Coronary cta image processing method, device, medium and clinical decision system

Abstract

The application aims to provide a coronary CTA image processing method, device, medium and auxiliary decision-making system. The method comprises the following steps: A, acquiring a real-time coronary CTA image sequence; B, preprocessing the real-time coronary CTA image sequence; C, inputting continuous N layers of real-time images into a pericardium coarse segmentation model to output a real-time pericardium coarse segmentation result of the N / 2th layer of real-time images; D, acquiring a real-time pericardium intermediate segmentation result according to the real-time pericardium coarse segmentation result and real-time heart coordinate information; E, inputting the continuous N layers of real-time images and the real-time pericardium intermediate segmentation result of the N / 2th layer of real-time images into a pericardium fine segmentation model to obtain a real-time pericardium fine segmentation result of the N / 2th layer of real-time images; F, repeating the steps C-E; and G, merging the obtained one or more real-time pericardium fine segmentation results along the hierarchical dimension. The method realizes accurate segmentation of the pericardium boundary.

Description

Technical Field

[0001] This application relates to the field of image processing, and more particularly to a technique for coronary CTA image processing. Background Technology

[0002] Coronary CTA (CTA) is a widely used clinical diagnostic method for cardiovascular diseases due to its advantages such as being non-invasive, safe, and convenient. Numerous clinical studies have shown that the density and volume of pericardial fat based on CTA are related to various cardiovascular diseases, making the calculation of pericardial fat parameters crucial. Because the pericardial tissue structure is complex, the pericardium needs to be segmented before calculating pericardial fat parameters to reduce interference from the lungs, liver, bone tissue, diaphragm, air bubbles, etc., thus facilitating the physician's judgment. Currently, there is a lack of intelligent tools in clinical practice for precise pericardial segmentation to assist in the automatic calculation of pericardial fat parameters. The usual practice is for physicians to manually delineate the pericardial boundaries on hundreds of slice images and calculate the relevant parameters based on the boundary delineation results. This process carries risks of interference from other tissues, fatigue, and is time-consuming and laborious, potentially leading to misdiagnosis and missing the optimal treatment period. Summary of the Invention

[0003] One objective of this application is to provide a coronary CTA image processing method, device, medium, and clinical decision system to address the problem of inaccurate pericardial segmentation results in the prior art.

[0004] According to one aspect of this application, a method for processing coronary CTA images is provided, the method comprising:

[0005] A. Acquire real-time coronary CTA image sequences, wherein the real-time coronary CTA image sequences include multi-slice real-time images;

[0006] B. Preprocess the real-time coronary CTA image sequence;

[0007] C. Input the N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into the pericardial coarse segmentation model to output the real-time pericardial coarse segmentation result of the N / 2th real-time image, where N is a positive integer;

[0008] D. Obtain the real-time pericardial coarse segmentation result of the N / 2 layer real-time image and the real-time heart coordinate information, wherein the real-time heart coordinate information is obtained by inputting the N / 2 layer real-time image into the heart localization model.

[0009] E. Input the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image into the pericardial fine segmentation model to obtain the real-time pericardial fine segmentation results of the N / 2th real-time image.

[0010] F. Repeat steps C-E above to obtain the real-time pericardial fine segmentation results corresponding to each layer of real-time image in the real-time coronary CTA image sequence;

[0011] G. Merge all the obtained real-time pericardial segmentation results along the hierarchical dimension to obtain the three-dimensional pericardial segmentation result corresponding to the real-time coronary CTA image sequence.

[0012] Preferably, step E includes:

[0013] E1. Input the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image into the pericardial fine segmentation model to obtain the real-time pericardial boundary mask;

[0014] E2. Fill the real-time pericardial boundary mask into the corresponding position of the real-time pericardial intermediate segmentation result to obtain the pericardial fine segmentation result of the N / 2 layer real-time image.

[0015] Preferably, the pericardial coarse segmentation model is trained using the following method:

[0016] Acquire multiple historical coronary CTA image sequences, each of which includes multi-layer historical images;

[0017] Historical coronary CTA image sequences were preprocessed and divided into training and testing sets.

[0018] Input the N consecutive layers of historical images from each historical coronary CTA image sequence in the training set into the UNet-based pericardial coarse segmentation model framework to obtain the historical pericardial coarse segmentation result of the N / 2th layer historical image and the trained pericardial coarse segmentation model.

[0019] Preferably, the method further includes:

[0020] The N / 2nd layer historical image is input into the heart localization model to output the historical heart coordinate information corresponding to the N / 2nd layer historical image;

[0021] Based on the historical coarse pericardial segmentation results and historical heart coordinate information, the historical intermediate pericardial segmentation results are obtained. The coordinate information of the historical pericardial pixels in the historical intermediate pericardial segmentation results is within the range of the historical heart coordinate information.

[0022] The pericardial fine segmentation model was trained using multiple historical pericardial intermediate segmentation results and N consecutive layers of historical images corresponding to each historical pericardial intermediate segmentation result.

[0023] Preferably, the pericardial fine segmentation model is trained using the following method:

[0024] Multiple historical pericardial boundary masks are extracted from the pericardial boundaries of the aforementioned historical intermediate pericardial results;

[0025] Based on the historical cardiac coordinate information corresponding to the multiple historical pericardial boundary masks, the original image blocks of the N consecutive historical pericardial boundaries are obtained from the N consecutive historical images.

[0026] The original image blocks of the N consecutive layers of historical pericardial boundaries and the historical pericardial boundary mask are stitched together along the hierarchical dimension to obtain the pericardial boundary image blocks;

[0027] The pericardial boundary image patch is input into the pericardial fine segmentation model composed of ENet. After the UNet network in the pericardial coarse segmentation model framework has trained multiple target parameters, the UNet network and the ENet network are trained simultaneously until the sum of the first parameter loss of the UNet network and the second parameter loss of the ENet network converges to the target threshold, thereby completing the training of the pericardial fine segmentation model.

[0028] Preferably, the method further includes:

[0029] Step 1: Obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on the pericardial fat threshold range;

[0030] Step J: Extract risk biomarkers based on the pericardial fat segmentation results, wherein the risk biomarkers include pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability.

[0031] According to another aspect of this application, a coronary CTA image processing device is provided, the device comprising:

[0032] The CTA image sequence receiving module is used to acquire real-time coronary CTA image sequences, which include multi-layer real-time images.

