Blood vessel plaque intelligent identification method and system based on optical coherence tomography

Abstract

The invention discloses a blood vessel plaque intelligent identification method and system based on optical coherence tomography, and relates to the technical field of medical image processing. Firstly, a three-dimensional blood vessel grid model is constructed through OCT volume data, vertex curvature parameters are quantified to generate a dynamic threshold template, the segmentation process is made to adapt to the local form of a blood vessel, and the problem of wrong segmentation of a traditional algorithm in a complex anatomical region is solved. Secondly, introducing a fractal dimension verification mechanism, quantitatively evaluating an initial segmentation result based on edge complexity, and when fractal features deviate from a preset threshold value, dynamically adjusting curvature weight parameters and iteratively optimizing segmentation, forming'segmentation-verification-correction 'closed-loop control, and remarkably suppressing artifact interference; and the finally output three-dimensional identification result comprises plaque space coordinates and morphological characteristics, and a visual decision basis is provided for clinic.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to an intelligent vascular plaque recognition method and system based on optical coherence tomography. Background Art

[0002] In the clinical diagnosis of cardiovascular disease, accurate identification of atherosclerotic plaques is key to assessing the degree of vascular stenosis and predicting acute cardiovascular events (such as plaque rupture and thrombosis). Optical coherence tomography (OCT) technology has become one of the "gold standards" for detecting intravascular lesions due to its high resolution (micrometer level) and deep tissue imaging capabilities. However, the high noise characteristics of OCT images, the complexity of plaque morphology (such as heterogeneous structures such as lipid core, calcified areas, and fibrous caps), and the limitations of traditional image processing technology have led to the following technical bottlenecks in existing methods for intelligent plaque identification:

[0003] Traditional methods often rely on grayscale thresholds or texture features for plaque segmentation. However, due to the high grayscale overlap between the physiological structures of the vessel wall (such as the media-adventitia interface) and the plaque area, fixed thresholds can easily lead to over-segmentation (misidentifying normal tissue as plaque) or under-segmentation (missing low-contrast plaque). For example, calcified plaques are easily detected due to their high reflectivity, but the lipid core is often overlooked due to signal attenuation, resulting in a high false-negative rate. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an intelligent vascular plaque identification method and system based on optical coherence tomography.

[0005] In order to achieve the above object, the technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention discloses a method for intelligently identifying vascular plaques based on optical coherence tomography, comprising the following steps:

[0007] Obtaining blood vessel OCT volume data collected by an optical coherence tomography device, wherein the OCT volume data includes three-dimensional spatial coordinates and corresponding grayscale values;

[0008] extracting point cloud data of the vascular wall surface based on the grayscale gradient distribution of the OCT volume data and generating a three-dimensional vascular mesh model;

[0009] Calculating the three-dimensional curvature parameters of each vertex in the three-dimensional blood vessel mesh model to generate a curvature distribution matrix;

[0010] Establishing a mapping relationship between a curvature weight map and a dynamic threshold template according to the numerical distribution range of the curvature distribution matrix;

[0011] Adopting the dynamic threshold template to perform adaptive region growing segmentation on the OCT volume data to obtain initial plaque contour data;

[0012] Extracting edge fractal dimension feature values ​​of the initial plaque contour data to generate a fractal dimension verification result;

[0013] When the fractal dimension verification result does not reach the preset fractal dimension threshold, dynamically correcting the adjustment coefficient of the curvature weight map and regenerating the dynamic threshold template according to the deviation between the current edge fractal dimension feature value and the preset fractal dimension threshold;

[0014] Repeat the adaptive region growing segmentation and fractal dimension verification steps until the preset conditions are met;

[0015] The output includes three-dimensional recognition results of effective patch boundary coordinates and morphological features.

[0016] In a second aspect, the present invention discloses an intelligent vascular plaque identification system based on optical coherence tomography, which implements the above-mentioned intelligent vascular plaque identification method based on optical coherence tomography, including:

[0017] An OCT data acquisition and preprocessing module is used to obtain vascular OCT volume data collected by an optical coherence tomography device, and based on the grayscale gradient distribution of the OCT volume data, extract the vascular wall surface point cloud data to generate a three-dimensional vascular mesh model;

[0018] a three-dimensional curvature feature calculation module, configured to calculate the three-dimensional curvature parameters of each vertex in the three-dimensional vascular mesh model, generate a curvature distribution matrix, and establish a mapping relationship between a curvature weight map and a dynamic threshold template based on the numerical distribution range of the curvature distribution matrix;

[0019] A dynamic segmentation module, configured to perform adaptive region growing segmentation on the OCT volume data using the dynamic threshold template to obtain initial plaque contour data;

[0020] A fractal dimension verification module, configured to extract edge fractal dimension eigenvalues ​​of the initial plaque contour data and generate a fractal dimension verification result;

[0021] a feedback control module, configured to dynamically correct the adjustment coefficient of the curvature weight map and regenerate a dynamic threshold template based on the deviation between the current edge fractal dimension eigenvalue and the preset fractal dimension threshold when the fractal dimension verification result does not reach the preset fractal dimension threshold;

[0022] The iterative management and result output module repeatedly executes the adaptive region growing segmentation and fractal dimension verification steps until the preset conditions are met, and outputs the three-dimensional recognition results including the valid patch boundary coordinates and morphological features.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. By calculating the 3D curvature parameters of each vertex in the 3D vascular mesh model, a curvature distribution matrix is ​​generated, and a mapping relationship between the curvature weight map and the dynamic threshold template is established. This allows the segmentation process to fully consider the local geometric characteristics of the blood vessels, especially in complex structures such as bifurcations and bends, and can more accurately identify plaque boundaries and avoid misjudgments caused by changes in vascular structure;

