A coronary cta image analysis system for plaque vulnerability assessment

Abstract

The present application relates to the technical field of image processing, and particularly relates to a coronary CTA image analysis system for plaque vulnerability evaluation, which comprises a frame construction unit for constructing a center line guide frame corresponding to the center point of each image layer. A feature extraction and screening unit is used for extracting a discrete pipe wall feature set including the coordinates of each feature node in each image layer, the density attenuation coefficient and the high-density barrier identifier, and screening the candidate feature nodes from the discrete pipe wall feature set. A space correction unit is used for mapping the candidate feature nodes in the previous image layer to the current image layer to obtain the curvature correction projection coordinates. A connection optimization unit is used for determining the optimal matching path of the candidate feature nodes between adjacent image layers. An entity reconstruction unit is used for reconstructing the target plaque entity according to the optimal matching path. A vulnerability evaluation unit is used for generating the vulnerability evaluation result of the target plaque entity. The present application realizes the vulnerability evaluation of the plaque structure in the coronary CTA image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a coronary CTA image analysis system for assessing plaque vulnerability. Background Technology

[0002] Rupture of coronary atherosclerotic plaques is a key pathological basis for serious cardiovascular events such as acute myocardial infarction. Accurate and non-invasive assessment of plaque vulnerability is of great value for clinical risk warning and intervention decisions. Computed tomography angiography (CTA), with its high spatial resolution and non-invasive characteristics, has become an important imaging tool for plaque morphology and composition analysis, and its application in cardiovascular disease screening and personalized diagnosis and treatment is increasingly widespread.

[0003] Current coronary CTA plaque analysis techniques mostly rely on semi-automatic segmentation or traditional image processing methods to extract plaque regions and macroscopic parameters. However, when dealing with natural vascular curvature, branching structures, and dynamic changes in the cardiac cycle, it is difficult to effectively maintain the spatial correlation of plaque features between consecutive slices. At the same time, it is easily affected by calcification artifacts, image noise, and vascular deformation, resulting in blurred plaque boundary recognition and insufficient stability of feature extraction.

[0004] Because the coronary arteries exhibit complex spatial motion patterns during cardiac pulsation, and are limited by the physical resolution of existing imaging systems, traditional analysis methods still need to improve the accuracy of continuous tracking and boundary localization of plaque structures when processing dynamically acquired coronary artery image data, making it difficult to assess plaque vulnerability. Summary of the Invention

[0005] To address the technical problems of low accuracy in plaque structure identification and difficulty in assessing plaque vulnerability in coronary CTA images, this invention aims to provide a coronary CTA image analysis system for plaque vulnerability assessment. The specific technical solution adopted is as follows: Firstly, a coronary CTA image analysis system for assessing plaque vulnerability is provided, comprising: a frame construction unit, used to extract the centerline of the target vessel based on diastolic image data of the coronary artery, determine multiple image layers along the centerline, and construct a centerline guiding frame corresponding to the center point of each image layer; a feature extraction and screening unit, used to perform radial sampling and feature fitting on the diastolic image data based on the centerline guiding frame corresponding to each image layer, to obtain a discrete vessel wall feature set including the coordinates, density attenuation coefficient, and high-density barrier marker of each feature node in each image layer, and to screen candidate feature nodes from the discrete vessel wall feature set according to the density attenuation coefficient and the high-density barrier marker, wherein the density attenuation coefficient is used to characterize the micro-gradient parameter of the vessel wall interface sharpness corresponding to the feature node, and the high-density barrier marker is used to indicate whether calcified tissue exists on the radial sampling path corresponding to the feature node; and a spatial correction unit, used to determine the spatial transformation relationship between adjacent image layers based on the centerline guiding frame corresponding to each image layer, and to map the candidate feature nodes in the previous image layer to the current image layer to obtain curvature-corrected projection coordinates. The connectivity optimization unit constructs a connectivity cost function based on the curvature-corrected projected coordinates and density attenuation coefficients of candidate feature nodes, determining the optimal matching path between candidate feature nodes in adjacent image layers. The entity reconstruction unit reconstructs the target patch entity based on the optimal matching path, outputting the spatial index range of the target patch entity in the diastolic image. The vulnerability assessment unit generates the vulnerability assessment results for the target patch entity.

[0006] In one possible design, the frame construction unit is specifically used for: segmenting the target blood vessel from diastolic image data and determining the centerline of the target blood vessel; resampling along the centerline at preset length intervals to determine multiple ordered center points, each center point corresponding to an image layer; and for each image layer, establishing a local orthogonal coordinate system with the coordinates of the center point corresponding to the image layer as the origin, the tangent of the centerline as the axial vector, and the preset reference direction within the cross-section as the radial vector, to form a centerline guiding frame.

[0007] In one possible design, the feature extraction and filtering unit is specifically used for: for each image layer, generating multiple radial sampling rays at preset angular intervals within a cross-section perpendicular to the axial vector, with the center point of the image layer as the origin; for each radial sampling ray, performing three-dimensional linear interpolation sampling along the radial sampling ray with a preset step size to obtain a continuous gray-level numerical sequence from the lumen center to the outer wall of the blood vessel; fitting the gray-level numerical sequence using a preset single exponential decay model, and determining the attenuation constant obtained from the fitting as the density attenuation coefficient; determining the coordinates of the feature nodes corresponding to the radial sampling rays based on the position offset parameters obtained from the single exponential attenuation model fitting, the coordinates of the center point of the image layer, and the unit direction vector corresponding to the radial sampling rays, where the feature nodes are the vessel wall boundary points corresponding to the radial sampling rays; determining the high-density barrier markers corresponding to the radial sampling rays based on the gray-level numerical sequence corresponding to the radial sampling rays and a preset calcification threshold; and storing the coordinates, density attenuation coefficients, and high-density barrier markers of the feature nodes corresponding to each radial sampling ray in each image layer in association to form a discrete vessel wall feature set.

[0008] In one possible design, candidate feature nodes are selected from the discrete pipe wall feature set based on the density attenuation coefficient and the high-density barrier marker. This includes: determining the attenuation screening threshold based on the distribution of the density attenuation coefficients of all feature nodes in the discrete pipe wall feature set; and identifying feature nodes whose density attenuation coefficient is greater than the attenuation screening threshold and whose high-density barrier marker indicates that there is no calcified tissue on the radial sampling path as candidate feature nodes.

[0009] In one possible design, the spatial correction unit is specifically used for: extracting the axial and radial vectors from the centerline guide frame corresponding to each adjacent image layer for each group of adjacent image layers; determining the interlayer rotation matrix to characterize the bending and twisting of blood vessels between adjacent image layers based on the axial and radial vectors from the centerline guide frame corresponding to each adjacent image layer, and defining the interlayer rotation matrix as the spatial transformation relationship between adjacent image layers; for each image layer other than the first image layer, rotating and spatially projecting the coordinates of the candidate feature nodes of the previous image layer through the interlayer rotation matrix to obtain the curvature-corrected projection coordinates of the candidate feature nodes in the current image layer, and the curvature-corrected projection coordinates characterize the theoretical projection position of the candidate feature nodes as they bend and twist synchronously with the blood vessels.

