Method and system for extracting nevus flammeus lesion features based on OCTA technology

Abstract

The invention relates to a nevus flammeus lesion feature extraction method and system based on an OCTA technology. The method comprises the following steps: acquiring a cross-sectional image and a projection image of a focus through an optical coherence tomography angiography technology; extracting the boundary of epidermis and air and the boundary of dermis and epidermis based on the cross-sectional image to obtain the thickness of an epidermis layer; performing threshold segmentation and median filtering on the projection image, generating a binary image, skeletonizing the binary image, and generating a vascular skeleton diagram; calculating the average blood vessel density based on the binarized image, and calculating the average blood vessel diameter and the ratio of the blood vessel lengths of different vessel diameters based on the binarized image and the blood vessel skeleton diagram; dividing a blood vessel layering area by taking a dermis and epidermis boundary as a depth reference plane, and calculating a superficial / deep blood vessel density ratio. According to the binarized image, vertical blood vessels are recognized, and the proportion of the vertical blood vessels is calculated; and all the features are integrated into nevus flammeus lesion features. Characteristic quantification of lesion blood vessels is realized, and objective analysis basis is provided for photodynamic therapy.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method and system for extracting port-wine stain lesion features based on OCTA technology. Background Technology

[0002] In the clinical diagnosis and treatment of port-wine stains, the traditional method of relying on physicians to assess the color changes of lesions by visual inspection has obvious limitations: it cannot penetrate the epidermis to observe the three-dimensional structural features of malformed blood vessels in the dermis, resulting in the inability to quantify key structural information such as the depth distribution of lesions, blood vessel density, and vessel diameter distribution.

[0003] While existing optical imaging technologies can provide partial images of blood vessels, they lack systematic algorithms to extract quantitative indicators of lesions. For example, they cannot accurately analyze the spatial distribution patterns of blood vessels in layers, nor can they establish a quantitative correlation between vascular structural features and treatment response. This technological deficiency has led to a long-term reliance on subjective experience for dosage selection and efficacy prediction in vascular-targeted photodynamic therapy, resulting in high treatment ineffectiveness and an inability to predict outcomes in advance.

[0004] Although optical coherence tomography (OCT) has recently provided a new tool for visualizing blood vessels in the skin, its application in port-wine stains remains at the imaging stage.

[0005] The lack of quantitative standards described above severely hinders the realization of precise treatment for port-wine stains and is a core bottleneck that urgently needs to be overcome in clinical practice. Summary of the Invention

[0006] Therefore, it is necessary to provide a method and system for extracting port-wine stain lesion features based on OCTA technology to address the aforementioned technical problems.

[0007] Firstly, this application provides a method for extracting port-wine stain lesion features based on OCTA technology, including:

[0008] S1. Use OCTA technology based on OMAG algorithm to scan the target tissue and obtain a three-dimensional tomographic image containing tissue depth structural information and a three-dimensional projection image containing blood flow information.

[0009] S2. Based on three-dimensional tomographic images, identify the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set;

[0010] S3. Calculate the epidermal layer thickness based on the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set to obtain an epidermal layer thickness distribution map.

[0011] S4. Perform adaptive threshold segmentation and median filtering on the 3D projection image to generate a denoised binarized blood vessel image, and then perform skeletonization on the denoised binarized blood vessel image to generate a blood vessel skeleton image with a single pixel width.

[0012] S5. Calculate the average blood vessel density based on the denoised binarized blood vessel image, and calculate the average blood vessel diameter and the proportion of blood vessel length of different diameters based on the denoised binarized blood vessel image and the blood vessel skeleton map.

[0013] S6. Divide the vascular layer region using the dermal-epidermal boundary coordinate set as the depth reference plane, and calculate the ratio of superficial vascular density to deep vascular density based on the vascular layer region.

[0014] S7. Based on the denoised binary vascular image, identify vertical blood vessels and calculate the proportion of vertical blood vessels.

[0015] S8. The distribution map of epidermal thickness, average blood vessel density, average blood vessel diameter, proportion of blood vessel length of different diameters, ratio of superficial blood vessel density to deep blood vessel density, and proportion of vertical blood vessels are used as the lesion characteristics of port-wine stains.

[0016] Secondly, this application also provides a port-wine stain lesion feature extraction system based on OCTA technology, applied to the method described in the first aspect, comprising:

[0017] The image acquisition module is used to scan the target tissue using OCTA technology based on the OMAG algorithm, and to acquire a three-dimensional tomographic image containing tissue depth and structural information as well as a three-dimensional projection image containing blood flow information.

[0018] The boundary recognition module is used to identify the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set based on three-dimensional tomographic images.

[0019] The thickness calculation module is used to calculate the epidermal layer thickness based on the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set, and obtain the epidermal layer thickness distribution map.

[0020] The image processing module is used to perform adaptive threshold segmentation and median filtering on the three-dimensional projection image to generate a denoised binarized blood vessel image, and to perform skeletonization processing on the denoised binarized blood vessel image to generate a blood vessel skeleton map with a single pixel width.

[0021] The vascular feature calculation module is used to calculate the average vascular density based on the denoised binarized vascular image, and to calculate the average vascular diameter and the proportion of vascular length of different diameters based on the denoised binarized vascular image and the vascular skeleton map.

[0022] The vascular layering module is used to divide the vascular layering region using the dermal-epidermal boundary coordinate set as the depth reference plane, and calculate the ratio of superficial vascular density to deep vascular density based on the vascular layering region.

[0023] The vertical blood vessel recognition module is used to identify vertical blood vessels and calculate the proportion of vertical blood vessels based on the denoised binarized blood vessel image.