[0025] 2. If the fractal dimension verification result fails to reach the preset threshold, the curvature weight map adjustment coefficient is dynamically modified and a dynamic threshold template is regenerated based on the deviation between the current edge fractal dimension eigenvalue and the preset fractal dimension threshold. This dynamic adjustment mechanism enables the method to adapt to different patient vascular conditions and varying OCT image quality. Even in the presence of interference factors such as noise and artifacts, the segmentation results can be continuously optimized through iteration, improving the stability and reliability of recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0027] Figure 1 is a flow chart of the method of the present invention;

[0028] Figure 2 A flow chart for updating the adjustment coefficient of the curvature weight map of the present invention;

[0029] Figure 3 A flow chart for establishing a dynamic threshold template of the present invention;

[0030] Figure 4 is a flow chart for calculating the three-dimensional curvature parameters of the present invention;

[0031] Figure 5 A flow chart for generating a curvature weight map of the present invention;

[0032] Figure 6 This is a flow chart of extracting edge fractal dimension feature values ​​according to the present invention;

[0033] Figure 7 This is a flowchart of the process when the fractal dimension verification of the present invention fails to meet the standards;

[0034] Figure 8 It is a diagram of the system module composition of the present invention;

[0035] Figure 9 This is a data flow chart of the system of the present invention. DETAILED DESCRIPTION

[0036] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0037] Application Overview:

[0038] Traditional methods rely on grayscale thresholds or texture features for plaque segmentation. However, due to the high grayscale overlap between the physiological structure of the vessel wall and the plaque region, fixed thresholds can easily lead to over-segmentation or under-segmentation. For example, calcified plaques are easily detected due to their high reflectivity, but the lipid core is often overlooked due to signal attenuation, resulting in a high false-negative rate. When dealing with complex plaques containing heterogeneous structures, existing technologies struggle to balance segmentation accuracy and robustness, and are particularly prone to missed detections or misclassifications in low-contrast areas.

[0039] To address these issues, a method is needed that can adaptively adjust segmentation parameters and verify segmentation results. Taking into account the morphological differences between vessel walls and plaques, three-dimensional curvature features can be used to characterize local structural properties. Combined with a dynamic threshold generation mechanism, this approach overcomes the limitations of fixed thresholds. Furthermore, fractal dimension verification can quantitatively assess the complexity of segmented edges. Parameter optimization can be achieved through a feedback adjustment mechanism, thereby improving the reliability and accuracy of the segmentation process.

[0040] Therefore, if Figure 1 As shown, this application proposes the following technical solutions:

[0041] The intelligent identification method of vascular plaque based on optical coherence tomography includes the following steps:

[0042] Obtain blood vessel OCT volume data collected by an optical coherence tomography device, wherein the OCT volume data includes three-dimensional spatial coordinates and corresponding grayscale values.

[0043] Based on the grayscale gradient distribution of the OCT volume data, point cloud data of the blood vessel wall surface is extracted to generate a three-dimensional blood vessel mesh model.

[0044] Calculate the three-dimensional curvature parameters of each vertex in the three-dimensional vascular mesh model to generate a curvature distribution matrix. The three-dimensional curvature parameters refer to the mean curvature and Gaussian curvature calculated by fitting local surface equations. Specifically, they can be achieved by fitting a quadratic surface to the local surface patches around the vertices using the least squares method. They are used to characterize the concave and convex characteristics of the vascular wall surface.

[0045] According to the numerical distribution range of the curvature distribution matrix, a mapping relationship between the curvature weight map and the dynamic threshold template is established. The curvature weight map refers to a continuous weight distribution map generated after normalizing the Gaussian curvature. Specifically, it can be achieved by smoothing the curvature values ​​of adjacent vertices using bilinear interpolation. It is used to quantify the impact of curvature differences in different areas on the segmentation threshold. The dynamic threshold template refers to a spatially adaptive segmentation threshold that combines the basic grayscale threshold and the curvature weight offset. Specifically, it can be achieved by multiplying the curvature weight value with the adjustment coefficient to generate an offset, and then superimposing it on the basic threshold. It is used to adapt to the grayscale differences in different curvature areas.

[0046] The dynamic threshold template is used to perform adaptive region growing segmentation on the OCT volume data to obtain initial plaque contour data.

[0047] Extract edge fractal dimension eigenvalues ​​from the initial patch contour data to generate a fractal dimension verification result. This fractal dimension verification result measures the complexity of the contour edge calculated using multi-resolution sampling. Specifically, this can be achieved using logarithmic scale contour length slope analysis to determine whether the segmented edge conforms to the morphological characteristics of the actual patch.

[0048] When the fractal dimension verification result does not reach the preset fractal dimension threshold, the adjustment coefficient of the curvature weight map is dynamically corrected and a dynamic threshold template is regenerated according to the deviation between the current edge fractal dimension feature value and the preset fractal dimension threshold.

[0049] Repeat the adaptive region growing segmentation and fractal dimension verification steps until the preset conditions are met.

[0050] The output includes three-dimensional recognition results of effective patch boundary coordinates and morphological features.

[0051] Specifically, the surface point cloud of the vascular wall is first extracted through the grayscale gradient information of the OCT volume data, and a three-dimensional mesh model is constructed to characterize the vascular morphology. Subsequently, the Gaussian curvature and mean curvature of each vertex are calculated, and the high curvature area is screened to generate a curvature weight map. Based on the weight map and the dynamic adjustment coefficient, a spatially related segmentation threshold is generated to guide the region growing algorithm to preferentially grow in areas where the curvature and grayscale features match. After the segmentation is completed, the fractal dimension analysis is performed on the contour edge. If the characteristic value does not meet the standard, the weight distribution of the curvature weight map is updated by the adjustment coefficient, and the threshold template is regenerated for iterative optimization. This process continues until the fractal dimension difference converges or the maximum number of iterations is reached, and finally a three-dimensional plaque recognition result that conforms to the morphological characteristics is output.

[0052] Compared to existing technologies, traditional methods rely on fixed thresholds or single features, resulting in limited segmentation accuracy. This solution, however, implements a spatially adaptive segmentation strategy by mapping curvature weights to dynamic thresholds. Combined with a feedback adjustment mechanism based on fractal dimensions, it effectively distinguishes between overlapping grayscale areas of the vessel wall and plaque. Furthermore, the iterative optimization process avoids the blindness of manual parameter adjustments, significantly improving the recognition rate of low-contrast plaques.