[0010] In one possible design, the connection optimization unit is specifically used to: for each image layer other than the first image layer, take the Euclidean distance between the curvature-corrected projection coordinates and the coordinates of each candidate feature node in the current image layer as a geometric bias term; take the absolute difference between the density attenuation coefficients of the candidate feature nodes in the previous image layer and the candidate feature nodes in the current image layer as an attribute difference term; and construct a connection cost function based on the geometric bias term and the attribute difference term. The connection cost function is used to quantify the matching cost of any two candidate feature nodes in adjacent image layers belonging to the same patch entity.

[0011] In one possible design, the connection optimization unit is specifically used for: constructing an augmented cost matrix based on the number of candidate feature nodes in the previous image layer and the number of candidate feature nodes in the current image layer; introducing annihilated virtual nodes in the row direction of the augmented cost matrix, with the same number of candidate feature nodes in the current image layer, and introducing new virtual nodes in the column direction of the augmented cost matrix, with the same number of candidate feature nodes in the previous image layer; configuring a first connection cost for annihilated virtual nodes and a second connection cost for new virtual nodes; the first connection cost is a preset annihilation threshold used to quantify the cost of candidate feature nodes terminating during vertical extension; the second connection cost is a preset germination threshold used to quantify the cost of candidate feature nodes starting to germinate during vertical extension; in the augmented cost matrix, determining the matching cost between real nodes based on the connection cost function, determining the matching cost between real nodes in the previous image layer and new virtual nodes in the column direction based on the first connection cost, and determining the matching cost between real nodes in the current image layer and annihilated virtual nodes in the row direction based on the second connection cost; and using a preset matching optimization algorithm to solve for the candidate feature node matching relationship with the minimum total connection cost corresponding to the augmented cost matrix, forming the optimal matching path.

[0012] In one possible design, the entity reconstruction unit is specifically used to: vertically cascade candidate feature nodes matched between adjacent image layers according to the optimal matching path to form multiple vertical feature chains; identify the vertical feature chains with more than a preset layer number threshold as target patch entities; extract the starting and ending image layer indices of the target patch entity in the vertical direction, as well as the minimum and maximum angle indices in the circumferential direction of the cross-section, wherein the starting, ending, minimum, and maximum angle indices constitute the spatial index range of the target patch entity; and associate and store the average density decay coefficient of all candidate feature nodes included in the target patch entity as the diastolic baseline gradient parameter of the target patch entity.

[0013] In one possible design, the vulnerability assessment unit includes a transperiod mapping module; the transperiod mapping module is used to identify the proximal and distal bifurcation points of the vascular branch where the target plaque entity is located; generate normalized position coefficients based on the distances from the center point of the target plaque entity along the vascular centerline to the proximal and distal bifurcation points respectively; locate the corresponding bifurcation structures of the same vascular branch in the systolic image data; determine the position corresponding to the center point of the target plaque entity in the systolic image data based on the normalized position coefficients, and determine the plaque matching region in the systolic image data based on the spatial index range of the target plaque entity.

[0014] In one possible design, the vulnerability assessment unit further includes an assessment module; a feature extraction and screening unit, used to determine candidate feature nodes and their corresponding density attenuation coefficients within the plaque matching region; an assessment module, used to determine the mean of the density attenuation coefficients of all candidate feature nodes within the plaque matching region as the systolic density attenuation parameter; to determine the change in density attenuation coefficient based on the diastolic baseline gradient parameter and the systolic density attenuation parameter; to determine the radial strain of the lumen based on the average diastolic lumen radius corresponding to the target plaque entity and the average systolic lumen radius corresponding to the plaque matching region; to determine the gradient strain response characteristics based on the change in density attenuation coefficient and the radial strain of the lumen; and to generate a vulnerability assessment result for the target plaque entity based on the gradient strain response characteristics and a preset vulnerability threshold. If the gradient strain response characteristics are greater than the preset vulnerability threshold, the vulnerability assessment result is used to indicate that the target plaque entity is a vulnerable plaque.

[0015] The present invention has the following beneficial effects: In the coronary CTA image analysis system for plaque vulnerability assessment provided by this invention, a local coordinate system is first established along the centerline of the target vessel using a frame construction unit. This provides a centerline-guided frame that can adaptively change with the vessel geometry for all subsequent analyses, fundamentally solving the problem of establishing a unified measurement benchmark in tortuous vessels. Based on this, the feature extraction and screening unit uses this local frame for high-density radial sampling and transforms the partially volumetrically blurred grayscale information into density attenuation coefficients with clear physical meaning through model fitting. Simultaneously, high-density barrier markers indicating calcified tissue are generated, thereby screening out candidate feature nodes that characterize high interface sharpness and are free from calcification interference. This achieves an effective transformation from blurred image data to microscopic physical features. The spatial correction unit further calculates the spatial transformation relationship using the centerline-guided frame of adjacent image layers, mapping candidate feature nodes from the previous image layer to the current image layer to obtain curvature-corrected projection coordinates. Through this geometric correction mechanism that changes with vessel posture, the misalignment of inter-layer feature points caused by macroscopic vessel curvature and axial torsion is effectively eliminated, allowing subsequent connection optimization to focus on the morphological continuity of the plaque itself rather than the macroscopic geometric changes of the vessel. The connectivity optimization unit constructs a connectivity cost function based on curvature-corrected projected coordinates and density attenuation coefficients. By quantifying geometric deviations and attribute differences, it solves for the optimal matching path of candidate feature nodes between adjacent image layers globally, ensuring that vertically connected nodes possess both spatial continuity and physical property consistency. Finally, the entity reconstruction unit selects a set of candidate feature nodes continuously distributed across image layers based on the optimal matching path, reconstructs them into the target plaque entity, and outputs its spatial index range in the diastolic image. The vulnerability assessment unit then performs a quantitative assessment of the vulnerability of this plaque entity. This system overcomes the technical defect of misjudging natural curvature as plaque rupture due to ignoring vessel posture, eliminates the problem of blurred vessel wall boundaries caused by partial volume effects, and achieves accurate identification and continuous tracking of plaque structures, significantly improving the accuracy and reliability of plaque analysis in coronary CTA images. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic diagram of the structure of a coronary CTA image analysis system for assessing plaque vulnerability, provided in one embodiment of the present invention. Figure 1 ; Figure 2A schematic diagram of the structure of a coronary CTA image analysis system for assessing plaque vulnerability, provided in one embodiment of the present invention. Figure 2 . Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a coronary CTA image analysis system for plaque vulnerability assessment proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0020] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] The following description, in conjunction with the accompanying drawings, details a specific scheme for a coronary CTA image analysis system for assessing plaque vulnerability provided by the present invention.