[0024] The feature integration module is used to integrate the epidermal thickness distribution map, average blood vessel density, average blood vessel diameter, proportion of blood vessel length with different diameters, ratio of superficial blood vessel density to deep blood vessel density, and proportion of vertical blood vessels as lesion features of port-wine stains.

[0025] The aforementioned method and system for extracting port-wine stain lesion features based on OCTA technology acquires three-dimensional tomographic and three-dimensional projection images of port-wine stain lesions using optical coherence tomography (OCT) vascular imaging technology. Based on the tomographic images, the boundary coordinate sets between the epidermis and air, and between the dermis and epidermis, are extracted to obtain an epidermal thickness distribution map. The projection image undergoes adaptive threshold segmentation and median filtering to generate a denoised binary vascular image. This denoised binary vascular image is then skeletonized to generate a single-pixel-width vascular skeleton map. The average vascular density is calculated based on the denoised binary vascular image, and the average vascular diameter and the proportion of different vascular lengths are calculated based on the denoised binary vascular image and the vascular skeleton map. Vascular stratification regions are divided using the dermal-epidermal boundary coordinate set as a depth reference plane, and the ratio of superficial to deep vascular density is calculated based on these stratified regions. Vertical vessels are identified based on the denoised binary vascular image, and their proportion is calculated. All obtained features are integrated as the port-wine stain lesion features. The above process enables the quantification of the characteristics of blood vessels in the lesion, providing an objective analytical basis for photodynamic therapy. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a method for extracting port-wine stain lesion features based on OCTA technology provided by the present invention;

[0028] Figure 2 This is a schematic diagram of the process for identifying the skin-air boundary coordinate set in an optional embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of the process for identifying the dermal-epidermal boundary coordinate set in an optional embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of a port-wine stain lesion feature extraction system based on OCTA technology provided by the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0032] S1. Use OCTA technology based on OMAG algorithm to scan the target tissue and obtain a three-dimensional tomographic image containing tissue depth and structural information as well as a three-dimensional projection image containing blood flow information.

[0033] Specifically, when using OCTA technology based on the OMAG algorithm to scan target tissue, the OMAG algorithm acquires reflected light signals at different depths of the tissue through rapid and continuous low-coherence optical interferometry, thereby constructing a three-dimensional tomographic image containing tissue depth and structural information. The gray value of each voxel in this three-dimensional tomographic image represents the intensity of reflection of incident light by the tissue at that location, thus reflecting the morphological characteristics of the tissue.

[0034] Meanwhile, OCTA technology is based on the Doppler effect. When red blood cells flow within blood vessels, they cause a frequency shift in reflected light. By analyzing and processing this frequency shift signal, a three-dimensional projection image containing blood flow information can be generated without the use of contrast agents. This image presents the blood flow signal in a specific encoding method, such as brightness variations or pseudo-color annotations, allowing visualization of the distribution, direction, and hemodynamic characteristics of blood vessels.

[0035] S2. Based on three-dimensional tomographic images, identify the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set.

[0036] Specifically, the skin-air boundary is the interface between the outermost layer of tissue and the surrounding air. It exhibits unique reflective characteristics in images, possessing high reflectivity and relatively regular boundaries. By setting specific grayscale thresholds and edge detection algorithms, such as using the Sobel or Canny operators, the pixels on this boundary can be identified, and their three-dimensional coordinates recorded, forming a skin-air boundary coordinate set.

[0037] The dermal-epidermal boundary is the interface between the epidermis and dermal tissues, and its reflective properties differ from those of the epidermal-air boundary. The optical properties of dermal tissue are more complex, containing more cells, fibers, and blood vessels, resulting in variations in reflected light signals. Semantic segmentation models from deep learning or texture-based classification algorithms can be used to classify each voxel in a 3D tomographic image, identifying pixels on the dermal-epidermal boundary and recording their 3D coordinates to construct a dermal-epidermal boundary coordinate set.

[0038] S3. Calculate the epidermal layer thickness based on the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set, and obtain the epidermal layer thickness distribution map.

[0039] Specifically, in three-dimensional space, for each corresponding lateral position (x, y), the epidermal thickness at that position can be obtained by calculating the coordinate difference between the epidermal-air boundary point and the dermal-epidermal boundary point along the depth direction (z-axis). That is, epidermal thickness d(x, y) = z_dermal-epidermal boundary(x, y) - z_epidermal-air boundary(x, y). Interpolating the thickness values ​​at all lateral positions generates a complete epidermal thickness distribution map. This map displays the thickness variation of the epidermis within the target tissue region in a two-dimensional or pseudo-three-dimensional form, with different colors or grayscale levels representing different thickness ranges.

[0040] S4. Perform adaptive threshold segmentation and median filtering on the 3D projection image to generate a denoised binarized blood vessel image, and then perform skeletonization processing on the denoised binarized blood vessel image to generate a blood vessel skeleton map with a single pixel width.

[0041] Specifically, due to the significant differences in vascular signal intensity across different regions of a 3D projection image, traditional fixed-threshold segmentation methods struggle to accurately separate blood vessels from background tissue. Adaptive threshold segmentation algorithms dynamically determine the threshold for each pixel based on the statistical characteristics of local image regions, such as mean and standard deviation. For example, in darker areas of the image, where vascular signals are relatively weak, the adaptive threshold is lowered to ensure detection of these subtle vascular signals; conversely, in brighter areas, the adaptive threshold is raised to avoid misinterpreting background noise as vascular signals. After adaptive threshold segmentation, a preliminary binary vascular image is obtained, where vascular pixels are labeled as foreground (e.g., white) and background pixels are labeled as background (e.g., black).

