Automated Recognition Method and System for Vascular Structures in OCT Imaging
By combining multi-scale Hessian matrices and topological analysis, the problem of inaccurate and inconsistent identification of the three-layer vascular structure in OCT technology was solved, achieving high-precision and standardized automated identification of vascular structures, thus improving the accuracy and consistency of diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- THE SECOND AFFILIATED HOSPITAL ARMY MEDICAL UNIV
- Filing Date
- 2025-09-22
- Publication Date
- 2026-06-30
Smart Images
Figure CN121190435B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of vascular structure recognition technology, specifically to an automated method and system for recognizing vascular structures in OCT images. Background Technology
[0002] In the field of medical imaging analysis of vascular structures, optical coherence tomography (OCT) technology, with its high-resolution imaging capabilities, has enabled the visualization of the three-layer structure of blood vessels (intima, media, and adventitia), providing important imaging evidence for the diagnosis and research of vascular diseases.
[0003] However, current OCT technology still heavily relies on the visual observation and subjective judgment of professionals in identifying the three layers of blood vessels. Due to differences in the experience level, observation angle, and judgment criteria of different operators, significant individual differences can easily arise in the structural determination of the same vascular image, making it difficult to establish a unified identification standard. This human-induced error and inconsistency directly reduces the accuracy of vascular structure analysis results. At best, it leads to blurred lesion localization and inaccurate assessment of lesion severity; at worst, it may delay treatment or lead to inappropriate diagnostic and treatment decisions due to misjudgment.
[0004] Even with the existence of some auxiliary identification systems or algorithms, they still face core bottlenecks in practical applications. On the one hand, these systems or algorithms lack sufficient accuracy in extracting vascular edge features, making it impossible to precisely distinguish the subtle boundaries of the three-layer structure; on the other hand, they lack standardized identification logic and still require human intervention to correct the results, making it difficult to achieve truly accurate and unified identification, and failing to meet the clinical demand for high precision and consistency in vascular structure analysis. Summary of the Invention
[0005] The present invention aims to provide an automated method and system for identifying vascular structures in OCT images, so as to achieve automated identification of vascular structures in OCT images and solve the problems of inaccurate and inconsistent identification and easy error in the existing technology.
[0006] To achieve the above objectives, the present invention employs the following technical solution: an automated identification method for vascular structures in OCT images, comprising the following steps:
[0007] Step 1, Extraction of vascular layer boundary features: Extract vascular layer boundary features from OCT images based on multi-scale Hessian matrix response, including geometric and physical features, to form a feature response map;
[0008] Step 2, Three-layer topology recognition: Construct the topological relationship of the vascular structure based on the feature response map, and extract the nested ring structure; identify and output the initial boundary coordinate point set and structure score of the three layers of the vascular intima, media and adventitia;
[0009] Step 3, Vascular Structure Optimization: The initial boundary coordinate point set is precisely optimized, and special cells are detected in the optimized region to generate a result report.
[0010] Meanwhile, this solution also provides an automated identification system for vascular structures in OCT images, applied to the aforementioned automated identification method for vascular structures in OCT images, including:
[0011] The image input module is used to connect to OCT devices and acquire OCT image information.
[0012] The preprocessing module is used to preprocess image information, suppress noise, and enhance structural boundaries;
[0013] The feature recognition module is used to identify features from OCT images and extract vascular layer boundary features to form a feature response map;
[0014] The boundary determination module is used to determine and calibrate the boundaries of the three-layered vascular structure based on the extracted features; and to identify pathological parameters.
[0015] The output module is used to output the recognition results in a visual form.
[0016] The principles and advantages of this scheme are:
[0017] Traditional OCT vascular analysis techniques primarily employ image segmentation algorithms, which heavily rely on the continuity of image grayscale or the salience of gradients. When calcification shadows, blood artifacts, noise, or blurred local boundaries exist in the vessel wall, these methods fail completely due to interruptions or distortions in pixel information, failing to yield complete and coherent structural results. Furthermore, existing methods are merely pure image processing tools; their outputs may be geometrically correct, but anatomically inaccurate, making it difficult to incorporate the prior medical knowledge that blood vessels are layered and have a reasonable thickness range into the analysis process.
