Vulnerable plaque identification method based on IVUS images, computer equipment, readable storage medium and program product

Through the training of IVUS images and OCT images registration and deep learning models, TCFA in IVUS images is automatically recognized, which solves the problem of low-resolution recognition of thin fiber caps in IVUS images, and realizes automatic TCFA recognition in the absence of OCT, reducing the number of TCFAs Diagnosis time.

CN114387464BActive Publication Date: 2024-11-08HANGZHOU ARTERYFLOW TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111453768.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-01
Publication Date
2024-11-08
Estimated Expiration
2041-12-01

AI Technical Summary

Technical Problem

In the prior art, the low resolution of IVUS images is difficult to directly identify thin fiber caps, making it difficult to accurately identify vulnerable plaque TCFA, and the vascular contour cannot be confirmed by OCT images alone.

Method used

Through registration of IVUS images and OCT images, classification labels are artificially marked, deep learning models are trained to automatically intercept the boundaries of the lumen and external elastic membrane, and use CNN or integrated learning models to automatically identify TCFA.

Benefits of technology

In the absence of OCT diagnosis, dangerous TCFA lesions can be automatically identified, effectively reducing the doctor's diagnosis time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114387464B_ABST
    Figure CN114387464B_ABST
Patent Text Reader

Abstract

The present application discloses a vulnerable plaque identification method based on IVUS images, a computer device, a readable storage medium and a program product, wherein the vulnerable plaque identification method based on IVUS images comprises: obtaining IVUS images and OCT images based on coronary arteries, and registering the IVUS images with the OCT images; manually annotating the IVUS images according to the corresponding OCT images to obtain classification labels of the IVUS images, wherein the classification labels include a normal class and a TCFA class; depicting the lumen boundary and the external elastic membrane boundary on the IVUS images to obtain a mask image; training a first deep learning model according to the IVUS images and the mask images to obtain a trained second deep learning model, wherein the second deep learning model is used to automatically intercept the lumen boundary and / or the external elastic membrane boundary and obtain the boundary of a preliminary ROI; and training a classifier by a deep learning algorithm or a machine learning algorithm, wherein the classifier automatically identifies TCFA on the IVUS images with the preliminary ROI boundary after training.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of medical engineering, and in particular to a vulnerable plaque identification method based on IVUS images, a computer device, a readable storage medium and a program product. Background Art

[0002] Acute coronary syndrome (ACS) is often associated with coronary thrombosis, mainly caused by plaque rupture (55-60%), plaque erosion (30-35%) and calcified nodules (2-7%). Patients with ACS are more likely to experience unstable angina, acute myocardial infarction and sudden death from coronary heart disease. The lesion that is most similar to plaque rupture in morphology, that is, the most common type of vulnerable plaque is thin-cap fibroatherosclerosis (TCFA), which has a necrotic core and a fibrous cap less than 65μm, infiltrated by a large number of macrophages. In order to observe this vulnerable plaque in the coronary artery, intravascular ultrasound (IVUS) and optical coherence tomography (OCT) are usually used.

[0003] IVUS provides a tomographic assessment of the lumen area and vessel wall condition. In addition, IVUS also provides plaque size, distribution, and composition. However, since the axial and lateral resolution of IUVS images exceeds 150μm, it is difficult to directly identify thin fibrous caps. Therefore, doctors mainly use OCT to identify TCFA lesions, because OCT has a spatial resolution of less than 16μm, lipid-rich plaques and necrotic cores can be seen, and it is the gold standard for identifying TCFA lesions. However, it is difficult to see the vessel wall through OCT images alone, so the contour of the vessel cannot be confirmed. Summary of the invention

[0004] Based on this, it is necessary to provide a vulnerable plaque identification method based on IVUS images to address the above technical issues.

[0005] The present application provides a vulnerable plaque identification method based on IVUS images, including:

[0006] Obtaining IVUS (intravascular ultrasound) images and OCT (optical coherence tomography) images based on the coronary arteries, and registering the IVUS images with the OCT images;

[0007] Manually annotate the IVUS images according to the corresponding OCT images to obtain classification labels of the IVUS images, wherein the classification labels include a normal class and a TCFA (thin cap fibroatherosclerosis) class;

[0008] Describing the lumen boundary and the external elastic membrane boundary on the IVUS image to obtain a mask image;

[0009] Training a first deep learning model according to the IVUS image and the mask image to obtain a trained second deep learning model, wherein the second deep learning model is used to automatically intercept the lumen boundary and / or the external elastic membrane boundary and obtain the boundary of a preliminary ROI (region of interest);

[0010] The classifier is trained by a deep learning algorithm or a machine learning algorithm, and after the training is completed, the classifier automatically recognizes TCFA on IVUS images with preliminary ROI boundaries.