[0033] The data preprocessing module is used to preprocess real-time coronary artery image sequences;

[0034] The pericardial segmentation module includes a coarse pericardial segmentation unit, an intermediate pericardial segmentation unit, a fine pericardial segmentation unit, and a 3D data synthesis unit. The coarse pericardial segmentation unit inputs N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into a coarse pericardial segmentation model to output the real-time coarse pericardial segmentation result of the N / 2th real-time image, where N is a positive integer. The intermediate pericardial segmentation unit obtains the real-time intermediate pericardial segmentation result of the N / 2th real-time image based on the coarse pericardial segmentation result of the N / 2th real-time image and real-time cardiac coordinate information, where the real-time cardiac coordinate information is obtained by inputting the N / 2th real-time image into a cardiac positioning model. The fine pericardial segmentation unit inputs the N consecutive real-time images and the real-time intermediate pericardial segmentation result of the N / 2th real-time image into a fine pericardial segmentation model to obtain the fine pericardial segmentation result of the N / 2th real-time image. The 3D data synthesis unit synthesizes the fine pericardial segmentation results of all real-time images into a three-dimensional pericardial segmentation result corresponding to the real-time image sequence.

[0035] According to another aspect of this application, an auxiliary decision-making system is provided, including the coronary CTA image processing device as described above, and further comprising:

[0036] The pericardial fat segmentation module is used to obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on the pericardial fat threshold range.

[0037] A cardiovascular disease risk-related biomarker extraction module is used to extract risk biomarkers based on the pericardial fat segmentation results, wherein the risk biomarkers include pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability.

[0038] The clinical decision support module receives the output from the cardiovascular disease risk-related biomarker extraction module, inputs the pericardial fat volume, fat density variability, and radiomics features into the cardiovascular disease risk prediction model built into the clinical decision support module, and outputs high-risk warnings and key high-risk factors for cardiovascular disease.

[0039] According to another aspect of this application, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of any of the methods described above.

[0040] According to one aspect of this application, a computer-readable medium is provided for storing instructions that, when executed, cause a system to perform any of the methods described above.

[0041] Compared with existing technologies, since the pericardial boundary is not visible in all layers and pericardial features are similar in consecutive layers, this application first inputs N consecutive layers of real-time images into a coarse pericardial segmentation model, outputting the real-time coarse pericardial segmentation result of the N / 2th layer. Based on the N layers of real-time images, the coarse pericardial segmentation model predicts the real-time coarse pericardial segmentation result of the N / 2th layer of these N layers of real-time images, solving the problem of difficult segmentation of the pericardium in layers with indistinct boundaries. In addition, since there are many tissues around the pericardium and the contrast between the features of the tissues inside and outside the pericardium is not obvious, direct segmentation of the pericardial region will lead to false positives. This solution further obtains the real-time intermediate pericardial segmentation result based on the real-time coarse pericardial segmentation result and real-time cardiac coordinate information to eliminate the influence of low-level false positives. The real-time cardiac coordinate information is obtained through a cardiac localization model. Furthermore, to prevent the loss of fine features at the pericardial boundary, which would lead to inaccurate boundary segmentation, this solution inputs the N consecutive layers of real-time images and the real-time intermediate pericardial segmentation result into a fine pericardial segmentation model to obtain the real-time fine pericardial segmentation result, accurately segmenting the pericardial boundary and making the pericardial segmentation more refined. Steps C through E are repeated to obtain the real-time pericardial fine segmentation result corresponding to each layer of the real-time coronary CTA image sequence. This allows for the merging of one or more real-time pericardial fine segmentation results along the layer dimension, thereby obtaining the three-dimensional pericardial segmentation result corresponding to the real-time coronary CTA image sequence. Using the method of this application, a three-dimensional pericardial segmentation result is obtained where each layer of the real-time image has undergone fine pericardial segmentation.

[0042] Furthermore, compared with existing technologies, this application also provides a coronary CTA image processing device and an auxiliary decision-making system, providing doctors with intelligent auxiliary tools. The auxiliary decision-making system includes the coronary CTA image processing device, as well as a pericardial fat segmentation module, a cardiovascular disease risk-related biomarker extraction module, and a clinical auxiliary decision-making module. After obtaining a precisely segmented three-dimensional pericardium using the coronary CTA image processing device, the pericardial fat segmentation module obtains the corresponding pericardial fat segmentation result based on the pericardial fat threshold range. The cardiovascular disease risk-related biomarker extraction module extracts risk biomarkers based on the pericardial fat segmentation result. The clinical auxiliary decision-making module inputs pericardial fat volume, fat density variability, and radiomics features into a built-in cardiovascular disease risk prediction model, outputting a high-risk warning for cardiovascular disease and key high-risk factors. Based on precise pericardial segmentation, the auxiliary decision-making system can thus better segment pericardial fat and extract risk biomarkers. Attached Figure Description

[0043] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0044] Figure 1 A flowchart of a method for coronary CTA image processing according to an embodiment of this application is shown;

[0045] Figure 2 shows a schematic diagram of the device structure of a coronary CTA image processing device according to an embodiment of the present application;

[0046] Figure 3 This diagram illustrates the system topology of a clinical decision-making system according to an embodiment of the present application.

[0047] Figure 4 Exemplary systems that can be used to implement the various embodiments described in this application are shown. Detailed Implementation

[0048] The present application will now be described in further detail with reference to the accompanying drawings.

[0049] In a typical configuration of this application, the terminal, the device of the service network, and the trusted party all include one or more processors (e.g., a central processing unit (CPU)), input / output interfaces, network interfaces, and memory.

[0050] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory. Memory is an example of computer-readable media.

[0051] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PCM), programmable random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0052] The devices referred to in this application include, but are not limited to, terminals, network devices, or devices formed by integrating terminals and network devices through a network. The terminals include, but are not limited to, any mobile electronic product capable of human-computer interaction (e.g., via a touchpad), such as smartphones and tablets. These mobile electronic products can use any operating system, such as Android or iOS. The network devices include electronic devices capable of automatically performing numerical calculations and information processing according to pre-set or stored instructions. Their hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), and embedded devices. The network devices include, but are not limited to, computers, network hosts, single network servers, multiple network server clusters, or clouds composed of multiple servers. Here, a cloud consists of a large number of computers or network servers based on cloud computing, where cloud computing is a type of distributed computing, consisting of a virtual supercomputer composed of a group of loosely coupled computer clusters. The network includes, but is not limited to, the Internet, wide area network, metropolitan area network, local area network, VPN network, wireless ad hoc network, etc. Preferably, the device can also be a program running on the terminal, network device, or a device formed by integrating the terminal and network device, network device, touch terminal, or network device and touch terminal through a network.