[0042] To further remove noise and improve the quality of vascular images, median filtering is employed. Median filtering uses a sliding window approach, sorting pixel values ​​within the window and replacing the center pixel's value with the median. This method is highly effective at removing random noise points such as salt-and-pepper noise, while preserving the integrity of blood vessel edges as much as possible. The size of the filtering window can be selected based on the resolution of the vascular image and the noise characteristics; for example, a smaller window (such as 3×3 or 5×5) can be used to remove small noise points, while a larger window can be used to handle denser noise areas.

[0043] Skeletonization involves gradually eroding the boundaries of the blood vessel region while preserving the vessel's topology, until only a single-pixel-wide skeleton remains. Skeletonization algorithms can be based on distance transforms or morphological operations. Distance transform-based methods first calculate the distance from each foreground pixel to the nearest background pixel, then find the pixel with the largest local distance as the skeleton point. Morphological operation-based methods, on the other hand, iteratively erode and dilate the vessel boundaries through repeated erosion and dilation operations, ultimately obtaining the skeleton.

[0044] S5. Calculate the average blood vessel density based on the denoised binarized blood vessel image, and calculate the average blood vessel diameter and the proportion of blood vessel length of different diameters based on the denoised binarized blood vessel image and the blood vessel skeleton map.

[0045] Specifically, when calculating the average blood vessel density based on a denoised binarized blood vessel image, the total area of ​​the analysis region is used. Within this region, the number of all foreground pixels belonging to blood vessels is counted. Dividing the number of blood vessel pixels by the total area of ​​the analysis region yields the average blood vessel density value. This indicator reflects the degree of occupancy of blood vessels per unit area and can be used to assess the richness and density of blood vessel distribution.

[0046] When calculating the average vessel diameter and the proportion of vessel lengths with different diameters, a denoised binarized vessel image and a vessel skeleton map are combined. The skeleton map is used to determine the centerline path of the vessel, and then the vessel diameter is calculated along these paths in the denoised binarized vessel image. For each vessel segment, the width of the vessel can be measured perpendicular to the skeleton. The diameter of the vessel segment is obtained by averaging the widths of multiple measurement points. The diameters of all vessel segments are statistically analyzed to calculate the average vessel diameter. Simultaneously, based on preset diameter intervals (such as small vessels, medium vessels, and large vessels), the proportion of vessel length in each interval to the total vessel length is calculated, thus obtaining the proportion of vessel lengths with different diameters.

[0047] S6. Divide the vascular layer region using the dermal-epidermal boundary coordinate set as the depth reference plane, and calculate the ratio of superficial vascular density to deep vascular density based on the vascular layer region.

[0048] Specifically, the criteria for dividing superficial and deep vessels can be determined based on clinical experience and anatomical knowledge. For example, vessels in and above the papillary dermis can be classified as superficial vessels, while those in the reticular layer and deeper can be classified as deep vessels. Alternatively, vessels with different functions and distribution characteristics can be stratified by analyzing their hemodynamic and morphological features. After determining the stratified regions, vessel density is calculated in both superficial and deep vessel regions. Similar to the method used to calculate the average vessel density, the ratio of the number of vessel pixels to the corresponding area in each stratified region is statistically analyzed to obtain the superficial and deep vessel densities. Finally, the ratio of superficial to deep vessel density is calculated, which reflects the differences in the distribution ratio of vessels at different depths.

[0049] S7. Based on the denoised binary vascular image, identify vertical blood vessels and calculate the proportion of vertical blood vessels.

[0050] Specifically, in denoised binary vascular images, blood vessels are represented by white pixels against a black background. When identifying vertical blood vessels, the direction of the vessels is first analyzed. Vertical blood vessels appear in a two-dimensional image as segments roughly perpendicular to the skin surface. Vertical blood vessels can be identified by statistically analyzing the direction of these segments. More specifically, in denoised binary vascular images, connected component analysis is performed on the blood vessels to determine each individual branch. For each branch, the direction is calculated by analyzing the distribution of its pixels. For example, the least squares method can be used to fit the pixels of the blood vessel branch to obtain its linear equation, thereby determining the vessel's orientation angle.

[0051] The orientation angle refers to the angle between a blood vessel and a preset reference direction (such as the horizontal or vertical axis of an image). In a two-dimensional image, if the horizontal axis of the image is used as a reference, the orientation angle of a vertical blood vessel should be close to 90 degrees. A range of angles is set, for example, from 80 to 100 degrees, and blood vessel branches with orientation angles within this range are identified as vertical blood vessels.

[0052] When calculating the proportion of vertical vessels, the total length of all vessel segments identified as vertical vessels is counted and then divided by the total length of all vessel segments in the denoised binarized vascular image. The resulting proportion of vertical vessels reflects the ratio of vertically oriented vessels in the overall vascular network within the two-dimensional image plane.

[0053] S8. The distribution map of epidermal thickness, average blood vessel density, average blood vessel diameter, proportion of blood vessel length of different diameters, ratio of superficial blood vessel density to deep blood vessel density, and proportion of vertical blood vessels are used as the lesion characteristics of port-wine stains.

[0054] Specifically, the epidermal thickness distribution map, average vessel density, average vessel diameter, proportion of vessel lengths of different diameters, ratio of superficial to deep vessel density, and proportion of vertical vessels obtained in the previous steps are integrated to form a complete set of port-wine stain lesion features. These features comprehensively describe the morphological and vascular characteristics of port-wine stain lesions from different perspectives, providing rich and accurate quantitative information for subsequent disease diagnosis, treatment planning, efficacy evaluation, and prognosis.

