An OCTA-based method and system for evaluating the proportion of vertical microvessels

Abstract

The present application relates to a kind of vertical microvessel proportion evaluation method and system based on OCTA, belong to image processing technical field, solve the problem that the fresh red macule diagnosis and treatment prediction is not accurate due to the space trend of lesion blood vessels is ignored in prior art.The specific steps include: the dermal-epidermal junction of target area skin OCTA three-dimensional image is identified and positioned;With dermal-epidermal junction as reference surface, first superficial layer data and second superficial layer data are intercepted to dermis deep side direction, and after respectively false image is removed, maximum density projection is carried out, and the corresponding two-dimensional vascular plan is obtained;Using adaptive threshold strategy, the blood vessel mask of each two-dimensional vascular plan is extracted, and the first mask graph and the second mask graph of binary are obtained, and the total blood vessel area of first mask graph is counted;The first mask graph and the second mask graph are carried out connected domain correlation analysis, vertical microvessel in the first mask graph is identified, and vertical microvessel proportion is obtained in combination with total blood vessel area, and the accuracy of vertical blood vessel proportion evaluation is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for assessing the proportion of vertical microvessels based on OCTA. Background Technology

[0002] Port-wine stains (PWS) are abnormal dilation of capillaries in the dermis, typically appearing as pink, red, or purple patches. They deepen and thicken with age, severely impacting patients' quality of life. Vascular-targeted photodynamic therapy (V-PDT) and pulsed dye laser (PDL) have become the mainstream treatments for port-wine stains in China.

[0003] However, clinical practice shows that the efficacy of V-PDT varies significantly among individuals, with some patients experiencing poor treatment responses and difficulty achieving ideal vascular clearance. Previous studies have found a close correlation between V-PDT efficacy and the structure of PWS lesions. In conventional vertically incident laser treatment, lesion vessels parallel to the skin surface (i.e., perpendicular to the incident light direction) expose the largest cross-sectional area in the laser irradiation direction, thus fully absorbing photons and achieving effective photocoagulation destruction. In contrast, vessels perpendicular to the skin surface (i.e., parallel to the incident light direction) have a very small exposed cross-section, and optical scattering from deeper tissues makes it difficult for vertical vessels to be fully photocoagulated. Therefore, the three-dimensional orientation of the lesion vessel network, especially the abundance of "vertically oriented microvessels," is a key anatomical factor contributing to the difficulty in uniformly penetrating laser or photodynamic energy and triggering treatment resistance.

[0004] Currently, optical coherence tomography (OCTA) is commonly used in the diagnosis and treatment of peripheral vascular disease (PWS) for non-invasive 3D imaging of microvessels before treatment. While it can present high-resolution three-dimensional vascular structures of lesions, existing OCTA image processing and parameter extraction schemes neglect the influence of spatial orientation on light targeting and absorption. That is, they mostly treat blood vessels as uniform tubular structures, failing to effectively identify the three-dimensional spatial characteristics and vertical distribution of blood vessels. Furthermore, in OCTA-based PWS diagnosis and treatment, when encountering highly resistant lesions, blindly increasing the energy of a single laser irradiation based solely on traditional experience not only fails to improve efficacy but also easily leads to epidermal thermal damage, scarring, or pigmentation. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a method and system for assessing the proportion of vertical microvessels based on OCTA, in order to solve the problem of inaccurate diagnosis and treatment prediction of port-wine stains due to neglecting the spatial orientation of lesion vessels in the prior art.

[0006] The objective of this invention is mainly achieved through the following technical solutions: On one hand, embodiments of the present invention provide a method for assessing the proportion of vertical microvessels based on OCTA, comprising the following steps: The OCTA three-dimensional image of the target area skin is processed to identify and locate the dermal-epidermal junction; Using the dermal-epidermal junction as a reference plane, first superficial layer data and second superficial layer data are extracted towards the deep dermis. After removing artifacts from the first superficial layer data and the second superficial layer data respectively, maximum density projection is performed to obtain the corresponding two-dimensional vascular planar image. An adaptive threshold strategy is used to extract vascular masks from each of the two-dimensional vascular plane images to obtain a binarized first mask image and a second mask image. The total vascular area of ​​the first mask image is then calculated. Connectivity analysis is performed on the first mask image and the second mask image to identify vertical microvessels in the first mask image, and the proportion of vertical microvessels is obtained by combining the total blood vessel area.

[0007] Furthermore, identifying the vertical microvessels includes: Eight-neighborhood connectivity analysis was performed on the first and second mask images respectively, and the corresponding blood vessel connected regions were marked. Overlapping blood vessels are identified in the vascular connected regions of the first mask image and the second mask image respectively, and the suspected vertical microvessels that are covered are separated to obtain the corresponding candidate connected regions. Traverse each candidate connected region in the first mask image. If the candidate connected region satisfies both the geometric morphology condition and the depth continuity condition, then the candidate connected region is identified as a vertical microvessel. The depth continuity condition is determined by correlation analysis with the candidate connected regions in the second mask image.