[0053] Of course, those skilled in the art should understand that the above-described devices are merely examples, and other existing or future devices that are applicable to this application should also be included within the scope of protection of this application, and are hereby incorporated by reference.

[0054] In the description of this application, "multiple" means two or more, unless otherwise expressly and specifically defined.

[0055] Figure 1A method for processing coronary CTA images according to an embodiment of this application is illustrated. The method includes steps A, B, C, D, E, F, and G. In step A, a real-time coronary CTA image sequence is acquired, wherein the real-time coronary CTA image sequence includes multiple real-time images. In step B, the real-time coronary CTA image sequence is preprocessed. In step C, N consecutive real-time images from the preprocessed real-time coronary CTA image sequence are input into a pericardial coarse segmentation model to output a real-time pericardial coarse segmentation result for the N / 2th real-time image, wherein N is a positive integer. In step D, a real-time pericardial intermediate segmentation result for the N / 2th real-time image is obtained based on the real-time pericardial coarse segmentation result of the N / 2th real-time image and real-time cardiac coordinate information. The pericardial coordinate information is obtained by inputting the N / 2th layer real-time image into the cardiac localization model; in step E, the real-time intermediate pericardial segmentation results of the N consecutive layers of real-time images and the N / 2th layer real-time image are input into the pericardial fine segmentation model to obtain the real-time fine pericardial segmentation results of the N / 2th layer real-time image; in step F, steps C-E are repeated to obtain the real-time fine pericardial segmentation results corresponding to each layer of real-time image in the real-time coronary CTA image sequence; in step G, all the obtained real-time fine pericardial segmentation results are merged along the hierarchical dimension to obtain the three-dimensional pericardial segmentation results corresponding to the real-time coronary CTA image sequence.

[0056] Specifically, in step A, a real-time coronary CTA image sequence is acquired, wherein the real-time coronary CTA image sequence includes multiple layers of real-time images. For example, the real-time coronary CTA image sequence is acquired using a CTA acquisition device. In some embodiments, the multiple layers of real-time images in the real-time coronary CTA image sequence are arranged sequentially along the hierarchical dimension.

[0057] In step B, the real-time coronary CTA image sequence is preprocessed. In some embodiments, the preprocessing includes, but is not limited to, normalization. For example, normalization performs a series of standard processing transformations on the multi-layer real-time images in the real-time coronary CTA image sequence, facilitating subsequent processing of the real-time coronary CTA image sequence.

[0058] In step C, the N consecutive real-time images from the preprocessed real-time coronary CTA image sequence are input into the pericardial coarse segmentation model to output the real-time pericardial coarse segmentation result of the N / 2th real-time image, where N is a positive integer. In some embodiments, "inputting the N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into the pericardial coarse segmentation model to output the real-time pericardial coarse segmentation result of the N / 2th real-time image" includes: inputting the N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into the pericardial coarse segmentation model to directly output the real-time pericardial coarse segmentation result through the pericardial coarse segmentation model, and using this real-time pericardial coarse segmentation result as the real-time pericardial coarse segmentation result of the N / 2th real-time image in the N consecutive real-time images. In some embodiments, the real-time pericardial coarse segmentation result includes a real-time pericardial coarse segmentation image, which includes the coarse segmentation result of the pericardium. Those skilled in the art will understand that the pericardial segmentation result includes the segmented pericardial region. For example, the real-time images of layers 1 to 4 of the preprocessed real-time coronary CTA image sequence are input into the pericardial coarse segmentation model, and the real-time coarse segmentation result of the second layer real-time image is output. In some embodiments, N / 2 is rounded up to the nearest integer. For example, the real-time images of layers 1 to 5 of the preprocessed real-time coronary CTA image sequence are input into the pericardial coarse segmentation model, and the real-time coarse segmentation result of the third layer real-time image is output. The real-time coarse segmentation result of the third layer real-time image is the third layer real-time image after pericardial coarse segmentation. In some embodiments, if N is greater than or equal to a target threshold (e.g., 3, 4, etc.), in order to obtain the real-time pericardial coarse segmentation results of the starting layer real-time image and the ending layer real-time image in the continuous N layers of images, 0 pixels are used as padding before the starting layer (layer 1) real-time image and after the ending layer (layer N) real-time image in the continuous N layers of real-time images. The real-time images of the layer before the starting layer real-time image, layer 1, and layer 2 are used as input to the pericardial coarse segmentation model to obtain the real-time pericardial coarse segmentation result of the starting layer real-time image; the real-time pericardial coarse segmentation result of the ending layer real-time image is obtained by using the N-1 layer, layer N, and layer N+1 real-time images as input to the pericardial coarse segmentation model.

[0059] In step D, the real-time pericardial coarse segmentation result of the N / 2 layer real-time image is obtained based on the real-time pericardial coarse segmentation result and the real-time heart coordinate information. The real-time heart coordinate information is obtained by inputting the N / 2 layer real-time image into a heart localization model. In some embodiments, the real-time pericardial coarse segmentation result includes a real-time pericardial region after coarse segmentation, which includes multiple real-time pericardial pixels, each with coordinate information. The N / 2 layer real-time image is input into the heart localization model to obtain a heart region, which includes multiple real-time heart pixels, each with real-time heart coordinate information. The real-time pericardial coarse segmentation result and the real-time heart coordinate information are compared. The pixel values ​​of real-time pericardial pixels in the real-time pericardial coarse segmentation result that are located outside the heart region in the real-time heart coordinate information are set to 0 to obtain the real-time pericardial intermediate segmentation result. The coordinate information of the real-time pericardial pixels in the real-time intermediate segmentation result is all within the range of the real-time heart coordinate information. By setting the pixel value of pericardial pixels located outside the heart region to 0 in this embodiment, most low-level false positives, including images of pericardial tissues such as bone, liver, and lungs, can be eliminated, thus eliminating low-level false positives. In some embodiments, the cardiac localization model includes existing cardiac localization models (e.g., using the cardiac localization model provided at https: / / arxiv.org / abs / 1804.02767).

[0060] In step E, the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image are input into the pericardial fine segmentation model to obtain the real-time pericardial fine segmentation result of the N / 2th real-time image. In some embodiments, the real-time pericardial boundary mask is directly output by the pericardial fine segmentation model. The real-time pericardial boundary mask is filled into the corresponding positions of the real-time pericardial intermediate segmentation results to obtain the real-time pericardial fine segmentation result. For a detailed description of this step, please refer to the corresponding embodiment below, which will not be repeated here.

