Left trunk lesion degree analysis system based on AI and intravascular ultrasound

Abstract

The invention provides a left trunk lesion degree analysis system based on AI and intravascular ultrasound, and relates to the field of medical image analys.The left trunk lesion degree analysis system comprises the steps that firstly, a coronary artery image is obtained through an IVUS catheter, and then noise reduction and contrast enhancement processing are conducted on the obtained coronary artery image; performing image semantic segmentation on the processed image to identify a blood vessel boundary and a plaque region, extracting medical image features including the plaque type, the plaque size, the minimum lumen diameter and the blood vessel area from the blood vessel boundary and the plaque region, and finally evaluating the lesion level of the left trunk lesion based on the extracted medical image features. Therefore, automatic and intelligent left trunk lesion degree analysis can be realized, dependence on doctors is reduced, and meanwhile, a more objective lesion degree evaluation result is provided.

Description

Technical Field

[0001] The present application relates to the field of medical image analysis, and more specifically, to a left main trunk lesion severity analysis system based on AI and intravascular ultrasound. Background Art

[0002] The left main (LM) coronary artery is an important branch in the coronary artery system, responsible for supplying approximately two-thirds of myocardial blood flow. If LM is diseased, it may lead to widespread ischemia and serious complications such as acute myocardial infarction, heart failure, and even sudden death. LM lesions usually manifest as the formation of atherosclerotic plaques, leading to vascular stenosis. Since LM has a wide range of blood supply, once severe stenosis occurs, it may lead to large areas of myocardial ischemia, so timely and accurate diagnosis and treatment are required.

[0003] However, a lot of manual intervention is required in the traditional analysis of left main lesions. Specifically, doctors need to manually mark the vascular boundaries and plaque areas on the coronary artery images. This step is not only time-consuming and laborious, but also easily affected by the doctor's personal experience and subjective judgment. In addition, doctors also need to carefully measure the changes in vascular diameter to assess the degree of stenosis, which often depends on the doctor's professional knowledge, resulting in differences in consistency and objectivity of diagnostic results between different doctors.

[0004] Therefore, an optimized left main coronary artery disease severity analysis system is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a left main trunk lesion degree analysis system based on AI and intravascular ultrasound.

[0006] According to one aspect of the present application, a left main trunk lesion degree analysis system based on AI and intravascular ultrasound is provided, which includes: A coronary artery data acquisition module, used for acquiring a coronary artery image, wherein the coronary artery image is obtained by extending an IVUS catheter into a cardiac catheterization laboratory; A coronary artery image preprocessing module, used for performing image noise reduction and contrast enhancement on the coronary artery image to obtain a preprocessed coronary artery image; An image semantic segmentation module is used to perform image semantic segmentation on the preprocessed coronary artery image to obtain a blood vessel boundary and plaque area image, wherein the image semantic segmentation module includes: a coronary artery multi-scale feature extraction unit, used to perform coronary artery multi-scale feature extraction on the preprocessed coronary artery image to obtain a coronary artery superficial feature map and a coronary artery deep feature map; a coronary artery superficial-deep feature fusion unit, used to perform sparse core association fusion on the coronary artery superficial feature map and the coronary artery deep feature map to obtain a coronary artery superficial-deep significant sparse interactive fusion feature map; a coronary artery superficial-deep feature pooling unit, used to perform pooling processing on the coronary artery superficial-deep significant sparse interactive fusion feature map to obtain a coronary artery superficial-deep significant sparse interactive fusion representation; a semantic segmentation unit, used to obtain the blood vessel boundary and plaque area image based on the coronary artery superficial-deep significant sparse interactive fusion representation; A feature extraction module, used to extract medical image features from the blood vessel boundary and plaque area images, wherein the medical image features include plaque type, plaque volume, minimum lumen diameter and blood vessel area; The intelligent auxiliary analysis module for the degree of lesion is used to determine the grade label of the degree of lesion of the left main trunk based on the medical image features.

[0007] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery multi-scale feature extraction unit is used to: input the preprocessed coronary artery image into the coronary artery multi-scale feature extractor based on the hollow pyramid network model to obtain the coronary artery superficial feature map and the coronary artery deep feature map.

[0008] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery superficial-deep feature fusion unit includes: a coronary artery core feature correlation subunit, which is used to perform core feature correlation analysis on the coronary artery superficial feature map and the coronary artery deep feature map to obtain the coronary artery core feature global coding matrix; a coronary artery feature sparse mask fusion subunit, which is used to perform kernel sparse correlation mask fusion on the coronary artery superficial feature map and the coronary artery deep feature map based on the coronary artery core feature global coding matrix to obtain the coronary artery superficial-deep significant sparse interactive fusion feature map.

[0009] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery core feature correlation subunit includes: a coronary artery kernel feature extraction secondary subunit, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map into the kernel feature distillation network to obtain a coronary artery superficial layer kernel feature vector and a coronary artery deep layer kernel feature vector; a coronary artery feature global association secondary subunit, which is used to perform global association encoding on the coronary artery superficial layer kernel feature vector and the coronary artery deep layer kernel feature vector to obtain the global encoding matrix between the coronary artery kernel features.

[0010] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery kernel feature extraction secondary subunit is used to: expand each feature matrix of the coronary artery superficial feature map along the channel dimension into a feature vector to obtain a set of coronary artery superficial feature vectors; calculate the feature difference coefficient of each coronary artery superficial feature vector in the set of coronary artery superficial feature vectors relative to other coronary artery superficial feature vectors to obtain a set of coronary artery superficial feature difference coefficients; normalize the set of coronary artery superficial feature difference coefficients to obtain a set of normalized coronary artery superficial feature difference coefficients; use each normalized coronary artery superficial feature difference coefficient in the set of normalized coronary artery superficial feature difference coefficients as a weight, calculate the position-weighted sum of the set of coronary artery superficial feature vectors to obtain the coronary artery superficial kernel feature vector.

[0011] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery feature global association secondary subunit is used to: use the matrix obtained by multiplying the transposed vector of the coronary artery superficial kernel feature vector and the coronary artery deep kernel feature vector as the global coding matrix between the coronary artery kernel features.