[0023] Please see Figure 1 This illustration shows a schematic diagram of a coronary CTA image analysis system for assessing plaque vulnerability according to an embodiment of the present invention, as shown below. Figure 1 As shown, the coronary CTA image analysis system 10 for plaque vulnerability assessment includes a frame construction unit 11, a feature extraction and screening unit 12, a spatial correction unit 13, a connectivity optimization unit 14, a solid reconstruction unit 15, and a vulnerability assessment unit 16.

[0024] The frame construction unit 11 is used to extract the centerline of the target vessel based on the diastolic image data of the coronary artery, determine multiple image layers along the centerline, and construct the centerline guide frame corresponding to the center point of each image layer.

[0025] As one possible implementation, firstly, the frame construction unit 11 acquires clinically collected coronary diastolic image data. To ensure image quality, the coronary diastolic image data typically selects the phase of the cardiac cycle with sufficient coronary artery filling and minimal motion artifacts, such as the 75% phase. Then, a pre-defined vessel extraction algorithm (e.g., Frangi filtering based on Hessian matrix eigenvalues ​​or a deep learning-based segmentation network) is used to segment the target vessel from the image data. A centerline extraction algorithm, such as a fast-traveling method, is then used to determine the centerline point sequence of the target vessel, denoted as the centerline of the target vessel. The centerline point sequence is continuously distributed along the direction of the target vessel, completely covering the main trunk and major branches of the target vessel.

[0026] To achieve uniform layering of the target blood vessel, the frame construction unit 11 further performs three equal-interval B-spline resamplings along the centerline at preset length intervals (e.g., 0.5 mm) to obtain a series of ordered spatial discrete points, denoted as center points. Each resampled center point corresponds to an image layer, and the image layer index is defined as... , ,in, This represents the total number of center points after resampling.

[0027] For each image layer The frame building unit 11 constructs a frame with the image layer. Centerline guide frame corresponding to the center point Centerline guide frame With image layer The coordinates of the corresponding center point The origin is the point where the sampled data is taken after resampling. The three-dimensional coordinates (x, y, z) of each center point in the world coordinate system; the axial vector is the tangent of the centerline at each center point. Its direction is consistent with the blood flow direction, used to indicate the longitudinal extension direction of the blood vessel; perpendicular to the axial vector The preset reference direction within the cross-section is the radial vector. The preset reference direction is used to indicate the image layer. The 0° starting direction on the cross-section serves as the angular reference for subsequent radial sampling.

[0028] In some embodiments, to ensure that the centerline guide frame changes smoothly along the vascular path and to avoid unnecessary torsion of the frame, the frame construction unit 11 uses a minimum rotation frame algorithm to generate a serialized radial vector. This ensures that the frame rotates smoothly as it bends along the centerline, thus providing a stable local coordinate system for subsequent units.

[0029] The feature extraction and filtering unit 12 is used to perform radial sampling and feature fitting on the diastolic image data based on the centerline guide frame corresponding to each image layer, to obtain a discrete pipe wall feature set including the coordinates, density attenuation coefficient and high-density barrier marker of each feature node in each image layer, and to filter candidate feature nodes from the discrete pipe wall feature set according to the density attenuation coefficient and high-density barrier marker.

[0030] Among them, the density attenuation coefficient is used to characterize the micro-gradient parameter of the pipe wall interface sharpness corresponding to the feature node, and the high-density barrier marker is used to indicate whether there is calcified tissue on the radial sampling path corresponding to the feature node.

[0031] As one possible implementation, firstly, for each image layer The feature extraction and filtering unit 12 uses the center point of the image layer as the reference point. With the origin as the point, and the vector perpendicular to the axis as the point. Establish a polar coordinate system within the cross-section, and define the cross-section angle index as... , The preset angle interval is set to 5°, that is... For each group Generate a line from Start, along the angle Radial sampling rays extending outward in the indicated radial direction The ray length can be set to 3 mm to ensure coverage of conventional pipe diameters and wall thicknesses.

[0032] For each radial sampling ray, the feature extraction and filtering unit 12 operates along each radial sampling ray. The path is obtained by performing three-dimensional linear interpolation sampling in the coronary artery diastolic image data with a preset step size (e.g., 0.1 mm) to obtain a one-dimensional grayscale value sequence that continuously changes grayscale value from the center of the lumen towards the outer wall of the vessel, denoted as . , Used to characterize the grayscale decay process as it transitions from the high-density lumen center to the low-density outer wall of the blood vessel.

[0033] In order to extract the true physical interface features from the blurred gray-scale transition band, the feature extraction and filtering unit 12 performs a process on each gray-scale value sequence. A pre-defined single exponential decay model is used for fitting. The fitting function is defined as follows: In the formula, Radial distance, Radial distance grayscale value at that location and The background and gain constants are used to adjust the grayscale value. This is the position offset parameter, used to characterize the radial distance from the origin to the pipe wall boundary point. Let be the density attenuation coefficient to be solved, used to characterize the sharpness of the pipe wall interface and the attenuation rate of the CT value radially away from the pipe wall boundary. The larger the value, the more severe the grayscale decay at the pipe wall interface, and the sharper the interface.

[0034] Furthermore, a predefined single-exponential decay model is solved using a nonlinear least squares method (such as the Levenberg-Marquardt algorithm) with radial distance. For independent variable, radial distance grayscale value at As the dependent variable, the solution that minimizes the sum of squared residuals is found simultaneously through iterative optimization. The optimal values ​​of the four parameters minimize the sum of squared residuals between the grayscale values ​​fitted by the model and the actual sampled values. As the amplitude and background offset of the model, it can adapt to different image contrasts and tissue backgrounds, ensuring the extracted density attenuation coefficient. It accurately reflects the true sharpness of the pipe wall interface.

[0035] Based on this, at each sampling location Corresponding density attenuation coefficient and position offset parameters Then, based on the position offset parameter Center point coordinates of the image layer and the unit direction vector corresponding to the radial sampling ray Determine the coordinates of the feature nodes corresponding to the radial sampling ray.

[0036] In some embodiments, the formula for calculating the coordinates of a feature node is as follows: In the formula, For the first In the image layer, the first The coordinates of the feature nodes corresponding to each radial sampling ray, i.e., the positions of the pipe wall boundary points in the world coordinate system. For the first The origin coordinates of the centerline guide frame of each image layer. To fit the obtained pipe wall position offset parameters, For the first In the image layer, the first The unit direction vector corresponding to each radial sampling ray.