[0018] While deep learning-based black-box models may be effective in certain situations, their decision-making process is untraceable. Doctors cannot trust a system that cannot explain why a certain area is identified as a boundary, rendering its clinical application significantly flawed and hindering its widespread adoption. This is considered a difficult-to-overcome technological bias.
[0019] This approach breaks away from the traditional mindset of "image segmentation," moving beyond the problem of "distinguishing between foreground and background pixels" to "identifying and verifying a complex topological structure that conforms to specific anatomical rules in a noisy environment." This approach transforms clinicians' visual diagnostic experience and anatomical knowledge into a computable mathematical model. Through a three-tiered progressive analysis framework of "feature enhancement - topology recognition - constraint optimization," it achieves automated, high-precision, and standardized identification of the three layers of blood vessels in OCT images.
[0020] This approach extracts features based on the tubular geometry of blood vessels, rather than simply relying on grayscale gradients. It adapts to blood vessels of different sizes through multi-scale analysis and suppresses noise responses from non-vascular edges using algorithms, improving recognition accuracy and ensuring precise subsequent quantization calculations. Secondly, by introducing topological analysis, it completely abandons the reliance on pixel-by-pixel continuity, instead identifying the existence and persistence of "annular voids" in the image. Even if the boundary is locally missing, as long as the topological structure remains stable, the system can automatically infer and complete the outline, ensuring reliable structural output even in complex diseased blood vessels. Thirdly, it encodes physician experience (such as inter-slice spacing, angles, and area ratios) into mathematical constraints. The optimization process not only pursues image feature matching but also mandates that the output results fall within an anatomically reasonable parameter space, ensuring biomechanical and anatomical reliability and solving the problem of traditional methods easily producing absurd results. Furthermore, each judgment has clear mathematical and physical basis, allowing traceability of the decision-making process at any boundary point, making it easily accepted and adopted by clinicians. This improves judgment efficiency and reliability, achieving fully automated OCT vascular structure recognition and making the diagnostic process more standardized and unified. Attached Figure Description
[0021] Figure 1 This is a flowchart illustrating the automated identification method for vascular structures in OCT imaging according to the present invention.
[0022] Figure 2 This is an existing OCT vascular image in the automated identification method for vascular structures in OCT imaging according to the present invention;
[0023] Figure 3 This is a flowchart of step 1 of the automated identification method for vascular structures in OCT images according to the present invention;
[0024] Figure 4 This is a cross-sectional view of a vascular structure in the automated identification method for vascular structures in OCT images according to the present invention;
[0025] Figure 5 This is a schematic diagram of the construction of the vascular structure distance field in step 2 of the automated identification method for vascular structures in OCT images according to the present invention;
[0026] Figure 6 This is a schematic diagram of a three-layered blood vessel constructing a distance field in the automated identification method for vascular structures in OCT imaging according to the present invention.
[0027] Figure 7 This is a schematic diagram of the automated vascular structure recognition system for OCT imaging according to the present invention. Detailed Implementation
[0028] The following detailed description illustrates the specific implementation method:
[0029] Example 1
[0030] The automated identification method for vascular structures in OCT imaging in this embodiment transforms prior anatomical knowledge (slice thickness range) into mathematical constraints and defines pathological quantitative parameters to achieve automated identification of vascular structures, improve identification accuracy and consistency, and reduce diagnostic risks.
[0031] In this embodiment, as shown in the appendix Figure 1 As shown, it includes the following steps:
[0032] S1, Vascular layer boundary feature extraction: Based on the multi-scale Hessian matrix response, vascular layer boundary features are extracted from OCT images, including geometric and physical features, to form a feature response map.