[0012] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery feature sparse mask fusion subunit includes: a coronary artery kernel feature sparse mask secondary subunit, which is used to input the global encoding matrix between the coronary artery kernel features into the sparse mask unit to obtain the sparse weight matrix associated between the coronary artery kernel features; a coronary artery superficial-deep layer position-by-position interaction secondary subunit, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map into the position-by-position association interaction network to obtain the coronary artery superficial-deep layer position-by-position interaction fusion feature map; the coronary artery superficial-deep layer fusion feature significant The sparse modulation secondary subunit is used to perform feature-significant sparse modulation on the coronary artery superficial-deep layer position-by-position interactive fusion feature map based on the associated sparse weight matrix between the coronary artery kernel features to obtain the coronary artery superficial-deep layer significant sparse interactive fusion feature map; wherein the coronary artery superficial-deep layer position-by-position interactive secondary subunit is used to: perform position-wise multiplication of the eigenvalues ​​of the corresponding positions in each feature matrix along the channel dimension in the coronary artery superficial feature map and each feature matrix along the channel dimension in the coronary artery deep feature map to obtain the coronary artery superficial-deep layer position-by-position interactive fusion feature map.

[0013] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery kernel feature sparse mask secondary subunit is used to: compare the eigenvalues ​​of each position in the global coding matrix between the coronary artery kernel features with a preset threshold to obtain the coronary artery kernel feature correlation sparse weight matrix composed of multiple updated eigenvalues; wherein, in response to the eigenvalue being greater than the second preset threshold, the product of the eigenvalue and constant three is used as the updated eigenvalue; in response to the eigenvalue being greater than the first preset threshold and less than or equal to the second preset threshold, the original value of the eigenvalue is used as the updated eigenvalue; in response to the eigenvalue being less than or equal to the first preset threshold, the value obtained by dividing the eigenvalue by constant three is used as the updated eigenvalue.

[0014] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the coronary artery superficial-deep feature pooling unit is used to: perform global maximum pooling along the channel dimension on the coronary artery superficial-deep significant sparse interaction fusion feature map to obtain the coronary artery superficial-deep significant sparse interaction fusion matrix as the coronary artery superficial-deep significant sparse interaction fusion representation.

[0015] In the above-mentioned left main trunk lesion degree analysis system based on AI and intravascular ultrasound, the semantic segmentation unit is used to: input the coronary artery superficial-deep significant sparse interaction fusion matrix into the image semantic segmentor based on the Softmax function to obtain the blood vessel boundary and plaque area image.

[0016] Compared with the prior art, the left main trunk lesion degree analysis system based on AI and intravascular ultrasound provided by the present application first obtains the coronary artery image of the heart by using an IVUS catheter, then performs noise reduction and contrast enhancement processing on the obtained coronary artery image of the heart, and then performs image semantic segmentation on the processed image to identify the vascular boundary and plaque area, and extracts medical image features including plaque type, volume, minimum lumen diameter and vascular area, and finally evaluates the lesion grade of the left main trunk lesion based on the extracted medical image features. In this way, an automated and intelligent analysis of the lesion degree of the left main trunk can be achieved, reducing the dependence on doctors and providing a more objective lesion degree assessment result. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 This is a system block diagram of a left main trunk lesion severity analysis system based on AI and intravascular ultrasound according to an embodiment of the present application.

[0018] Figure 2 This is a data flow diagram of a left main trunk lesion severity analysis system based on AI and intravascular ultrasound according to an embodiment of the present application.

[0019] Figure 3 It is a block diagram of the image semantic segmentation module in the left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application.

[0020] Figure 4 It is a block diagram of the coronary artery superficial-deep feature fusion unit in the left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application.

[0021] Figure 5 It is a block diagram of the subunit associated with the core features of the coronary artery in the left main stem lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application.

[0022] Figure 6 It is a block diagram of the coronary artery feature sparse mask fusion subunit in the left main stem lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] The left main (LM) artery plays a vital role in the coronary artery system, providing about two-thirds of the blood supply to the heart muscle. Lesions of the LM artery, such as the formation of atherosclerotic plaques, may lead to vascular stenosis, which in turn may cause serious consequences such as myocardial ischemia, acute myocardial infarction, heart failure, and even sudden death. Since the LM artery has a wide range of blood supply, once severe stenosis occurs, it may cause large areas of myocardial ischemia. Therefore, the diagnosis and treatment of LM artery lesions must be rapid and accurate. In the traditional analysis of left main lesions, manual operations account for a large part. Doctors need to manually identify vascular boundaries and plaque areas in coronary artery images. This step is not only time-consuming and labor-intensive, but also easily affected by the doctor's personal experience and subjective judgment. In addition, assessing the degree of vascular stenosis requires doctors to carefully measure changes in vascular diameter. This process relies heavily on the doctor's professional knowledge, which may lead to differences in consistency and objectivity in the diagnosis results between different doctors.

[0025] Intravascular ultrasound (IVUS) is a technique that uses a catheter to insert into a blood vessel for imaging. It can provide high-resolution images of the vessel wall and plaques, and is used to evaluate the structure and nature of vascular lesions. In the analysis of the degree of left main trunk lesions, IVUS can accurately display vascular stenosis, plaque characteristics and their effects on blood flow. When combined with AI, it can realize automated image analysis and feature extraction, improve the efficiency and accuracy of diagnosis, and reduce the doctor's subjective judgment.

[0026] Based on this, the present application proposes a left main trunk lesion severity analysis system based on AI and intravascular ultrasound, which realizes objective auxiliary analysis of the left main trunk lesion severity through automated and intelligent means by adopting technologies based on AI and intravascular ultrasound (IVUS). Figure 1 This is a system block diagram of a left main trunk lesion severity analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. Figure 1 and Figure 2As shown, in the left main trunk lesion degree analysis system 100 based on AI and intravascular ultrasound, it includes: a coronary artery data acquisition module 110, used to obtain a coronary artery image, which is obtained by inserting an IVUS catheter into the cardiac catheterization room; a coronary artery image preprocessing module 120, used to perform image noise reduction and contrast enhancement on the coronary artery image to obtain a preprocessed coronary artery image; an image semantic segmentation module 130, used to perform image semantic segmentation on the preprocessed coronary artery image to obtain a vascular boundary and plaque area image; a feature extraction module 140, used to extract medical image features from the vascular boundary and plaque area image, the medical image features including plaque type, plaque volume, minimum lumen diameter and vascular area; a lesion degree intelligent auxiliary analysis module 150, used to determine the grade label of the left main trunk lesion degree based on the medical image features.