[0008] Furthermore, overlapping vessel identification is performed on the vascular connected regions of the first mask image or the second mask image to obtain the candidate connected regions, including: The vessel centerline skeleton is extracted from the vessel connected regions in the first mask and the second mask respectively to obtain the skeleton points, main skeleton direction, and skeleton bifurcation points of the corresponding vessel connected regions. Distance transformations are performed on the first and second mask images respectively to obtain the distance transformation values ​​of each pixel in the corresponding mask images, and the distance transformation values ​​are sampled along the skeleton points to obtain the corresponding local widths; Traverse each skeleton point to identify local expansion candidate regions in overlapping blood vessels. If a region meets any of the following conditions, it is identified as the local expansion candidate region: the local width change rate of adjacent skeleton points exceeds a preset change rate threshold, or there is a local peak in the distance transformation in the neighborhood of each skeleton bifurcation point. Along the main skeletal direction, local width minima are searched before and after all local enlargement candidate regions, and the regions are segmented from the local width minima. The local enlargement candidate regions are used as the vascular connected regions where the currently overlapping vessels are occluded. The candidate connected regions are obtained from all the segmented vascular connected regions.

[0009] Furthermore, determining whether a candidate connected component satisfies the depth continuity condition includes: Using the center coordinates of the candidate connected region in the first mask as a reference, a search is performed within a preset neighborhood window of the second mask. If a corresponding region exists and the center position of the corresponding region satisfies a preset offset, and the area difference between the candidate connected region and the corresponding region satisfies a preset area threshold, then the candidate connected region satisfies the depth continuity condition.

[0010] Furthermore, if a candidate connected component satisfies the geometric morphology conditions, then the aspect ratio of the candidate connected component satisfies a preset aspect ratio range, and both the density and pixel area are greater than the corresponding preset threshold.

[0011] Furthermore, a binarized first mask or second mask is obtained, including: The grayscale values ​​of the two-dimensional blood vessel planar image are statistically analyzed, and initial candidate background regions are selected based on the grayscale peak values. The highlighted areas of the two-dimensional blood vessel planar image are extracted based on a preset highlight threshold, and the highlighted areas are expanded by a preset range to obtain an exclusion area; Using the initial candidate background region outside the exclusion region as the second candidate background region, the mean and standard deviation of the background pixels of the two-dimensional blood vessel planar image are calculated based on the pixel intensity within the second candidate background region. A global adaptive segmentation threshold is set based on the mean and standard deviation of the background pixels, and joint constraints are constructed. Each pixel in the two-dimensional blood vessel planar image is compared with the adaptive segmentation threshold. Pixels that are greater than or equal to the adaptive segmentation threshold and satisfy the joint constraint condition are set as blood vessel regions, and the rest are set as background, thus obtaining a binarized first mask image or second mask image.

[0012] Furthermore, the joint constraint condition includes a local contrast constraint; wherein, if the difference between the gray level of a pixel and the average gray level of the background in a preset local neighborhood is greater than a preset gray level difference, then the pixel satisfies the local contrast constraint.

[0013] Furthermore, artifact removal from the first and second shallow layer data includes: Using the first and second shallow layer data as input, a projection artifact prediction is performed using an artifact removal neural network to obtain the artifact confidence map of the first and second shallow layer data. Regions exceeding the preset confidence threshold in the two artifact confidence maps are treated as artifacts and removed from the corresponding shallow layer data.

[0014] Furthermore, locating the dermal-epidermal junction includes: The connection weights between adjacent pixels are calculated based on the gradient map of the OCTA 3D image. The shortest path algorithm is used to search for the path with the minimum cumulative connection weight to obtain the skin-air boundary. A horizontal A-line signal intensity integral projection is performed in the region below the skin-air boundary to obtain a first signal intensity curve distributed along the depth. Based on the peak value of the first signal intensity curve, the coordinate system of the OCTA three-dimensional image is mapped back to obtain the dermal-epidermal boundary.

[0015] On the other hand, embodiments of the present invention provide a vertical microvessel proportion assessment system based on OCTA, comprising: The acquisition module is used to acquire OCTA three-dimensional images of the skin in the target area; The positioning module is used to locate the dermal-epidermal junction; it is also used to use the dermal-epidermal junction as a reference plane to extract the first superficial layer data and the second superficial layer data in the direction of deep dermis, and to perform maximum density projection on the two superficial layers data after removing artifacts to obtain the corresponding two-dimensional vascular planar image. The blood vessel segmentation module is used to extract blood vessel masks from each of the two-dimensional blood vessel planar images using an adaptive threshold strategy, to obtain a binarized first mask image and a second mask image, and to count the total blood vessel area of ​​the first mask image. The vertical blood vessel recognition module is used to perform connected component correlation analysis on the first mask image and the second mask image to identify vertical microvessels in the first mask image. An evaluation module is used to determine the percentage of vertical microvessels based on the identified vertical microvessels combined with the total vascular area.