[0033] In this embodiment, since the cross-section of the blood vessel presents an approximately elliptical tubular structure in the OCT image, its main features are: continuity along the direction of the blood vessel, symmetrical gradient changes perpendicular to the direction of the blood vessel, and local curvature characteristics, which are extracted as geometric features.
[0034] And as attached Figure 2 As shown, in OCT images, the blood vessel layer appears as alternating bright and dark rings. The difference in scattering characteristics between the blood vessel wall and the blood / adventitia results in higher gray levels (strong scattering) in the blood vessel wall region and lower gray levels (weak scattering) in the lumen. There is a clear gray level transition at the interlayer boundaries, from which physical features can be extracted.
[0035] Based on the original OCT image, the extracted tubular structure features are quantized. In this embodiment, as shown in the attached figure... Figure 3As shown, Gaussian filtering is applied to OCT images using the tubular geometric characteristics and gray-level gradient properties of the vascular layer boundaries to construct a multi-scale space and enhance vascular boundary features. Traditional OCT vascular recognition mainly relies on gradient information, which has limited effectiveness for blurred inter-layer boundaries. Therefore, in this embodiment, Hessian matrix quantization of local structural curvature is used, and the minimum eigenvalue response to the tubular structure is extracted. Combined with gradient attenuation terms to suppress non-vascular edge noise, the brightness of the vascular layer boundaries is significantly enhanced, strengthening the inter-layer boundaries of vessels with different diameters. This achieves multi-scale fusion, improves the accuracy of vascular structure recognition, and can adapt to vessels of different diameters. The extracted feature response map fusion can then be expressed as:
[0036] ;
[0037] In the formula, These are the two-dimensional spatial coordinates of pixels in the image; For scale parameters (vessel size); For scale The Hessian matrix is used to detect local curvature; For tubular reinforcement calculations, This represents the smallest eigenvalue of the matrix; the larger the negative value, the more pronounced the tubular structure. For gradient suppression calculation; This is the OCT image after Gaussian filtering. ; The original image information of the OCT image is represented as... ; This is the attenuation coefficient.
[0038] Geometric features are used to represent the tubular structure of blood vessels, clarifying their directional characteristics to indicate the normal direction of the vessel wall. Physical features reflect the boundary strength of the blood vessels, ensuring that the identified tubular structures represent true boundaries. Simultaneously, multi-scale analysis allows for the acquisition of scale features, identifying the scale data with the strongest tubular feature response at each location to obtain approximate thickness information of the blood vessel structure at that location.
[0039] The above feature information is fused into a feature response map to clearly show the boundary location of the potential vascular layer.
[0040] S2, Three-layer topology recognition: Construct the topological relationship of the vascular structure based on the feature response map, extract the nested ring structure; identify and output the initial boundary coordinate point set and structure score of the three layers of the vascular intima, media and adventitia.
[0041] In traditional methods, when a local rupture occurs in the blood vessel wall, the integrity of the vascular ring structure is compromised, making it difficult to guarantee the integrity and accuracy of the vascular structure during reconstruction. To overcome this problem, we broke away from conventional thinking and returned to the initial view of the normal three-layer structure of a blood vessel as three nested rings. We discovered that the actual vascular layer forms "holes" in the image, as shown in the attached diagram. Figure 4 As shown in the shaded area, artifacts or noise cannot form stable holes. Therefore, this solution creatively uses topological theory to solve the problem of vascular ring structure integrity, and adopts a continuous cohomology method to penetrate failure interference, solving the misjudgment problem of traditional segmentation methods when there is vascular rupture or noise interference, thereby constructing a complete and accurate vascular topology structure.
[0042] In this embodiment, the topological relationship of the vascular structure is constructed by distance transformation, the nested ring structure with significant persistence is extracted, and the initial boundary coordinate point set and structural integrity score of the three-layer structure are preliminarily confirmed.
[0043] Specifically, it includes the following sub-steps:
[0044] S2.1, Construct a distance field model of the vascular structure.