[0027] In an embodiment of the present application, the coronary artery data acquisition module 110 is used to obtain a coronary artery image, and the coronary artery image is obtained by inserting an IVUS catheter into the cardiac catheter room. It should be understood that intravascular ultrasound (IVUS) is a medical imaging technology that combines ultrasound technology and catheter technology. By inserting a miniaturized ultrasound probe into the cardiovascular cavity, high-resolution images of the vascular wall and the intracavitary structure can be obtained. A specific implementation method is to insert the IVUS catheter into the patient's blood vessel (usually the femoral artery or radial artery) through the catheter room, and then the ultrasound probe in the catheter emits ultrasound to obtain real-time coronary artery images. The acquired image can show the anatomical structure of the blood vessel, the nature of the plaque, and the degree of vascular stenosis in detail, thereby helping doctors to make more accurate diagnoses and evaluations.

[0028] In the embodiment of the present application, the coronary artery image preprocessing module 120 is used to perform image noise reduction and contrast enhancement on the coronary artery image to obtain a preprocessed coronary artery image. Accordingly, considering that in the process of acquiring the coronary artery image, various types of noise may be introduced into the image due to factors such as equipment limitations, patient movement, and electromagnetic interference, and at the same time, in order to make important features such as vascular boundaries and plaques more obvious and facilitate subsequent accurate semantic segmentation and feature extraction, in the technical solution of the present application, image noise reduction and contrast enhancement are performed on the coronary artery image to obtain a preprocessed coronary artery image.

[0029] In the embodiment of the present application, the image semantic segmentation module 130 is used to perform image semantic segmentation on the preprocessed coronary artery image to obtain a blood vessel boundary and plaque area image.

[0030] Accordingly, in the image semantic segmentation module, the technical concept of the present application is to extract multi-scale features from the pre-processed coronary artery image by using AI-based image recognition and semantic segmentation technology, so as to intelligently obtain the semantic segmentation results of the vascular boundary and plaque area image based on the sparse core association and feature pooling between the superficial features of the coronary artery and the deep features of the coronary artery. In this way, various manifestations of lesions can be captured more comprehensively to automatically identify vascular boundaries and plaque areas, thereby reducing the need for manual annotation, avoiding errors caused by differences in personal experience and subjective judgment of doctors, and improving the accuracy of image segmentation.

[0031] Specifically, Figure 3 FIG. 4 is a block diagram of an image semantic segmentation module in a left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. Figure 3 As shown, the image semantic segmentation module 130 includes: a coronary artery multi-scale feature extraction unit 131, which is used to perform coronary artery multi-scale feature extraction on the preprocessed coronary artery image to obtain a coronary artery superficial feature map and a coronary artery deep feature map; a coronary artery superficial-deep feature fusion unit 132, which is used to perform sparse core association fusion on the coronary artery superficial feature map and the coronary artery deep feature map to obtain a coronary artery superficial-deep significant sparse interactive fusion feature map; a coronary artery superficial-deep feature pooling unit 133, which is used to perform pooling processing on the coronary artery superficial-deep significant sparse interactive fusion feature map to obtain a coronary artery superficial-deep significant sparse interactive fusion representation; and a semantic segmentation unit 134, which is used to obtain the vascular boundary and plaque area image based on the coronary artery superficial-deep significant sparse interactive fusion representation.

[0032] In an embodiment of the present application, the coronary artery multi-scale feature extraction unit 131 is used to perform coronary artery multi-scale feature extraction on the pre-processed coronary artery image to obtain a coronary artery superficial feature map and a coronary artery deep feature map. Specifically, in an embodiment of the present application, the coronary artery multi-scale feature extraction unit is used to: input the pre-processed coronary artery image into a coronary artery multi-scale feature extractor based on a hollow pyramid network model to obtain the coronary artery superficial feature map and the coronary artery deep feature map. It should be understood that the pre-processed coronary artery image contains the pre-processed coronary artery image, such as superficial features of the contour of the vascular wall, the texture of the tissue or lesion, or more complex plaque features, vascular branch points and their interconnection methods and other deep features, and this information is crucial for understanding and evaluating the health of the coronary arteries. Considering that the dilated pyramid network model can extract features in parallel through dilated convolution layers with different dilation rates, it can capture detailed information of coronary artery images at different scales, which is helpful to identify vascular boundaries and plaque areas. At the same time, by inserting holes in the convolution kernel, the dilated convolution can expand the receptive field without increasing the number of parameters, so that the model can capture contextual information in a larger range, thereby deeply understanding the global characteristics of vascular structure and lesion areas. Based on this, in the technical solution of the present application, the preprocessed coronary artery image is input into the coronary artery multi-scale feature extractor based on the dilated pyramid network model to capture and mine feature information of different scales and levels, and obtain the coronary artery superficial feature map and the coronary artery deep feature map.

[0033] In the embodiment of the present application, the coronary artery superficial-deep layer feature fusion unit 132 is used to perform sparse core association fusion on the coronary artery superficial layer feature map and the coronary artery deep layer feature map to obtain a coronary artery superficial-deep layer significant sparse interactive fusion feature map. Specifically, Figure 4 FIG. 1 is a block diagram of a coronary artery superficial-deep feature fusion unit in a left main coronary artery lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. Figure 4 As shown, the coronary artery superficial-deep feature fusion unit 132 includes: a coronary artery core feature correlation subunit 1321, which is used to perform core feature correlation analysis on the coronary artery superficial feature map and the coronary artery deep feature map to obtain a coronary artery inner kernel feature global coding matrix; a coronary artery feature sparse mask fusion subunit 1322, which is used to perform kernel sparse correlation mask fusion on the coronary artery superficial feature map and the coronary artery deep feature map based on the coronary artery inner kernel feature global coding matrix to obtain the coronary artery superficial-deep significant sparse interactive fusion feature map.