[0061] In step F, steps C through E are repeated to obtain the real-time pericardial fine segmentation result corresponding to each real-time image layer in the real-time coronary CTA image sequence. For example, the real-time coronary CTA image sequence includes 100 real-time images. The real-time images from layers 1 to 4 are input into the pericardial coarse segmentation model, and the output is the real-time pericardial coarse segmentation result. This real-time pericardial coarse segmentation result is used as the real-time pericardial coarse segmentation result for the second real-time image. Further, using the methods in steps C through E, the real-time pericardial fine segmentation result for the second real-time image is obtained through a cardiac localization model and a pericardial fine segmentation model. The real-time images from layers 1 to 5 (or layers 2 to 4) are input into the pericardial coarse segmentation model, and the output is the real-time pericardial coarse segmentation result. This real-time pericardial coarse segmentation result is used as the real-time pericardial coarse segmentation result for the third real-time image. This process is repeated until the real-time pericardial fine segmentation result for each real-time image layer in the real-time coronary CTA image sequence is obtained.

[0062] In step G, all the obtained real-time pericardial fine segmentation results are merged along the hierarchical dimension to obtain the three-dimensional pericardial segmentation result corresponding to the real-time coronary CTA image sequence. For example, after obtaining the real-time pericardial fine segmentation result of each layer, one or more real-time pericardial fine segmentation results are combined into 3D data to obtain the three-dimensional pericardial segmentation result of the real-time coronary CTA image sequence.

[0063] In some embodiments, step E includes steps E1 (not shown) and E2 (not shown). In step E1, the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image are input into the pericardial fine segmentation model to obtain a real-time pericardial boundary mask. In step E2, the real-time pericardial boundary mask is filled into the corresponding position of the real-time pericardial intermediate segmentation result to obtain the pericardial fine segmentation result of the N / 2th real-time image. In some embodiments, the pericardial fine segmentation model achieves the effect of preserving fine-grained features during pericardial segmentation. For example, the real-time pericardial boundary mask includes the position information (e.g., coordinate position information) of the real-time pericardial boundary mask in the real-time pericardial intermediate segmentation result. Based on this position information, the real-time pericardial boundary mask can be filled into the corresponding position of the real-time pericardial intermediate segmentation result to obtain the real-time pericardial fine segmentation result. In some embodiments, the real-time pericardial boundary mask includes a finely segmented pericardial boundary obtained by the pericardial fine segmentation model. The obtained real-time pericardial boundary mask, including finely segmented pericardial boundaries, is filled into the corresponding position of the real-time pericardial intermediate segmentation result, thereby obtaining a more accurate and refined real-time pericardial fine segmentation result with more precise pericardial boundary contour position.

[0064] In some embodiments, the pericardial coarse segmentation model is trained using the following method: acquiring multiple historical coronary CTA image sequences, each of which includes multiple layers of historical images; preprocessing the historical coronary CTA image sequences and dividing them into training and testing sets; inputting the N consecutive layers of historical images from each historical coronary CTA image sequence in the training set into the UNet-based pericardial coarse segmentation model framework to obtain the historical pericardial coarse segmentation result of the N / 2th layer historical image and the trained pericardial coarse segmentation model. For example, the pericardial coarse segmentation model is trained using multiple historical coronary CTA image sequences. In some embodiments, the preprocessing includes, but is not limited to, normalization processing. Through normalization processing, a series of standard processing transformations are performed on multiple layers of real-time images in the real-time coronary CTA image sequence to facilitate subsequent processing of the real-time coronary CTA image sequence. For example, after preprocessing the historical coronary CTA image sequences, the training and testing sets are randomly divided. In some embodiments, the training and testing sets are divided in a 7:3 ratio. Of course, those skilled in the art will understand that the 7:3 division ratio described above is merely an example, and other existing or future possible division ratios may be applicable to this embodiment and are also within the scope of protection of this application, and are incorporated herein by reference. In some embodiments, the pericardial coarse segmentation model is trained based on a UNet network. For example, a 2D UNet segmentation network is first defined, i.e., a pre-defined pericardial coarse segmentation model framework based on a 2D UNet network is established. Since the pericardial boundary is not visible on all layers of the coronary CTA image sequence, and pericardial features are similar in consecutive layers, the imaging characteristics of the pericardium in coronary CTA images are used to derive features of adjacent layers. Each consecutive N-layer historical image of a historical coronary CTA image sequence in the training set is taken as a 2D image with N channels. This 2D image with N channels is used as the input of the network, and the N / 2th layer is used as the output of the network (for example, the output historical pericardial coarse segmentation result is taken as the historical pericardial coarse segmentation result of the N / 2th layer historical image in the N-layer historical image). Multiple historical pericardial coarse segmentation results and a trained pericardial coarse segmentation model are obtained by training multiple historical coronary CTA image sequences.

[0065] In some embodiments, the method further includes step H (not shown), in which the N / 2th layer historical image is input into the cardiac localization model to output historical cardiac coordinate information corresponding to the historical pericardial coarse segmentation result; historical pericardial intermediate segmentation results are obtained based on the historical pericardial coarse segmentation result and the historical cardiac coordinate information, wherein the coordinate information of the historical pericardial pixels in the historical pericardial intermediate segmentation result is within the range of the historical cardiac coordinate information; and a pericardial fine segmentation model is trained using multiple historical pericardial intermediate segmentation results and N consecutive layers of historical images corresponding to each historical pericardial intermediate segmentation result. In some embodiments, historical pericardial intermediate segmentation results need to be generated during the training of the pericardial fine segmentation model so that the pericardial fine segmentation model can be trained subsequently using the historical pericardial intermediate segmentation results and the corresponding N consecutive layers of historical images. In some embodiments, the historical pericardial coarse segmentation result includes the coordinate information of the historical pericardial region after coarse pericardial segmentation. For example, the historical pericardial region includes multiple historical pericardial pixels, each of which has coordinate information. The N / 2 layer historical image is input into the heart localization model to obtain the heart region. This heart region includes multiple historical heart pixels, and the coordinate information of each historical heart pixel is historical heart coordinate information. The historical pericardial coarse segmentation result is compared with the historical heart coordinate information. The pixel values ​​of historical pericardial pixels located outside the heart region in the historical pericardial coarse segmentation result but within the historical heart coordinate information are set to 0 to obtain the historical pericardial intermediate segmentation result. The coordinate information of the historical pericardial pixels in the historical pericardial intermediate segmentation result is all within the range of the historical heart coordinate information. Through this embodiment, setting the pixel values ​​of historical pericardial pixels located outside the heart region to 0 can eliminate most low-level false positives, including images of pericardial tissues such as bone, liver, and lungs, thus eliminating false positives.