[0053] Through the above technical solution, this application solves the over-segmentation and under-segmentation problems caused by grayscale overlap in traditional segmentation methods, reducing the risk of missed detection of low-reflection areas such as lipid cores. Through the synergistic effect of dynamic thresholding and fractal validation, it achieves accurate identification of complex plaque structures, providing a more reliable three-dimensional morphological analysis basis for clinical diagnosis.

[0054] like Figure 2 As shown in the figure, it is a flow chart for updating the adjustment coefficient of the curvature weight map of the present application; the present application further proposes a specific method for dynamically correcting the curvature weight map when the fractal dimension verification result does not meet the standard: calculate the ratio k = N / of the current edge fractal dimension eigenvalue N and the preset fractal dimension threshold M. M When the ratio is less than 1, an adjustment coefficient increment Δα=β*(1-k) positively correlated with the degree of deviation is generated, and the increment is superimposed on the original adjustment coefficient of the curvature weight map to complete the update, that is, α'=α+Δα.

[0055] The current edge fractal dimension eigenvalue, N, is a quantitative indicator derived from multiscale contour analysis. Specifically, it can be implemented as the slope of the linear regression between contour length and sampling scale in a logarithmic coordinate system. It characterizes the complexity of the patch edge. This eigenvalue reflects the detail richness of the patch morphology and serves as an evaluation metric for segmentation accuracy during the verification phase.

[0056] The preset fractal dimension threshold M is a pre-set standard value for the fractal dimension. This value can be determined through clinical data statistics or expert experience and is used to determine whether the plaque segmentation results meet morphological requirements. This threshold serves as the target parameter for iterative optimization, guiding the dynamic adjustment of the curvature weight map.

[0057] The adjustment coefficient increment Δα is a compensation amount based on the degree of deviation from the fractal dimension. Specifically, it is generated by multiplying the scale factor by the deviation. It is used to adjust the influence of the curvature weight on the dynamic threshold template. This increment achieves adaptive optimization of the segmentation parameters through a linear compensation mechanism.

[0058] Specifically, during each iteration, the ratio of the fractal dimension eigenvalue of the current patch outline to the preset threshold is first calculated. When the ratio is less than 1, it indicates that the segmentation result has missing edge details or is over-smoothed. At this time, an increment is generated according to the degree of deviation and superimposed on the original adjustment coefficient. For example, when the fractal dimension eigenvalue is only 80% of the preset threshold, the adjustment coefficient will increase the compensation amount corresponding to the 20% deviation. Through this mechanism, the weight of the high curvature area in the curvature weight map is enhanced, prompting the dynamic threshold template to give priority to morphologically complex areas in subsequent iterations, thereby improving the segmentation accuracy of the patch edge.

[0059] Compared to existing technologies, traditional methods typically use fixed compensation coefficients or manually adjust parameters empirically, making them incapable of addressing the dynamic changes in plaque type and imaging quality. For example, the fractal dimensions of calcified and lipid plaques differ significantly, but existing methods are unable to automatically adapt parameters, resulting in fluctuations in segmentation performance. This approach establishes a closed-loop feedback mechanism by quantifying the mathematical relationship between fractal dimension deviation and adjustment coefficients, making parameter adjustment more goal-oriented and computationally reproducible.

[0060] Through the above technical solution, this application can dynamically optimize segmentation parameters based on the actual morphological characteristics of the plaque, effectively resolving the over-segmentation or under-segmentation issues caused by fixed adjustment coefficients in traditional methods. While maintaining algorithm stability, this solution significantly improves the segmentation accuracy of heterogeneous plaques, demonstrating greater adaptability when dealing with low-contrast lipid cores and high-curvature fibrous cap edges.

[0061] like Figure 3 As shown, this is a flow chart for establishing a dynamic threshold template of the present application; the present application further proposes that the establishment of a dynamic threshold template includes: setting a basic grayscale threshold as a preset value; generating a threshold offset based on the product of the curvature weight value and the current adjustment coefficient α; adding the basic grayscale threshold and the threshold offset to generate a dynamic segmentation threshold related to the spatial position; wherein the current adjustment coefficient is updated according to the fractal dimension verification result at each iteration.

[0062] The base grayscale threshold is a preset baseline value for initial image segmentation. It can be set empirically or through statistical analysis, for example, based on OCT device imaging parameters or the median of historical data distribution. This threshold serves as the initial baseline for the segmentation process and controls the baseline sensitivity of region growing.

[0063] The curvature weight is a quantitative indicator of the complexity of the local geometric structure of the vessel wall. It can be obtained by normalizing the absolute value of Gaussian curvature, for example, by mapping the absolute value of vertex Gaussian curvature to the interval [0, 1]. The curvature weight is used to identify areas of plaque on the vessel wall surface. Areas of high curvature typically correspond to plaque edges or locations of structural abrupt changes.

[0064] The threshold offset is a dynamically generated threshold adjustment based on the curvature weight and the current adjustment coefficient α. This is achieved by calculating the linear product of the curvature weight and the adjustment coefficient. For example, as the adjustment coefficient increases, the threshold offset increases for regions with high curvature. The threshold offset is used to increase segmentation sensitivity in regions with high curvature and reduce the risk of misclassification in regions with low curvature.

[0065] The dynamic segmentation threshold refers to a grayscale judgment standard that changes with spatial position. It can be generated by superimposing the basic grayscale threshold with the threshold offset, for example, forming a higher segmentation threshold at the vascular bifurcation or plaque edge area. The dynamic segmentation threshold can adapt to the differences in tissue characteristics of different anatomical parts and avoid the problems of over-segmentation or under-segmentation caused by the use of fixed thresholds. The adjustment coefficient refers to a dynamic parameter that controls the degree of influence of the curvature weight on the threshold offset. It can be iteratively updated according to the fractal dimension verification results. For example, when the fractal dimension does not meet the standard, the coefficient value is adjusted by the proportional factor. The adjustment coefficient is used to balance the influence of the curvature feature on the segmentation process and achieve adaptive optimization of the algorithm.