[0034] It should be understood that the coronary artery superficial feature map and the coronary artery deep feature map respectively express the information of the coronary artery at different depth levels, and there is a significant interaction between the two. To this end, the most representative and discriminative feature information in the original features can be extracted, and those features that are most important for lesion identification can be selectively highlighted, while ignoring those unimportant or redundant information, so as to achieve a comprehensive analysis of the coronary artery image features. In the technical solution of the present application, the coronary artery superficial feature map and the coronary artery deep feature map are sparsely core-correlated fused to obtain a coronary artery superficial-deep significant sparse interactive fusion feature map.

[0035] In detail, first, the core features of the coronary artery superficial feature map and the coronary artery deep feature map are associated to obtain the global encoding matrix between the coronary artery kernel features. In this way, the intrinsic connection between them can be revealed to obtain a global representation of a coronary artery at different depths. Specifically, by inputting the coronary artery superficial feature map and the coronary artery deep feature map into the kernel feature distillation network to utilize the powerful representation ability of the network to identify and enhance the essential information and attributes in the features, reduce the influence of redundant and irrelevant information, and obtain the coronary artery superficial kernel feature vector and the coronary artery deep kernel feature vector. Then, in order to further reflect the relationship strength and pattern between the two, so as to discover potential data patterns and information, the coronary artery superficial kernel feature vector and the coronary artery deep kernel feature vector are globally associated and encoded to obtain the global encoding matrix between the coronary artery kernel features, so that the information between the most relevant features can be ensured to be retained, providing data support for subsequent sparse calculations.

[0036] Next, based on the global coding matrix between the coronary artery kernel features, the kernel sparse correlation mask fusion is performed on the coronary artery superficial feature map and the coronary artery deep feature map to obtain the coronary artery superficial-deep significant sparse interaction fusion feature map. Specifically, the global coding matrix between the coronary artery kernel features is input into the sparse mask unit to screen and remove those relatively minor connections or weights by adopting a sparse strategy, and to emphasize those parts that play a core role in the information transmission process, so as to form a sparse coronary artery kernel feature correlation sparse weight matrix. At the same time, in order to capture and more carefully refine the interaction and integration of information at each position between the coronary artery superficial feature map and the coronary artery deep feature map, the coronary artery superficial feature map and the coronary artery deep feature map are input into the position-by-position correlation interaction network to focus more on the feature interaction between the two in local details and structures to obtain the coronary artery superficial-deep position-by-position interaction fusion feature map. Finally, the sparse weight matrix associated with the kernel features of the coronary arteries is used as the weight matrix to perform feature-significant sparse modulation on the coronary artery superficial-deep position-by-position interactive fusion feature map. That is, the various feature matrices along the channel dimension in the coronary artery superficial-deep position-by-position interactive fusion feature map are multiplied with the sparse weight matrix associated with the kernel features of the coronary artery to highlight and intensify the most significant and representative areas and features in the shallow and deep features, and the coronary artery superficial-deep significant sparse interactive fusion feature map is obtained.

[0037] Specifically, Figure 5 FIG. 4 is a block diagram of a subunit associated with core features of coronary arteries in a left main coronary artery lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. Figure 5 As shown, the coronary artery core feature correlation subunit 1321 includes: a coronary artery kernel feature extraction secondary subunit 13211, used to input the coronary artery superficial feature map and the coronary artery deep feature map into the kernel feature distillation network to obtain a coronary artery superficial kernel feature vector and a coronary artery deep kernel feature vector; a coronary artery feature global correlation secondary subunit 13212, used to perform global correlation coding on the coronary artery superficial kernel feature vector and the coronary artery deep kernel feature vector to obtain a global coding matrix between the coronary artery kernel features.

[0038] More specifically, in an embodiment of the present application, the coronary artery kernel feature extraction secondary subunit is used to: expand each feature matrix of the coronary artery superficial feature map along the channel dimension into a feature vector to obtain a set of coronary artery superficial feature vectors; calculate the feature difference coefficient of each coronary artery superficial feature vector in the set of coronary artery superficial feature vectors relative to other coronary artery superficial feature vectors to obtain a set of coronary artery superficial feature difference coefficients; normalize the set of coronary artery superficial feature difference coefficients to obtain a set of normalized coronary artery superficial feature difference coefficients; use each normalized coronary artery superficial feature difference coefficient in the set of normalized coronary artery superficial feature difference coefficients as a weight, and calculate the position-weighted sum of the set of coronary artery superficial feature vectors to obtain the coronary artery superficial kernel feature vector.

[0039] More specifically, in an embodiment of the present application, the characteristic difference coefficient of each coronary artery superficial feature vector in the set of coronary artery superficial feature vectors relative to other coronary artery superficial feature vectors is calculated to obtain a set of coronary artery superficial feature difference coefficients, including: extracting a predetermined coronary artery superficial feature vector from the set of coronary artery superficial feature vectors; subtracting the predetermined coronary artery superficial feature vector from other coronary artery superficial feature vectors in the set of coronary artery superficial feature vectors by position, and calculating a norm of each coronary artery superficial feature difference vector in the obtained set of coronary artery superficial feature difference vectors to obtain a set of coronary artery superficial feature norm values; and calculating the global mean of the set of coronary artery superficial feature norm values ​​to obtain the coronary artery superficial feature difference coefficient.

[0040] More specifically, in an embodiment of the present application, the set of coronary artery superficial characteristic difference coefficients is normalized to obtain a set of normalized coronary artery superficial characteristic difference coefficients, including: calculating an exponential function with a natural constant e as the base and the negative number of each coronary artery superficial characteristic difference coefficient in the set of coronary artery superficial characteristic difference coefficients as the exponent to obtain a set of coronary artery superficial characteristic difference indices; calculating the position-by-position sum of the set of coronary artery superficial characteristic difference indices to obtain a total sum of coronary artery superficial characteristic difference indices; and dividing the set of coronary artery superficial characteristic difference indices by the total sum of coronary artery superficial characteristic difference indices by position to obtain the set of normalized coronary artery superficial characteristic difference coefficients.

[0041] More specifically, in an embodiment of the present application, the coronary artery feature global association secondary subunit is used to: use the matrix obtained by multiplying the coronary artery superficial kernel feature vector and the transposed vector of the coronary artery deep kernel feature vector as the global encoding matrix between the coronary artery kernel features.