[0066] In some embodiments, the pericardial fine segmentation model is trained using the following method: Multiple historical pericardial boundary masks are extracted from the pericardial boundaries of the historical intermediate pericardial segmentation results; based on the historical cardiac coordinate information corresponding to the multiple historical pericardial boundary masks, original image blocks of the N consecutive historical pericardial boundaries are obtained from the N consecutive historical images; the original image blocks of the N consecutive historical pericardial boundaries and the historical pericardial boundary masks are stitched together along the hierarchical dimension to obtain the pericardial boundary image blocks; the pericardial boundary image blocks are input into the pericardial fine segmentation model composed of ENet. After the UNet network in the pericardial coarse segmentation model framework has trained multiple target parameters, the UNet network and the ENet network are trained simultaneously until the sum of the first parameter loss of the UNet network and the second parameter loss of the ENet network converges to the target threshold, thereby completing the training of the pericardial fine segmentation model. Those skilled in the art will understand that model training can be performed based on ENet networks and UNet networks. For example, https: / / arxiv.org / pdf / 1606.02147.pdf and https: / / arxiv.org / abs / 1505.04597 introduce the ENet and UNet networks, respectively. For instance, firstly, several small historical pericardial boundary mask patches are extracted from the pericardial boundary of the historical pericardial intermediate segmentation result after removing low-level false positives in the N / 2th layer generated in step H. These historical pericardial boundary masks include the pericardial boundary. Then, redundant mask patches are filtered out using NMS (an existing algorithm). The historical cardiac coordinate information (bbox) of the remaining historical pericardial boundary mask patches is mapped to the corresponding N consecutive layers of historical images of the historical pericardial intermediate segmentation result, resulting in N layers of original historical pericardial boundary image patches. The original image patch of the Nth layer of historical pericardial boundary and the mask patch of the N / 2th layer of historical pericardial boundary are concatenated along the channel dimension (e.g., the layer dimension) to obtain a pericardial boundary image patch with N+1 channels. This patch is used as input to the ENet network to train the ENet network and outputs the real pericardial mask corresponding to the mask patch of the N / 2th layer of historical pericardial boundary, i.e., the boundary segmentation result.After the UNet network has been trained for multiple epochs (e.g., the hyperparameters mentioned above), it is trained together with the ENet network until the sum of the first parameter loss (loss1) of the UNet network and the second parameter loss (loss2) of the ENet network converges to the target threshold (the target threshold is within a certain range and can be set as needed, without a fixed value). At this point, the training of the ENet network is complete, generating a trained pericardial fine segmentation model. In this embodiment, the pericardial fine segmentation model is a Multi-BPR (Boundary Patch Refinement) model. Based on the improved Boundary Patch Refinement (Multi-BPR) network, the pericardial boundary is finely predicted, which can improve the accuracy of the model's segmentation at the pericardial boundary. In application, we use the continuous N layers of images from step A and the coarse pericardial segmentation result of layer N / 2 in step B as input to the Multi-BPR network model. The output is the refine mask of the boundary fine segmentation result of layer N / 2 of the pericardial segmentation result. Finally, the refine mask is filled into the corresponding position of the first pericardial segmentation image of layer N / 2 in step B to obtain the real-time pericardial fine segmentation result of layer N / 2. In this embodiment, the lightweight ENet network is used as the base network, which can reduce the time the model spends during inference.

[0067] In some embodiments, the method further includes steps I (not shown) and J (not shown). In step I, the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result is obtained according to the pericardial fat threshold range. In step J, risk biomarkers are extracted based on the pericardial fat segmentation result, wherein the risk biomarkers include pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability. For example, based on the three-dimensional pericardial segmentation result obtained in step G, the pericardial fat segmentation result is obtained using the pericardial fat threshold range (e.g., [-200HU, 0]). Specifically, for example, the threshold range for pericardial fat is [-200HU, 0]. HU values ​​less than -200 in the three-dimensional pericardial segmentation result are set to -200, and values ​​greater than 0 are set to 0, resulting in an image within the threshold range of [-200HU, 0]. The three-dimensional pericardial segmentation result is multiplied by the image within the threshold range to obtain the pericardial fat segmentation result. Further, cardiovascular disease risk-related biomarkers (e.g., the risk biomarkers) are extracted. In some embodiments, the risk markers include, but are not limited to, intracardiac fat volume, intracardiac fat density variability, and intracardiac fat radiomics features. Based on the intracardiac fat segmentation results, the number of fat pixels and CT density values ​​can be statistically determined, thereby calculating the intracardiac fat volume and intracardiac fat density variability. The volume V of fat in three-dimensional space (e.g., the intracardiac fat volume) can be calculated using the formula V = number of pixels * pixelpacing * pixelpacing * slicespacing. Here, pixelpacing is the distance between pixels in medical image data, and slicespacing is the vertical spacing of medical image data in two-dimensional parallel layers. The intracardiac fat density variability can be calculated from the distribution of CT density values. Furthermore, cardiovascular disease risk can be predicted by inputting the intracardiac fat radiomics features, intracardiac fat volume, and intracardiac fat density variability into an existing cardiovascular disease risk prediction model, outputting the cardiovascular disease risk probability and key high-risk factors (e.g., the target high-risk factors). The existing cardiovascular disease risk prediction model is a model already in the art, and it is based on a probabilistic logistic regression model. The risk biomarkers can be used as input to the cardiovascular disease risk prediction model, and the output is the probability of cardiovascular disease risk and the target high-risk factors.

[0068] Figure 2A structural diagram of a coronary CTA image processing device according to an embodiment of this application is shown. The device includes a CTA image sequence receiving module for acquiring a real-time coronary CTA image sequence, the real-time coronary CTA image sequence including multiple real-time images; a data preprocessing module for preprocessing the real-time coronary image sequence; and a pericardial segmentation module, including a coarse pericardial segmentation unit, a mid-level pericardial segmentation unit, a fine pericardial segmentation unit, and a 3D data synthesis unit. The coarse pericardial segmentation unit is used to input N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into a coarse pericardial segmentation model to output the real-time coarse pericardial segmentation result of the N / 2th real-time image, where N is a positive integer; the mid-level pericardial segmentation unit is used to… The system obtains the real-time intermediate pericardial segmentation result of the N / 2 layer real-time image based on the real-time coarse pericardial segmentation result and the real-time heart coordinate information, wherein the real-time heart coordinate information is obtained by inputting the N / 2 layer real-time image into a heart positioning model; the pericardial fine segmentation unit inputs the real-time intermediate pericardial segmentation result of the N consecutive layers of real-time images and the N / 2 layer real-time image into the pericardial fine segmentation model to obtain the pericardial fine segmentation result of the N / 2 layer real-time image; and the 3D data synthesis unit is used to synthesize the pericardial fine segmentation results of all layers of real-time images into a three-dimensional pericardial segmentation result corresponding to the real-time image sequence.

