target region image in the initial three-dimensional medical image.

[0092] Here, the input features of the first coronary artery classification model are composed of the target region image and the location information of the target region image in the initial three-dimensional medical image. In this way, the first coronary artery classification model processes the connected components of the plain CT scan, rather than an entire plain CT scan image.

[0093] The first coronary artery classification model is built on ResNet with added image self-attention SENet. The introduction of the self-attention mechanism allows the feature weights to be reassigned, making the model more focused on the main coronary vessel categories of interest.

[0094] The categories of coronary vessels include: left anterior descending artery (LAD), right coronary artery (RCA), left circumflex artery (LCX), left main coronary artery (LM), and branch vessels. Among these, coronary vessels other than the left anterior descending artery (LAD), right coronary artery (RCA), left circumflex artery (LCX), and left main coronary artery (LM) are referred to as branch vessels.

[0095] Specifically, when determining the category of coronary vessels in each target region image using the first coronary classification model—that is, determining which vessel contains calcified plaques—the process involves first using a high-dimensional tensor constructed from the connected components of interest (target region image) and their coordinate information (location information) as input to the first classification model to learn and extract abstract features with location information. Then, the abstract features are compressed to obtain global receptive field information, and the output dimension matches the number of input feature channels. This represents the global distribution of responses across feature channels and allows layers closer to the input to also obtain the global receptive field. Next is the grasping operation, which is similar to a gate mechanism in a recurrent neural network. Weights are generated for each feature channel using weight parameters, which are learned to explicitly model the correlation between feature channels. Finally, a weight reassignment operation is performed. The extracted output weights are treated as the weights of each feature channel after feature selection, and then multiplied and weighted channel-by-channel to recalibrate the original features in the channel dimension. Finally, a fully connected layer outputs the predicted coronary vessel category.

[0096] It should be noted that the first coronary artery classification is used to identify local features in plain CT images.

[0097] S104. Input the initial three-dimensional medical image into the pre-constructed second coronary artery classification model to determine the calcification prediction region of the calcified plaque included in the initial three-dimensional medical image, the location information of each calcification prediction region in the initial three-dimensional medical image, and the corresponding coronary artery category.

[0098] Here, the input feature of the second coronary artery classification model is the entire image, so the second coronary artery classification model is used to identify global features in plain CT images.

[0099] The second coronary artery classification model consists of three parts: the backbone, the neck, and the prediction. Furthermore, an SPP structure is introduced into this second coronary artery classification model.

[0100] The backbone is based on Darknet, with a structure similar to ResNet, stacking multiple residual modules primarily for extracting more efficient feature maps. The entire Darknet is a fully convolutional network that heavily utilizes residual skip connections. It also uses varying kernel strides to downsampling and mitigate the negative gradient effects of pooling. These residual connections enhance the detailed features of calcified patches on plain CT scans, enabling the backbone module to better learn the context and texture information of these patches.

[0101] The Neck section uses PANet, an improved version based on FPN. The main method of FPN is to downsample shallow large-scale feature maps from top to bottom and fuse them with deep feature maps to output multiple feature maps of different scales for prediction. PANet adds an upsampling operation on deep feature maps, upsamples deep small-scale feature maps and concatenates them together, proposing a top-down + bottom-up feature fusion method. This pyramid-like structure is more adaptable to calcified patches of different sizes to establish pixel-level connections and build the dependencies between patch features of different scales through global and local information of calcified patches. PANet can be regarded as a multi-scale input and multi-scale output encoder. It has two main aspects that have the greatest impact on network performance: (1) multi-scale feature fusion; (2) divide and conquer: in short, it outputs multiple feature maps of different scales with different receptive fields.

[0102] The connection between the Neck and Prediction parts uses a convolutional set with an inserted SPP structure. The SPP structure enables feature fusion at different scales. Learning features at different scales better enhances the network's ability to acquire receptive fields of calcified patches of different sizes. Finally, the tensors after the three pooling operations are concatenated by channel. To ensure that the tensor size of each branch is the same during concatenation, different levels of padding are applied during max pooling. For an example, please refer to [link to example]. Figure 2 , Figure 2 A schematic diagram of the SPP structure provided in this application.