[0011] Optionally, the classifier is a CNN (convolutional neural network) model, and the CNN model is trained using IVUS images with the preliminary ROI boundary and classification labels;

[0012] After training, the CNN model performs TCFA recognition and classification on the IVUS image with the preliminary ROI boundary to obtain the classification label.

[0013] Optionally, after training, the CNN model performs TCFA recognition and classification on the IVUS image with the preliminary ROI boundary, specifically including:

[0014] After training, the CNN model iteratively processes the convolution layer and the pooling layer on the IVUS image with the preliminary ROI boundary to extract the feature map, outputs the probabilities of the normal class and the TCFA class respectively through the fully connected layer, and selects the class with the highest probability as the classification result.

[0015] Optionally, training the CNN model using IVUS images with the preliminary ROI boundary and classification labels further includes:

[0016] During the training process, weighted binary cross entropy is used as the loss function, and the weights of samples with TCFA are increased.

[0017] Optionally, training the CNN model using IVUS images with the preliminary ROI boundary and classification labels further includes:

[0018] Before training, data augmentation was performed on IVUS images with the classification label of TCFA.

[0019] Optionally, the area between the lumen boundary and the external elastic membrane boundary on the IVUS image is further divided into a plurality of boundary areas;

[0020] According to the ratio of any two boundary areas, feature extraction is performed on IVUS images with multiple boundary areas;

[0021] Using the extracted features to train the classifier and the classification label, the classifier is an integrated learning model;

[0022] After the ensemble learning model is trained, TCFA recognition and classification are performed on IVUS images with multiple boundary areas to obtain the classification labels.

[0023] Optionally, training the classifier using the extracted features and the classification labels specifically includes:

[0024] The chi-square test is used to select relevant features, and the classifier is trained using the selected features and the classification labels.

[0025] The present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the vulnerable plaque identification method based on IVUS images described in the present application.

[0026] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the vulnerable plaque identification method based on IVUS images described in the present application are implemented.

[0027] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the vulnerable plaque identification method based on IVUS images described in the present application.

[0028] The vulnerable plaque identification method based on IVUS images in this application has at least the following effects:

[0029] This embodiment uses the trained second deep learning model and the trained classifier to realize the automatic recognition of TCFA on IVUS images, so that dangerous TCFA lesions can be automatically identified even if the patient has not undergone OCT diagnosis, effectively reducing the doctor's diagnosis time. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of a vulnerable plaque identification method based on IVUS images in one embodiment of the present application;

[0031] Figure 2 This is a schematic diagram of an IVUS image marking the lumen boundary and the EEM boundary in one embodiment of the present application;

[0032] Figure 3 is a schematic diagram of a mask image in an embodiment of the present application;

[0033] Figure 4 This is a schematic diagram of an IVUS image with multiple boundary areas in one embodiment of the present application;

[0034] Figure 5It is a structural block diagram of a computer device in one embodiment of the present application. DETAILED DESCRIPTION

[0035] In the prior art, due to the limitation of resolution, it is difficult for doctors to directly identify TCFA (thin-cap fibroatherosclerotic plaque) on IVUS (intravascular ultrasound) images. Common vulnerable plaque TCFA (thin-cap fibroatherosclerotic plaque) is observed by using IVUS technology and OCT (optical coherence tomography) technology. In the prior art, the axial and lateral resolution of IUVS images is greater than that of OCT images. It is difficult to directly identify TCFA thin fibrous caps only through IUVS images, and the contour of blood vessels cannot be confirmed only through OCT images.

[0036] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0037] See also Figure 1 to Figure 3 , an embodiment of the present application provides a vulnerable plaque identification method based on IVUS images, comprising steps S100 to S500. Among them:

[0038] Step S100, obtaining IVUS images and OCT images based on the coronary arteries, and registering the IVUS images and the OCT images.

