A plaque recognition method based on coronary CT images

Abstract

This invention discloses a plaque recognition method based on coronary artery CT images. First, a dataset is prepared, and the coronary artery centerline is marked. Then, a cross-sectional image of the coronary artery is reconstructed based on the centerline, and the contours of the intima and endothelium are manually drawn on the cross-sectional image. Next, a convolutional neural network is constructed and trained based on the prepared dataset. Using this convolutional neural network, the intersection points of the rays and the intima and endothelium are predicted after inputting the coronary artery cross-sectional image. Connecting these points sequentially yields the contours of the intima and endothelium. Further, by calculating the area difference between the intima and endothelium, the presence of plaques on the cross-section can be identified. This method can better recognize the contours of the intima and endothelium and plaques, significantly improving accuracy and repeatability. Plaque recognition time is shortened, efficiency is significantly improved, and it has high auxiliary value for clinicians' diagnosis.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and more specifically to a method for plaque recognition based on coronary artery CT images. Background Technology

[0002] Coronary CT angiography is central to the diagnosis and risk stratification of coronary artery disease. From quantifying coronary artery calcification using the Agaston Score to classifying the degree of luminal stenosis, to describing plaque composition, identifying high-risk plaque characteristics, and analyzing plaque radiomics parameters, the role of coronary CT in plaque assessment has been continuously evolving. Currently, in clinical practice, plaque recognition algorithms primarily rely on the gradient features of image grayscale values, resulting in large errors in the identified coronary artery contours. Extensive manual optimization is often required to achieve satisfactory results for physicians, a time-consuming process susceptible to inter-observer variability. With the development of deep learning technology, intelligent medical image analysis has made significant progress, offering new possibilities for plaque recognition tasks. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the present invention aims to provide a plaque identification method based on coronary artery CT images.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A plaque identification method based on coronary artery CT images, the specific process of which is as follows:

[0006] S1. Select a set number of coronary CT images, manually mark the center lines of the three major coronary artery branches in each coronary CT image, and use the marked center lines to perform surface reconstruction for each vessel in the image to generate a coronary artery cross-sectional image; then manually mark the contours of the intima and epicardium of the coronary artery in each coronary artery cross-sectional image.

[0007] The coronary artery cross-sectional images with manually marked intricate outlines are randomly divided into training and testing sets according to a set ratio.

[0008] For each coronary artery cross-sectional image, rays are emitted in all directions from the image center as the origin. The distance from the intersection of the ray with the coronary artery intima and endothelium to the origin is used as the gold standard for model prediction. The coronary artery cross-sectional image is input into a convolutional neural network model, which extracts features over multiple layers and outputs the predicted distances from the intersection of the ray with the coronary artery intima and endothelium at each angle to the origin. For training set images, the network parameters are adjusted using stochastic gradient descent by comparing the convolutional neural network's predictions with the gold standard to reduce the prediction error of the convolutional neural network model. For test set images, the performance of the convolutional neural network model is measured by the difference between its predictions and the gold standard.

[0009] S2. Using the trained convolutional neural network model, predict the intersection points of the rays emitted from the image center in each direction with the coronary artery intima and endothelium from the input coronary artery cross-sectional image, and connect the intersection points in sequence to obtain the prediction results of the coronary artery intima and endothelium.

[0010] S3. Plaque Recognition: The coronary artery contour is obtained by connecting the rays predicted by the convolutional neural network model to the intersections of the intima and epicardium of the coronary artery. The area difference between the intima and epicardium should be greater than 0.1 mm. 2 Furthermore, the patch prediction result is obtained by cutting off sections with a length greater than 2 mm along the centerline.

[0011] Furthermore, in step S2, the set ratio is 8:2.

[0012] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0013] The present invention also provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program; and the processor is used to execute the computer program to implement the above-described method.

[0014] The beneficial effects of this invention are as follows: the method of this invention can better achieve the delineation of the intima and epicardium of the coronary arteries and the identification of plaques, with significantly improved accuracy and repeatability. Plaque identification time is shortened, efficiency is significantly improved, and it has high auxiliary value for clinicians' diagnosis. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the implementation of step S1.1 in an embodiment of the present invention;

[0016] Figure 2 This is a schematic diagram illustrating the implementation of step S1.3 in an embodiment of the present invention;

[0017] Figure 3This is a schematic diagram illustrating the implementation of step S3 in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings. It should be noted that this embodiment is based on the present technical solution and provides detailed implementation methods and specific operation processes, but the protection scope of the present invention is not limited to this embodiment.