[0103] The Prediction part is a decoder structure. Its main structure is a Conv+Bn+Act module and a 1×1×1 convolutional classification layer. The main reasons for using a 1×1×1 convolution instead of a fully connected layer for classification are twofold: (1) the input scale of the fully connected layer is limited, while the convolutional layer only needs to limit the number of input and output channels; (2) the output of the fully connected layer is one-dimensional or two-dimensional, and the feature map needs to be filtered when it is input to the fully connected layer, which to some extent destroys the spatial information of the feature map. The output of the convolutional layer is four-dimensional (c,z,x,y), where c is the confidence level, and x,y,z are the coordinates of the center point on the XYZ axis, respectively. Choosing the convolutional layer as the decoder greatly preserves the spatial structure information of the original image on the feature map.

[0104] It should be noted that the first and second coronary artery classification models are run simultaneously, but they have different input formats. The second coronary artery classification model takes the entire plain CT scan as input, and its expressive power is more reflected in global information. The first coronary artery classification model, based on a self-attention mechanism, predicts the coronary artery category using the simply connected regions and their location information from a single plain CT scan, learning category features more from local information.

[0105] S105. Using the first coronary artery classification model and the second coronary artery classification model, the regions of calcified plaques with the same location information and the same category of coronary arteries are identified and annotated in the initial three-dimensional medical image to obtain the target three-dimensional medical image.

[0106] It should be noted that the first coronary artery classification model is used to perform the first identification of calcified plaques in plain CT images from local features, and inputs a set of identification results; while the second coronary artery classification model is used to perform the second identification of calcified plaques in plain CT images from global features simultaneously, and inputs a set of identification results.

[0107] Here, the target 3D medical image is determined by annotating the initial 3D medical image with regions (regions where calcified plaques exist) that are identified as true by both coronary artery classification models (coronary vessels with the same location information and the same category).

[0108] In this way, the marked areas in the target three-dimensional medical image are the areas for the final calculation of calcification integral, and the type of coronary vessels corresponding to each area can also be determined.

[0109] S106. Based on the calcified plaque regions marked in the target three-dimensional medical image, calculate the calcification integral to obtain the total coronary artery calcification integral.

[0110] In one embodiment of this application, the step of calculating the total coronary artery calcification score based on the calcified plaque region marked in the target three-dimensional medical image includes: calculating the Agaston integral, volume integral, and mass integral of the calcified plaque region for each coronary artery in each slice; summing the Agaston integral, volume integral, and mass integral of each coronary artery in each slice to obtain the Agaston integral, volume integral, and mass integral of each coronary artery; and summing the Agaston integral, volume integral, and mass integral of each coronary artery to obtain the total coronary artery calcification score.

[0111] Here, we first calculate the Agaston integral, volume integral, and mass integral of the calcified regions of the coronary arteries (LM, LAD, LCX, RCA, and branch vessels) in each slice. Then, we sum these integrals separately to obtain the Agaston integral, volume integral, and mass integral of each coronary artery (LM, LAD, LCX, RCA, and branch vessels). Finally, we sum these integrals again to obtain the total coronary artery calcification score. The total coronary artery calcification score includes three types of integrals: Agaston integral, volume integral, and mass integral.

[0112] For an example, please refer to Figure 3 , Figure 3 This is a schematic diagram of the technical architecture of the method for calculating coronary artery calcium scores provided in this application. Figure 3 As shown, the calculation process of this scheme can be simply summarized as follows: First, the cardiac region of the target three-dimensional coronary artery plain CT image is segmented; then, the segmented cardiac plain CT image is used to simultaneously perform classification tasks based on local features and global features; finally, the plaque attribution is determined by AND gate logic, and the Agassone integral, mass integral and volume integral are calculated, thus completing the closed loop of automatic calculation of calcification integral.

[0113] Thus, this application improves the efficiency and accuracy of cardiac region image segmentation by using an image segmentation model trained based on a hybrid loss function. Furthermore, it achieves comprehensive recognition of local and global features of the three-dimensional image of the cardiac region through two coronary artery classification models, thereby improving the accuracy of calcified region and coronary artery category recognition results, and consequently improving the accuracy of calcification integral calculation results.