[0039] In step S100, images of patients with stable or unstable angina pectoris who have received both IVUS and OCT diagnostic techniques are first collected. Secondly, within the target segment of the image, each OCT image frame with an interval of 0.4 mm is co-registered with its comparable IVUS image frame (approximately every 24 IVUS image frames) by using anatomical landmarks such as vessel shape, side branches, calcium, perivascular structures, and distance to the opening. It can be understood that the ratio of the number of OCT image frames with the same interval to its comparable IVUS image frame is allowed to fluctuate due to the difference in spatial resolution.

[0040] Step S200: manually annotate the IVUS image according to the corresponding OCT image to obtain a classification label of the IVUS image.

[0041] In step S200, each IVUS image frame is annotated according to whether the corresponding OCT image frame has TCFA. The classification labels of IVUS images include normal class (class without TCFA) and TCFA class. The TCFA class (class with TCFA) can also be recorded as "OCT-TCFA". For example, the OCT image is analyzed by an radiologist to obtain an accurately annotated IVUS image. Annotating IVUS images by analyzing OCT images can improve the accuracy of the annotator's annotation of IVUS images, thereby ensuring the accuracy of the samples in each step.

[0042] Step S300, depicting the lumen boundary and the external elastic membrane boundary on the IVUS image to obtain a mask image.

[0043] In step S300, IVUS images of patients with coronary artery disease are collected. Depending on the length of the lesion, one case usually has 2000 to 3500 frames. A specialized radiologist then delineates the lumen boundary and the external elastic membrane (EEM) boundary on the IVUS image to obtain a mask image.

[0044] See also Figure 2 The white arrows in the figure point to the lumen boundary, and the white triangles point to the EEM boundary. The white arrows and white triangles in the figure are only used to illustrate the IVUS images.

[0045] See also Figure 3 In the figure, "2:lumen" refers to the lumen boundary and the area within the lumen boundary; "1:EEM" refers to the area between the lumen boundary and the EEM boundary; "0:_background" refers to the area outside the EEM boundary. As shown in the figure, the media layer contains smooth cells, which appear as dark, low-echo, annular areas in the IVUS image. As can be seen in the figure, the mask image clearly demarcates the lumen boundary and the external elastic membrane boundary. In the mask image, the plaque is located in the area within the EEM boundary.

[0046] Step S400, training a first deep learning model according to the IVUS image and the mask image to obtain a trained second deep learning model, and the second deep learning model is used to automatically capture the lumen boundary and / or the external elastic membrane boundary and obtain the boundary of the preliminary ROI.

[0047] In step S400, different from the medical image segmentation model with CNN as the core in the prior art, the first deep learning model in this embodiment is a Swin-Unet structure with Swin Transformer as the core. Swin Transformer combines the advantages of convolutional neural network (CNN) and ViT model (Vision Transformer). It can not only generate hierarchical feature maps for segmentation or detection tasks like CNN, but also solve the problem of excessive computational complexity of ViT model. Specifically, as a 2D image segmentation model, the first deep learning model adopts a symmetric encoder-decoder U-type network with jump connections. The U-type network is a typical structure in the field of medical image segmentation. It can upsample the global features extracted from Swin Transformer to pixel-level predictions of the same size as the input data, effectively solving the problem of insufficient learning of global information due to the locality of CNN convolution operations.

[0048] In step S400, after the training is completed, the second deep learning model receives the IVUS image and can automatically capture the ROI (Region of Interest) boundary, that is, automatically segment the lumen area and / or the area within the EEM to obtain an IVUS image with the ROI boundary. It can be understood that after the first deep learning model is trained, it is converted into the second deep learning model.

[0049] Step S500: training a classifier by a deep learning algorithm or a machine learning algorithm. After the training is completed, the classifier automatically recognizes TCFA on the IVUS image with the preliminary ROI boundary.

[0050] It can be understood that the purpose of automatic identification is to automatically identify vulnerable plaques in low-resolution IVUS images. This embodiment uses an AI algorithm (including a trained second deep learning model and a trained classifier) ​​to achieve automatic identification of TCFA on IVUS images, so that dangerous TCFA lesions can be automatically identified even if the patient has not undergone OCT diagnosis, effectively reducing the doctor's diagnosis time.