[0016] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. This invention proposes an artifact removal and adaptive thresholding strategy for extracting superficial data at different depths from the dermis-epidermis region. This strategy obtains vascular masks for each superficial layer, enabling more stable differentiation between real vertical vessels and projected trailing artifacts under various imaging conditions. It reduces the risk of misjudgment caused by fixed thresholds or single-rule judgments, laying the foundation for accurate vertical microvessel identification. Through connected component correlation analysis of superficial layers at different depths, not only are independent vertical microvessels identified, but also obscured vertical microvessels within overlapping vessels, further improving the accuracy of vertical microvessel assessment.

[0017] 2. The introduction of skeleton-guided overlapping vessel segmentation and recovery, through local expansion recognition and skeleton width mutation detection, can reduce the missed and false judgments in overlapping occlusion scenarios. The combined mechanism of geometric morphology conditions and depth connectivity conditions avoids the error caused by single morphology condition judgment, and improves the stability and clinical reliability of VVP statistical results.

[0018] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0020] Figure 1 This is a flowchart of the vertical microvessel proportion assessment method based on OCTA according to an embodiment of the present invention. Detailed Implementation

[0021] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0022] Example 1 A specific embodiment of the present invention discloses a method for assessing the proportion of vertical microvessels based on OCTA, such as... Figure 1 As shown, it includes the following steps: Step S1: Recognize and process the OCTA three-dimensional image of the target area skin to locate the dermal-epidermal junction; Step S2: Using the dermal-epidermal junction as a reference plane, first superficial layer data and second superficial layer data are extracted towards the deep dermis. After removing artifacts from the first superficial layer data and the second superficial layer data, maximum density projection is performed to obtain the corresponding two-dimensional vascular planar image. Step S3: Extract vascular masks from each of the two-dimensional vascular plane images using an adaptive threshold strategy to obtain a binarized first mask image and a second mask image, and count the total vascular area of ​​the first mask image. Step S4: Perform connected component correlation analysis on the first mask image and the second mask image to identify vertical microvessels in the first mask image, and obtain the proportion of vertical microvessels by combining the total blood vessel area.

[0023] Using the above method, superficial layer data at different depths are extracted from the located dermis-epidermis. Artifact removal and adaptive thresholding strategies are employed to obtain vascular masks for each superficial layer, improving mask accuracy and laying the foundation for accurate vertical microvessel identification. Through connected component correlation analysis of superficial layers at different depths, not only are independent vertical microvessels identified, but also obscured vertical microvessels within overlapping vessels, further improving the accuracy of vertical microvessel assessment. Accurate identification of the proportion of vertical microvessels can assist physicians in the diagnosis and treatment of port-wine stains, thereby improving treatment efficacy.

[0024] Specifically, in step S1, three-dimensional data of the target skin area, i.e., the lesion area, is acquired using an OCTA imaging system. Preferably, a 1300nm band swept frequency source OCTA is used, and the scanning environment is controlled at room temperature (e.g., 23–26℃). Volumetric data with sufficient signal-to-noise ratio is obtained through adjustable A-line density.

[0025] Furthermore, the acquired OCTA 3D data contains vascular structure information at different depths. Using a shortest path algorithm based on image gradient information combined with A-line integral energy, the epidermal-air interface of the vascular OCTA 3D image is automatically located, and the dermal-epidermal junction (DEJ) is then located inwards from this interface. The specific process is as follows: S11. Calculate the connection weights between adjacent pixels based on the gradient map of the OCTA 3D image, and use the shortest path algorithm to search for the path with the minimum cumulative connection weight to obtain the skin-air boundary. For example, using the first column of pixels in the OCTA 3D data as the starting point and the last column as the ending point, Dijkstra's algorithm is used iteratively to search from each starting point to locate the path with the minimum cumulative connection weight as the skin-air boundary. The connection weights between adjacent pixels are calculated based on the following formula:

[0026] Where m and n are the coordinates of adjacent pixels, that is, the coordinates of pixel m are (x, n) and (n) respectively. m ,y m ), the coordinates of pixel n are (x n ,y n ); Here, grady(m) represents the connection weight between pixel m and pixel n, indicating the connection cost between adjacent pixels; grady(m) = y m+1 -y m-1 , representing the vertical gradient of pixel m; gradx(m) = x m+1 -x m-1 , representing the horizontal gradient of pixel m; grady(n) = y n+1 -y n-1 , representing the vertical gradient of pixel n; gradx(n) = x n+1 -x n-1 λ represents the horizontal gradient of pixel n; λ1 is the vertical gradient weight, λ2 is the horizontal gradient weight, and σ is the smoothing constant.