[0037] Meanwhile, to eliminate the interference of calcified tissue on the analysis of soft plaques, the feature extraction and screening unit 12 detects the gray value sequence corresponding to each radial sampling ray. Check if there are any sampling points with CT values ​​exceeding a preset calcification threshold (an empirical value of 400 HU (Henness units)). If so, generate a high-density barrier marker with a first logic value (e.g., 1). This indicates the presence of calcified tissue on the radial sampling path; otherwise, a high-density barrier identifier with a second logic value (such as 0) is generated.

[0038] After completing all image layers of the target blood vessel ( All angles in ) After the calculation, the feature extraction and filtering unit 12 calculates the coordinates of the feature nodes corresponding to each radial sampling ray in each image layer. Density attenuation coefficient and high-density barrier markings Associative storage constitutes a discrete pipe wall feature set. .

[0039] Furthermore, the feature extraction and screening unit 12 determines the attenuation screening threshold based on the distribution of the density attenuation coefficients of all feature nodes in the discrete pipe wall feature set.

[0040] In some embodiments, the feature extraction and filtering unit 12 can select the 85th percentile of non-zero values ​​as the attenuation filtering threshold by statistically analyzing the histogram distribution of the density attenuation coefficients of all feature nodes. Alternatively, the attenuation screening threshold can be determined by summing the mean and standard deviation of the density attenuation coefficients of all feature nodes. .

[0041] Finally, the discrete pipe wall feature set is traversed. For each feature node, the density decay coefficient must be greater than the decay screening threshold. Feature nodes whose high-density barrier markers indicate that there is no calcified tissue on the radial sampling path (i.e., the high-density barrier marker is the second logical value) are identified as candidate feature nodes.

[0042] In some embodiments, for each image layer This image layer is constructed from all candidate feature nodes of that image layer. The corresponding set of candidate feature nodes ,in, For image layer The number of candidate feature nodes.

[0043] Understandably, in this embodiment of the invention, firstly, using a centerline guide frame as a reference, high-density radial sampling rays are generated at preset angular intervals within the cross-section, and three-dimensional linear interpolation sampling is performed with a sub-voxel-level step size. This overcomes the problem of blurred vessel wall boundaries caused by the limited spatial resolution of CT images, obtaining a vessel wall density profile that can finely depict the continuous grayscale changes from the lumen center to the vessel wall. Based on this, a single exponential decay model is introduced to fit the density profile, transforming some volumetric effect information that is difficult to directly utilize in conventional image processing into a density decay coefficient that can characterize the sharpness of the vessel wall interface. The fitted position offset parameters, combined with the centerline guide frame, accurately locate the coordinates of the vessel wall boundary points, achieving digital characterization of sub-millimeter-level microstructures such as thin fibrous caps. Simultaneously, by detecting whether there are grayscale values ​​exceeding the calcification threshold on the sampling path and generating high-density barrier markers, interference from calcified tissue on soft plaque analysis is effectively identified and eliminated. After obtaining the discrete feature set of the entire blood vessel wall, the attenuation screening threshold is further determined based on the statistical distribution of the density attenuation coefficients of all feature nodes. Feature nodes with a density attenuation coefficient greater than the threshold and no calcification obstruction are selected as candidate feature nodes and grouped by image layer. This allows the massive voxel data to be focused on potential lesion areas with high interface sharpness and no calcification interference, which greatly reduces the computational complexity of subsequent spatial correction and connectivity optimization. It also fundamentally avoids the interference of calcification artifacts or low-gradient normal tissues on the plaque recognition process, providing pure and reliable input data for the accurate reconstruction of anatomically continuous plaque entities.

[0044] The spatial correction unit 13 is used to determine the spatial transformation relationship of each adjacent image layer based on the centerline guide frame of each image layer in each adjacent image layer, and to map the candidate feature nodes in the previous image layer to the current image layer to obtain curvature correction projection coordinates.

[0045] As one possible implementation, for each pair of adjacent image layers, for example, the 1st... The image layer and the first Image layers ( Spatial correction unit 13 first extracts the first image from the centerline guide frame corresponding to the two image layers. Axial vector of each image layer and radial vector and the Axial vector of each image layer and radial vector Then, the inter-layer rotation matrix used to characterize the tortuosity and twisting of blood vessels between adjacent image layers is calculated. The interlayer rotation matrix represents the spatial transformation relationship between adjacent image layers.

[0046] In some embodiments, the rotation axis is first determined based on the cross product of the axial vectors of the two image layers, expressed by the following formula: Its direction is perpendicular to the bending plane; then the rotation angle is determined based on the dot product of the axial vectors of the two image layers, expressed by the formula: Then, the bending rotation matrix was constructed using the Rodrigues rotation formula. Bending rotation matrix Describes the axial vector Rotate to The required rigid body rotation. This will then lead to the... radial vector of each image layer By bending rotation matrix Perform a rotational transformation to obtain the intermediate radial vector, which is expressed by the following formula: The intermediate radial vector represents the first value when only the bending effect is considered. The radial vector of the image layer in the first... The theoretical direction in the coordinate system of each image layer. At this point, the intermediate radial vector... and All located perpendicular to Within the same cross-section, the included angle between the two is the torsional angle that needs to be compensated, and its formula is expressed as follows: After obtaining the twist angle Subsequently, the spatial correction unit 13 constructs a vector around the axis using Rodriguez's rotation formula. Torsional compensation rotation matrix Finally, the principal bending rotation matrix is ​​multiplied by the torsional compensation rotation matrix to obtain the complete interlayer rotation matrix. The composite order of this matrix ensures that bending alignment is completed first, followed by torsional compensation of the radial basis vectors, thereby fully characterizing the spatial pose changes of blood vessels between adjacent image layers.

[0047] After obtaining the interlayer rotation matrix corresponding to the adjacent image layers, the spatial correction unit 13 performs curvature correction projection on each candidate feature node in the previous image layer.

[0048] In some embodiments, it is assumed that the first A candidate feature node in an image layer The coordinates are , No. The coordinates of the center point of each image layer are , No. The coordinates of the center point of each image layer are For candidate feature nodes The formula for calculating curvature-corrected projection is as follows: In the formula, For the first Candidate feature nodes in each image layer After curvature correction projection, at the first The coordinates in each image layer coordinate system represent the theoretical position that the candidate feature node should appear in the next image layer coordinate system, assuming that the node is attached to the blood vessel wall and moves synchronously with the macroscopic bending and twisting of the blood vessel.