[0055] The aforementioned method for extracting port-wine stain lesion features based on OCTA technology acquires three-dimensional tomographic and three-dimensional projection images of port-wine stain lesions using optical coherence tomography (OCT) angiography. Based on the tomographic images, it extracts the boundary coordinate sets between the epidermis and air, and between the dermis and epidermis, to obtain an epidermal thickness distribution map. The projection image undergoes adaptive thresholding and median filtering to generate a denoised binary vascular image, which is then skeletonized to generate a single-pixel-width vascular skeleton map. The average vascular density is calculated based on the denoised binary vascular image, and the average vascular diameter and the proportion of different vascular lengths are calculated based on the denoised binary vascular image and the vascular skeleton map. The vascular stratification region is divided using the dermal-epidermal boundary coordinate set as a depth reference plane, and the ratio of superficial to deep vascular density is calculated based on the stratified vascular regions. Vertical vessels are identified based on the denoised binary vascular image, and their proportion is calculated. All obtained features are integrated as the port-wine stain lesion features. The above process enables the quantification of the characteristics of blood vessels in the lesion, providing an objective analytical basis for photodynamic therapy.

[0056] refer to Figure 2 In an optional embodiment, step S2, identifying the skin-air boundary coordinate set, includes the following steps:

[0057] S211. Calculate the vertical gradient intensity and horizontal gradient intensity for each pixel column of the three-dimensional tomographic image.

[0058] S212. Based on the vertical and horizontal gradient intensities, calculate the connection weights between adjacent pixels using the gradient fusion algorithm; the expression for the gradient fusion algorithm is:

[0059] w(m,n)=λ1[(1-grady(m))+(1-grady(n))]+λ2[(1-gradx(m))+(1-gradx(n))]+σ;

[0060] Where m and n are the pixel coordinates of adjacent path points, and m represents the pixel coordinate (x, y). m ,y m ), where n represents the pixel coordinates (x, y). n ,y nw(m,n) represents the connection weight between pixel m and pixel n, indicating the connection cost between adjacent pixels m and n; 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 Let represent 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.

[0061] S213. Starting from the leftmost pixel of the three-dimensional tomographic image, iteratively search for the path with the minimum cumulative connection weight in five neighborhood directions: right, upper right, lower right, upper right, and lower right, to obtain the skin-air boundary coordinate set.

[0062] Specifically, when identifying the skin-air boundary coordinate set, gradient calculation is first performed for each pixel column of the 3D tomographic image, that is, calculating the vertical and horizontal gradient intensities for each pixel. The vertical gradient intensity grady(m) reflects the rate of gray-level change of the pixel in the vertical direction, and its calculation formula is grady(m) = y m+1 -y m-1 , where y m+1 and y m-1 These represent the grayscale values ​​of the pixel m in the vertical direction, preceding and following each other. Similarly, the horizontal gradient intensity gradx(m) reflects the rate of change of grayscale value of a pixel in the horizontal direction, and its calculation formula is gradx(m) = x m+1 -x m-1 , where x m+1 and x m-1 These represent the grayscale values ​​of the pixel m in the horizontal direction, one before and one after the previous pixel.

[0063] After obtaining the vertical and horizontal gradient intensities of each pixel, the connection weights between adjacent pixels are calculated based on the gradient fusion algorithm. The expression for the gradient fusion algorithm is w(m,n)=λ1[(1-grady(m))+(1-grady(n))]+λ2[(1-gradx(m))+(1-gradx(n))]+σ. Here, m and n are the coordinates of adjacent pixels, and m represents the pixel coordinates (x, y, y). m ,y m ), where n represents the pixel coordinates (x, y). n ,y nw(m,n) represents the connection weight between pixel m and pixel n, used to measure the connection cost between adjacent pixels. λ1 and λ2 are the weight coefficients of the vertical and horizontal gradients, respectively, used to adjust the influence of the vertical and horizontal gradients in the connection weight calculation. σ is a smoothing constant used to avoid the connection weights being too small due to excessively large gradients, ensuring the stability of the algorithm.

[0064] The core idea of ​​gradient fusion algorithms is to comprehensively consider the gradient information of adjacent pixels to determine the degree of connection between them. Adjacent pixel pairs with smaller gradient values ​​are more likely to belong to the boundaries of the same region, and therefore have relatively larger connection weights. In this way, boundary regions in an image can be effectively identified.

[0065] To apply this algorithm in practice, we can start from the leftmost pixel of the 3D tomographic image and iteratively search in five neighborhood directions: right, upper right, lower right, upper right, and lower right. The goal of the search is the path with the minimum cumulative connection weight, which can be achieved using a dynamic programming algorithm. The dynamic programming algorithm records the minimum cumulative weight of each pixel reaching the starting point and gradually expands the path to the right until it finds a path with the minimum cumulative connection weight. This path represents the boundary between the skin and the air.

[0066] In the implementation, a weight accumulation matrix with the same size as the image is first initialized, and the accumulated weight of the first pixel is set to zero. Then, each pixel is processed sequentially, and the accumulated weights of the pixels to its right and lower right are updated according to the connection weights calculated by the gradient fusion algorithm. In this way, the path is gradually expanded to the right until all pixels in the image have been processed. Finally, by backtracking the accumulated weight matrix, the path with the minimum accumulated weight is determined, thus obtaining the skin-air boundary coordinate set.

[0067] refer to Figure 3 In an optional embodiment, step S2, identifying the dermal-epidermal boundary coordinate set, includes the following steps:

[0068] S221. For a three-dimensional tomographic image, perform horizontal A-line signal intensity integral projection on the image region below the skin-air boundary coordinate set to obtain the first signal intensity curve distributed along the depth.

[0069] S222. If the first signal intensity curve has a single second peak, then the depth coordinates corresponding to the single second peak are taken as the dermal-epidermal interface position.

[0070] S223. If multiple stray peaks appear in the first signal intensity curve due to skin surface distortion, a bicubic spline interpolation algorithm is used to flatten the three-dimensional tomographic image. The horizontal A-line signal intensity integral projection is performed on the flattened image to obtain the second signal intensity curve. The second peak is located in the second signal intensity curve.