[0051] In one embodiment, step S500 is performed by a deep learning algorithm, and step S500 adopts steps S510 to S512, wherein:

[0052] Step S510: Before training, data enhancement processing is performed on IVUS images with classification labels of TCFA.

[0053] In the field of medical engineering, most of the input data are negative samples without TCFA, while there are relatively few positive samples with TCFA. In this case, the classifier trained with the data set is prone to high specificity and low sensitivity. Therefore, when using the CNN model as a classifier, data enhancement processing is performed on the images of positive samples (TCFA class exists) in the training data, which can effectively reduce problems such as distribution imbalance and model overfitting. Data enhancement processing can, for example, randomly rotate the image several times at a rotation angle of 30° each time, and the number of times can be, for example, 1 to 3 times.

[0054] In step S511, the classifier is a CNN (convolutional neural network) model, and the CNN model is trained using IVUS images with preliminary ROI boundaries and classification labels.

[0055] Furthermore, in step S511, it also includes using weighted binary cross entropy as a loss function during the training process, and increasing the weight of samples with TCFA. The loss function is Weighted Cross Entropy, and the formula is as follows:

[0056]

[0057] in,

[0058] WCE is weighted cross entropy;

[0059] β is the weight of the custom positive sample (TCFA class exists);

[0060] p represents the actual value of the probability that the sample is a positive sample category;

[0061] (1-p) represents the probability that the sample belongs to the negative sample category (TCFA category does not exist);

[0062] Represents the predicted value of the probability that the sample is a positive sample class.

[0063] When β>1, the proportion of False Negative (normal samples without TCFA) can be reduced, and when β<1, the proportion of False Positive (samples with TCFA) can be reduced. For normal samples and samples with TCFA, the larger the relative weight value, the lower the misclassification rate. In this embodiment, the value range of β can be, for example, 0<β<2.

[0064] It can be understood that the loss function calculates the cross entropy of each pixel separately and then averages all pixels, which means that by default, each type of pixel contributes equally to the loss. However, if the number of each type of pixel in the image is unbalanced, for example, the number of normal class samples in this embodiment is greater than the number of TCFA class samples, it will affect the loss function, thereby dominating the training process and affecting the training effect. In other words, increasing the weight of samples with TCFA class alleviates the problem of sample imbalance.

[0065] In step S512, after training, the CNN model performs TCFA recognition and classification on the IVUS images with preliminary ROI boundaries.

[0066] Specifically, after training, the CNN model iteratively processes the convolution layer and the pooling layer on the IVUS image with the preliminary ROI boundary to extract the feature map, and outputs the probabilities of the normal class and the TCFA class respectively through the fully connected layer, and selects the class with the highest probability as the classification result. The classification result obtains the classification label (normal class and TCFA class). Different from the manual labeling method before training, the CNN model after training in this embodiment can automatically identify and classify to obtain the classification label.

[0067] In steps S510 to S512, after the IVUS image is initially intercepted by ROI, the CNN model is used for classification to determine whether TCFA exists in the area. Currently, the CNN model is an advanced deep learning model in the field of pattern recognition. In terms of image classification, the CNN model can be used, for example, MobileNet, ResNet, and EfficientNet. This embodiment deals with the problem of sample imbalance, and uses data enhancement and a custom loss function to solve the impact of too many normal class samples without TCFA on model training.

[0068] In one embodiment, step S500 is performed by a machine learning algorithm, and step S500 adopts steps S520 to S5213. Among them:

[0069] Step S520: further segment the area between the lumen boundary and the external elastic membrane boundary on the IVUS image into a plurality of boundary areas.

[0070] See also Figure 4 In step S520, segmenting into multiple boundary regions refers to performing more accurate image region segmentation, and the basis for segmentation is the distance from the lumen boundary.

[0071] Specifically, it is divided into five regions S1 to S5, including regions S1, S2, S3, S4 and S5. Among them, S1 is defined as a region with a thickness of 2 pixels (32.2 μm) adjacent to the lumen boundary, S2 is defined as a region with a thickness of 4 pixels (64.4 μm) away from the S1 boundary, S3 is defined as a region with a thickness of 7 pixels (109.2 μm) away from the S2 boundary, S4 is defined as a region with a thickness of 13 pixels (202.8 μm) away from the S3 boundary, and S5 is defined as the remaining region adjacent to the external elastic membrane boundary.