[0027] S12. Perform horizontal A-line signal intensity integral projection in the region below the skin-air boundary to obtain the first signal intensity curve distributed along the depth. S13. Based on the peak value of the first signal intensity curve, map it back to the coordinate system of the OCTA three-dimensional image to obtain the dermal-epidermal boundary.

[0028] Specifically, in step S2, firstly, using the located DEJ as the reference plane, i.e., the zero point of the depth coordinate, at least two data slices with preset depth ranges are extracted towards the deep dermis to obtain the first and second superficial layer data constituting adjacent depth layers. The preset depth is 0–200 μm by default, and can be adjusted according to the lesion thickness, such as to 100–300 μm. Then, after artifact removal from the first and second superficial layer data, maximum intensity projection (MIP) is performed to obtain the corresponding two-dimensional vascular planar image. The artifact removal process is as follows: Using the first and second shallow layer data as inputs, respectively, a projection artifact prediction is performed using an artifact removal neural network to obtain the artifact confidence maps of the first and second shallow layer data. Regions exceeding the preset confidence threshold in the two artifact confidence maps are treated as artifacts and removed from the corresponding shallow layer data.

[0029] The artifact removal neural network can be any of a convolutional neural network, an encoder-decoder network, an attention network, or a 3D segmentation network. Training samples consist of manually labeled artifact regions, unstable high-reflectivity regions verified by repeated scans, and suspected projection tail regions that are significantly inconsistent with adjacent depth layers. A pre-set confidence threshold is determined based on the verification results of the artifact removal neural network training set. By suppressing the tail and projection artifacts caused by strong surface reflections on deep tissues, artifacts are prevented from being misidentified as vertical blood vessels in subsequent scans.

[0030] Specifically, in step S3, on the one hand, an adaptive threshold strategy is used to extract vascular masks from the two-dimensional vascular plane images of the first and second superficial layer data, respectively. The specific process is as follows: S31. Perform grayscale statistics on the two-dimensional blood vessel planar image, and select initial candidate background regions based on the grayscale peak values; S32. Extract the bright areas of the two-dimensional blood vessel planar image based on a preset brightness threshold, and expand the bright areas by a preset range to obtain an exclusion area; For example, a preliminary grayscale statistical analysis is performed on the two-dimensional blood vessel planar image to obtain a grayscale histogram. The set of pixels corresponding to the low-intensity main peak on the left side of the grayscale histogram is selected as the initial candidate background region. The grayscale values ​​of all pixels in the blood vessel planar image are sorted from smallest to largest. The grayscale values ​​located at the 85%-95% position are used as the preset highlight threshold, and the preset range is set to 3-7 pixels to obtain the exclusion area. S33, the initial candidate background region outside the exclusion area is used as the second candidate background region. Based on the pixel intensity in the second candidate background region, the mean and standard deviation of the background pixels of the two-dimensional blood vessel planar image are calculated. S34. Set a global adaptive segmentation threshold based on the mean and standard deviation of the background pixels, and construct joint constraints; the joint constraints are used to suppress interference from tissue texture, speckle noise and isolated bright spots.

[0031] S35. Compare each pixel in the two-dimensional blood vessel planar image with the adaptive segmentation threshold, set the pixels that are greater than or equal to the adaptive segmentation threshold and satisfy the joint constraint condition as blood vessel regions, and set the rest as background, to obtain the binarized first mask image or second mask image.

[0032] For example, firstly, preliminary grayscale statistics are performed on the two-dimensional blood vessel planar image, and the set of pixels corresponding to the low-intensity main peak on the left side of the grayscale histogram is selected as the initial candidate background region. Then, based on the pixels in the initial candidate background region, a pre-set high-brightness threshold is used to coarsely extract pixels in the two-dimensional blood vessel planar image that are higher than the pre-set high-brightness threshold, resulting in bright bands, bright spots, and suspected blood vessel areas that are significantly higher than the overall background. The coarsely extracted region is then expanded outwards by a pre-set range to form an exclusion zone. The initial candidate background region outside the exclusion zone is then used as the second candidate background region. Based on the pixel intensity within the sampling area of ​​this second candidate background region, the background mean μ_noise and standard deviation σ_noise can be calculated. To reduce the impact of local abnormal bright spots or dark lines on the estimation results, robust statistical methods such as truncated mean, median absolute deviation, quantile truncation, or iterative outlier removal can be used to correct μ_noise and σ_noise and construct an adaptive segmentation threshold T, expressed as: T = μ_noise + k·σ_noise, Where k is a system coefficient that can be adjusted based on skin color, device signal-to-noise ratio, scan depth, image gain, or lesion contrast. A vascular mask M(x,y) is constructed based on an adaptive segmentation threshold. When the pixel intensity I(x,y) is greater than or equal to the threshold T and the joint additional constraint condition is met, M(x,y)=1; otherwise, M(x,y)=0. This yields a binary vascular mask, where the vascular mask value of 1 represents the foreground (vascular region), which is white; the background region is black.