[0069] Here, the specific implementation methods of the CTA image sequence receiving module, data preprocessing module, pericardial coarse segmentation unit, pericardial intermediate segmentation unit, pericardial fine segmentation unit, and 3D data synthesis unit are the same as or similar to the specific embodiments of steps A, B, C, D, E, and F, and therefore will not be repeated here, but are included by reference.

[0070] In some embodiments, the pericardial intermediate segmentation unit includes a first pericardial intermediate segmentation unit (not shown) and a second pericardial intermediate segmentation unit (not shown). The first pericardial intermediate segmentation unit is used to input the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image into the pericardial fine segmentation model to obtain a real-time pericardial boundary mask. The second pericardial intermediate segmentation unit is used to fill the real-time pericardial boundary mask into the corresponding position of the real-time pericardial intermediate segmentation result to obtain the pericardial fine segmentation result of the N / 2th real-time image.

[0071] Here, the specific implementation methods corresponding to the first and second pericardial intermediate segmentation units are the same as or similar to the specific embodiments of steps E1 and E2, and therefore will not be repeated here, but are included by reference.

[0072] In some embodiments, the pericardial coarse segmentation model is trained by the following method: acquiring multiple historical coronary CTA image sequences, each historical coronary CTA image sequence including multiple layers of historical images; preprocessing the historical coronary CTA image sequences to divide them into training and testing sets; inputting the N consecutive layers of historical images from each historical coronary CTA image sequence in the training set into the UNet-based pericardial coarse segmentation model framework to obtain the historical pericardial coarse segmentation result of the N / 2th layer historical image and the trained pericardial coarse segmentation model.

[0073] The training process of the pericardial coarse segmentation model described in this embodiment is the same as or similar to the corresponding specific embodiment described above, and therefore will not be repeated here, but is included by reference.

[0074] In some embodiments, the device further includes an H module (not shown), which is used to input the N / 2th layer historical image into the heart localization model to output historical heart coordinate information corresponding to the historical pericardial coarse segmentation result; obtain historical pericardial intermediate segmentation results based on the historical pericardial coarse segmentation result and the historical heart coordinate information, wherein the historical heart coordinate information is obtained by inputting the N / 2th layer historical image into the heart localization model, and the coordinate information of historical pericardial pixels in the historical pericardial intermediate segmentation result is within the range of the historical heart coordinate information; and train a pericardial fine segmentation model using multiple historical pericardial intermediate segmentation results and N consecutive layers of historical images corresponding to each historical pericardial intermediate segmentation result.

[0075] Here, the specific implementation of the H module is the same as or similar to the specific embodiment of step H, and therefore will not be repeated here, but is included by reference.

[0076] In some embodiments, the pericardial fine segmentation model is trained by the following method: extracting multiple historical pericardial boundary masks from the pericardial boundaries of the historical intermediate pericardial results; obtaining the original image blocks of the N consecutive historical pericardial boundaries from the N consecutive historical images based on the historical cardiac coordinate information corresponding to the multiple historical pericardial boundary masks; stitching the original image blocks of the N consecutive historical pericardial boundaries and the historical pericardial boundary masks along the hierarchical dimension to obtain the pericardial boundary image blocks; inputting the pericardial boundary image blocks into the pericardial fine segmentation model composed of ENet; after the UNet network in the pericardial coarse segmentation model framework has trained multiple target parameters, the UNet network and the ENet network are trained simultaneously until the sum of the first parameter loss of the UNet network and the second parameter loss of the ENet network converges to the target threshold, thereby completing the training of the pericardial fine segmentation model.

[0077] The training process of the pericardial fine segmentation model described in this embodiment is the same as or similar to the corresponding specific embodiment described above, and therefore will not be repeated here, but is included by reference.

[0078] In some embodiments, the device further includes an I module (not shown) and a J module (not shown). The I module is used to obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on the pericardial fat threshold range. The J module is used to extract risk biomarkers based on the pericardial fat segmentation result, wherein the risk biomarkers include pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability.

[0079] Here, the specific implementation methods corresponding to modules I and J are the same as or similar to the specific embodiments of steps I and J, and therefore will not be repeated here, but are included by reference.

[0080] Figure 3 This application illustrates an auxiliary decision-making system according to one embodiment of the present application. The system includes a CTA image sequence receiving module, a data preprocessing module, a pericardial segmentation module, and a 3D data synthesis unit as described in the above embodiment; it also includes a pericardial fat segmentation module, a cardiovascular disease risk-related biomarker extraction module, and a clinical auxiliary decision-making module. The pericardial fat segmentation module is used to obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on a pericardial fat threshold range. The cardiovascular disease risk-related biomarker extraction module is used to extract risk biomarkers based on the pericardial fat segmentation result, wherein the risk biomarkers include pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability. The clinical auxiliary decision-making module receives the output of the cardiovascular disease risk-related biomarker extraction module, inputs the pericardial fat volume, fat density variability, and radiomics features into a cardiovascular disease risk prediction model built into the clinical auxiliary decision-making module, and outputs a high-risk warning for cardiovascular disease and key high-risk factors.

[0081] Specifically, the pericardial fat segmentation module is used to obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on the pericardial fat threshold range.

[0082] The cardiovascular disease risk-related biomarker extraction module is used to extract risk biomarkers based on the pericardial fat segmentation results, wherein the risk biomarkers include pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability.

[0083] The clinical decision support module receives the output of the cardiovascular disease risk-related biomarker extraction module, inputs the pericardial fat volume, fat density variability, and radiomics features into the cardiovascular disease risk prediction model built into the clinical decision support module, and outputs high-risk warnings and key high-risk factors for cardiovascular disease.