[0066] Specifically, a base grayscale threshold is set as the initial segmentation benchmark, for example, 80% of the typical grayscale value of the tunica media layer of the vessel wall is used as a default value. During the segmentation process, the curvature weight reflects the structural complexity of the local vessel wall. For example, high-curvature regions of calcified plaques have a weight close to 1, while smooth vessel wall regions have a weight close to 0. The adjustment coefficient is dynamically adjusted based on the fractal dimension verification results. For example, when the plaque edge roughness is insufficient, the adjustment coefficient is increased to increase the segmentation priority of high-curvature regions. The threshold offset is calculated by multiplying the curvature weight by the adjustment coefficient. For example, when the adjustment coefficient is 0.5 and the curvature weight is 0.8, the offset is 0.4. The final dynamic segmentation threshold is the sum of the base threshold and the offset. For example, when the base threshold is 120 and the offset is 40, the segmentation threshold at that location is 160. As a result, a higher segmentation threshold is used in high-curvature regions, such as plaque edges, to reduce misclassification of normal tissue; a lower threshold is maintained in low-curvature regions to avoid missing low-contrast plaques.

[0067] Compared with existing technologies, traditional methods use a fixed grayscale threshold for segmentation, which cannot distinguish structural differences between different regions of the vessel wall. For example, the same threshold is used for the high-reflectivity area of ​​calcified plaque and the low-signal area of ​​the lipid core, resulting in reduced segmentation accuracy. However, this solution introduces curvature weights and dynamic adjustment coefficients to generate a spatially adaptive segmentation threshold. For example, the segmentation threshold is increased in curved areas to suppress noise interference, while the threshold is lowered in straight segments to enhance the detection of weak signal plaques. This dynamic threshold mechanism effectively overcomes the limitations of fixed threshold methods in complex anatomical structures.

[0068] Through the above-mentioned technical solution, the present application can dynamically adjust the segmentation sensitivity based on the local geometric characteristics of the vessel wall. For example, the segmentation threshold is automatically increased at the edge of the plaque to suppress noise interference, while the threshold is lowered in uniform areas to reduce the risk of missed detection. This solution significantly improves the segmentation accuracy of heterogeneous plaques, for example, achieving precise contour delineation at the junction of the lipid core and the fibrous cap, while reducing the probability of missegmentation at the media-adventitia interface. The dynamic adjustment coefficient mechanism further ensures that the algorithm remains adaptable at different iteration stages, for example, quickly locating the main plaque area in the initial stage and gradually refining the edge details in subsequent iterations.

[0069] like Figure 4 The figure shows a flow chart for calculating the three-dimensional curvature parameters of the present application. The present application further proposes that the three-dimensional curvature parameters of each vertex in the three-dimensional blood vessel mesh model be calculated as follows:

[0070] Construct a local surface patch for each vertex in the three-dimensional blood vessel mesh model;

[0071] The local quadratic surface equation is fitted by the least squares method to calculate the average curvature and Gaussian curvature of the vertex; the vertex whose absolute value of Gaussian curvature is greater than the preset curvature threshold is marked as a high curvature feature point;

[0072] The preset curvature threshold is set as the initial threshold in the initial stage, and is reduced to γ ​​times the original value with each additional iteration, where 0<γ<1.

[0073] A local surface patch is a local geometric structure centered on a target vertex and encompassing adjacent vertices. It can be constructed using Delaunay triangulation or the k-nearest neighbor algorithm. Its purpose is to capture the local curvature variations of the vascular wall surface and provide spatially correlated data for subsequent curvature calculations. Least-squares fitting of the local quadratic surface equation involves finding the optimal quadratic surface parameters based on local vertex coordinates. This method can be implemented using polynomial fitting. This method eliminates noise interference and accurately describes the surface geometry.

[0074] The absolute value of Gaussian curvature is a quantitative indicator of the degree of surface curvature obtained through differential geometry calculations. It can be derived from the first- and second-order derivatives of the quadratic surface equation. This indicator can effectively distinguish physiologically smooth areas of the vessel wall from areas of high curvature caused by plaque. Dynamic adjustment of the preset curvature threshold involves gradually lowering the initial threshold during the iterative process. For example, the initial threshold can be set to 0.25, and then decayed by γ = 0.8 with each iteration. This mechanism prioritizes the capture of significant curvature features during the initial segmentation and gradually refines the detection sensitivity in subsequent iterations.

[0075] Specifically, after obtaining the three-dimensional vascular mesh model, the surrounding adjacent vertices are selected for each vertex to construct a local surface patch. The local vertex coordinates are fitted with a quadratic surface using the least squares method, the principal curvature parameters are derived, and the absolute value of the Gaussian curvature is calculated. When the Gaussian curvature of a vertex exceeds the initial threshold, it is marked as a feature point of the potential plaque area. In subsequent iterations, the initial threshold is attenuated according to a predetermined ratio, so that each segmentation process can detect more subtle curvature changes. For example, using a higher initial threshold in the first iteration can avoid misjudging normal wrinkles in the vascular endothelium, while gradually lowering the initial threshold in subsequent iterations can help capture tiny protrusions at the edge of calcified plaques.

[0076] Compared to existing methods, which typically use a fixed curvature threshold for feature point detection, this approach is unable to adapt to the varying curvature distributions across different vascular segments. This approach, by iteratively lowering the threshold, can gradually improve detection accuracy while preserving significant features, avoiding missed or false detections caused by improper threshold settings. For example, when analyzing fibrous plaques with minute calcification deposits, the dynamic threshold mechanism can effectively distinguish physiological curvature from pathological structures.