[0114] Please see Figure 4 , Figure 5 , Figure 4 One of the schematic diagrams of a coronary artery calcium integration calculation device provided in the embodiments of this application; Figure 5 This is a second schematic diagram of a coronary artery calcium integration calculation device provided in an embodiment of this application. Figure 4 As shown, the computing device 400 includes:

[0115] The segmentation module 410 is used to input the acquired target three-dimensional coronary artery plain CT image of the target patient into a pre-trained image segmentation model for image segmentation, and to obtain the initial three-dimensional medical image of the heart region of the target patient.

[0116] The identification module 420 is used to identify the regions of calcified plaques in the initial three-dimensional medical image, and determine the target region image corresponding to at least one calcified plaque and the position information of each target region image in the initial three-dimensional medical image.

[0117] The first determining module 430 is used to input the location information of each target region image identified from the initial three-dimensional medical image into a pre-constructed first coronary artery classification model to determine the category of coronary arteries in each target region image in the initial three-dimensional medical image;

[0118] The second determining module 440 is used to input the initial three-dimensional medical image into a pre-constructed second coronary artery classification model to determine the calcification prediction region of the calcified plaque included in the initial three-dimensional medical image, the location information of each calcification prediction region in the initial three-dimensional medical image, and the corresponding coronary artery category.

[0119] The annotation module 450 is used to annotate the regions of calcified plaques with the same location information and the same category of coronary vessels identified by the first coronary artery classification model and the second coronary artery classification model in the initial three-dimensional medical image to obtain the target three-dimensional medical image.

[0120] The calculation module 460 is used to calculate the calcification integral based on the calcified plaque region marked in the target three-dimensional medical image, and obtain the total calcification integral of the coronary artery.

[0121] Optional, such as Figure 5 As shown, the computing device 400 further includes a model building module 470, which is used for:

[0122] Multiple 3D coronary artery plain CT images to be trained are acquired; the cardiac region in each 3D coronary artery plain CT image to be trained is segmented in advance to determine the true segmentation result of the cardiac region in each 3D coronary artery plain CT image to be trained.

[0123] The multiple three-dimensional coronary artery plain CT images to be trained are input into a segmentation neural network to determine the predicted segmentation result of each three-dimensional coronary artery plain CT image to be trained; wherein, the segmentation neural network is built based on 3D-U-net;

[0124] The predicted segmentation results of each 3D coronary artery plain CT image to be trained are compared with the actual segmentation results of the same 3D coronary artery plain CT image to determine the value of the hybrid loss function.

[0125] Based on the value of the hybrid loss function, the segmentation neural network is iteratively trained using the backpropagation algorithm to obtain the image segmentation model.

[0126] Optionally, when the recognition module 420 identifies the regions of calcified plaques in the initial three-dimensional medical image and determines at least one target region image corresponding to a calcified plaque and the position information of each target region image in the initial three-dimensional medical image, the recognition module 420 is used to:

[0127] The initial three-dimensional medical image is binarized according to a preset CT threshold to generate a three-dimensional binary medical image; the three-dimensional binary medical image is used to identify the region of suspected calcified plaque in the coronary artery.

[0128] Based on the three-dimensional binary medical image, connected components are constructed, and each constructed connected component is identified as a region of suspected calcified plaques included in the three-dimensional binary medical image, thereby identifying at least one initial calcified region.

[0129] Based on the threshold method, at least one initial calcification region that meets the threshold requirement is selected from the at least one initial calcification region, and the selected initial calcification region is determined as the target calcification region, thus obtaining at least one target calcification region;

[0130] For each target calcified region, based on the value of the bounding box of the target calcified region, a target region image of the corresponding size is cropped from the initial three-dimensional medical image, and the position information of each cropped target region image in the initial three-dimensional medical image is determined.

[0131] Optionally, when the calculation module 460 calculates the total coronary artery calcification score based on the calcified plaque region marked in the target three-dimensional medical image, the calculation module 460 is used to:

[0132] Calculate the Agaston integral, volume integral, and mass integral of the calcified plaque region for each coronary artery in each slice.