[0045] In this embodiment, combined with the appendix Figure 2 As shown, instead of directly processing grayscale images, it transforms them into a topographic map through topological structure. In this map, vessel boundaries (bright rings) are transformed into "valleys," and vessel lumens and interlayer regions are transformed into "peaks." The "cavities" described in this scheme are "peaks" surrounded by "valleys." In this embodiment, the feature response map generated by S1 has significantly enhanced the vessel boundary information (boundary response map), as shown in the attached figure. Figure 5 As shown, the high-brightness (white) areas in the image correspond to the vessel boundaries (such as the internal limiting membrane, media, and external membrane), while the low-brightness (black) areas correspond to the vessel lumen and background tissue.
[0046] Subsequently, the characteristic response Figure 2 Value-based mapping is used, setting the blood vessel boundary region as the target. Based on this, the Euclidean distance from each pixel in the image to the nearest blood vessel boundary point is calculated, generating a distance field topographic map similar to "elevation". Combined with the attached... Figure 6 As shown (boundaries unclear), in this topographic map, distance values are mapped to altitude. The distance values are smallest at the vessel boundary itself and in its adjacent areas, even reaching 0, and are represented as "valleys." Conversely, areas far from the vessel boundary (such as the center of the vessel lumen or the center of the adventitia) have the largest distance values, and are represented as "peaks." This transforms geometric features into structural relationships. The distance field can then be calculated as follows:
[0047] ;
[0048] In the formula, For a binary boundary point set, when the distance field The larger the value, the closer it is to the center of the blood vessel. =0 indicates the location of the blood vessel boundary.
[0049] S2.2 Perform continuous homology analysis on the model to screen out the void feature group and calculate the void persistence.
[0050] In this embodiment, assuming there is a complete vascular layer (bright ring) in the original OCT image, its range field should resemble a "mountain peak surrounded by a ring valley". A water level threshold is set. The algorithm increments from 0 to the maximum value, performing a "water level" scan. When the water level is low (e.g., at the threshold), the algorithm detects the water level. =2), at this point, most of the mountain peak is above water and may be connected to other surrounding areas, not forming an isolated island. When the water level rises (e.g., to the threshold), If the water level is ≤5, then the area ≤5 is submerged, and the originally connected peaks are completely separated by a ring-shaped valley (transformed from a bright ring), forming an independent "island." This is the birth of a "cavity." As the water level continues to rise (e.g., the threshold),... =8), meaning that the area ≤8 is submerged, the original island is reduced to just a tiny top, and eventually sinks completely underwater, at which point the "hole" disappears.
[0051] The generated binarized region can be represented as:
[0052] ;
[0053] For each Calculate the 1-dimensional homology group It detects "hole" structures in images and records the formation and disappearance of holes, i.e., when... When adding, if a new cavity is created, it is recorded. This indicates the "water level" at which the vascular layer begins to appear; when the cavity disappears, it is recorded... This represents the water level at which the vascular layer is completely submerged, thus generating a group of void features, represented as:
[0054] .
[0055] And based on this, the persistence of the void is calculated:
[0056] By determining the duration of cavities, structural stability is quantified and used as an indicator of the stability of the vascular layer structure, ultimately generating a set of cavity features. .
[0057] S2.3 performs a three-layer structure selection based on persistence and generates an initial set of boundary coordinate points.
[0058] Because islands formed by real blood vessel layers can remain independent during larger water level changes, The noise level is relatively high, and the small fluctuations caused by noise are quickly submerged. Based on the calculated persistence, voids that meet the anatomical constraints are selected to form a nested relationship of three layers. In this embodiment, when >When the persistence threshold T is reached, it indicates a stable cavity that has existed for a long time, which is a true vascular layer; when When the value is less than the persistence threshold T, it is considered a short-lived cavity, quickly submerged, and thus an artifact noise. Therefore, a cavity set is formed by selecting the true blood vessel layer and constructing a nested relationship. It can be represented as:
[0059] ;
[0060] In the formula, This is the topology persistence threshold, which can be set based on practical experience, such as within the range of 0-10. =5. That is, by assessing the persistence of the "i-th cavity," we determine whether the overall structure of this cavity is sufficiently stable and significant, thereby judging its structural stability. Even if the original image has local blurring, it can still obtain complete, continuous, and anatomically consistent boundaries, solving the problem of recognition failure caused by local breaks.