[0084] For example, based on the three-dimensional pericardial segmentation result obtained in step G, the pericardial fat segmentation result is obtained using a pericardial fat threshold range (e.g., [-200HU, 0]). Specifically, for example, the threshold range for pericardial fat is [-200HU, 0]. In the three-dimensional pericardial segmentation result, HU values ​​less than -200 are set to -200, and values ​​greater than 0 are set to 0, resulting in an image within the threshold range of [-200HU, 0]. The three-dimensional pericardial segmentation result is multiplied by the image within the threshold range to obtain the pericardial fat segmentation result. Further, cardiovascular disease risk-related biomarkers (e.g., the risk biomarkers) are extracted. In some embodiments, the risk biomarkers include, but are not limited to, pericardial fat volume, pericardial fat density variability, and pericardial fat radiomics features. Based on the pericardial fat segmentation result, the number of fat pixels and CT density value can be statistically determined, thereby calculating the pericardial fat volume and pericardial fat density variability. The volume V of fat in three-dimensional space (e.g., the volume of pericardial fat) can be calculated using the formula V = number of pixels * pixelpacing * pixelpacing * slicespacing. Here, pixelpacing is the distance between pixels in medical image data, and slicespacing is the vertical spacing of medical image data in two-dimensional parallel layers. The density variability of pericardial fat can be calculated from the distribution of CT density values. Furthermore, cardiovascular disease risk can be predicted by inputting the pericardial fat radiomics features, pericardial fat volume, and pericardial fat density variability into an existing cardiovascular disease risk prediction model, outputting the cardiovascular disease risk probability and key high-risk factors (e.g., the target high-risk factor). The existing cardiovascular disease risk prediction model is a model already in the prior art, based on a probabilistic logistic regression model. The risk biomarkers can be used as subsequent inputs into the cardiovascular disease risk prediction model to output the cardiovascular disease risk probability and the target high-risk factor.

[0085] Of course, those skilled in the art will understand that the above description is merely an example, and other existing or future first configuration information that may be applicable to this application is also within the scope of protection of this application and is incorporated herein by reference.

[0086] In addition to the methods and devices described in the above embodiments, this application also provides a computer-readable storage medium storing computer code that, when executed, performs the method described in any of the preceding embodiments.

[0087] This application also provides a computer program product that, when executed by a computer device, performs the method described in any of the preceding claims.

[0088] This application also provides a computer device, the computer device comprising:

[0089] One or more processors;

[0090] Memory, used to store one or more computer programs;

[0091] When the one or more computer programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method as described in any of the preceding methods.

[0092] Figure 4 Exemplary systems that can be used to implement the various embodiments described in this application are shown;

[0093] like Figure 4 As shown in some embodiments, system 300 can function as any of the devices described in each of the embodiments. In some embodiments, system 300 may include one or more computer-readable media having instructions (e.g., system memory or NVM / storage device 320) and one or more processors (e.g., one or more processors 305) coupled to the one or more computer-readable media and configured to execute the instructions to implement the module and thus perform the actions described in this application.

[0094] In one embodiment, the system control module 310 may include any suitable interface controller to provide any suitable interface to at least one of the processors 305 and / or any suitable device or component communicating with the system control module 310.

[0095] The system control module 310 may include a memory controller module 330 to provide an interface to the system memory 315. The memory controller module 330 may be a hardware module, a software module, and / or a firmware module.

[0096] System memory 315 can be used, for example, to load and store data and / or instructions for system 300. In one embodiment, system memory 315 may include any suitable volatile memory, such as suitable DRAM. In some embodiments, system memory 315 may include double data rate type quad synchronous dynamic random access memory (DDR4 SDRAM).

[0097] In one embodiment, the system control module 310 may include one or more input / output (I / O) controllers to provide interfaces to the NVM / storage device 320 and (one or more) communication interfaces 325.

[0098] For example, NVM / storage device 320 may be used to store data and / or instructions. NVM / storage device 320 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (one or more) non-volatile storage devices (e.g., one or more hard disk drives (HDDs), one or more optical disc drives (CDs), and / or one or more digital universal optical disc (DVD) drives).

[0099] NVM / storage device 320 may include storage resources that are physically part of a device on which system 300 is mounted, or that can be accessed by the device without necessarily being part of it. For example, NVM / storage device 320 may be accessed via a network through one or more communication interfaces 325.

[0100] One or more communication interfaces 325 may provide the system 300 with an interface to communicate over one or more networks and / or with any other suitable device. The system 300 may wirelessly communicate with one or more components of a wireless network in accordance with any of one or more wireless network standards and / or protocols.

[0101] In one embodiment, at least one of the processors 305 may be logically packaged with one or more controllers of the system control module 310 (e.g., memory controller module 330). In one embodiment, at least one of the processors 305 may be logically packaged with one or more controllers of the system control module 310 to form a system-in-package (SiP). In one embodiment, at least one of the processors 305 may be integrated with the logic of one or more controllers of the system control module 310 on the same die. In one embodiment, at least one of the processors 305 may be integrated with the logic of one or more controllers of the system control module 310 on the same die to form a system-on-a-chip (SoC).

[0102] In various embodiments, system 300 may be, but is not limited to, a server, workstation, desktop computing device, or mobile computing device (e.g., laptop computing device, handheld computing device, tablet computer, netbook, etc.). In various embodiments, system 300 may have more or fewer components and / or different architectures. For example, in some embodiments, system 300 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0103] It should be noted that this application can be implemented in software and / or a combination of software and hardware, for example, using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In one embodiment, the software program of this application can be executed by a processor to implement the steps or functions described above. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, a magnetic or optical drive, a floppy disk, or similar devices. Furthermore, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0104] Furthermore, a portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0105] Communication media include media through which communication signals containing, for example, computer-readable instructions, data structures, program modules, or other data are transmitted from one system to another. Communication media can include guided transmission media (such as cables and wires (e.g., optical fibers, coaxial cables, etc.)) and wireless (unguided transmission) media capable of propagating energy waves, such as sound, electromagnetic, RF, microwave, and infrared. Computer-readable instructions, data structures, program modules, or other data can be embodied as modulated data signals in, for example, wireless media (such as carrier waves or similar mechanisms embodied as part of spread spectrum technology). The term "modulated data signal" refers to a signal whose one or more characteristics are altered or set in a manner that encodes information in the signal. Modulation can be analog, digital, or a hybrid modulation technique.

[0106] By way of example and not limitation, computer-readable storage media may include volatile and non-volatile, removable and non-removable media implemented by any method or technique for storing information such as computer-readable instructions, data structures, program modules or other data. For example, computer-readable storage media include, but are not limited to, volatile memories such as random access memory (RAM, DRAM, SRAM); and non-volatile memories such as flash memory, various read-only memories (ROM, PROM, EPROM, EEPROM), magnetic and ferromagnetic / ferroelectric memories (MRAM, FeRAM); and magnetic and optical storage devices (hard disks, magnetic tapes, CDs, DVDs); or other media now known or hereafter developed capable of storing computer-readable information / data for use by a computer system.