[0042] Specifically, Figure 6 FIG. 4 is a block diagram of a coronary artery feature sparse mask fusion subunit in a left main coronary artery lesion degree analysis system based on AI and intravascular ultrasound according to an embodiment of the present application. Figure 6 As shown, the coronary artery feature sparse mask fusion subunit 1322 includes: a coronary artery kernel feature sparse mask secondary subunit 13221, which is used to input the global encoding matrix between the coronary artery kernel features into the sparse mask unit to obtain the sparse weight matrix associated between the coronary artery kernel features; a coronary artery superficial-deep layer position-by-position interaction secondary subunit 13222, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map into the position-by-position association interaction network to obtain the coronary artery superficial-deep layer position-by-position interaction fusion feature map; a coronary artery superficial-deep layer fusion feature significant sparse modulation secondary subunit 13223, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map into the position-by-position association interaction network to obtain the coronary artery superficial-deep layer position-by-position interaction fusion feature map; a coronary artery superficial-deep layer fusion feature significant sparse modulation secondary subunit 13224, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map into the position-by-position association interaction network to obtain the coronary artery superficial-deep layer position-by-position interaction fusion feature map; a coronary artery superficial-deep layer fusion feature significant sparse modulation secondary subunit 13225, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map into the position-by-position association interaction network to obtain the coronary artery superficial-deep layer position-by-position interaction fusion feature map; a coronary artery superficial-deep layer fusion feature significant sparse modulation secondary subunit 13226, which is used to input the coronary artery superficial layer feature map and the coronary artery deep layer feature map Element 13223 is used to perform feature significant sparse modulation on the coronary artery superficial-deep layer position-by-position interactive fusion feature map based on the correlation sparse weight matrix between the coronary artery kernel features to obtain the coronary artery superficial-deep layer significant sparse interactive fusion feature map; wherein the coronary artery superficial-deep layer position-by-position interactive secondary subunit is used to: perform position point multiplication on the eigenvalues ​​of the corresponding positions in each feature matrix along the channel dimension in the coronary artery superficial feature map and each feature matrix along the channel dimension in the coronary artery deep feature map to obtain the coronary artery superficial-deep layer position-by-position interactive fusion feature map.

[0043] More specifically, in an embodiment of the present application, the coronary artery kernel feature sparse mask secondary subunit is used to: compare the eigenvalues ​​of each position in the global encoding matrix between the coronary artery kernel features with a preset threshold to obtain the coronary artery kernel feature correlation sparse weight matrix composed of multiple updated eigenvalues; wherein, in response to the eigenvalue being greater than a second preset threshold, the product of the eigenvalue and a constant three is used as the updated eigenvalue; in response to the eigenvalue being greater than a first preset threshold and less than or equal to the second preset threshold, the original value of the eigenvalue is used as the updated eigenvalue; in response to the eigenvalue being less than or equal to the first preset threshold, the value obtained by dividing the eigenvalue by the constant three is used as the updated eigenvalue.

[0044] More specifically, in an embodiment of the present application, the coronary artery superficial-deep fusion feature significant sparse modulation secondary subunit is used to: matrix multiply each feature matrix along the channel dimension in the coronary artery superficial-deep position-by-position interactive fusion feature map with the associated sparse weight matrix between the coronary artery kernel features to obtain the coronary artery superficial-deep significant sparse interactive fusion feature map.

[0045] In an embodiment of the present application, the coronary artery superficial-deep feature fusion unit is used to: perform sparse core association fusion processing on the coronary artery superficial feature map and the coronary artery deep feature map according to the following formula to obtain a coronary artery superficial-deep significant sparse interactive fusion feature map; wherein the formula is:

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] in, and are respectively the coronary artery superficial layer characteristic map and the coronary artery deep layer characteristic map, For the feature map Perform kernel feature distillation. and are the superficial kernel feature vector of the coronary artery and the deep kernel feature vector of the coronary artery, For the feature map Perform feature fine-grained decomposition operations. is the first, second, ..., first eigenvector in the set of decomposed eigenvectors ,..., ,..., After decomposition, the eigenvectors represents the norm of the eigenvector, is the set of decomposed eigenvectors, represents the number of vectors in the set of decomposed eigenvectors minus one, yes The corresponding characteristic difference coefficient, Indicated by natural constant The exponential function with base , is the number of vectors in the set of decomposed eigenvectors, is the kernel eigenvector, yes The transposed vector of is matrix multiplication, is the global encoding matrix between the coronary artery kernel features, is a sparse masking operation, yes The eigenvalues ​​at each position in , is the preset threshold, is the sparse weight matrix of the correlation between the coronary artery kernel features, For point-by-position multiplication, is the coronary artery superficial-deep layer position-by-position interactive fusion feature map, It is the coronary artery superficial-deep significant sparse interactive fusion feature map.

[0053] In an embodiment of the present application, the coronary artery shallow-deep layer feature pooling unit 133 is used to perform pooling processing on the coronary artery shallow-deep layer significant sparse interactive fusion feature map to obtain a coronary artery shallow-deep layer significant sparse interactive fusion representation. Specifically, in an embodiment of the present application, the coronary artery shallow-deep layer feature pooling unit is used to: perform global maximum pooling along the channel dimension on the coronary artery shallow-deep layer significant sparse interactive fusion feature map to obtain a coronary artery shallow-deep layer significant sparse interactive fusion matrix as the coronary artery shallow-deep layer significant sparse interactive fusion representation. It should be understood that in order to further achieve feature compression and significance enhancement to highlight those features that are most important for identifying lesion areas and provide assistance for subsequent semantic segmentation, in the technical solution of the present application, the coronary artery shallow-deep layer significant sparse interactive fusion feature map is subjected to global maximum pooling along the channel dimension to obtain a coronary artery shallow-deep layer significant sparse interactive fusion matrix.