[0061] The voids selected from the void set are sorted in descending order of void area, and the sorted voids are checked to verify the extracted nesting relationships. For each valid void, its average radius position is taken as the initial boundary coordinate value, i.e.
[0062] Then the set of boundary coordinate points of the i-th layer can be represented as: .
[0063] The boundary coordinate point set of each membrane layer is an ordered, discrete two-dimensional coordinate point sequence, and the line connecting each point sequence represents a closed contour line. Each boundary layer is a list of N points, each point being a (x, y) coordinate pair. For example, the boundary point set of the inner membrane can be represented as: ={(x 11 ,y 11 ), (x 12 ,y 12 ),...,(x) 1n ,y 1n The boundary point set of the middle membrane can be represented as: ={(x 21 ,y 21 ), (x 22 ,y 22 ),...,(x) 2n ,y2n The outer membrane boundary point set can be represented as: ={(x 31 ,y 31 ), (x 32 ,y 32 ),...,(x) 3n ,y 3n )}.
[0064] The initial boundary coordinate point set of the three-layer structure can then be represented as:
[0065] In the formula, The inner membrane boundary; The boundary of the middle membrane; This is the outer membrane boundary.
[0066] Simultaneously, the confirmed structure is scored based on the calculated durability, and the score is calculated as follows:
[0067] In the formula, Score the intima boundary structure; Score the boundary structure of the middle membrane; Scoring of the outer membrane boundary structure.
[0068] In this embodiment, by determining the persistence of voids, blurry images that are difficult to interpret with the naked eye are transformed into quantized images with clear boundaries and annotations. This can solve the problems in traditional methods, such as segmentation interruption, loss of layer structure, misjudgment as independent structure, generation of false boundaries, and layer count errors caused by local vascular rupture, calcified plaque occlusion, blood turbulence artifacts, and interference from branch vessels. It ensures that complete rings can be identified and automatically filters out short-lived voids, thereby improving the identification and accurate establishment of nested topological relationships.
[0069] S3, Vascular Structure Optimization: Precisely optimizes the initial boundary coordinate point set, detects special cells within the optimized region, and generates a results report.
[0070] In this embodiment, after smoothing the calculated initial boundary coordinate point set, a three-layer initial boundary contour is established, represented as follows: , represent the boundary curves of the intima, median, and adventitia, respectively. Simultaneously, using the initial boundary coordinates as the starting point for optimization, the structural score is used to set the optimization weights and initialize the energy function parameters, with their weight coefficients being respectively... And set the constraint parameters.
[0071] The energy function is constructed in this way and expressed as:
[0072] ;
[0073] In the formula, This is a uniform term for the region (consistency); This refers to the interlayer spacing constraint term; This is the boundary smoothing term.
[0074] in, The weights for regional uniformity can be adjusted based on persistence. The three-layer boundary needs to be optimized; Let be the grayscale variance within the k-th layer region; This is a penalty function that increases when the interlayer spacing deviates from the normal range; The average distance between the i-th layer and the j-th layer; This represents the total length of the boundary, used to control boundary smoothing.
[0075] In this embodiment, the constraint parameters are set according to the anatomical constraint conditions. These constraints include the allowable range of distances between adjacent membranes, the maximum allowable angle between the boundary normals of adjacent layers, and the range of area ratios between adjacent layers. Specifically, this can be expressed as:
[0076] ;
[0077] In the formula, This refers to the permissible range of distance between the intima and the media. This refers to the allowable range of distance between the middle membrane and the outer membrane, such as 1.0 mm to 15.0 mm. The angle between the normal vectors of the boundary between adjacent layers; This is the maximum allowable angle between the normal vectors, such as 15°, to ensure that the boundaries of adjacent layers have similar orientations. The ratio of the areas of adjacent layers; The area ratio is set within a range, for example, 0.7 to 1.3, to prevent abnormal expansion or contraction.