[0107] Herein, one embodiment of this application includes an apparatus comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the apparatus is triggered to run a method and / or technical solution based on the foregoing embodiments of this application.

[0108] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within this application. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in the apparatus claims may also be implemented by a single unit or device in software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.

Claims

1. A method for processing coronary CTA images, characterized in that, The method includes: A. Acquire real-time coronary CTA image sequences, wherein the real-time coronary CTA image sequences include multi-slice real-time images; B. Preprocess the real-time coronary CTA image sequence; C. Input the N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into the pericardial coarse segmentation model to output the real-time pericardial coarse segmentation result of the N / 2th real-time image, where N is a positive integer; D. Obtain the real-time pericardial coarse segmentation result of the N / 2 layer real-time image and the real-time heart coordinate information, wherein the real-time heart coordinate information is obtained by inputting the N / 2 layer real-time image into the heart localization model. E. Input the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image into the pericardial fine segmentation model to obtain the real-time pericardial fine segmentation results of the N / 2th real-time image. F. Repeat steps C-E above to obtain the real-time pericardial fine segmentation results corresponding to each layer of real-time image in the real-time coronary CTA image sequence; G. Merge all the obtained real-time pericardial segmentation results along the hierarchical dimension to obtain the three-dimensional pericardial segmentation result corresponding to the real-time coronary CTA image sequence.

2. The method according to claim 1, characterized in that, Step E includes: E1. Input the real-time pericardial intermediate segmentation results of the N consecutive real-time images and the N / 2th real-time image into the pericardial fine segmentation model to obtain the real-time pericardial boundary mask; E2. Fill the real-time pericardial boundary mask into the corresponding position of the real-time pericardial intermediate segmentation result to obtain the pericardial fine segmentation result of the N / 2 layer real-time image.

3. The method according to claim 1, characterized in that, The pericardial coarse segmentation model was trained using the following method: Acquire multiple historical coronary CTA image sequences, each of which includes multi-layer historical images; Historical coronary CTA image sequences were preprocessed and divided into training and testing sets. Input the N consecutive layers of historical images from each historical coronary CTA image sequence in the training set into the UNet-based pericardial coarse segmentation model framework to obtain the historical pericardial coarse segmentation result of the N / 2th layer historical image and the trained pericardial coarse segmentation model.

4. The method according to claim 3, characterized in that, The method further includes: The N / 2nd layer historical image is input into the heart localization model to output the historical heart coordinate information corresponding to the N / 2nd layer historical image; Based on the historical coarse pericardial segmentation results and historical heart coordinate information, the historical intermediate pericardial segmentation results are obtained. The coordinate information of the historical pericardial pixels in the historical intermediate pericardial segmentation results is within the range of the historical heart coordinate information. The pericardial fine segmentation model was trained using multiple historical pericardial intermediate segmentation results and N consecutive layers of historical images corresponding to each historical pericardial intermediate segmentation result.

5. The method according to claim 4, characterized in that, The pericardial fine segmentation model was trained using the following method: Multiple historical pericardial boundary masks are extracted from the pericardial boundary of the historical pericardial intermediate segmentation result; Based on the historical cardiac coordinate information corresponding to the multiple historical pericardial boundary masks, the original image blocks of the N consecutive historical pericardial boundaries are obtained from the N consecutive historical images. The original image blocks of the N consecutive layers of historical pericardial boundaries and the historical pericardial boundary mask are stitched together along the hierarchical dimension to obtain the pericardial boundary image blocks; The pericardial boundary image patch is input into the pericardial fine segmentation model composed of ENet. After the UNet network in the pericardial coarse segmentation model framework has trained multiple target parameters, the UNet network and the ENet network are trained simultaneously until the sum of the first parameter loss of the UNet network and the second parameter loss of the ENet network converges to the target threshold, thereby completing the training of the pericardial fine segmentation model.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Step 1: Obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on the pericardial fat threshold range; Step J: Extract risk biomarkers based on the pericardial fat segmentation results, wherein the risk biomarkers include pericardial fat radiomics features, which include pericardial fat volume and pericardial fat density variability.

7. A coronary CTA image processing device, characterized in that, The device includes: The CTA image sequence receiving module is used to acquire real-time coronary CTA image sequences, which include multi-layer real-time images. The data preprocessing module is used to preprocess real-time coronary CTA image sequences; The pericardial segmentation module includes a coarse pericardial segmentation unit, a medium pericardial segmentation unit, a fine pericardial segmentation unit, and a 3D data synthesis unit. The pericardial coarse segmentation unit is used to input N consecutive real-time images from the preprocessed real-time coronary CTA image sequence into the pericardial coarse segmentation model to output the real-time pericardial coarse segmentation result of the N / 2th real-time image, where N is a positive integer. The pericardial intermediate segmentation unit is used to obtain the real-time pericardial intermediate segmentation result of the N / 2 layer real-time image based on the real-time pericardial coarse segmentation result of the N / 2 layer real-time image and the real-time heart coordinate information, wherein the real-time heart coordinate information is obtained by inputting the N / 2 layer real-time image into the heart localization model. The pericardial fine segmentation unit inputs the real-time pericardial intermediate segmentation results of the continuous N layers of real-time images and the N / 2th layer of real-time images into the pericardial fine segmentation model to obtain the pericardial fine segmentation results of the N / 2th layer of real-time images; The 3D data synthesis unit is used to synthesize the pericardial fine segmentation results of all layers of real-time images into the three-dimensional pericardial segmentation results corresponding to the real-time coronary CTA image sequence.

8. A decision support system, characterized in that, The coronary CTA image processing device as described in claim 7 further includes: The pericardial fat segmentation module is used to obtain the pericardial fat segmentation result corresponding to the three-dimensional pericardial segmentation result based on the pericardial fat threshold range. A cardiovascular disease risk-related biomarker extraction module is used to extract risk biomarkers based on the pericardial fat segmentation results, wherein the risk biomarkers include pericardial fat radiomics features, and the pericardial fat radiomics features include pericardial fat volume and pericardial fat density variability. The clinical decision support module receives the output from the cardiovascular disease risk-related biomarker extraction module, inputs the pericardial fat volume, fat density variability, and radiomics features into the cardiovascular disease risk prediction model built into the clinical decision support module, and outputs high-risk warnings and key high-risk factors for cardiovascular disease.

9. A computer device for coronary CTA image processing, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 6.

Images