[0077] Through the above technical solution, this application can accurately identify areas of abnormal local curvature caused by plaque on the surface of the blood vessel wall, and is particularly suitable for detecting calcified plaques with complex morphological features. By dynamically adjusting the detection threshold, the efficiency of the initial segmentation and the accuracy of subsequent iterations can be balanced, reducing the impact of noise interference on feature extraction and improving the accuracy of plaque boundary positioning.

[0078] like Figure 5 As shown in FIG, it is a flow chart of generating a curvature weight map of the present application; the present application further proposes generating a curvature weight map, including:

[0079] Normalize the absolute value of Gaussian curvature to generate a curvature weight value in the interval [0,1];

[0080] The curvature weight values ​​of adjacent vertices are bilinearly interpolated to generate a continuous curvature weight map.

[0081] Normalization refers to mapping the absolute value of the original Gaussian curvature to a fixed interval through a linear or nonlinear transformation. This can be achieved by using the maximum and minimum normalization method. For example, the absolute value of the curvature is divided by the maximum curvature value of the region in which it is located, so that the curvature weights of different regions are comparable. Bilinear interpolation refers to interpolation calculations in two-dimensional or three-dimensional space based on the curvature weight values ​​of adjacent vertices. This can be achieved by taking a weighted average of the weight values ​​of four adjacent vertices. For example, the interpolation weight coefficient is determined by the ratio of the distances between vertices, thereby generating a smooth transition continuous curvature distribution.

[0082] Specifically, in a three-dimensional vascular mesh model, the absolute value of the Gaussian curvature of each vertex may be non-uniformly distributed. Directly using the original curvature value will result in discrete jumps in the weight map. Through normalization, the curvature differences of different vascular segments can be unified to the same scale range, eliminating the weight deviation caused by excessive differences in the absolute value of the curvature. Furthermore, performing bilinear interpolation on adjacent vertices can fill the gaps between vertices in the curvature weight map, generating a continuous spatial distribution map, avoiding weight mutations caused by the discreteness of the mesh vertices, and thus providing a stable and reliable spatial weight basis for the subsequent construction of the dynamic threshold template.

[0083] Compared with existing technologies, traditional methods typically directly use discrete vertex curvature values ​​to generate weight maps, resulting in spatially stepped changes in segmentation thresholds, which can easily lead to boundary misjudgments during region growing. However, this application uses the collaborative processing of curvature normalization and bilinear interpolation to enable the curvature weight map to accurately reflect the continuous curvature variation characteristics of the blood vessel surface, effectively suppressing local weight anomalies caused by noise, and improving the spatial consistency of plaque segmentation boundaries.

[0084] Through the above technical solution, this application solves the problem of plaque segmentation boundary breakage or adhesion caused by discontinuous curvature weight in traditional methods, and realizes the smooth transition expression of vascular surface curvature characteristics, thereby significantly improving the geometric continuity of plaque contour recognition and reducing the risk of over-segmentation or under-segmentation caused by sudden curvature changes.

[0085] This application further proposes that adaptive region growing segmentation includes:

[0086] When selecting seed points, the weighted sum of the grayscale gradient amplitude and curvature weight value is comprehensively evaluated;

[0087] The similarity judgment threshold of adjacent voxels is dynamically adjusted during the growth process, and the judgment threshold is negatively correlated with the current growth depth;

[0088] The slope coefficient of the negative correlation is dynamically compensated according to the fractal dimension verification results.

[0089] The weighted sum refers to the linear combination of the grayscale gradient amplitude and the curvature weight value to form a comprehensive score. Specifically, it can be achieved by multiplying the normalized grayscale gradient amplitude by the first weight coefficient and adding the normalized curvature weight value by the second weight coefficient. This method can balance the influence of different features on the seed point selection.

[0090] Dynamically adjusting the similarity judgment threshold means gradually reducing the allowable range of grayscale differences between adjacent voxels as the region grows deeper into the vascular wall. This can be achieved by using the difference between the initial threshold and the current growth depth multiplied by the attenuation factor. This method can prevent the spread of errors caused by the complex vascular structure.

[0091] Dynamic compensation refers to the real-time correction of the slope coefficient of the negative correlation based on the fractal dimension verification results. Specifically, the slope coefficient can be increased when the fractal dimension value is lower than the preset threshold, so that the judgment threshold decreases faster with increasing depth. This method can optimize the segmentation sensitivity for patch edge features of different shapes.

[0092] Specifically, during the plaque segmentation phase, high-probability initial seed points are screened by calculating the weighted sum of the grayscale gradient amplitude and the curvature weight value, avoiding selection bias caused by relying solely on a single feature. During the region growing process, the initial judgment threshold is set to a high value to ensure segmentation accuracy in the core area. Subsequently, as the growth depth increases, the threshold is gradually reduced according to a preset negative correlation, thereby adapting to the grayscale variation characteristics of different layers of the vessel wall. When the fractal dimension verification results indicate insufficient edge complexity, the system automatically increases the slope coefficient of the negative correlation, so that the judgment threshold decreases faster with increasing depth in subsequent iterations, thereby enhancing the ability to capture low-contrast plaque edges.

[0093] Compared with existing technologies, traditional methods that use fixed thresholds or single features for region growing are unable to account for variations in grayscale distribution at different depths in the vessel wall and heterogeneous plaque structures, resulting in over-segmentation of calcified areas and under-segmentation of the lipid core. This solution effectively addresses the error accumulation problem caused by fixed thresholds by selecting seed points through weighted and multi-feature fusion, combined with a dynamic threshold adjustment mechanism and fractal dimension feedback control.

[0094] Through the above technical solution, the present application can adaptively adjust the segmentation parameters according to the morphological complexity of the plaque edge, while ensuring the accuracy of calcified plaque segmentation, significantly improving the recognition ability of low-contrast lipid cores, avoiding the false negative problem caused by fixed thresholds in traditional methods, and enhancing the robustness of plaque segmentation results.

[0095] like Figure 6 FIG. 1 is a flowchart of extracting edge fractal dimension eigenvalues ​​of the present application. The present application further proposes steps for extracting edge fractal dimension eigenvalues ​​of initial plaque contour data, including:

[0096] Perform multi-resolution sampling on the initial patch contour data to obtain contour length sequences at different scales;

[0097] The logarithmic slope of the contour length and the sampling scale is calculated by the least squares method to obtain the edge fractal dimension eigenvalue.