[0078] Minimize the energy function using gradient descent or constrained optimization algorithms. Simultaneously satisfying the constraints, the iteration continues until convergence, thus optimizing the solution.
[0079] Within each optimized layer region, special cells (i.e., aberrant scattering cells: a population of abnormal cells inferred from OCT image features that differ from normal vascular smooth muscle cells; these cells are typically closely related to the occurrence, development, and vulnerability of vascular lesions). In this embodiment, the blood vessel is divided into three layers using the optimized boundaries, and then the aberration degree of each pixel within each region is calculated, which can be expressed as:
[0080] ;
[0081] In the formula, , These represent the mean and standard deviation of grayscale values within the k-th layer region, respectively. A threshold is also set based on the grayscale fluctuation range of normal cells. Typically, a value of 2.5 to 3.0 can be used. > When that happens, the location is marked as a special cell.
[0082] Finally, the optimized three-layer boundary coordinates will be output, including parameters such as the thickness and area of each layer, and the set of marked special cell locations will also be output, which can be represented as follows: .
[0083] By optimizing boundary coordinates, the boundaries are made clearer and smoother. Simultaneously, constraint optimization ensures the structure conforms to anatomical rules, and the interlayer spacing remains stable within the normal range. Furthermore, it enables rapid detection of special cells, improving detection accuracy and effectiveness, and enhancing overall automation efficiency.
[0084] Example 2
[0085] This embodiment provides an automated recognition system for vascular structures in OCT imaging, applied to the aforementioned automated recognition method for vascular structures in OCT imaging, as shown in the attached figure. Figure 7 As shown, it includes:
[0086] The image input module is used to connect to OCT devices and acquire OCT image information.
[0087] In this embodiment, the image input module is connected to the OCT imaging device, enabling it to parse and be compatible with the data protocols of devices from different manufacturers, achieving lossless acquisition and high-speed transmission of raw image data. Simultaneously, this module automatically extracts and binds metadata information for each frame of image, including but not limited to acquisition time, spatial resolution, device parameters, and patient information, providing a precise scale calibration and traceability basis for all subsequent processing steps, ensuring the measurement accuracy of the analysis results.
[0088] The preprocessing module is used to preprocess image information, suppress noise, and enhance structural boundaries.
[0089] In raw OCT data, interference factors such as tissue scattering noise, optical artifacts, and residual blood signals can severely obscure valuable microstructures. In this embodiment, the preprocessing module implements anisotropic filtering, adaptively adjusting the filtering intensity based on the local content of the image. Stronger smoothing is applied to uniform regions to effectively suppress noise, while weaker smoothing is maintained in critical structural regions such as vascular layer boundaries to preserve their sharpness and detail. After preprocessing, the signal-to-noise ratio of the image is significantly improved, providing a more accurate image architecture for subsequent feature recognition and extraction, rather than simply blurring the image.
[0090] The feature recognition module is used to identify features from OCT images and extract vascular layer boundary features to form a feature response map.
[0091] In this embodiment, the feature recognition module transforms the preprocessed grayscale image into a feature map that highlights the geometric characteristics of the blood vessel wall. Based on the fact that blood vessel cross-sections appear as unique tubular or layered structures in OCT images, these structures can be precisely quantified mathematically by calculating their local curvature. Simultaneously, the module employs multi-scale analysis to address blood vessels of different sizes; a small analysis scale is used to capture subtle intima boundaries, while a large analysis scale is used to perceive robust adventitia structures. Finally, a feature response map is identified and extracted, where the brightness value of each pixel no longer represents grayscale, but rather the probability or salience that the location belongs to a blood vessel layer boundary, thus clearly highlighting the blood vessel structure from the complex background.
[0092] The boundary determination module is used to determine and calibrate the boundaries of the three-layered vascular structure based on the extracted features; and to identify pathological parameters.