[0071] S224. Map the single second peak of the first signal intensity curve or the second peak of the second signal intensity curve back to the coordinate system of the three-dimensional tomographic image according to the lateral position to obtain the dermal-epidermal boundary coordinate set.

[0072] Specifically, A-line signal strength integral projection refers to the accumulation of signal strength at each depth along the horizontal direction. This operation can compress a two-dimensional image region into a one-dimensional signal strength curve, which reflects the variation of the total signal strength at different depths below the skin-air boundary.

[0073] Next, the obtained first signal intensity curve is analyzed. If a single second peak exists in the curve, it indicates that the signal intensity reaches a significant local maximum at that depth. Based on the optical properties of skin tissue, the dermal-epidermal interface typically experiences significant changes in signal intensity due to variations in tissue composition and structure. Therefore, the depth coordinates corresponding to this single second peak can be determined as the location of the dermal-epidermal interface.

[0074] However, in practical applications, factors such as skin surface distortion may cause multiple stray peaks in the first signal intensity curve. In such cases, it is difficult to accurately determine the dermal-epidermal interface location directly based on the first signal intensity curve. To solve this problem, a bicubic spline interpolation algorithm can be used to flatten the three-dimensional tomographic image. Bicubic spline interpolation is a commonly used image interpolation method that estimates pixel values ​​at non-integer coordinates based on the values ​​of surrounding pixels, thereby achieving image smoothing and geometric correction. This algorithm can correct image deformation caused by skin surface distortion to a certain extent, making the boundaries of each skin layer straighter and more regular.

[0075] The flattened image is then subjected to horizontal A-line signal intensity integral projection again to obtain a second signal intensity curve. Because the image has been flattened, stray peaks in the second signal intensity curve are suppressed, and the signal intensity distribution more closely matches the actual tissue structure characteristics. The second peak is located within the second signal intensity curve, and similarly, the depth coordinates corresponding to this second peak are determined as the location of the dermal-epidermal interface.

[0076] Finally, the single second peak in the first signal intensity curve or the second peak in the second signal intensity curve is mapped back to the coordinate system of the three-dimensional tomographic image according to its lateral position. This lateral position mapping refers to converting the peak position into a specific coordinate point in the three-dimensional tomographic image based on the correspondence between the peak position in the signal intensity curve and the lateral position in the image. In this way, a complete dermal-epidermal boundary coordinate set is obtained, which describes the positional distribution of the dermal-epidermal interface in the three-dimensional image space.

[0077] In an optional embodiment, step S4 includes the following steps:

[0078] Calculate the average intensity of the N×N neighboring pixels of each pixel in the 3D projected image.

[0079] If the pixel value of a pixel is greater than the average intensity of the corresponding N×N neighboring pixels, then the pixel value of the pixel is set to 1; otherwise, it is set to 0, generating an initial binary image; where N is a preset positive integer.

[0080] An N×N neighborhood median filter is applied to the initial binary image, and the median value of the center pixel is replaced with the median value of the neighboring pixels to obtain a denoised binary blood vessel image.

[0081] The Hilditch iterative algorithm is applied to the denoised binarized blood vessel image to remove boundary pixels until only pixels with a width of 1 remain, thus obtaining the blood vessel skeleton map.

[0082] Specifically, the average intensity of N×N neighboring pixels is calculated for each pixel. This average reflects the average intensity of pixels within the local area. The current pixel's value is then compared to this neighborhood average. If the pixel value is greater than the neighborhood average, it indicates that the pixel's intensity is relatively high in the local area, potentially indicating a blood vessel signal; therefore, its pixel value is set to 1 (white), representing the foreground. Otherwise, it is set to 0 (black), representing the background. In this way, blood vessels and background areas can be initially distinguished, generating an initial binary image. Here, N is a preset positive integer, which can be selected based on the expected width of the blood vessels and the image resolution. For example, N can be an odd number such as 3, 5, or 7 to ensure the central symmetry of the neighborhood window.

[0083] Next, median filtering of the N×N neighborhood is performed on the initial binary image. Median filtering is a non-linear filtering technique that effectively removes random noise points such as salt-and-pepper noise while maintaining the integrity of blood vessel edges. During the filtering process, for each pixel in the image, the pixel values ​​within its N×N neighborhood are considered. These pixel values ​​are sorted, and the median value is taken, replacing the original value of the center pixel. After median filtering, isolated noise points and small artifacts in the initial binary image are removed, and the continuity and connectivity of blood vessels are enhanced, resulting in a denoised binary blood vessel image.

[0084] Finally, the Hilditch iterative algorithm was applied to the denoised binarized blood vessel image to generate a single-pixel-wide blood vessel skeleton map. The Hilditch algorithm is a skeleton extraction method whose basic idea is to iteratively remove boundary pixels, gradually refining the blood vessel structure until only a single-pixel-wide skeleton remains. In each iteration, the algorithm identifies and removes boundary pixels that can be removed without disrupting blood vessel connectivity. Specifically, the algorithm checks the neighboring pixels of each blood vessel pixel, determining whether the pixel can be deleted based on preset deletion conditions (such as the number of neighboring connections and the distribution of surrounding pixels). This process is repeated until no more deletable pixels remain in the image. In the final blood vessel skeleton map, blood vessels are represented as thin lines of single-pixel width, preserving the topological structure and main branch information of the blood vessels, providing a foundation for subsequent blood vessel feature analysis.

[0085] In one optional embodiment, calculating the average vascular density includes:

[0086] The total number of pixels with a value of 1 in the denoised binarized vascular image is counted as A. The average vascular density is obtained by dividing the total number of pixels A by the total number of pixels M in the denoised binarized vascular image.