[0054] In the embodiment of the present application, the semantic segmentation unit 134 is used to obtain the vascular boundary and plaque area image based on the coronary artery superficial-deep significant sparse interactive fusion representation. Specifically, in the embodiment of the present application, the semantic segmentation unit is used to: input the coronary artery superficial-deep significant sparse interactive fusion matrix into the image semantic segmenter based on the Softmax function to obtain the vascular boundary and plaque area image. That is, semantic segmentation processing is performed using the coronary artery superficial-deep significant sparse interactive fusion representation after pooling the coronary artery superficial-deep significant sparse interactive fusion feature map, so as to intelligently obtain the semantic segmentation result of the vascular boundary and plaque area image. Specifically, the coronary artery superficial-deep significant sparse interaction fusion matrix is ​​input into the image semantic segmenter based on the Softmax function. For each pixel in the matrix, the Softmax function calculates the probability that it belongs to the vascular boundary or plaque area, and then outputs a probability map. Each pixel in the probability map represents the possibility that it belongs to different categories (such as blood vessels, plaques, etc.). Finally, based on the output probability map, the threshold processing technology is applied to convert the probability map into a binary map to clearly identify the vascular boundary and plaque area, and finally the vascular boundary and plaque area image is obtained. In this way, the various manifestations of the lesion can be captured more comprehensively to automatically identify the vascular boundary and plaque area, thereby reducing the need for manual annotation, avoiding errors caused by differences in doctors' personal experience and subjective judgment, and improving the accuracy of image segmentation.

[0055] In particular, in the technical solution of the present application, the coronary artery superficial feature map and the coronary artery deep feature map represent the image superficial features and image deep semantic features of the preprocessed coronary artery image, respectively. When the coronary artery superficial feature map and the coronary artery deep feature map are subjected to sparse core association fusion, the feature receptive field deviation and feature depth difference between the coronary artery superficial feature map and the coronary artery deep feature map will cause the feature manifold of the coronary artery superficial-deep significant sparse interactive fusion feature map obtained by feature significant joint interactive coding to have expression sparsity and uneven distribution in the high-dimensional feature space, thereby affecting the fine-grained accuracy of the image semantic style of the coronary artery superficial-deep significant sparse interactive fusion matrix input based on the Softmax function image semantic segmentor.

[0056] Based on this, in a preferred embodiment, the coronary artery superficial-deep significant sparse interaction fusion matrix is ​​input into an image semantic segmenter based on a Softmax function to obtain a vascular boundary and plaque area image, including: flattening the coronary artery superficial-deep significant sparse interaction fusion matrix to obtain a coronary artery superficial-deep significant sparse interaction fusion vector; performing L random sampling on the coronary artery superficial-deep significant sparse interaction fusion vector to obtain L coronary artery image representation random perturbation eigenvalues; arranging the L coronary artery image representation random perturbation eigenvalues ​​in order from large to small to obtain a coronary artery image representation random perturbation ordered vector; calculating the inner product of the coronary artery superficial-deep significant sparse interaction fusion vector to obtain a coronary artery image representation energy modulation eigenvalue; based on the coronary artery image The image representation energy modulation feature value is used to perform semantic intensity scaling on the randomly perturbed ordered vector of coronary artery image representation to obtain a semantically enhanced randomly perturbed ordered vector of coronary artery image representation; the product between the semantically enhanced randomly perturbed ordered vector of coronary artery image representation and the transposed vector of the coronary artery superficial-deep significant sparse interactive fusion vector is calculated to obtain an orthogonal modulation matrix for coronary artery image representation best sample hint; the matrix product between the orthogonal modulation matrix for coronary artery image representation best sample hint and the coronary artery superficial-deep significant sparse interactive fusion vector is calculated to obtain an optimized coronary artery superficial-deep significant sparse interactive fusion vector; the optimized coronary artery superficial-deep significant sparse interactive fusion vector is input into the image semantic segmentor based on the Softmax function to obtain the vascular boundary and plaque area image.

[0057] The above optimization process is expressed as follows:

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] in, is the coronary artery superficial-deep significant sparse interaction fusion vector, It means to perform mean operation on vector. is the mean of the coronary artery superficial-deep significant sparse interactive fusion vectors, represents the length of the coronary artery superficial-deep significant sparse interaction fusion vector, is the first significant sparse interaction fusion vector of the coronary artery superficial-deep layer The eigenvalues ​​at the positions, is the variance of the coronary artery superficial-deep significant sparse interaction fusion vector, represents the Gaussian density probability function, express Random sampling, and Indicates the significant sparse interaction fusion vector of the superficial-deep layer of the coronary artery The random sampling The coronary artery images represent the random perturbation eigenvalues, Express Arrange them in order from large to small. is a randomly disturbed ordered vector representing the coronary artery image, represents the square of the vector's two-norm, Characterizing energy modulation characteristic values ​​for the coronary artery image, It means point multiplication by position. represents matrix multiplication, express The transposed vector of The best sample for characterizing the coronary artery image suggests an orthogonal modulation matrix, The optimized coronary artery superficial-deep layer significantly sparse interaction fusion vector.

[0066] Accordingly, by performing L random sampling on the coronary artery superficial-deep significant sparse interaction fusion vector to obtain a set of new feature values ​​based on the original features but containing slight changes, randomness is introduced to explore different aspects of the feature space, and by simulating the noise or variation in the actual data, it helps the model to better generalize to unseen data. Then, these coronary artery image representation random perturbation feature values ​​are arranged in order from large to small to form a coronary artery image representation random perturbation ordered vector to identify which perturbation directions may be the most important for the current problem. Such an ordering not only helps the algorithm focus on the feature differences that can best distinguish different categories, but also helps to reveal hidden data structures or patterns. Subsequently, the inner product of the coronary artery superficial-deep significant sparse interactive fusion vector is calculated to obtain the energy modulation eigenvalue of the coronary artery image representation, which reflects the overall strength or "energy" of the original feature vector and provides a benchmark for measuring the importance of the feature. Using this coronary artery image representation energy modulation eigenvalue as a reference point, the previously formed coronary artery image representation random perturbation ordered vector is semantically scaled to construct a semantically enhanced coronary artery image representation random perturbation ordered vector. In this way, those disturbances that appear to be more important in the original feature space are amplified, thereby enhancing the model's sensitivity to changes in key features, while also retaining relatively small but potentially useful disturbance information.