[0093] The boundary determination module receives the extracted feature response map and completes the decision-making process from pixel features to anatomical structures and then to clinical parameters.
[0094] In this embodiment, the boundary determination module includes a structure layer determination unit and a parameter extraction unit. The structure layer determination unit extracts structural features and verifies the formation of a clear three-layer structural boundary. The parameter extraction unit identifies specific cells within each formed structural region, performs pathological parameter identification, and then marks them.
[0095] The structural layer determination unit utilizes topological analysis principles, no longer treating the vascular layer as an isolated collection of edge pixels, but rather recognizing it as a nested, complete ring structure (including the intima, media, and adventitia). This approach gives it strong anti-interference capabilities; even if the vascular boundary shows local breaks due to calcification or plaque obscuring it, the system can still infer a complete, anatomically consistent contour based on the overall structure, ensuring the accuracy and integrity of the structure.
[0096] After the initial contour is acquired, it undergoes constrained optimization. Clinical knowledge, such as typical thickness ranges, relative positional relationships, and smoothness between layers, is transformed into mathematical constraints to finely adjust the initial contour, resulting in a final boundary with sub-pixel precision that best conforms to biomechanical characteristics. Based on this, the parameter extraction unit automatically performs quantitative analysis, calculating a series of key pathological parameters, such as layer thickness, thickness variability, luminal area, and plaque load. Furthermore, based on image features such as scattering characteristics, it automatically identifies and labels vulnerable plaque features, including suspected lipid-rich necrotic cores and microcalcifications.
[0097] The output module is used to output the recognition results in a visual form.
[0098] In this embodiment, the output module transforms complex analysis results into an intuitive, reliable form that can be directly used for clinical decision-making. It generates a structured, comprehensive report that includes not only visual images of the three-layer boundaries and specific cells (highlighted in different colors), but also integrates all quantitative parameters and structural integrity scores.
[0099] In this embodiment, the report is designed to align with clinical reading habits, allowing physicians to conduct interactive reviews. For example, clicking anywhere on an image displays the precise thickness value at that location, or focusing on specific labeled cells. Ultimately, all images, data, and reports can be archived or transmitted to the hospital information system (HIS / PACS) in accordance with medical data standards, seamlessly integrating into clinical workflows.
[0100] In this embodiment, an automated and standardized analysis pipeline is constructed from raw data to clinical insights. Through core algorithmic innovations in the feature recognition and boundary determination modules, physicians' visual diagnostic experience (such as identifying tubular structures, inferring complete outlines, and judging based on anatomical common sense) is encoded into a computable mathematical model. This model ultimately provides precise, quantifiable, and objective evidence through the output module. This fundamentally changes the traditional OCT vascular assessment model, which relies on physicians' "visual observation" and "manual measurement," significantly improving analysis efficiency, repeatability, and accuracy. It provides strong data support for the precise diagnosis of vascular diseases, treatment planning (such as stent size selection), and postoperative evaluation.