[0033] By first coarsely excluding high-signal structures and then performing robust statistics on the low-signal main peak region, the background noise estimation can be made closer to the actual imaging noise level. By incorporating coarse extraction and expansion processes, a safety buffer is set during vessel extraction to avoid the accidental deletion of marginal or weak vessels, which would affect the accuracy of subsequent vessel identification.

[0034] Meanwhile, to improve segmentation accuracy, in addition to judging solely based on the global threshold, joint constraints are further applied, including local contrast constraints. Specifically, if the grayscale of a pixel is higher than the average grayscale of the background in its surrounding local neighborhood by a preset difference, or higher than a preset multiple of the local background fluctuation level, then the pixel satisfies the local contrast constraint.

[0035] Furthermore, the joint constraints also include connectivity constraints. If a candidate region needs to form a continuous foreground structure in its 8-neighborhood, and the number of its continuous foreground pixels is greater than a preset lower limit, then the connectivity constraints are satisfied. By combining joint constraints, the risk of tissue textures, speckle noise, and isolated bright spots being misidentified as blood vessels can be reduced, thus improving the accuracy of vascular masks.

[0036] On the other hand, the total vessel area (TVA) of the first mask image is calculated. This is obtained by multiplying the total number of foreground vessel pixels (N_vessel) in the mask by the physical area of ​​a single pixel, i.e.: TVA = N_vessel × (S_x × S_y), Where S_x and S_y are the pixel physical dimensions in the x and y directions, respectively. To improve accuracy, the first mask image can be first subjected to hole filling, morphological opening and closing operations, boundary burr smoothing, and abnormal small region removal before calculating the total blood vessel area.

[0037] Specifically, in step S4, the steps for performing connected component correlation analysis on the first mask image and the second mask image to identify vertical microvessels in the first mask image are as follows: S41. Perform 8-neighborhood connectivity analysis on the first mask image and the second mask image respectively, and mark the corresponding blood vessel connected regions; For example, when performing 8-neighborhood connectivity analysis on the first or second mask image, each foreground pixel is first scanned row by row, and temporary numbers are assigned to each pixel in the same real connected region. That is, if there is no label in the 8-neighborhood of a pixel, a new number is assigned; if there is a number in the neighborhood, the smallest number is taken as the temporary number of the current pixel, and it is recorded that the number belongs to the same connected region. Then, a second scan is performed, and the temporary labels of the recorded connected regions are merged to obtain the independent blood vessel connected regions corresponding to the first or second mask image.

[0038] S42. Perform overlapping vessel identification on the vascular connected regions of the first mask image and the second mask image respectively, separate the suspected vertical microvessels that are covered, and obtain the corresponding candidate connected regions; specifically including: S421. Extract the vessel centerline skeleton from the vessel connected regions in the first mask and the second mask respectively to obtain the skeleton points, main direction of the skeleton, and bifurcation points of the corresponding vessel connected regions. For example, firstly, a skeleton thinning algorithm is used to shrink connected components to a single-pixel-wide centerline while preserving the original topological structure of the connected components, resulting in a binarized skeleton map. Then, all endpoints in the skeleton map are traversed (i.e., each node has only one skeleton point in its neighborhood). Starting from an endpoint, the skeleton is backtracked until a bifurcation point is encountered. If the backtracking length is less than a preset branch threshold (e.g., 5 pixels), the short branch is deleted to eliminate false bifurcations caused by spikes and noise. Secondly, spline interpolation or moving average is performed on the skeleton point coordinates to make subsequent direction calculations more stable. Finally, based on the thinned skeleton map, principal component analysis is performed on the skeleton points of each connected component to calculate the first principal component vector, obtaining the main skeleton direction. For each pixel in the skeleton map, the number of skeleton points within its 8-neighborhood is counted. When the number is greater than or equal to 3, the pixel is marked as a skeleton bifurcation point.