[0072] Step S521, performing feature extraction on the IVUS image with multiple boundary regions according to the ratio of any two boundary regions.

[0073] Specifically, feature extraction was performed in the training data, and the extracted IVUS image features were generated after filtering the scattered and noisy images (using a 4×4 median filter). Specifically, the extracted IVUS image features included: (1) 2D geometry-based features, including plaque burden, minimum lumen diameter, lumen area, and lumen eccentricity; (2) probability distribution parameters, including gamma-beta, Nakagami-μ, Nakagami-ω, and Rayleigh-b; (3) texture-based features, including variance, moment, smoothness, kurtosis, entropy, and homogeneity.

[0074] Furthermore, the above features are calculated for the whole of multiple boundary regions and each region in S1 to S5. Including the ratio of features calculated between two different regions (S1:S2, S1:S3, S1:S4, S1:S5, S2:S3, S2:S4, S2:S5, S3:S4, S3:S5 and S4:S5), a total of 1449 features were extracted. The feature categories are shown in the following table:

[0075]

[0076]

[0077] Step S522, using the extracted feature training and classification label classifier, the classifier is an integrated learning model. Specifically, in step S522, the chi-square test is used to select the relevant features (the number is N) shown in the above table. It can be understood that the value of N is determined by the result of the chi-square test, and all features with a result probability P value <0.05 obtained by the chi-square test are taken (P>0.05 represents no statistical significance). The classifier is trained using the features selected by the chi-square test and the corresponding classification labels (artificially marked classification labels). In this embodiment, the classifier is an integrated learning model XGBoost. When the classifier is trained, a five-fold cross-validation is also included to prevent the model from overfitting.

[0078] Step S523, after the ensemble learning model is trained, TCFA recognition and classification are performed on the IVUS images with multiple boundary areas to obtain classification labels. Different from the manual labeling method before training, the ensemble learning model XGBoost after training in this embodiment can automatically recognize and classify to obtain classification labels.

[0079] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0080] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a vulnerable plaque identification method based on IVUS images is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0081] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0082] Step S100, obtaining IVUS images and OCT images based on the coronary arteries, and registering the IVUS images with the OCT images;

[0083] Step S200, manually labeling the IVUS image according to the corresponding OCT image to obtain a classification label of the IVUS image;

[0084] Step S300, depicting the lumen boundary and the external elastic membrane boundary on the IVUS image to obtain a mask image;

[0085] Step S400, training a first deep learning model according to the IVUS image and the mask image to obtain a trained second deep learning model, wherein the second deep learning model is used to automatically intercept the lumen boundary and / or the external elastic membrane boundary and obtain the boundary of the preliminary ROI;

[0086] Step S500: training a classifier by a deep learning algorithm or a machine learning algorithm. After the training is completed, the classifier automatically recognizes TCFA on the IVUS image with the preliminary ROI boundary.

[0087] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0088] Step S100, obtaining IVUS images and OCT images based on the coronary arteries, and registering the IVUS images with the OCT images;

[0089] Step S200, manually labeling the IVUS image according to the corresponding OCT image to obtain a classification label of the IVUS image;

[0090] Step S300, depicting the lumen boundary and the external elastic membrane boundary on the IVUS image to obtain a mask image;

[0091] Step S400, training a first deep learning model according to the IVUS image and the mask image to obtain a trained second deep learning model, wherein the second deep learning model is used to automatically intercept the lumen boundary and / or the external elastic membrane boundary and obtain the boundary of the preliminary ROI;

[0092] Step S500: training a classifier by a deep learning algorithm or a machine learning algorithm. After the training is completed, the classifier automatically recognizes TCFA on the IVUS image with the preliminary ROI boundary.

[0093] In one embodiment, a computer program product is provided, comprising computer instructions, which when executed by a processor implement the following steps:

[0094] Step S100, obtaining IVUS images and OCT images based on the coronary arteries, and registering the IVUS images with the OCT images;

[0095] Step S200, manually labeling the IVUS image according to the corresponding OCT image to obtain a classification label of the IVUS image;

[0096] Step S300, depicting the lumen boundary and the external elastic membrane boundary on the IVUS image to obtain a mask image;

[0097] Step S400, training a first deep learning model according to the IVUS image and the mask image to obtain a trained second deep learning model, wherein the second deep learning model is used to automatically intercept the lumen boundary and / or the external elastic membrane boundary and obtain the boundary of the preliminary ROI;

[0098] Step S500: training a classifier by a deep learning algorithm or a machine learning algorithm. After the training is completed, the classifier automatically recognizes TCFA on the IVUS image with the preliminary ROI boundary.