[0087] Specifically, the total number of pixels A with a value of 1 in the denoised binarized vascular image is counted. In a binary image, pixels with a value of 1 represent vascular regions, so A reflects the total area covered by blood vessels in the image. The total number of pixels M in the denoised binarized vascular image is obtained by multiplying the image's width (number of pixels horizontally) by its height (number of pixels vertically). Dividing the total number of pixels A by the total number of pixels M, i.e., D = A / M, gives the average vascular density. This density value represents the proportion of vascular pixels to the total number of pixels in the image, and is a quantitative indicator of the richness of blood vessels.

[0088] In one optional embodiment, calculating the average vessel diameter includes:

[0089] Count the total number of pixels L with a value of 1 in the vascular skeleton image, and calculate the ratio R1 of the total number of pixels A to the total number of pixels L.

[0090] Multiplying the ratio R1 by the scale factor yields the average blood vessel diameter. The scale factor represents the number of micrometers corresponding to one pixel.

[0091] Specifically, first, the total number of pixels with a value of 1 (L) in the vascular skeleton image is counted. This represents the total length of the vascular skeleton, expressed in pixels. Then, the total number of pixels with a value of 1 (A) in the denoised binarized vascular image is obtained, representing the total area of ​​the blood vessel. Next, the ratio R1 of the total number of pixels A to the total number of pixels L is calculated: R1 = A / L. This ratio R1 actually reflects the average width of the blood vessel. Finally, the ratio R1 is multiplied by a scale factor to obtain the average blood vessel diameter. This scale factor, representing the number of micrometers corresponding to one pixel, is determined by the calibration parameters of the OCTA device or the image resolution and is used to convert pixel units to actual length units (such as micrometers). For example, if the scale factor is 2 micrometers / pixel, the calculated average blood vessel diameter will be in micrometers. This calculation process provides a quantitative indicator describing the average diameter of the blood vessel by converting the ratio of blood vessel area to length into actual length units.

[0092] In one optional embodiment, calculating the proportion of blood vessel lengths with different diameters includes the following steps:

[0093] Calculate the maximum Euclidean distance from each skeleton pixel in the vascular skeleton image to the nearest background pixel; where the background pixel is the pixel with a pixel value of 0 in the vascular skeleton image.

[0094] The local blood vessel diameter is calculated based on the maximum Euclidean distance, and then grouped into multiple diameter intervals according to preset diameter value intervals. The formula for calculating the local blood vessel diameter is as follows:

[0095] d = 2 × d max ×c L ;

[0096] Where d is the local blood vessel diameter, d max For the maximum Euclidean distance, c L This is the scale factor.

[0097] For different pipe diameter ranges, calculate the value per 1 mm using the following formula. 2 R2: Percentage of vessel lengths of different diameters

[0098]

[0099] Where A is the total number of pixels with a value of 1 in the denoised binarized vascular image, L is the total number of pixels with a value of 1 in the vascular skeleton image, and M is the total number of pixels in the denoised binarized vascular image.

[0100] Specifically, for each pixel in the vascular skeleton image, the maximum Euclidean distance to the nearest background pixel is calculated. Here, background pixels refer to pixels with a value of 0 in the vascular skeleton image. The distance calculation can be implemented using a distance transformation algorithm, such as using Euclidean distance transformation to calculate the Euclidean distance to the nearest background pixel for each foreground pixel (pixel value of 1). Among these distance values, the maximum value is taken as the maximum Euclidean distance d corresponding to the blood vessel position of that pixel. max .

[0101] Next, using the formula d = 2 × d max ×c L Calculate the local blood vessel diameter. Where d represents the local blood vessel diameter. max It is the maximum Euclidean distance calculated previously, c L This is the scale factor, used to convert pixel distances into actual length units (such as micrometers or millimeters). Multiplying by 2 is because the maximum Euclidean distance represents the radius of the blood vessel, and the diameter is twice the radius.

[0102] Then, based on the calculated local blood vessel diameter values, they are grouped according to preset diameter intervals. For example, diameter intervals can be set as 0-25μm, 25-50μm, 50-75μm, etc., and each local blood vessel diameter is assigned to the corresponding diameter interval.

[0103] For each pipe diameter range, calculate every 1 mm. 2 The percentage of vessel length for different diameters, R2. The specific formula is as follows: R2=[A / (L×M)]×c L Where A is the total number of pixels with a value of 1 in the denoised binarized blood vessel image. L is the total number of pixels with a value of 1 in the blood vessel skeleton image. M is the total number of pixels in the denoised binarized blood vessel image. L This is the scale factor, used to convert pixel units to actual length units. Here, A / (L×M) is actually the ratio of blood vessel area density to blood vessel length density, multiplied by the scale factor c. L To obtain every 1mm 2 The percentage of blood vessel length within a given diameter range within an area. This method allows for the quantification of the length distribution of blood vessels of different diameters within a unit area.

[0104] In an optional embodiment, step S6 includes the following steps:

[0105] The ordinate of the dermal-epidermal boundary coordinate set is used as the zero point of the depth reference, z0.

[0106] In the denoised binarized blood vessel image, the number of blood vessel pixels V within the depth layer from z0 to z0+Δz1 is counted. shallow and the number of blood vessel pixels V in the depth layer from z0+Δz1 to z0+Δz2deep ; where Δz1 and Δz2 are preset depth values.

[0107] The density ratio R3 of superficial blood vessels to deep blood vessels is calculated using the following formula:

[0108]

[0109] Among them, M shallow M represents the total number of pixels in the superficial region of a denoised binarized vascular image. deep This represents the total number of pixels in the deep region of a denoised binarized vascular image.

[0110] Specifically, a reference point is first determined, that is, the ordinate of the dermal-epidermal boundary coordinate set is taken as the zero point of the depth reference z0, and all subsequent depth measurements are calculated from this reference point.