[0133] The Agaston integral, volume integral, and mass integral of each coronary artery in each slice are added together to obtain the Agaston integral, volume integral, and mass integral of each coronary artery.

[0134] The Agaston integral, volume integral, and mass integral of each coronary artery are added together to obtain the total coronary artery calcification integral.

[0135] Optionally, when the identification module 420 is used to select at least one initial calcification region that meets the threshold requirement from the at least one initial calcification region based on the threshold method, and to determine the selected initial calcification region as the target calcification region, the identification module 420 is used to:

[0136] The initial calcified regions with a volume smaller than the first volume threshold, an area greater than the first area threshold, and a number of voxels greater than the first number are defined as the target calcified regions.

[0137] Optionally, the hybrid loss function includes the cross-entropy loss function, the structural similarity loss function, and the IOU loss function.

[0138] Optionally, the types of coronary vessels include: left anterior descending artery, right coronary artery, left circumflex artery, left main coronary artery, and branch vessels.

[0139] Optionally, the second coronary artery classification model includes three parts: a backbone part, a neck part, and a prediction part.

[0140] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 600 includes a processor 610, a memory 620, and a bus 630.

[0141] The memory 620 stores machine-readable instructions executable by the processor 610. When the electronic device 600 is running, the processor 610 and the memory 620 communicate via the bus 630. When the machine-readable instructions are executed by the processor 610, they can perform the operations described above. Figures 1 to 3 The steps in the method embodiment shown are specifically implemented in the method embodiment and will not be repeated here.

[0142] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figures 1 to 3 The steps in the method embodiment shown are specifically implemented in the method embodiment and will not be repeated here.

[0143] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0144] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0145] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0146] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0147] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for calculating coronary artery calcium scores, characterized in that, The calculation method includes: The acquired three-dimensional coronary artery plain CT image of the target patient is input into a pre-trained image segmentation model for image segmentation to obtain the initial three-dimensional medical image of the cardiac region of the target patient. The regions of calcified plaques in the initial three-dimensional medical image are identified to determine at least one target region image corresponding to a calcified plaque and the location information of each target region image in the initial three-dimensional medical image. Each target region image identified from the initial 3D medical image and its location information are input into a pre-built first coronary artery classification model to determine the category of coronary vessels in each target region image of the initial 3D medical image. The first coronary artery classification model is built on ResNet with added image self-attention SENet. The introduction of the self-attention mechanism reassigns feature weights, making the model more focused on the main coronary vessel categories of interest. The initial 3D medical image is input into a pre-constructed second coronary artery classification model to determine the calcification prediction region of the calcified plaques included in the initial 3D medical image, the location information of each calcification prediction region in the initial 3D medical image, and the corresponding coronary artery category. The input feature of the second coronary artery classification model is the entire image. The second coronary artery classification model is used to identify global features in the plain CT image. The second coronary artery classification model includes three parts: Backbone part, Neck part, and Prediction part, and introduces an SPP structure in the second coronary artery classification model. The regions of calcified plaques with the same location information and the same category of coronary arteries identified by the first coronary artery classification model and the second coronary artery classification model are labeled in the initial three-dimensional medical image to obtain the target three-dimensional medical image. Based on the calcified plaque regions marked in the target three-dimensional medical image, the calcification integral is calculated to obtain the total coronary artery calcification integral.

2. The calculation method according to claim 1, characterized in that, The image segmentation model is constructed using the following steps: Multiple 3D coronary artery plain CT images to be trained are acquired; the cardiac region in each 3D coronary artery plain CT image to be trained is segmented in advance to determine the true segmentation result of the cardiac region in each 3D coronary artery plain CT image to be trained. The multiple three-dimensional coronary artery plain CT images to be trained are input into a segmentation neural network to determine the predicted segmentation result of each three-dimensional coronary artery plain CT image to be trained; wherein, the segmentation neural network is built based on 3D-U-net; The predicted segmentation results of each 3D coronary artery plain CT image to be trained are compared with the actual segmentation results of the same 3D coronary artery plain CT image to determine the value of the hybrid loss function. Based on the value of the hybrid loss function, the segmentation neural network is iteratively trained using the backpropagation algorithm to obtain the image segmentation model.