[0039] S422. Perform distance transformation on the first mask and the second mask respectively to obtain the distance transformation value of each pixel in the corresponding mask, and sample the distance transformation value along the skeleton point respectively. Multiply the distance transformation value by 2 to obtain the corresponding local width. S423. Traverse each skeleton point to identify local enlargement candidate regions in overlapping blood vessels. If a region meets any of the following conditions: the local width change rate of adjacent skeleton points exceeds a preset change rate threshold, i.e., there is a local width increase region; or there is a local peak in the distance transformation in the neighborhood of each skeleton bifurcation point, i.e., there is a bright clump at the skeleton bifurcation point; then it is identified as the local enlargement candidate region.

[0040] It should be noted that, due to the simple horizontal vessels or overlapping vessels running in the same direction, they usually appear as slender, continuous structures with a relatively consistent main direction on the skeleton; while occluded vertical vessels are more likely to form local near-circular high-signal cores on the slender main trunk. Based on this, identifying local enlargement candidate regions can improve the ability to identify vertical vessels in cases of overlapping, adhesion, or occlusion. Specifically, when it is determined that there is a local peak in the distance transformation within the neighborhood of the skeleton bifurcation point, a circular window with radius R is taken with the bifurcation point P as the center. The mean and maximum values ​​of the distance transformation values ​​within the circular window are calculated. If the maximum value is significantly greater than the mean value (e.g., greater than 1.5 times), and the pixel distance values ​​around the point corresponding to the maximum value decrease, then a peak region is determined to exist.

[0041] S424. Along the main trunk direction, search for local width minimum points before and after all local expansion candidate regions, and divide them from the local width minimum points into the main trunk part and the local expansion candidate part. The local expansion candidate region is used as the vascular connected region where the currently overlapping blood vessels are occluded. The candidate connected region is obtained from all the segmented vascular connected regions.

[0042] S43. Traverse each candidate connected region in the first mask image. If the candidate connected region satisfies both the geometric morphology condition and the depth continuity condition, then the candidate connected region is identified as a vertical microvessel. The depth continuity condition is determined by correlation analysis with the candidate connected regions in the second mask image.

[0043] On the one hand, morphological feature parameters are extracted for each candidate connected component, including the aspect ratio, solidity (i.e., the ratio of area to convex hull area) of the bounding rectangle of the connected component, and the area of ​​the connected component pixels. If a candidate connected component exhibits an approximately circular geometric feature, that is, the aspect ratio, solidity, and pixel area are all greater than the corresponding preset thresholds, then the geometric morphological condition is satisfied.

[0044] For example, the number of pixels in each candidate connected component is counted to obtain the pixel area of ​​the connected component. For each candidate connected component, the area of ​​the smallest convex polygon containing that component is calculated to obtain the convex hull area. The ratio of the pixel area of ​​the connected component to the convex hull area yields the density. The minimum row, maximum column, minimum column, and maximum column corresponding to each candidate connected component are found. The difference between the maximum and minimum columns yields the width of the bounding rectangle, and the difference between the maximum and minimum rows yields the height of the bounding rectangle. The aspect ratio of the connected component is then calculated using the formula: Aspect Ratio = max(width, height) / min(width, height). When the aspect ratio is within the range of 0.85–1.15, the density is ≥ 0.75, and the area is greater than the preset minimum pixel cutoff value, the geometric morphology condition is met. The threshold ranges can be adaptively adjusted according to the scanning resolution.

[0045] On the other hand, the depth continuity condition judgment for a candidate connected component includes: Using the center coordinates of the candidate connected regions in the first mask image as a reference, a search is performed within a preset neighborhood window of the second mask image. If a corresponding region exists and its center position satisfies a preset offset, and the area difference between the candidate connected region and the corresponding region satisfies a preset area threshold, then the candidate connected region satisfies the depth continuity condition; otherwise, the candidate connected region does not satisfy the depth continuity condition. For example, if it only appears in the first mask image (i.e., a single layer), or if there is significant center drift or area jump in adjacent layers, it is more likely to be an artifact, noise, or a false circular structure formed by overlapping blood vessels, and therefore does not satisfy the depth continuity condition. By judging the depth continuity condition, the accuracy of vertical blood vessel recognition is further improved.

[0046] Finally, based on the connected region area of ​​the identified vertical vessels, the vertical vessel percentage (VVP) is calculated and expressed as: VVP = (Σ area of ​​connected regions identified as vertical vessels / TVA) × 100%; Furthermore, based on VVP, the diagnosis and treatment of port-wine stains can be predicted. If the VVP value is greater than or equal to a preset risk threshold (e.g., 5%), the patient is determined to have a high VVP phenotype (treatment resistance tendency), and a specific treatment plan of "multiple treatment courses, gradual increase of low energy, and avoidance of single high-energy irradiation" is recommended. If the VVP is less than the threshold, a "standard energy treatment course" plan is recommended. The plan can also be further weighted and adjusted based on the patient's age, location, and past medical history to generate a final recommended plan, which is then uploaded to the display terminal or medical information system as an auxiliary prompt for doctors to decide on the final treatment plan based on the clinical situation.