[0111] Next, the number of blood vessel pixels in two regions is counted in the denoised binarized blood vessel image. One is the number of blood vessel pixels V in the shallow region from z0 to z0+Δz1. shallow Δz1 is a pre-set depth value; the other is the number of blood vessel pixels V in the deep region from z0+Δz1 to z0+Δz2. deep Δz2 is also a pre-set depth value. For example, the shallow layer is 0-400μm below the dermal-epidermal boundary, and the deep layer is 400-600μm.

[0112] Then, we also need to calculate the total number of pixels M for the shallow and deep regions. shallow and M deep This step involves calculating the total number of pixels within the two depth ranges mentioned above in the denoised binarized blood vessel image.

[0113] After obtaining the above data, first calculate the density of superficial blood vessels, that is, using the number of pixels V of superficial blood vessels. shallow Divide by the total number of pixels in the shallow region M shallow This yields the proportion of superficial blood vessels in the total pixels. Similarly, the density of deep blood vessels is the number of deep blood vessel pixels, V. deep Divide by the total number of pixels in the deep region M deep .

[0114] Finally, calculate the ratio R3 of superficial to deep vessel density by dividing the superficial vessel density by the deep vessel density. The ratio R3 reflects the relative relationship between superficial and deep vessel densities. If R3 is greater than 1, it indicates that the superficial vessel density is higher; if it is less than 1, the deep vessel density is higher.

[0115] In an optional embodiment, step S7 includes the following steps:

[0116] In the denoised binary vascular image, non-continuous blood vessels are outlined with rectangular boxes and identified as non-continuous blood vessels. If the target conditions are met, the non-continuous blood vessels are judged to be circular.

[0117] The ratio of the total number of pixels in the near-circular shape to the total number of pixels in the denoised binarized blood vessel image is used as the vertical blood vessel proportion.

[0118] The target conditions are: the difference between the length and width of the rectangle does not exceed δ1, the area ratio of blood vessels in the rectangle is greater than δ2, and the proportion of non-continuous blood vessels in the denoised binary blood vessel image is greater than δ3. δ1, δ2, and δ3 are preset parameters.

[0119] Specifically, this involves identifying discontinuous blood vessels in the image. These vessels appear discontinuous in the image due to their vertical orientation. To identify these vessels, a rectangular bounding box can be used to select the suspected area.

[0120] Next, morphological analysis is performed on the blood vessels within the selected area to determine if they conform to the characteristics of vertical blood vessels. Three conditions are set to determine whether a vertical blood vessel is circular or nearly circular. First, the difference between the length and width of the rectangle cannot exceed a preset threshold δ1. This condition ensures that the selected area is roughly square, as vertical blood vessels often appear close to circular or square in 2D projection. Second, the area occupied by the blood vessel within the rectangle needs to be greater than another preset parameter δ2. This means that the blood vessel must occupy a sufficiently large proportion within the rectangle to exclude smaller, non-featured structures that may simply be accidental. Finally, the proportion of discontinuous blood vessels in the entire image must also be greater than a third parameter δ3. This condition ensures that the analyzed blood vessels have a certain degree of salience within the overall vascular network, preventing misjudgments due to local noise or small vascular branches.

[0121] When a blood vessel region meets all the above conditions, it is identified as a near-circular vertical blood vessel. Subsequently, the proportion of vertical blood vessels is obtained by calculating the sum of the pixels of all such vertical blood vessels and comparing it with the total number of pixels in the entire denoised binarized blood vessel image.

[0122] The aforementioned method for extracting port-wine stain lesion features based on OCTA technology acquires three-dimensional tomographic and three-dimensional projection images of port-wine stain lesions using optical coherence tomography (OCT) angiography. Based on the tomographic images, it extracts the boundary coordinate sets between the epidermis and air, and between the dermis and epidermis, to obtain an epidermal thickness distribution map. The projection image undergoes adaptive thresholding and median filtering to generate a denoised binary vascular image, which is then skeletonized to generate a single-pixel-width vascular skeleton map. The average vascular density is calculated based on the denoised binary vascular image, and the average vascular diameter and the proportion of different vascular lengths are calculated based on the denoised binary vascular image and the vascular skeleton map. The vascular stratification region is divided using the dermal-epidermal boundary coordinate set as a depth reference plane, and the ratio of superficial to deep vascular density is calculated based on the stratified vascular regions. Vertical vessels are identified based on the denoised binary vascular image, and their proportion is calculated. All obtained features are integrated as the port-wine stain lesion features. The above process enables the quantification of the characteristics of blood vessels in the lesion, providing an objective analytical basis for photodynamic therapy.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] Based on the same inventive concept, this application also provides a system for implementing the above-mentioned method for extracting port-wine stain lesion features based on OCTA technology. The solution provided by this system is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the port-wine stain lesion feature extraction system based on OCTA technology provided below can be found in the limitations of the port-wine stain lesion feature extraction method based on OCTA technology described above, and will not be repeated here.

[0125] In one exemplary embodiment, such as Figure 4 As shown, a port-wine stain lesion feature extraction system 40 based on OCTA technology is provided, applied to the methods in the above-described method embodiments. The system includes:

[0126] The image acquisition module 41 is used to scan the target tissue using OCTA technology based on the OMAG algorithm, and acquire a three-dimensional tomographic image containing tissue depth structural information and a three-dimensional projection image containing blood flow information.

[0127] Boundary recognition module 42 is used to identify the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set based on three-dimensional tomographic images.

[0128] The thickness calculation module 43 is used to calculate the epidermal layer thickness based on the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set, and obtain the epidermal layer thickness distribution map.