[0049] Understandably, in this embodiment of the invention, the axial and radial vectors in the centerline guide frames of adjacent image layers are first extracted to characterize the blood flow direction and cross-sectional reference orientation of the blood vessel at that location, respectively. Then, the rotational component describing the main curvature direction of the blood vessel is calculated based on the axial vector, and the compensation component describing the axial torsion of the blood vessel is calculated based on the radial vector. These two components are then fused to construct a complete inter-layer rotation matrix. This matrix fully characterizes the combined effect of spatial curvature and axial torsion experienced by the blood vessel from the previous image layer to the current image layer. Based on this, the coordinates of candidate feature nodes in the previous image layer are rotated and spatially projected using this rotation matrix to obtain their curvature-corrected projection coordinates in the current image layer. The physical meaning of these coordinates is that they precisely represent the theoretical position that should appear if the candidate feature node were attached to the blood vessel wall and moved synchronously with it. Compared to traditional methods that directly track between layers based on absolute coordinates, this approach effectively eliminates the feature point position offset caused by the three-dimensional curvature and axial torsion of blood vessels by using dynamic correction based on the posture of the blood vessels themselves. This allows subsequent connection optimization to focus on the morphological continuity of the plaque itself rather than the macroscopic geometric changes of the blood vessels, thereby avoiding misjudging natural blood vessel curvature as plaque rupture or misconnecting spatially adjacent artifacts as continuous lesions. This provides a high-precision geometric correction benchmark for accurately reconstructing plaque entities that conform to anatomical rules.

[0050] The connection optimization unit 14 is used to construct a connection cost function based on the curvature correction projection coordinates and density attenuation coefficient of the candidate feature nodes, and to determine the optimal matching path between candidate feature nodes in each adjacent image layer.

[0051] As one possible implementation, for each image layer (excluding the first one), the Euclidean distance between the curvature-corrected projection coordinates and the coordinates of each candidate feature node in the current image layer is used as a geometric bias term. This geometric bias term reflects the natural continuity of the plaque's position during longitudinal extension; the smaller the distance, the more likely the two points spatially belong to the same plaque entity. The absolute difference between the density attenuation coefficients of candidate feature nodes in the previous image layer and those in the current image layer is used as an attribute difference term. This attribute difference term measures the longitudinal stability of the plaque's contents; a smaller attribute difference term value means that the plaque tissue properties change gently along the blood vessel direction, which is more consistent with the anatomical features of the actual lesion. Subsequently, the geometric bias term and the attribute difference term are fused to construct a connection cost function. This connection cost function is used to quantify the matching cost of any two candidate feature nodes in adjacent image layers belonging to the same plaque entity.

[0052] In some embodiments, the connection cost function is expressed as follows: In the formula, To select candidate feature nodes from the previous image layer With candidate feature nodes in the current image layer The matching cost required to determine that they belong to the same patch entity. These are candidate feature nodes from the previous image layer. These are candidate feature nodes in the current image layer. Candidate feature nodes Coordinates after curvature correction projection Candidate feature nodes coordinates Indicates calculation The Euclidean distance between them, i.e., the geometric deviation term; This is a dimensionless normalization coefficient, with a value that is the reciprocal of the average diameter of the entire blood vessel (e.g., ...). This is used to normalize the dimensions of attribute differences to the distance space, ensuring that the two are comparable in the cost function. Candidate feature nodes The density attenuation coefficient, Candidate feature nodes The density attenuation coefficient, Indicates calculation The absolute difference between them is the attribute difference item.

[0053] In practical applications, the number of candidate feature nodes in adjacent image layers is often unequal. This difference may stem from the natural growth or regression of patches in the vertical direction, or from fluctuations in image quality causing some nodes to go undetected. To address this imbalanced assignment problem, the connection optimization unit 14 constructs a network of size [missing information]. The augmented cost matrix, where It is the sum of the number of candidate feature nodes in the previous image layer and the number of candidate feature nodes in the current image layer.

[0054] In some embodiments, the augmented cost matrix is ​​specifically constructed by introducing the same number of annihilated virtual nodes as the number of candidate feature nodes in the current image layer along the row directions of the augmented cost matrix, with a total of [number missing] nodes along the row directions. For each node, a new virtual node is introduced along the column direction of the augmented cost matrix, with the same number of virtual nodes as the candidate feature nodes of the previous image layer, totaling [number] nodes along the column direction. There are 10 nodes. The matrix regions corresponding to each real node are filled with the connection cost function. The calculated matching cost is configured as follows: For the matching between real nodes in the previous image layer and newly generated virtual nodes in the column direction, a preset first connection cost (i.e., the extinction threshold, for example, twice the average voxel spacing) is used to quantify the cost of candidate feature nodes terminating (i.e., matching to newly generated virtual nodes) during the vertical extension process; For the matching between real nodes in the current image layer and extinct virtual nodes in the row direction, a preset second connection cost (i.e., the extinction threshold, for example, twice the average voxel spacing) is configured to quantify the cost of candidate feature nodes starting to be generated (i.e., matching to extinct virtual nodes) during the vertical extension process; The matching cost between virtual nodes is usually set to zero or a maximum value to avoid interfering with the optimization results.

[0055] Subsequently, the connection optimization unit 14 uses a preset matching optimization algorithm (such as the Hungarian algorithm or the Jonker-Volgenant algorithm) to solve the augmented cost matrix globally at the minimum cost, so as to efficiently find the matching relationship that minimizes the total connection cost, thereby determining the optimal matching path between candidate feature nodes between adjacent image layers. These matching relationships connect candidate feature nodes with high spatial continuity and attribute consistency, forming a series of vertical feature node connection chains spanning multiple layers.

[0056] Understandably, in this embodiment of the invention, a dual-constraint connection cost function containing a geometric deviation term and an attribute difference term is first constructed. The geometric deviation term, based on the Euclidean distance between the theoretical and measured positions after curvature correction projection, effectively eliminates the interference of macroscopic tortuosity and twisting of blood vessels on inter-layer feature localization, allowing the cost function to focus on the morphological continuity of the plaque itself rather than the geometric changes of the blood vessels. The attribute difference term, by quantifying the absolute difference in density attenuation coefficients between candidate feature nodes in adjacent layers, incorporates the compositional stability of the plaque contents into the matching decision, ensuring that vertically connected nodes have similar physical properties. Furthermore, to address the unbalanced assignment problem where the number of candidate feature nodes in adjacent layers is unequal in practical applications, a virtual node augmentation mechanism is introduced. This involves constructing a square augmentation cost matrix with the sum of the number of nodes in the two layers as its side length. In the row direction, an equal number of extinct virtual nodes as the current layer's nodes are introduced; in the column direction, an equal number of newly generated virtual nodes as the previous layer's nodes are introduced. Preset extinction and generation thresholds are configured for the matching between real and virtual nodes, respectively, thereby transforming the unbalanced assignment problem into a standard optimal matching problem. This mechanism not only handles changes in the number of nodes caused by the natural growth or regression of patches in the longitudinal direction, but also effectively addresses missed node detections due to image quality fluctuations, avoiding artifacts introduced by forced connections or disconnections due to mismatched node numbers. Finally, a pre-defined combinatorial optimization algorithm is used to solve the augmented cost matrix at the global minimum cost. The resulting optimal matching path achieves optimality in both spatial continuity and physical property consistency, significantly improving the accuracy and robustness of inter-layer feature matching. This lays a solid foundation for subsequently reconstructing patch entities that conform to real anatomical structures.