[0047] Compared with existing technologies, this embodiment provides an OCTA-based method for assessing the proportion of vertical microvessels. It identifies vertical microvessels by extracting superficial data at different depths from the dermis-epidermis layer based on localization and performing connected component association analysis using a vascular mask based on this superficial data. On one hand, it introduces skeleton-guided overlapping vessel segmentation and recovery, multi-parameter connected component geometry discrimination, and depth connectivity discrimination, which can reduce missed and false positives in overlapping occlusion scenarios, improving the stability and clinical reliability of VVP statistical results. On the other hand, it utilizes deep learning artifact probability maps and adaptive threshold strategies to more stably distinguish between real vertical vessels and projection tail artifacts under different imaging conditions, reducing the risk of misjudgment caused by fixed thresholds or single-rule judgments, ultimately achieving accurate assessment of the proportion of vertical microvessels.

[0048] Example 2 Another specific embodiment of the present invention discloses a vertical microvessel proportion assessment system based on OCTA, comprising: The acquisition module is used to acquire OCTA three-dimensional images of the skin in the target area; The positioning module is used to extract first and second superficial layer data in the direction of deep dermis, with the dermal-epidermal junction as the reference plane. After removing artifacts from the two superficial layer data, maximum density projection is performed to obtain the corresponding two-dimensional vascular planar image. The blood vessel segmentation module is used to extract blood vessel masks from each of the two-dimensional blood vessel planar images using an adaptive threshold strategy, to obtain a binarized first mask image and a second mask image, and to count the total blood vessel area of ​​the first mask image. The vertical blood vessel recognition module is used to perform connected component correlation analysis on the first mask image and the second mask image to identify vertical microvessels in the first mask image. An evaluation module is used to determine the percentage of vertical microvessels based on the identified vertical microvessels combined with the total vascular area.

[0049] The system can be used to assess the proportion of vertical microvessels according to any of the methods described in Embodiment 1. Related aspects can be referenced from each other, but are not repeated in this embodiment.

[0050] Compared with existing technologies, the OCTA-based vertical microvessel proportion assessment system provided in this embodiment achieves objective and accurate quantification of the proportion of vertical microvessels in the total vascular area through the linkage of various modules, providing a reliable imaging basis for the design of personalized treatment plans such as vascular-targeted photodynamic therapy.

[0051] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0052] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing the proportion of vertical microvessels based on OCTA, characterized in that, Includes the following steps: The OCTA three-dimensional image of the target area skin is processed to identify and locate the dermal-epidermal junction; Using the dermal-epidermal junction as a reference plane, first superficial layer data and second superficial layer data are extracted towards the deep dermis. After removing artifacts from the first superficial layer data and the second superficial layer data respectively, maximum density projection is performed to obtain the corresponding two-dimensional vascular planar image. An adaptive threshold strategy is used to extract vascular masks from each of the two-dimensional vascular plane images to obtain a binarized first mask image and a second mask image. The total vascular area of ​​the first mask image is then calculated. Connectivity analysis is performed on the first and second mask images to identify vertical microvessels in the first mask image, and the proportion of vertical microvessels is obtained by combining the total vessel area; identifying the vertical microvessels includes: Eight-neighborhood connectivity analysis was performed on the first and second mask images respectively, and the corresponding blood vessel connected regions were marked. Overlapping blood vessels are identified in the vascular connected regions of the first mask image and the second mask image respectively, and the suspected vertical microvessels that are covered are separated to obtain the corresponding candidate connected regions. Traverse each candidate connected region in the first mask image. If the candidate connected region satisfies both the geometric morphology condition and the depth continuity condition, then the candidate connected region is identified as a vertical microvessel. The depth continuity condition is determined by correlation analysis with the candidate connected regions in the second mask image.

2. The method according to claim 1, characterized in that, Overlapping vessel identification is performed on the vascular connected regions of the first mask image or the second mask image to obtain the candidate connected regions, including: The vessel centerline skeleton is extracted from the vessel connected regions in the first mask and the second mask respectively to obtain the skeleton points, main skeleton direction, and skeleton bifurcation points of the corresponding vessel connected regions. Distance transformations are performed on the first and second mask images respectively to obtain the distance transformation values ​​of each pixel in the corresponding mask images, and the distance transformation values ​​are sampled along the skeleton points to obtain the corresponding local widths; Traverse each skeleton point to identify local enlargement candidate regions in overlapping blood vessels. If a region meets any of the following conditions, it is identified as the local enlargement candidate region: the local width change rate of adjacent skeleton points exceeds a preset change rate threshold, or there is a local peak in the distance transformation in the neighborhood of each skeleton bifurcation point. Along the main skeletal direction, local width minima are searched before and after all local enlargement candidate regions, and the regions are segmented from the local width minima. The local enlargement candidate regions are used as the vascular connected regions where the currently overlapping vessels are occluded. The candidate connected regions are obtained from all the segmented vascular connected regions.