[0098] Multi-resolution sampling refers to the discretization of contour data at different spatial scales. This can be achieved by gradually increasing the sampling interval, for example by scaling the spacing between original contour points by a proportional factor to generate a sequence of contours from coarse to fine grains. The fractal dimension eigenvalue is a metric that quantifies the complexity of a curve. This can be achieved by calculating the trend of contour length at different sampling scales, for example by using the Hausdorff dimension calculation method from fractal geometry theory.

[0099] Specifically, the edge fractal dimension feature values ​​of the initial patch contour data are quantified using a multi-resolution analysis framework. First, the three-dimensional patch contour is spatially discretized, and contour lines are extracted at different sampling intervals. For example, the interval can be set to 1, 2, or 4 times the spacing of the original data points. After calculating the contour length at different scales, a logarithmic coordinate relationship between the sampling scale and the contour length is established, and the slope of the straight line is fitted using the least squares method as the fractal dimension feature. This slope reflects the complexity change of the contour line at different observation scales. For example, high-dimensional values ​​indicate that the edge has more microscopic details, while low-dimensional values ​​correspond to smooth boundaries.

[0100] Compared to existing technologies, traditional methods typically rely solely on single-scale morphological features for plaque verification, such as using contour length or area change rate as criteria. This makes it difficult to distinguish true plaque edges from artifacts caused by noise. However, this method, through multi-scale fractal dimension analysis, effectively captures the self-similar properties of edge structures. For example, the sharp edges of calcified plaques and the fuzzy boundaries of lipid cores exhibit significant differences in fractal dimension, thereby improving the reliability of verification results.

[0101] Through the above technical solution, the present application can accurately quantify the morphological complexity of plaque edges, overcome the problem of traditional single-scale analysis methods being sensitive to noise, effectively distinguish real plaques from imaging artifacts, reduce the probability of normal vascular wall structures being misjudged as plaques, and at the same time improve the recognition accuracy of low-contrast plaque edges.

[0102] The present application further proposes a preset condition that the difference in the edge fractal dimension eigenvalues ​​of two consecutive iterations is less than a preset convergence threshold, the number of iterations reaches a preset maximum number, and the current edge fractal dimension eigenvalue exceeds a preset fractal dimension threshold, any one or more of the following combinations.

[0103] The preset condition refers to the standard for stopping the iteration process, which can be implemented by using a method of parallel judgment of multiple preset independent conditions. When any condition is triggered, the calculation process is terminated.

[0104] The preset convergence threshold refers to the allowable range of the difference between two iteration results, which can be implemented by using an empirical value set based on statistical analysis of historical data, for example, it can be set to a numerical range between 0.05 and 0.1.

[0105] The preset maximum number refers to a safety limit on the number of iterations, which can be implemented by using a fixed value set based on the balance between computing resource consumption and processing efficiency, for example, an integer value within the range of 10-20 times.

[0106] The preset fractal dimension threshold refers to the minimum standard for the complexity of the patch edge morphology, which can be specifically achieved by using the median or average value of the statistical distribution of the fractal dimension of typical patch samples.

[0107] Specifically, during the iterative optimization process, after each fractal dimension verification, three independent conditions must be checked simultaneously: calculating whether the difference between the current and previous fractal dimensions is below the convergence threshold, counting whether the cumulative number of iterations has reached the maximum, and determining whether the current fractal dimension has exceeded a preset threshold. If any of these conditions are met, the iteration is terminated immediately to avoid an unbounded loop.

[0108] For example, in the patch contour optimization stage, when the fractal dimension value reaches 0.85 in the third iteration and exceeds the preset threshold of 0.82, the result is directly output; if the fractal dimension difference for five consecutive iterations remains below 0.08, it is determined to have converged and the calculation is terminated.

[0109] Compared to existing technologies, traditional methods typically only set a fixed number of iterations or a single convergence condition, failing to adapt to the dynamic optimization requirements of varying patch morphologies. For example, existing technologies may prematurely terminate iterations if morphological criteria are not met, or continue to perform ineffective calculations for complex morphologies. This solution, through a multi-condition hybrid judgment mechanism, prevents ineffective calculations caused by local optimization stagnation while ensuring immediate cessation when morphological criteria are met, balancing computational efficiency and accuracy.

[0110] Through the above technical solution, this application effectively solves the problem of over-segmentation or under-segmentation caused by a single preset condition in traditional segmentation algorithms, improves the stability of patch boundary recognition through a dynamic hybrid judgment mechanism, and significantly reduces computing resource consumption while ensuring the accuracy of the results, avoiding the waste of system resources caused by infinite iterations.

[0111] like Figure 7 As shown, this is a processing flow chart for when the fractal dimension verification of this application does not meet the standard; this application further proposes recording the parameter combination of the current dynamic threshold template when the fractal dimension verification result does not meet the standard; when the standard is not met for P consecutive iterations, it falls back to the historical optimal parameter combination; the historical optimal parameter combination is defined as the previous iteration parameter with the highest fractal dimension value.

[0112] Among them, the parameter combination refers to the adjustment coefficient, basic grayscale threshold and related weight parameters used by the dynamic threshold template in each iteration process. It can be implemented in the form of database storage or cache queue to track the trajectory of parameter adjustment and provide a data basis for subsequent operations. The fallback mechanism means that when the preset fractal dimension threshold is not reached after multiple consecutive iterations, the system automatically switches to the historical optimal parameter combination. It can be implemented by setting a loop counter and a parameter version management module to prevent the algorithm from falling into a local optimal solution and improve iteration efficiency. The historical optimal parameter combination refers to the parameter configuration with the highest fractal dimension value generated during the executed iteration process. It can be implemented by comparing the fractal dimension feature values ​​generated by each iteration and sorting and screening them to ensure that subsequent segmentation operations are always based on the current optimal threshold template parameters.