3. The calculation method according to claim 1, characterized in that, The step of identifying calcified plaque regions in the initial three-dimensional medical image, determining at least one target region image corresponding to a calcified plaque, and the location information of each target region image in the initial three-dimensional medical image includes: The initial three-dimensional medical image is binarized according to a preset CT threshold to generate a three-dimensional binary medical image; the three-dimensional binary medical image is used to identify the region of suspected calcified plaque in the coronary artery. Based on the three-dimensional binary medical image, connected components are constructed, and each constructed connected component is identified as a region of suspected calcified plaques included in the three-dimensional binary medical image, thereby identifying at least one initial calcified region. Based on the threshold method, at least one initial calcification region that meets the threshold requirement is selected from the at least one initial calcification region, and the selected initial calcification region is determined as the target calcification region, thereby obtaining at least one target calcification region; For each target calcified region, based on the value of the bounding box of the target calcified region, a target region image of the corresponding size is cropped from the initial three-dimensional medical image, and the position information of each cropped target region image in the initial three-dimensional medical image is determined.

4. The calculation method according to claim 1, characterized in that, The step of calculating the total coronary artery calcium score based on the calcified plaque regions marked in the target three-dimensional medical image includes: Calculate the Agaston integral, volume integral, and mass integral of the calcified plaque region for each coronary artery in each slice. The Agaston integral, volume integral, and mass integral of each coronary artery in each slice are added together to obtain the Agaston integral, volume integral, and mass integral of each coronary artery. The Agaston integral, volume integral, and mass integral of each coronary artery are added together to obtain the total coronary artery calcification integral.

5. The calculation method according to claim 3, characterized in that, The threshold-based method, which selects at least one initial calcification region that meets the threshold requirement from the at least one initial calcification region, and determines the selected initial calcification region as the target calcification region, includes: The initial calcified regions with a volume smaller than the first volume threshold, an area greater than the first area threshold, and a number of voxels greater than the first number are defined as the target calcified regions.

6. The calculation method according to claim 2, characterized in that, The hybrid loss function includes the cross-entropy loss function, the structural similarity loss function, and the IOU loss function.

7. A device for calculating coronary artery calcium integrals, characterized in that, The computing device includes: The segmentation module is used to input the acquired target three-dimensional coronary artery plain CT image of the target patient into a pre-trained image segmentation model for image segmentation, thereby obtaining an initial three-dimensional medical image of the target patient's cardiac region. The identification module is used to identify the regions of calcified plaques in the initial three-dimensional medical image, and determine the target region image corresponding to at least one calcified plaque and the position information of each target region image in the initial three-dimensional medical image; The first determining module is used to input the image of each target region identified from the initial three-dimensional medical image and the location information of each target region image into a pre-constructed first coronary artery classification model to determine the category of coronary vessels in each target region image in the initial three-dimensional medical image; wherein, the first coronary artery classification model is constructed based on ResNet with added image self-attention SENet. The introduction of the self-attention mechanism allows the feature weights to be reassigned, making the model more focused on the main coronary vessel categories of interest; The second determining module is used to input the initial three-dimensional medical image into a pre-constructed second coronary artery classification model to determine the calcification prediction region of the calcified plaques included in the initial three-dimensional medical image, the location information of each calcification prediction region in the initial three-dimensional medical image, and the corresponding coronary artery category; wherein, the input feature of the second coronary artery classification model is the entire image, the second coronary artery classification model is used to identify global features in the plain CT image, the second coronary artery classification model includes three parts: Backbone part, Neck part and Prediction part, and introduces an SPP structure in the second coronary artery classification model; The annotation module is used to annotate the regions of calcified plaques with the same location information and the same category of coronary vessels identified by the first coronary artery classification model and the second coronary artery classification model in the initial three-dimensional medical image to obtain the target three-dimensional medical image. The calculation module is used to calculate the calcification integral based on the calcified plaque regions marked in the target three-dimensional medical image, and obtain the total calcification integral of the coronary artery.

8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the machine-readable instructions are executed by the processor to perform the steps of the computation method as described in any one of claims 1 to 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, performs the steps of the computation method as described in any one of claims 1 to 6.

Images