[0067] Then, the product of the random perturbation ordered vector of semantically enhanced coronary image representation and the transpose of the coronary artery superficial-deep significant sparse interaction fusion vector is calculated to obtain the best sample prompt orthogonal modulation matrix of coronary image representation. The purpose is to find the interaction mode between features under two different perspectives. The formed matrix can be regarded as a conversion tool for capturing and representing the complex linear relationship between features. Finally, the final feature structure modulation is completed by applying the best sample prompt orthogonal modulation matrix of coronary image representation constructed above to the original coronary artery superficial-deep significant sparse interaction fusion vector, so that the features are more in line with the requirements of the objective function, reducing redundant information, while maximizing the retention of useful information, so as to improve learning efficiency and prediction performance.

[0068] In summary, the feature manifold optimization process of the coronary artery superficial-deep significant sparse interactive fusion vector is clarified, which extracts the best prompt sample based on the fine-grained target retrieval of random perturbations, and then guides the adversarial robust learning based on the best prompt sample to complete the structure of the high-dimensional manifold of the feature vector in the feature space, so as to improve the detail and structure of the feature expression. In this way, the fine-grained accuracy of the image semantic segmentation of the vascular boundary and plaque area image obtained by the image semantic segmentation device based on the Softmax function is improved. In this way, various manifestations of lesions can be captured more comprehensively to automatically identify vascular boundaries and plaque areas, thereby reducing the need for manual annotation, avoiding errors caused by differences in doctors' personal experience and subjective judgment, and improving the accuracy of image segmentation.

[0069] Therefore, the image semantic segmentation module 130 is explained, which uses AI-based image recognition and semantic segmentation technology to extract multi-scale features from the pre-processed coronary artery image, so as to intelligently obtain the semantic segmentation results of the vascular boundary and plaque area image based on the sparse core association and feature pooling between the coronary artery superficial features and the coronary artery deep features. In this way, various manifestations of lesions can be captured more comprehensively to automatically identify vascular boundaries and plaque areas, thereby reducing the need for manual annotation, avoiding errors caused by differences in doctors' personal experience and subjective judgment, and improving the accuracy of image segmentation.

[0070] In the embodiment of the present application, the feature extraction module 140 is used to extract medical image features from the vascular boundary and plaque area image, and the medical image features include plaque type, plaque volume, minimum lumen diameter and vascular area. It should be understood that in order to obtain detailed information about the lesion from the vascular boundary and plaque area image after semantic segmentation to help doctors more accurately judge the condition, it is necessary to extract medical image features from the vascular boundary and plaque area image in the present application. Specifically, the plaque type (such as calcification, fiber, lipid plaque) can help identify the stability of the plaque, the plaque volume and minimum lumen diameter can evaluate the degree of stenosis, and the vascular area helps to understand the overall condition of the blood vessel. These features can effectively quantify the lesion and provide a scientific basis for clinical treatment. In particular, in one feasible implementation of the present application, the plaque type can be obtained by analyzing the grayscale value, texture and shape characteristics of the plaque area in the image, the plaque volume can be obtained by calculating the pixel area of ​​the plaque area and converting it into the actual volume according to the resolution of the image, the morphological operation can be used to calculate the minimum diameter of the blood vessel in the vascular boundary image to obtain the minimum lumen diameter, and the total area of ​​the blood vessel can be calculated by counting the number of pixels in the vascular area and combining it with the actual area of ​​the pixels.

[0071] In an embodiment of the present application, the lesion degree intelligent auxiliary analysis module 150 is used to determine the grade label of the left main trunk lesion degree based on the medical image features. It should be understood that in order to establish a standardized evaluation system and reduce the errors caused by subjective judgment, the grade label of the lesion degree can be determined by medical image features. Specifically, the medical image features can be passed through a classifier-based lesion degree result generator to obtain a lesion grade result, and the lesion grade result represents the grade label of the left main trunk lesion degree. In this way, the grade label obtained by classification provides doctors with clear lesion degree information, which can assist them in making more scientific and reasonable clinical decisions.

[0072] In summary, the left main trunk lesion degree analysis system 100 based on AI and intravascular ultrasound according to the embodiment of the present application is explained, which first uses an IVUS catheter to obtain a coronary artery image of the heart, then performs noise reduction and contrast enhancement processing on the obtained coronary artery image of the heart, and then performs image semantic segmentation on the processed image to identify the vascular boundary and plaque area, and extracts medical image features including plaque type, volume, minimum lumen diameter and vascular area, and finally evaluates the lesion grade of the left main trunk lesion based on the extracted medical image features. In this way, an automated and intelligent analysis of the lesion degree of the left main trunk can be achieved, reducing dependence on doctors and providing a more objective lesion degree evaluation result.

[0073] As described above, the left main trunk lesion degree analysis system 100 based on AI and intravascular ultrasound according to the embodiment of the present application can be implemented in various terminal devices, such as a server for left main trunk lesion degree analysis based on AI and intravascular ultrasound. In one example, the left main trunk lesion degree analysis system 100 based on AI and intravascular ultrasound according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the left main trunk lesion degree analysis system 100 based on AI and intravascular ultrasound can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the left main trunk lesion degree analysis system 100 based on AI and intravascular ultrasound can also be one of the many hardware modules of the terminal device.

[0074] Alternatively, in another example, the left main trunk lesion extent analysis system 100 based on AI and intravascular ultrasound and the terminal device may also be separate devices, and the left main trunk lesion extent analysis system 100 based on AI and intravascular ultrasound may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0075] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details of the above application are only for the purpose of illustration and ease of understanding, rather than limitation, and the above details do not limit the present application to the use of the above specific details to be implemented.