[0113] Specifically, when the fractal dimension verification result fails to reach the preset threshold, the system stores the current dynamic threshold template's adjustment coefficient, grayscale threshold offset, and other parameters in a temporary storage area and starts an iteration counter. If the fractal dimension requirement is still not met after P consecutive iterations, the parameter combination corresponding to the iteration with the highest fractal dimension value in the historical parameter record is retrieved and reloaded into the dynamic threshold template generation module. During this process, the parameter rollback operation is implemented by overwriting the current parameter configuration and resetting the iteration counter to restart the verification process.

[0114] In some specific embodiments, the record of the parameter combination can be saved in the form of a log file marked with a timestamp, for example, a record entry containing the parameter value and the fractal dimension characteristic value is generated each time an iteration is completed. The number of consecutive times P in the fallback condition can be set to 3 to 5 times, for example, when the standard is not met for 3 consecutive iterations, the fallback operation is triggered. The screening of the historical optimal parameters can be achieved by updating the parameter index corresponding to the maximum fractal dimension value in real time, for example, comparing the current value with the historical maximum value after each fractal dimension verification, and updating the index to point to the current parameter if it is higher.

[0115] Compared to existing technologies, traditional methods often use random restarts or fixed parameter reset strategies after multiple iterations fail, resulting in low segmentation efficiency and inability to guarantee stable results. However, this solution, by recording parameter combinations and introducing a historically optimal fallback mechanism, retains valid parameter configurations during iterations, avoiding repeated invalid adjustment paths, thereby improving the convergence speed and robustness of the segmentation process.

[0116] Through the above technical solution, the present application solves the problems of reduced segmentation efficiency and unstable results caused by multiple iteration failures in the existing technology. The parameter tracking and fallback mechanism effectively reduces the number of invalid calculations, ensuring that the algorithm quickly converges to a plaque segmentation result that meets the fractal dimension requirements within a limited number of iterations, while improving the accuracy of vascular plaque recognition and system robustness.

[0117] like Figure 8 、 Figure 9 As shown, the present application further proposes an intelligent vascular plaque recognition system based on optical coherence tomography, including an OCT data acquisition and preprocessing module, a three-dimensional curvature feature calculation module, a dynamic segmentation module, a fractal dimension verification module, a feedback control module, and an iterative management and result output module.

[0118] The OCT data acquisition and preprocessing module is used to acquire vascular OCT volume data collected by the optical coherence tomography device and extract point cloud data on the vascular wall surface based on the grayscale gradient distribution of the OCT volume data to generate a three-dimensional vascular mesh model. This can be achieved by grayscale gradient analysis and point cloud reconstruction algorithms. Its function is to convert the original volume data into a structured mesh model, providing a geometric basis for subsequent curvature analysis. The three-dimensional curvature feature calculation module is used to calculate the three-dimensional curvature parameters of each vertex in the three-dimensional vascular mesh model and generate a curvature distribution matrix. This can be achieved through local surface fitting and curvature calculation algorithms. Its function is to quantify the local curvature characteristics of the vascular wall surface morphology and provide a spatially related curvature weight basis for dynamic threshold generation.

[0119] The dynamic segmentation module uses a dynamic threshold template to perform adaptive region growing segmentation on OCT volume data. This is achieved by combining a seed point selection strategy with a dynamic similarity threshold adjustment algorithm. Its purpose is to adaptively expand the plaque region based on curvature weights and grayscale features, avoiding over-segmentation or under-segmentation caused by fixed thresholds.

[0120] The fractal dimension verification module is used to extract the edge fractal dimension eigenvalues ​​of the initial patch contour data, which can be achieved through multi-resolution contour sampling and logarithmic slope calculation. Its function is to quantify the edge complexity through fractal dimension and verify the morphological rationality of the segmentation results.

[0121] The feedback control module is used to dynamically modify the adjustment coefficient of the curvature weight map according to the fractal dimension verification results. Specifically, it can be implemented by using a proportional factor adjustment and parameter fallback mechanism. Its function is to adjust the control intensity of the curvature weight on the segmentation process through iterative optimization, thereby improving the robustness of the algorithm.

[0122] The iteration management and result output module is used to repeatedly execute the segmentation and verification steps until the preset conditions are met. This can be achieved through convergence threshold judgment and iteration number limitation; its function is to ensure the stability and reliability of the output results.

[0123] Specifically, the system converts raw volume data into a three-dimensional vascular mesh model through the OCT data acquisition and preprocessing module. The three-dimensional curvature feature calculation module then generates a curvature weight map based on vertex curvature to guide the construction of a dynamic threshold template. The dynamic segmentation module performs region growing segmentation based on the dynamic threshold, and the fractal dimension verification module evaluates the edge complexity of the segmentation results. If the fractal dimension does not meet the standard, the feedback control module optimizes the dynamic threshold template by adjusting the coefficients of the curvature weight map. The iterative management and result output module loops through this process until convergence conditions are met, ultimately outputting a three-dimensional recognition result that includes plaque boundaries and morphological features.

[0124] Compared with existing technologies, traditional systems rely on fixed thresholds or single-feature segmentation methods, making it difficult to distinguish between grayscale overlap between normal vessel wall tissue and plaque areas, resulting in a high missegmentation rate. However, this system, through spatially adaptive mapping of curvature weights and dynamic thresholds, combined with iterative feedback adjustment of fractal dimensions, can effectively distinguish plaque areas with different morphologies and grayscale characteristics, such as the high-reflectivity areas of calcified plaques and the low-contrast areas of the lipid core, thereby reducing the risk of misidentification.

[0125] Through the above technical solution, this application solves the segmentation error problem caused by grayscale overlap in traditional methods. Through the closed-loop control mechanism verified by dynamic threshold and fractal dimension, the recognition accuracy of heterogeneous plaque structures is improved, especially the detection rate of low-contrast plaques is improved. At the same time, the risk of divergence in the iteration process is avoided through the parameter fallback mechanism, ensuring the stability of the system output.