[0057] The entity reconstruction unit 15 is used to reconstruct the target patch entity according to the optimal matching path and output the spatial index range of the target patch entity in the diastolic image.

[0058] As one possible implementation, the entity reconstruction unit 15 first receives the optimal matching path output by the connection optimization unit 14. This path contains the globally optimal matching relationship between candidate feature nodes between adjacent image layers, i.e., which nodes are vertically connected to form a potential continuous structure. The candidate feature nodes that match each other between adjacent image layers are then vertically concatenated to form multiple vertical feature chains spanning multiple image layers. Each vertical feature chain consists of a series of spatially continuous candidate feature nodes with consistent attributes, representing the vertical extension trajectory of the potential patch structure.

[0059] To eliminate short interference chains caused by image noise, artifacts, or isolated high gradient points, the entity reconstruction unit 15 performs length filtering on each longitudinal feature chain.

[0060] In some embodiments, the entity reconstruction unit 15 counts the number of consecutive image layers spanned by each longitudinal feature chain and compares this number with a preset layer threshold. This preset layer threshold can be set based on clinical experience or image resolution, for example, to 3 layers (corresponding to a physical length of approximately 1.5 mm) to eliminate random noise chains that are too short. Only when the number of consecutive image layers spanned by a longitudinal feature chain is greater than the preset layer threshold is it identified as a target plaque entity.

[0061] For each identified target patch entity, the entity reconstruction unit 15 further extracts its spatial positioning information to form the spatial index range of the entity.

[0062] In some embodiments, in the longitudinal dimension, the starting image layer index and the ending image layer index covered by the target plaque entity are extracted to determine its start and end positions in the longitudinal direction of the blood vessel; in the cross-sectional circumferential direction, based on the radial angles of all candidate feature nodes contained in the target plaque entity in each image layer, its minimum angle index and maximum angle index in the cross-section are determined to determine its coverage range on the circumference of the vessel wall.

[0063] It should be noted that the aforementioned starting image layer index, ending image layer index, minimum angle index, and maximum angle index together constitute the spatial index range of the target patch entity in the diastolic image. This index range does not contain specific pixel data, but rather serves as a lightweight anatomical localization pointer, capable of precisely locating the spatial region where the patch is situated.

[0064] Simultaneously, the entity reconstruction unit 15 calculates the arithmetic mean of the density decay coefficients of all candidate feature nodes contained in the target patch entity, and uses this mean as the diastolic baseline gradient parameter of the target patch entity, storing it in association with the spatial index range. This diastolic baseline gradient parameter represents the average interface sharpness of the patch in the diastolic state, and is a key benchmark data for subsequent dual-temporal dynamic mechanical verification.

[0065] Finally, the entity reconstruction unit 15 outputs the spatial index range of the target patch entity and the associated stored diastolic baseline gradient parameters.

[0066] Understandably, in this embodiment of the invention, firstly, based on the optimal matching path output by the connection optimization unit, candidate feature nodes that match each other between adjacent image layers are vertically concatenated to form a vertical feature chain spanning multiple layers, thereby reconstructing the discrete matching relationship between layers into a potential structure with a clear vertical extension trajectory. Based on this, the length of the vertical feature chain is filtered by a preset layer threshold, effectively eliminating short interference chains formed by image noise, isolated high-gradient points, or artifacts, ensuring that the structure identified as the target patch entity has sufficient anatomical continuity in the vertical direction, avoiding misjudging random noise as a lesion structure. For each target patch entity that passes the screening, the scheme further extracts its starting and ending image layer indices in the vertical direction and its minimum and maximum angle indices in the circumferential direction of the cross-section. These four indices together constitute a lightweight spatial index range, which does not depend on specific pixel data and can accurately locate the anatomical region where the patch is located, providing precise targeted input for subsequent inter-period mapping or dynamic analysis. Simultaneously, the scheme calculates and associates the average density attenuation coefficient of all candidate feature nodes within the target patch entity as a diastolic baseline gradient parameter. This parameter quantitatively characterizes the average interface sharpness of the patch during diastole, providing a crucial comparative benchmark for subsequent dual-temporal dynamic mechanical verification. Through these techniques, not only is the reconstruction from discrete feature nodes to continuous patch entities achieved, but a standardized patch description with both spatial positioning accuracy and physical property characterization capabilities is also output, significantly improving the system's accuracy in identifying patch structures and the comparability of subsequent analyses.

[0067] To assess the vulnerability of plaques, please refer to [link / reference]. Figure 2 It shows a structural schematic diagram of a fragility assessment unit 16 provided in an embodiment of the present invention, combined with Figure 1 ,like Figure 2 As shown, the vulnerability assessment unit 16 includes an inter-period mapping module 161 and an assessment module 162.

[0068] The interphase mapping module 161 is used to accurately map the target plaque entity determined during diastole to the image data during systole, so as to lock the corresponding position of the same anatomical region during systole.

[0069] The interphase mapping module 161 first identifies anatomical landmarks of the vascular branch where the target plaque is located in the diastolic image data. These anatomical landmarks specifically include the proximal and distal bifurcation points of the vascular segment. These two bifurcation points, as natural anatomical landmarks, are relatively stable in position during the cardiac cycle and are suitable for establishing interphase mapping relationships.

[0070] Subsequently, the inter-period mapping module 161 calculates the distances from the center point of the target plaque entity along the vessel centerline to the proximal bifurcation point and the distal bifurcation point, respectively, denoted as . and Based on these two distances, a normalized position coefficient is generated. Its formula is expressed as Normalized position coefficients It is used to characterize the relative longitudinal position of the center point of the target plaque with respect to the proximal and distal bifurcation points. Its value ranges from 0 to 1 and is independent of the absolute length of the blood vessel. Therefore, it is invariant to the longitudinal expansion and contraction of the blood vessel caused by heartbeat.

[0071] Next, the interphase mapping module 161 reads the systolic image data of the coronary artery and locates the same vascular branch as in the diastolic phase, identifying the proximal and distal bifurcation points of this vascular branch during systole. This is done based on the same normalized location coefficient. The position of the target plaque's solid center point in the systolic image is calculated, that is, the distance moved distally along the vessel centerline from the proximal bifurcation point during systole. Multiply the length of the vessel segment during systole by the total length of the segment, and the resulting location is the corresponding center point during systole.

[0072] Based on the location of the center point and the spatial index range of the target plaque entity (including the longitudinal length and circumferential angle range), the interphase mapping module 161 determines the plaque matching region in the systolic image data corresponding to the target plaque entity. This region includes the image layer range and circumferential angle range corresponding to the anatomical location of the diastolic plaque entity, providing precise targeting for subsequent systolic feature extraction.