[0019] This embodiment provides a plaque identification method based on coronary artery CT images. The specific process is as follows:

[0020] S1, such as Figure 1 As shown, 100 high-quality coronary CT images were selected. The center lines of the three major coronary artery branches in each coronary CT image were manually marked. For each vessel in the image, the marked center lines were used to perform surface reconstruction to generate a coronary artery cross-sectional image. Then, the contours of the intima and endothelium of the coronary artery were manually marked in each coronary artery cross-sectional image.

[0021] It should be noted that the three major coronary artery branches include the left anterior descending coronary artery, the left circumflex coronary artery, and the right coronary artery.

[0022] One hundred coronary artery cross-sectional images with manually annotated intima and endothelial contours were randomly divided into a training set and a test set at an 8:2 ratio.

[0023] like Figure 2 As shown, for each coronary artery cross-sectional image, rays are emitted in all directions from the image center as the origin. The distance from the intersection of the ray with the coronary artery intima and endothelium to the origin is used as the gold standard for model prediction. The coronary artery cross-sectional image is input into a convolutional neural network model, and after multi-layer feature extraction, the predicted distances from the intersection of the ray with the coronary artery intima and endothelium at each angle to the origin are output. For training set images, the difference between the prediction results of the convolutional neural network on the training set images and the gold standard is compared, and the network parameters are adjusted using the stochastic gradient descent algorithm to reduce the prediction error of the convolutional neural network model. For test set images, the difference between the prediction results of the convolutional neural network on the test set images and the gold standard is used to measure the performance of the convolutional neural network model. The model constructed in this embodiment has an average error of less than 0.52 mm in the prediction results of the coronary artery intima and endothelium on the test set, with an average processing time of 47.2 seconds per case. The average error of the results manually labeled by two independent observers on the test set is 0.54 mm, with an average processing time of 32.5 minutes per case. Experiments have shown that the model constructed in this embodiment outperforms the inter-observer difference on the test set.

[0024] S2. Using the trained convolutional neural network model, predict the intersection points of the rays emitted from the image center in each direction with the intima and endothelium of the coronary artery from the input coronary artery cross-sectional image, and connect the intersection points sequentially to obtain the prediction results of the intima and endothelium of the coronary artery.

[0025] S3, Patch Recognition: such as Figure 3 As shown, the coronary artery contour can be obtained by connecting the rays predicted by the convolutional neural network model to the intersections of the intima and epicardium of the coronary artery. The intersection point is selected when the area difference between the intima and epicardium is greater than 0.1 mm. 2 Furthermore, the predicted patch area is defined as a section with a length greater than 2 mm along the centerline. Further calculations can be made for parameters such as volume, load, and remodeling index of the patch area.

[0026] For those skilled in the art, various corresponding changes and modifications can be made based on the above technical solutions and concepts, and all such changes and modifications should be included within the protection scope of the claims of this invention.

Claims

1. A plaque recognition method based on a coronary CT image, characterized by, The specific process is as follows: S1. Select a set number of coronary CT images, manually label the center line of the three major coronary artery branches in each coronary CT image, and use the labeled center line to perform curved surface reconstruction to generate coronary cross-sectional images for each blood vessel in the image; then manually label the endocardial and epicardial contours of the coronary artery in each coronary cross-sectional image; The coronary cross-sectional images with manually labeled endocardial and epicardial contours are randomly divided into a training set and a test set according to a set ratio; For each coronary cross-sectional image, a ray is emitted in each direction with the image center as the origin, and the distance from the intersection point of the ray and the endocardial and epicardial contours to the origin is taken as the gold standard for model prediction; the coronary cross-sectional image is input into the convolutional neural network model, and the prediction results of the distance from the intersection point of the ray and the endocardial and epicardial contours to the origin at each angle are output after multi-layer feature extraction; for the training set images, the differences between the prediction results of the convolutional neural network for the training set images and the gold standard are compared, and the network parameters are adjusted using the stochastic gradient descent algorithm to reduce the prediction error of the convolutional neural network model; for the test set images, the differences between the prediction results of the convolutional neural network for the test set images and the gold standard are used to measure the performance of the convolutional neural network model; S2. Using the trained convolutional neural network model, the intersection points of the rays emitted in each direction with the endocardial and epicardial contours of the input coronary cross-sectional image are predicted, and the intersection points are sequentially connected to obtain the prediction results of the endocardial and epicardial contours of the coronary artery. S3, Plaque recognition: the intersection of the predicted ray and the intima and adventitia of the coronary artery can obtain the coronary artery profile. Select the cut-off with the area difference between the intima and adventitia greater than 0.1mm 2 and the length along the center line is greater than 2mm as the plaque prediction result.

2. The method of claim 1, wherein, In step S2, the set ratio is 8:

2.

3. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method of any one of claims 1-2.

4. A computer device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method of any one of claims 1-2. The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method of any one of claims 1-2.

Images