3. The method according to claim 1, characterized in that, Determining whether a candidate connected component satisfies the depth continuity condition includes: Using the center coordinates of the candidate connected region in the first mask as a reference, a search is performed within a preset neighborhood window of the second mask. If a corresponding region exists and the center position of the corresponding region satisfies a preset offset, and the area difference between the candidate connected region and the corresponding region satisfies a preset area threshold, then the candidate connected region satisfies the depth continuity condition.

4. The method according to claim 1, characterized in that, If a candidate connected component satisfies the geometric morphology conditions, then the aspect ratio of the candidate connected component satisfies the preset aspect ratio range, and both the density and pixel area are greater than the corresponding preset threshold.

5. The method according to any one of claims 1-4, characterized in that, Obtain the binarized first or second mask image, including: The grayscale values ​​of the two-dimensional blood vessel planar image are statistically analyzed, and initial candidate background regions are selected based on the grayscale peak values. The highlighted areas of the two-dimensional blood vessel planar image are extracted based on a preset highlight threshold, and the highlighted areas are expanded by a preset range to obtain an exclusion area; Using the initial candidate background region outside the exclusion region as the second candidate background region, the mean and standard deviation of the background pixels of the two-dimensional blood vessel planar image are calculated based on the pixel intensity within the second candidate background region. A global adaptive segmentation threshold is set based on the mean and standard deviation of the background pixels, and joint constraints are constructed. Each pixel in the two-dimensional blood vessel planar image is compared with the adaptive segmentation threshold. Pixels that are greater than or equal to the adaptive segmentation threshold and satisfy the joint constraint condition are set as blood vessel regions, and the rest are set as background, thus obtaining a binarized first mask image or second mask image.

6. The method according to claim 5, characterized in that, The joint constraint conditions include local contrast constraints; wherein, if the difference between the gray level of a pixel and the average gray level of the background in a preset local neighborhood is greater than a preset gray level difference, then the pixel satisfies the local contrast constraints.

7. The method according to claim 1, characterized in that, Artifact removal from the first and second shallow layer data includes: Using the first and second shallow layer data as input, a projection artifact prediction is performed using an artifact removal neural network to obtain the artifact confidence map of the first and second shallow layer data. Regions exceeding the preset confidence threshold in the two artifact confidence maps are treated as artifacts and removed from the corresponding shallow layer data.

8. The method according to any one of claims 1-4, 6, and 7, characterized in that, Locating the dermal-epidermal junction includes: The connection weights between adjacent pixels are calculated based on the gradient map of the OCTA 3D image. The shortest path algorithm is used to search for the path with the minimum cumulative connection weight to obtain the skin-air boundary. A horizontal A-line signal intensity integral projection is performed in the region below the skin-air boundary to obtain a first signal intensity curve distributed along the depth. Based on the peak value of the first signal intensity curve, the coordinate system of the OCTA three-dimensional image is mapped back to obtain the dermal-epidermal boundary.

9. A vertical microvessel proportion assessment system based on OCTA, characterized in that, include: The acquisition module is used to acquire OCTA three-dimensional images of the skin in the target area; The positioning module is used to locate the dermal-epidermal junction; It is also used to extract the first superficial layer data and the second superficial layer data in the direction of deep dermis, with the dermal-epidermal junction as the reference plane, and to perform maximum density projection on the two superficial layers data after removing artifacts to obtain the corresponding two-dimensional vascular plane map. The blood vessel segmentation module is used to extract blood vessel masks from each of the two-dimensional blood vessel planar images using an adaptive threshold strategy, to obtain a binarized first mask image and a second mask image, and to count the total blood vessel area of ​​the first mask image. The vertical blood vessel recognition module is used to perform connected component correlation analysis on the first mask image and the second mask image to identify vertical microvessels in the first mask image. An evaluation module is used to determine the percentage of vertical microvessels based on the identified vertical microvessels combined with the total vascular area; Identifying the vertical microvessels includes: Eight-neighborhood connectivity analysis was performed on the first and second mask images respectively, and the corresponding blood vessel connected regions were marked. Overlapping blood vessels are identified in the vascular connected regions of the first mask image and the second mask image respectively, and the suspected vertical microvessels that are covered are separated to obtain the corresponding candidate connected regions. Traverse each candidate connected region in the first mask image. If the candidate connected region satisfies both the geometric morphology condition and the depth continuity condition, then the candidate connected region is identified as a vertical microvessel. The depth continuity condition is determined by correlation analysis with the candidate connected regions in the second mask image.

Images