Claims

1. A left main trunk lesion degree analysis system based on AI and intravascular ultrasound, characterized in that: include: A coronary artery data acquisition module, used for acquiring a coronary artery image, wherein the coronary artery image is obtained by extending an IVUS catheter into a cardiac catheterization laboratory; A coronary artery image preprocessing module, used for performing image noise reduction and contrast enhancement on the coronary artery image to obtain a preprocessed coronary artery image; An image semantic segmentation module is used to perform image semantic segmentation on the preprocessed coronary artery image to obtain a blood vessel boundary and plaque area image, wherein the image semantic segmentation module includes: a coronary artery multi-scale feature extraction unit, used to perform coronary artery multi-scale feature extraction on the preprocessed coronary artery image to obtain a coronary artery superficial feature map and a coronary artery deep feature map; a coronary artery superficial-deep feature fusion unit, used to perform sparse core association fusion on the coronary artery superficial feature map and the coronary artery deep feature map to obtain a coronary artery superficial-deep significant sparse interactive fusion feature map; a coronary artery superficial-deep feature pooling unit, used to perform pooling processing on the coronary artery superficial-deep significant sparse interactive fusion feature map to obtain a coronary artery superficial-deep significant sparse interactive fusion representation; a semantic segmentation unit, used to obtain the blood vessel boundary and plaque area image based on the coronary artery superficial-deep significant sparse interactive fusion representation; A feature extraction module, used to extract medical image features from the blood vessel boundary and plaque area images, wherein the medical image features include plaque type, plaque volume, minimum lumen diameter and blood vessel area; The intelligent auxiliary analysis module for the degree of lesion is used to determine the grade label of the degree of lesion of the left main trunk based on the medical image features.

2. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 1 is characterized in that: The coronary artery multi-scale feature extraction unit is used to: input the preprocessed coronary artery image into a coronary artery multi-scale feature extractor based on a hollow pyramid network model to obtain the coronary artery superficial feature map and the coronary artery deep feature map.

3. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 2 is characterized in that: The coronary artery superficial-deep feature fusion unit comprises: A coronary artery core feature correlation subunit, used for performing core feature correlation analysis on the coronary artery superficial layer feature map and the coronary artery deep layer feature map to obtain a global coding matrix between coronary artery core features; The coronary artery feature sparse mask fusion subunit is used to perform kernel sparse correlation mask fusion on the coronary artery superficial feature map and the coronary artery deep feature map based on the global coding matrix between the coronary artery kernel features to obtain the coronary artery superficial-deep significant sparse interactive fusion feature map.

4. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 3 is characterized in that: The coronary artery core feature association subunit includes: A secondary subunit for extracting coronary artery kernel features, used for inputting the coronary artery superficial layer feature map and the coronary artery deep layer feature map into a kernel feature distillation network to obtain a coronary artery superficial layer kernel feature vector and a coronary artery deep layer kernel feature vector; The coronary artery feature global association secondary subunit is used to perform global association coding on the coronary artery superficial kernel feature vector and the coronary artery deep kernel feature vector to obtain a global coding matrix between the coronary artery kernel features.

5. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 4 is characterized in that: The coronary artery kernel feature extraction secondary subunit is used for: Expanding each feature matrix of the coronary artery superficial layer feature map along the channel dimension into a feature vector to obtain a set of coronary artery superficial layer feature vectors; Calculating the characteristic difference coefficient of each coronary artery superficial layer characteristic vector relative to other coronary artery superficial layer characteristic vectors in the set of coronary artery superficial layer characteristic vectors to obtain a set of coronary artery superficial layer characteristic difference coefficients; Normalizing the set of coronary artery superficial layer characteristic difference coefficients to obtain a set of normalized coronary artery superficial layer characteristic difference coefficients; Taking each normalized coronary artery superficial feature difference coefficient in the set of normalized coronary artery superficial feature difference coefficients as a weight, a position-weighted sum of the set of coronary artery superficial feature vectors is calculated to obtain the coronary artery superficial kernel feature vector.

6. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 5 is characterized in that: The coronary artery feature global association secondary subunit is used to: use the matrix obtained by multiplying the coronary artery superficial kernel feature vector and the transposed vector of the coronary artery deep kernel feature vector as the coronary artery kernel feature global encoding matrix.

7. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 6 is characterized in that: The coronary artery feature sparse mask fusion subunit comprises: A coronary artery kernel feature sparse mask secondary subunit, used for inputting the global encoding matrix between the coronary artery kernel features into the sparse mask unit to obtain a sparse weight matrix associated with the coronary artery kernel features; A coronary artery superficial layer-deep layer position-by-position interactive secondary subunit, used for inputting the coronary artery superficial layer feature map and the coronary artery deep layer feature map into a position-by-position associated interactive network to obtain a coronary artery superficial layer-deep layer position-by-position interactive fusion feature map; A secondary subunit of coronary artery superficial-deep fusion feature significant sparse modulation is used to perform feature significant sparse modulation on the coronary artery superficial-deep position-by-position interactive fusion feature map based on the correlation sparse weight matrix between the coronary artery kernel features to obtain the coronary artery superficial-deep significant sparse interactive fusion feature map; Among them, the coronary artery superficial-deep layer position-by-position interactive secondary subunit is used to: multiply the eigenvalues ​​of the corresponding positions in each feature matrix along the channel dimension in the coronary artery superficial feature map with the eigenvalues ​​of the corresponding positions in each feature matrix along the channel dimension in the coronary artery deep layer feature map by position point to obtain the coronary artery superficial-deep layer position-by-position interactive fusion feature map.

8. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 7 is characterized in that: The coronary artery kernel feature sparse mask secondary subunit is used to: Comparing the eigenvalues ​​of each position in the global encoding matrix between the coronary artery kernel features with a preset threshold to obtain the sparse weight matrix of the coronary artery kernel feature association composed of a plurality of updated eigenvalues; wherein, in response to the eigenvalue being greater than a second preset threshold, the product of the eigenvalue and a constant three is used as the updated eigenvalue; In response to the feature value being greater than a first preset threshold and less than or equal to the second preset threshold, taking the original value of the feature value as the updated feature value; In response to the eigenvalue being less than or equal to the first preset threshold, a value obtained by dividing the eigenvalue by constant three is used as the updated eigenvalue.

9. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 8, characterized in that: The coronary artery superficial-deep feature pooling unit is used to perform global maximum pooling along the channel dimension on the coronary artery superficial-deep significant sparse interaction fusion feature map to obtain a coronary artery superficial-deep significant sparse interaction fusion matrix as the coronary artery superficial-deep significant sparse interaction fusion representation.

10. The left main trunk lesion degree analysis system based on AI and intravascular ultrasound according to claim 9, characterized in that: The semantic segmentation unit is used to input the coronary artery superficial-deep significant sparse interactive fusion matrix into an image semantic segmentor based on a Softmax function to obtain the blood vessel boundary and plaque area image.

Images