[0101] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. An automated method for identifying vascular structures in OCT imaging, characterized in that, Includes the following steps, Step 1, Extraction of vascular layer boundary features: Extract vascular layer boundary features from OCT images based on multi-scale Hessian matrix response, including the tubular geometric characteristics and gray-level gradient physical characteristics of the vascular layer boundary, to form a feature response map; Step 2, Three-layer topology identification: Construct the topological relationship of the vascular structure based on the feature response map, and use persistent homology analysis to extract the persistent nested ring structure; identify and output the initial boundary coordinate point set of the three layers of the vascular intima, media and adventitia, and score the confirmed structure according to its persistence; Includes the following sub-steps: Step 2.1: Construct a distance field model of the vascular structure, transforming geometric features into structural relationships; Step 2.2: Perform continuous cohomology analysis on the model, screen out the void feature group, and calculate the void persistence; Step 2.3: Perform three-layer structure filtering based on persistence and generate an initial set of boundary coordinate points; Step 3, Vascular Structure Optimization: Based on the structural score, optimization weights are set, and combined with the set constraint parameters, an energy function including interlayer spacing constraints is constructed, expressed as follows: ； In the formula, For the region uniform term; This refers to the interlayer spacing constraint term; For boundary smoothing terms; in, For regional uniformity weights; The three-layer boundary needs to be optimized; Let be the grayscale variance within the k-th layer region; For the penalty function, The average distance between the i-th layer and the j-th layer; This is the total length of the boundary. The constraint parameters are set based on the anatomical constraints, including the allowable range of distances between adjacent membranes, the maximum allowable angle between the boundary normals of adjacent layers, and the range of area ratios of adjacent layers, expressed as follows: ； In the formula, This refers to the permissible range of distance between the intima and the media. This refers to the allowable range of distance between the middle membrane and the outer membrane; The angle between the normal vectors of the boundary between adjacent layers; This represents the maximum permissible angle between the normal vectors; The ratio of the areas of adjacent layers; The range is for area ratios; The initial boundary coordinate point set is precisely optimized to obtain a precise boundary contour, so that the optimized three-layer boundary conforms to anatomical rules. Specific cells are detected in the optimized area, and a result report is generated.
2. The automated identification method for vascular structures in OCT imaging according to claim 1, characterized in that, The feature response map is represented as follows: ； In the formula, These are the two-dimensional spatial coordinates of pixels in the image; For scale parameters; For tubular reinforcement calculations, For scale The Hessian matrix below, Represents the smallest eigenvalue of a matrix; For gradient suppression calculation; This is the OCT image after Gaussian filtering; ; The original image information of the OCT image is represented as... ; This is the attenuation coefficient.
3. The automated identification method for vascular structures in OCT imaging according to claim 1, characterized in that: In step 3, detecting special cells involves calculating the anomaly score for each pixel within each optimized region. A pixel with an anomaly score greater than a threshold is marked as a special cell. The anomaly score is expressed as... ； In the formula, This refers to the original image information of the OCT image; , denoted as the mean and standard deviation of grayscale values within the k-th layer region, respectively.
4. The automated identification method for vascular structures in OCT imaging according to claim 1, characterized in that: In step 2.2, a threshold is set. Increasing from 0 to the maximum value, the void feature group is filtered out by recording the generation and disappearance states of voids, and is represented as... ； The persistence of voids is then represented as In the formula, Threshold for generating voids; The threshold for void disappearance.
5. The automated identification method for vascular structures in OCT imaging according to claim 4, characterized in that: In step 2.3, based on the persistence combined with anatomical constraints, voids that meet the conditions are selected, forming a nested relationship of three layers. The void set can then be represented as... ； In the formula, This is the topology persistence threshold.
6. The automated identification method for vascular structures in OCT imaging according to claim 5, characterized in that: It also includes sorting the voids in the void set in descending order of void area, and checking the sorted voids; for each valid void, the average radius position is taken as the initial boundary coordinate value to generate an initial boundary coordinate point set; Then the set of boundary coordinate points of the i-th layer is represented as ; .
7. The automated identification method for vascular structures in OCT imaging according to claim 6, characterized in that: In step 2, the confirmed structure is scored based on the calculated durability, denoted as follows: ； In the formula, Score the intima boundary structure; Score the boundary structure of the middle membrane; Scoring of the outer membrane boundary structure.
8. An automated system for recognizing vascular structures in OCT imaging, characterized in that, The automated identification method for vascular structures in OCT imaging, applied to any one of claims 1-7, includes: The image input module is used to connect to OCT devices and acquire OCT image information. The preprocessing module is used to preprocess image information, suppress noise, and enhance structural boundaries; The feature recognition module is used to identify features from OCT images and extract vascular layer boundary features to form a feature response map; The boundary determination module is used to determine and calibrate the boundaries of the three-layered vascular structure based on the extracted features; and to identify pathological parameters. The output module is used to output the recognition results in a visual form.