[0073] After completing the mapping of the patch matching region, the evaluation module 162 initiates the dynamic mechanical verification process. The evaluation module 162 first calls the feature extraction and screening unit 12 to perform the same radial sampling, feature fitting, and candidate feature node screening operations as in the diastolic phase within the systolic patch matching region determined by the inter-phase mapping module 161, thereby obtaining all candidate feature nodes and their corresponding density attenuation coefficients within the region.

[0074] Evaluation module 162 calculates the arithmetic mean of the density attenuation coefficients of all candidate feature nodes within the patch matching region during the contraction period, and uses this as the density attenuation parameter during the contraction period, denoted as... Simultaneously, the evaluation module 162 obtains the associated stored diastolic baseline gradient parameters from the entity reconstruction unit 15, denoted as... .

[0075] Furthermore, the evaluation module 162 determines the change in density decay coefficient based on the diastolic baseline gradient parameter and the systolic density decay parameter, denoted as Its formula is expressed as Change in density attenuation coefficient It reflects the degree of change in the sharpness of the micro-interface of plaque contents due to changes in luminal pressure during the cardiac cycle.

[0076] Simultaneously, the evaluation module 162 obtains the average lumen radius of the target plaque entity in the diastolic image data, denoted as... And the average lumen radius corresponding to the patch matching region in the contraction phase image data, denoted as The average lumen radius can be calculated from the lumen boundary position parameters obtained during feature extraction.

[0077] Furthermore, the evaluation module 162 assesses the target plaque entity based on the average diastolic lumen radius. and the average lumen radius during the contraction period corresponding to the plaque matching region. Determine the radial strain of the cavity, denoted as . Its formula is expressed as , For a very small positive number, an empirical value of 0.0001 can be taken to avoid a denominator of zero, and the radial strain of the tube is... It reflects the relative change in the cross-sectional dimensions of the lumen from the diastolic to the systolic phase.

[0078] Subsequently, the evaluation module 162 assesses the change in the density attenuation coefficient. radial strain of the lumen Calculate the gradient strain response characteristics Its formula is expressed as , For a very small positive number, an empirical value of 0.0001 can be taken to avoid a denominator of zero, thus reflecting the gradient strain response characteristics. The dynamic response intensity of the plaque interface gradient characteristics under unit lumen deformation is quantified, which is a key indicator for distinguishing vulnerable plaques from stable plaques.

[0079] Finally, the evaluation module 162 will calculate the gradient strain response characteristics. With preset vulnerability threshold ( For example, an empirical value of 1.5 can be used for comparison to generate a fragility assessment result. This indicates that the target plaque underwent a significant change in interfacial gradient under unit lumen deformation, consistent with the characteristics of fluid redistribution in the lipid core under pressure. Therefore, the vulnerability assessment results indicate that the target plaque is a vulnerable plaque; conversely, if... This indicates that the plaque material has high rigidity, and the vulnerability assessment results indicate that the target plaque is a stable plaque.

[0080] In the coronary CTA image analysis system for plaque vulnerability assessment provided in this embodiment of the invention, the system first utilizes the natural anatomical landmark of the vascular bifurcation point through the interphase mapping unit to construct a longitudinal scaling mapping relationship based on normalized position coefficients. This accurately locks the reconstructed target plaque entity in the diastolic image to the corresponding region in the systolic image data. This mapping mechanism effectively overcomes the interference of nonlinear vascular deformation and rigid displacement caused by cardiac pulsation on interphase localization, ensuring that the plaque matching region analyzed in the systolic phase strictly corresponds to the target plaque entity in the diastolic phase in anatomical position, providing an accurate spatial benchmark for subsequent dynamic comparative analysis. Based on this, the vulnerability assessment unit further calls the feature extraction and screening unit to perform the same microscopic feature extraction operation within the plaque matching region in the systolic phase, obtaining the systolic density attenuation parameter, and comparing it with the diastolic benchmark gradient parameter to obtain the change in density attenuation coefficient. Simultaneously, the radial strain of the lumen is calculated based on the average lumen radius in the diastolic and systolic phases. Finally, the microscopic gradient change is fused with the macroscopic geometric variables to construct the gradient strain response characteristic, a core evaluation index. This index fully utilizes the natural changes in luminal pressure during the cardiac cycle, transforming the differences in physical response of different tissue components under dynamic stress into quantifiable values: the liquid lipid core undergoes fluid redistribution under pressure, leading to a significant change in the interfacial gradient, thus exhibiting a high response ratio; while dense fibrous tissue has greater stiffness, relatively stable interfacial characteristics, and a lower response ratio. By comparing the gradient strain response characteristics with a preset vulnerability threshold, this scheme can effectively distinguish tissue components with similar CT values ​​but vastly different mechanical properties, achieving specific confirmation of high-risk vulnerable plaques that are morphologically concealed but mechanically unstable, significantly improving the accuracy and reliability of plaque vulnerability assessment.

[0081] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A coronary CTA image analysis system for assessing plaque vulnerability, characterized in that, include: The frame construction unit is used to extract the centerline of the target vessel based on the diastolic image data of the coronary artery, determine multiple image layers along the centerline, and construct the centerline guide frame corresponding to the center point of each image layer. The feature extraction and filtering unit is used to perform radial sampling and feature fitting on the diastolic image data based on the centerline guide frame corresponding to each image layer, to obtain a discrete tube wall feature set including the coordinates, density attenuation coefficient, and high-density barrier marker of each feature node in each image layer, and to filter candidate feature nodes from the discrete tube wall feature set according to the density attenuation coefficient and the high-density barrier marker. The density attenuation coefficient is used to characterize the micro-gradient parameter of the tube wall interface sharpness corresponding to the feature node, and the high-density barrier marker is used to indicate whether there is calcified tissue on the radial sampling path corresponding to the feature node. The spatial correction unit is used to determine the spatial transformation relationship between adjacent image layers based on the centerline guide frame corresponding to each image layer in each adjacent image layer, and to map the candidate feature nodes in the previous image layer to the current image layer to obtain curvature correction projection coordinates. The connection optimization unit is used to construct a connection cost function based on the curvature correction projection coordinates of the candidate feature nodes and the density attenuation coefficient, and to determine the optimal matching path between candidate feature nodes in each adjacent image layer. The entity reconstruction unit is used to reconstruct the target patch entity according to the optimal matching path and output the spatial index range of the target patch entity in the diastolic image. A vulnerability assessment unit is used to generate vulnerability assessment results for the target patch entity.

2. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 1, characterized in that, The standard frame construction unit is specifically used for: The target vessel is segmented from the diastolic image data, and the centerline of the target vessel is determined. Resampling is performed along the center line at preset length intervals to determine multiple ordered center points, each center point corresponding to an image layer; For each image layer, a local orthogonal coordinate system is established with the coordinates of the center point corresponding to the image layer as the origin, the tangent of the center line as the axial vector, and the preset reference direction within the cross section as the radial vector, thus forming the center line guide frame.

3. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 1, characterized in that, The feature extraction and filtering unit is specifically used for: For each image layer, with the center point of the image layer as the origin, multiple radial sampling rays are generated at preset angle intervals within a cross section perpendicular to the axial vector; For each radial sampling ray, three-dimensional linear interpolation sampling is performed along the radial sampling ray with a preset step size to obtain a continuous gray value sequence from the center of the lumen to the outer wall of the blood vessel. The grayscale numerical sequence is fitted using a preset single exponential decay model, and the decay constant obtained from the fitting is determined as the density decay coefficient. Based on the position offset parameters obtained by fitting the single exponential decay model, the coordinates of the center point of the image layer, and the unit direction vector corresponding to the radial sampling ray, the coordinates of the feature node corresponding to the radial sampling ray are determined, and the feature node is the pipe wall boundary point corresponding to the radial sampling ray. The high-density barrier marker corresponding to the radial sampling ray is determined based on the gray value sequence corresponding to the radial sampling ray and the preset calcification threshold. The coordinates of the feature nodes corresponding to each radial sampling ray in each image layer, the density attenuation coefficient, and the high-density barrier identifier are associated and stored to form the discrete pipe wall feature set.

4. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 3, characterized in that, Candidate feature nodes are selected from the discrete pipe wall feature set based on the density attenuation coefficient and the high-density barrier identifier, including: The attenuation screening threshold is determined based on the distribution of the density attenuation coefficients of all feature nodes in the discrete pipe wall feature set. Feature nodes whose density attenuation coefficient is greater than the attenuation screening threshold and whose high-density barrier marker indicates that there is no calcified tissue on the radial sampling path are identified as candidate feature nodes.

5. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 1, characterized in that, The spatial correction unit is specifically used for: For each group of adjacent image layers, extract the axial and radial vectors from the centerline guide frame corresponding to each image layer in the adjacent image layers; Based on the axial and radial vectors in the centerline guide frame corresponding to each of the adjacent image layers, an interlayer rotation matrix is ​​determined to characterize the tortuosity and twisting of blood vessels between adjacent image layers, and the interlayer rotation matrix is ​​determined as the spatial transformation relationship corresponding to the adjacent image layers; For each image layer other than the first image layer, the coordinates of the candidate feature nodes in the previous image layer are rotated and spatially projected by the inter-layer rotation matrix to obtain the curvature-corrected projection coordinates of the candidate feature nodes in the current image layer. The curvature-corrected projection coordinates represent the theoretical projection position of the candidate feature nodes as they bend and twist synchronously with the blood vessels.

6. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 1, characterized in that, The connection optimization unit is specifically used for: For each image layer other than the first image layer, the Euclidean distance between the curvature-corrected projection coordinates and the coordinates of each candidate feature node in the current image layer is used as a geometric deviation term. The absolute difference between the density decay coefficients of candidate feature nodes in the previous image layer and candidate feature nodes in the current image layer is used as the attribute difference term. Based on the geometric deviation term and the attribute difference term, the connection cost function is constructed. The connection cost function is used to quantify the matching cost of any two candidate feature nodes in adjacent image layers belonging to the same patch entity.

7. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 6, characterized in that, The connection optimization unit is specifically used for: Construct an augmented cost matrix based on the number of candidate feature nodes in the previous image layer and the number of candidate feature nodes in the current image layer; In the row direction of the augmented cost matrix, the same number of annihilated virtual nodes as the current image layer candidate feature nodes are introduced, and in the column direction of the augmented cost matrix, the same number of newly generated virtual nodes as the previous image layer candidate feature nodes are introduced. A first connection cost is configured for the extinct virtual node, and a second connection cost is configured for the newly generated virtual node; the first connection cost is a preset extinction threshold, used to quantify the cost of candidate feature nodes terminating during the vertical extension process; The second connection cost is a preset new generation threshold, used to quantify the cost of candidate feature nodes starting to generate new nodes during the vertical extension process; In the augmented cost matrix, the matching cost between real nodes is determined according to the connection cost function, the matching cost between real nodes of the previous image layer and newly generated virtual nodes in the column direction is determined according to the first connection cost, and the matching cost between real nodes of the current image layer and annihilated virtual nodes in the row direction is determined according to the second connection cost. A preset matching optimization algorithm is used to find the candidate feature node matching relationship with the minimum total connection cost corresponding to the augmented cost matrix, thereby forming the optimal matching path.

8. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 1, characterized in that, The entity reconstruction unit is specifically used for: Based on the optimal matching path, candidate feature nodes matched between adjacent image layers are vertically concatenated to form multiple vertical feature chains; The vertical feature chain in which the number of consecutive image layers is greater than a preset layer number threshold is identified as the target patch entity; Extract the starting and ending image layer indices of the target patch entity in the longitudinal direction, and the minimum and maximum angle indices in the circumferential direction of the cross section. The starting image layer index, the ending image layer index, the minimum angle index, and the maximum angle index constitute the spatial index range of the target patch entity. The average density decay coefficient of all candidate feature nodes included in the target patch entity is associated and stored as the diastolic baseline gradient parameter of the target patch entity.

9. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 8, characterized in that, The vulnerability assessment unit includes a period mapping module; The inter-period mapping module is used to identify the proximal and distal bifurcation points of the vascular branch where the target plaque entity is located; Based on the distances from the center point of the target plaque along the blood vessel centerline to the proximal bifurcation point and the distal bifurcation point, a normalized position coefficient is generated. Locate the corresponding bifurcation structures of the same vascular branch in systolic image data; The position corresponding to the center point of the target patch entity in the contraction period image data is determined based on the normalized position coefficient, and the patch matching region in the contraction period image data is determined based on the spatial index range of the target patch entity.

10. The coronary CTA image analysis system for plaque vulnerability assessment according to claim 9, characterized in that, The vulnerability assessment unit also includes an assessment module; The feature extraction and filtering unit is also used to determine the candidate feature nodes and the corresponding density attenuation coefficients within the patch matching area; The evaluation module is used to determine the average density decay coefficient of all candidate feature nodes within the patch matching area as the shrinkage period density decay parameter. The change in density decay coefficient is determined based on the diastolic baseline gradient parameter and the systolic density decay parameter. The radial strain of the lumen is determined based on the average lumen radius during the diastolic phase corresponding to the target plaque entity and the average lumen radius during the systolic phase corresponding to the plaque matching region. The gradient strain response characteristics are determined based on the change in the density attenuation coefficient and the radial strain of the lumen. Based on the gradient strain response characteristics and the preset vulnerability threshold, a vulnerability assessment result for the target patch entity is generated. If the gradient strain response characteristics are greater than the preset vulnerability threshold, the vulnerability assessment result is used to indicate that the target patch entity is a vulnerable patch.

Images