[0099] In this embodiment, the computer program product includes a program code portion for executing the steps of the vulnerable plaque identification method based on IVUS images in each embodiment of the present application when the computer program product is executed by one or more computing devices. The computer program product can be stored on a computer-readable recording medium. The computer program product can also be provided via a data network (e.g., via a RAN, via the Internet and / or via an RBS) for downloading. Alternatively or additionally, the method can be encoded in a field programmable gate array (FPGA) and / or an application specific integrated circuit (ASIC), or the functionality can be provided for downloading with the aid of a hardware description language.

[0100] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0101] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0102] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. When the technical features in different embodiments are embodied in the same figure, it can be regarded that the figure also discloses the combination examples of the various embodiments involved.

[0103] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A vulnerable plaque identification method based on IVUS images, characterized in that: include: Obtaining IVUS (intravascular ultrasound) images and OCT (optical coherence tomography) images based on the coronary arteries, and registering the IVUS images with the OCT images; Manually annotate the IVUS images according to the corresponding OCT images to obtain classification labels of the IVUS images, wherein the classification labels include a normal class and a TCFA (thin cap fibroatherosclerosis) class; Describing the lumen boundary and the external elastic membrane boundary on the IVUS image to obtain a mask image; Training a first deep learning model according to the IVUS image and the mask image to obtain a trained second deep learning model, wherein the second deep learning model is used to automatically intercept the lumen boundary and the external elastic membrane boundary and obtain the boundary of a preliminary ROI (region of interest); On the IVUS image, the area between the lumen boundary and the external elastic membrane boundary is further divided into a plurality of boundary areas according to the distance from the lumen boundary, and feature extraction is performed on the IVUS image with the plurality of boundary areas, wherein the extracted features include texture features; The feature ratio between any two boundary areas is calculated, and the classifier is trained using the feature ratio and the classification label. After the training is completed, the classifier automatically recognizes TCFA on the IVUS image with the preliminary ROI boundary.

2. The vulnerable plaque identification method based on IVUS images according to claim 1, characterized in that: The extracted features include two-dimensional geometric features and probability distribution parameters.

3. The vulnerable plaque identification method based on IVUS images according to claim 2, characterized in that: The texture features include variance, moment, smoothness, kurtosis, entropy and homogeneity; The two-dimensional geometric features include plaque burden, minimum lumen diameter, lumen area, and lumen eccentricity.

4. The vulnerable plaque identification method based on IVUS images according to claim 1, characterized in that: The classifier is a CNN (convolutional neural network) model, and the CNN model is trained using IVUS images with the preliminary ROI boundary and classification labels; After training, the CNN model performs TCFA recognition and classification on the IVUS image with the preliminary ROI boundary to obtain the classification label.

5. The vulnerable plaque identification method based on IVUS images according to claim 4, characterized in that: After training, the CNN model performs TCFA recognition and classification on the IVUS image with the preliminary ROI boundary, specifically including: After training, the CNN model iteratively processes the convolution layer and the pooling layer on the IVUS image with the preliminary ROI boundary to extract the feature map, outputs the probabilities of the normal class and the TCFA class respectively through the fully connected layer, and selects the class with the highest probability as the classification result.

6. The vulnerable plaque identification method based on IVUS images according to claim 5, characterized in that: Training the CNN model using IVUS images with the preliminary ROI boundaries and classification labels also includes: During the training process, weighted binary cross entropy is used as the loss function, and the weights of samples with TCFA are increased.

7. The vulnerable plaque identification method based on IVUS images according to claim 1, characterized in that: The classifier is an integrated learning model based on XGBoost; The first deep learning model is a Swin-Unet structure with Swin Transformer as the core.

8. A computer device 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 vulnerable plaque identification method based on IVUS images as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the vulnerable plaque identification method based on IVUS images described in any one of claims 1 to 7 are implemented.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the steps of the vulnerable plaque identification method based on IVUS images described in any one of claims 1 to 7 are implemented.