[0126] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. An intelligent method for identifying vascular plaques based on optical coherence tomography, characterized by: The steps include: Obtaining blood vessel OCT volume data collected by an optical coherence tomography device, wherein the OCT volume data includes three-dimensional spatial coordinates and corresponding grayscale values; extracting point cloud data of the vascular wall surface based on the grayscale gradient distribution of the OCT volume data to generate a three-dimensional vascular mesh model; Calculating the three-dimensional curvature parameters of each vertex in the three-dimensional blood vessel mesh model to generate a curvature distribution matrix; Establishing a mapping relationship between a curvature weight map and a dynamic threshold template according to the numerical distribution range of the curvature distribution matrix; Adopting the dynamic threshold template to perform adaptive region growing segmentation on the OCT volume data to obtain initial plaque contour data; Extracting edge fractal dimension feature values ​​of the initial plaque contour data to generate a fractal dimension verification result; When the fractal dimension verification result does not reach the preset fractal dimension threshold, dynamically correcting the adjustment coefficient of the curvature weight map and regenerating the dynamic threshold template according to the deviation between the current edge fractal dimension feature value and the preset fractal dimension threshold; Repeat the adaptive region growing segmentation and fractal dimension verification steps until at least one preset condition is met; Output three-dimensional recognition results including effective patch boundary coordinates and morphological features; The preset conditions include that the difference in edge fractal dimension eigenvalues ​​of two consecutive iterations is less than a preset convergence threshold and the number of iterations reaches a preset maximum number.

2. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 1, characterized in that: The adjustment coefficient for dynamically modifying the curvature weight map includes: Calculate the ratio k=N of the current edge fractal dimension feature value N and the preset fractal dimension threshold M M ; When k is less than 1, an adjustment coefficient increment Δα=β*(1-k) is generated, where β is a preset proportional factor; The adjustment coefficient of the curvature weight map is updated to α′=α+Δα, where α is the original adjustment coefficient of the curvature weight map.

3. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 2, characterized in that: The establishment of the dynamic threshold template includes: Set the basic grayscale threshold to the preset value; Generate a threshold offset according to the product of the curvature weight value and the current adjustment coefficient α; Adding the basic grayscale threshold and the threshold offset to generate a dynamic segmentation threshold related to the spatial position; The current adjustment coefficient is updated according to the fractal dimension verification result at each iteration.

4. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 1, wherein: The calculating of the three-dimensional curvature parameters of each vertex in the three-dimensional blood vessel mesh model includes: constructing a local surface patch for each vertex in the three-dimensional blood vessel mesh model; Fitting the local quadratic surface equation by the least squares method to calculate the average curvature and Gaussian curvature of the vertex; marking the vertex whose absolute value of Gaussian curvature is greater than a preset curvature threshold as a high curvature feature point; The preset curvature threshold is set as the initial threshold in the initial stage, and is reduced to γ ​​times the original value with each additional iteration, where 0<γ<1.

5. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 4, characterized in that: The generation of the curvature weight map includes: Normalizing the absolute value of the Gaussian curvature to generate a curvature weight value in the interval [0, 1]; The curvature weight values ​​of adjacent vertices are bilinearly interpolated to generate a continuous curvature weight map.

6. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 1, characterized in that: The adaptive region growing segmentation includes: When selecting seed points, the weighted sum of the grayscale gradient amplitude and curvature weight value is comprehensively evaluated; Dynamically adjust the similarity determination threshold of adjacent voxels during the growth process, wherein the determination threshold is negatively correlated with the current growth depth; The slope coefficient of the negative correlation is dynamically compensated according to the fractal dimension verification result.

7. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 1, characterized in that: The step of extracting the edge fractal dimension feature value of the initial plaque contour data comprises: Performing multi-resolution sampling on the initial plaque contour data to obtain contour length sequences at different scales; The logarithmic slope of the contour length and the sampling scale is calculated by the least square method to obtain the edge fractal dimension eigenvalue.

8. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 1, characterized in that: The preset condition also includes that the current edge fractal dimension feature value exceeds the preset fractal dimension threshold.

9. The method for intelligently identifying vascular plaques based on optical coherence tomography according to claim 1, characterized in that: When the fractal dimension verification result does not meet the standards: Record the parameter combination of the current dynamic threshold template; When the criteria are not met after P consecutive iterations, the system will revert to the historical optimal parameter combination. The historical optimal parameter combination is defined as the previous iteration parameter with the highest fractal dimension value.

10. An intelligent vascular plaque identification system based on optical coherence tomography, characterized by: The method for intelligently identifying vascular plaques based on optical coherence tomography according to any one of claims 1 to 9 is implemented, comprising: An OCT data acquisition and preprocessing module is used to obtain vascular OCT volume data collected by an optical coherence tomography device, and based on the grayscale gradient distribution of the OCT volume data, extract the vascular wall surface point cloud data to generate a three-dimensional vascular mesh model; a three-dimensional curvature feature calculation module, configured to calculate the three-dimensional curvature parameters of each vertex in the three-dimensional vascular mesh model, generate a curvature distribution matrix, and establish a mapping relationship between a curvature weight map and a dynamic threshold template based on the numerical distribution range of the curvature distribution matrix; A dynamic segmentation module, configured to perform adaptive region growing segmentation on the OCT volume data using the dynamic threshold template to obtain initial plaque contour data; A fractal dimension verification module, configured to extract edge fractal dimension eigenvalues ​​of the initial plaque contour data and generate a fractal dimension verification result; a feedback control module, configured to dynamically correct the adjustment coefficient of the curvature weight map and regenerate a dynamic threshold template based on the deviation between the current edge fractal dimension eigenvalue and the preset fractal dimension threshold when the fractal dimension verification result does not reach the preset fractal dimension threshold; The iterative management and result output module repeatedly executes the adaptive region growing segmentation and fractal dimension verification steps until the preset conditions are met, and outputs the three-dimensional recognition results including the valid patch boundary coordinates and morphological features.

Images