[0129] Image processing module 44 is used to perform adaptive threshold segmentation and median filtering on the three-dimensional projection image to generate a denoised binarized blood vessel image, and to perform skeletonization processing on the denoised binarized blood vessel image to generate a blood vessel skeleton map with a single pixel width.

[0130] The vascular feature calculation module 45 is used to calculate the average vascular density based on the denoised binarized vascular image, and to calculate the average vascular diameter and the proportion of vascular length of different diameters based on the denoised binarized vascular image and the vascular skeleton map.

[0131] The vascular layering module 46 is used to divide the vascular layering region using the dermal-epidermal boundary coordinate set as the depth reference plane, and to calculate the ratio of superficial vascular density to deep vascular density based on the vascular layering region.

[0132] The vertical blood vessel recognition module 47 is used to identify vertical blood vessels and calculate the proportion of vertical blood vessels based on the denoised binarized blood vessel image.

[0133] The feature integration module 48 is used to integrate the epidermal thickness distribution map, average blood vessel density, average blood vessel diameter, proportion of blood vessel length with different diameters, ratio of superficial blood vessel density to deep blood vessel density, and proportion of vertical blood vessels as features of port-wine stain lesions.

[0134] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0135] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for extracting port-wine stain lesion features based on OCTA technology, characterized in that, The method includes: S1. Use OCTA technology based on OMAG algorithm to scan the target tissue and obtain a three-dimensional tomographic image containing tissue depth structural information and a three-dimensional projection image containing blood flow information. S2. Based on the three-dimensional tomographic image, identify the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set; S3. Calculate the epidermal layer thickness based on the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set to obtain an epidermal layer thickness distribution map; S4. Perform adaptive threshold segmentation and median filtering on the three-dimensional projection image to generate a denoised binarized blood vessel image, and perform skeletonization processing on the denoised binarized blood vessel image to generate a blood vessel skeleton map with a single pixel width. S5. Calculate the average blood vessel density based on the denoised binarized blood vessel image, and calculate the average blood vessel diameter and the proportion of blood vessel length of different diameters based on the denoised binarized blood vessel image and the blood vessel skeleton map. S6. Using the dermal-epidermal boundary coordinate set as the depth reference plane, divide the vascular layer region, and calculate the ratio of superficial vascular density to deep vascular density based on the vascular layer region. S7. Based on the denoised binary vascular image, identify vertical blood vessels and calculate the proportion of vertical blood vessels. S8. The epidermal thickness distribution map, the average blood vessel density, the average blood vessel diameter, the proportion of blood vessel length with different diameters, the ratio of superficial blood vessel density to deep blood vessel density, and the proportion of vertical blood vessels are used as the lesion characteristics of port-wine stains.

2. The method according to claim 1, characterized in that, The identification of the skin-air boundary coordinate set in S2 includes: Calculate the vertical gradient intensity and horizontal gradient intensity for each pixel column of the three-dimensional tomographic image; Based on the vertical gradient intensity and the horizontal gradient intensity, the connection weights between adjacent pixels are calculated according to the gradient fusion algorithm; the expression of the gradient fusion algorithm is: w(m,n)=λ1[(1-grady(m))+(1-grady(n))]+λ2[(1-gradx(m))+(1-gradx(n))]+σ; Where m and n are the pixel coordinates of adjacent path points, and m represents the pixel coordinate (x, y). m y m ), where n represents the pixel coordinates (x, y). n y n w(m, n) represents the connection weight between pixel m and pixel n, indicating the connection cost between adjacent pixels m and n; 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 Let represent the vertical gradient of pixel n; gradx(n) = x n+1 -x n-1 λ1 represents the horizontal gradient of pixel n; λ2 is the vertical gradient weight, λ1 is the horizontal gradient weight, and σ is the smoothing constant. Starting from the leftmost pixel of the three-dimensional tomographic image, the path with the minimum cumulative connection weight is iteratively searched in five neighborhood directions: right, upper right, lower right, upper right, and lower right, to obtain the skin-air boundary coordinate set.

3. The method according to claim 1, characterized in that, The identification of the dermal-epidermal boundary coordinate set in S2 includes: For the three-dimensional tomographic image, a horizontal A-line signal intensity integral projection is performed on the image region below the skin-air boundary coordinate set to obtain a first signal intensity curve distributed along the depth. If the first signal intensity curve has a single second peak, then the depth coordinates corresponding to the single second peak are taken as the dermal-epidermal interface position. If multiple stray peaks appear in the first signal intensity curve due to skin surface distortion, a bicubic spline interpolation algorithm is used to flatten the three-dimensional tomographic image. A-line signal intensity integral projection is performed on the flattened image in the horizontal direction to obtain a second signal intensity curve. The second peak is located in the second signal intensity curve. The single second peak of the first signal intensity curve or the second peak of the second signal intensity curve is mapped back to the coordinate system of the three-dimensional tomographic image according to its lateral position to obtain the dermal-epidermal boundary coordinate set.

4. The method according to claim 1, characterized in that, S4 includes: Calculate the average intensity of the N×N neighboring pixels of each pixel in the three-dimensional projected image; If the pixel value of the pixel is greater than the average intensity of the corresponding N×N neighboring pixels, then the pixel value of the pixel is set to 1; otherwise, it is set to 0, and an initial binary image is generated; where N is a preset positive integer. Perform N×N neighborhood median filtering on the initial binary image, and replace the center pixel value with the median value of the neighboring pixels to obtain a denoised binary blood vessel image; The Hilditch iterative algorithm is applied to the denoised binarized blood vessel image to remove boundary pixels until only pixels with a width of 1 remain, thus obtaining the blood vessel skeleton image.

5. The method according to claim 1, characterized in that, Calculating the average vascular density includes: The total number of pixels with a value of 1 in the denoised binarized vascular image is counted as A. The total number of pixels A is divided by the total number of pixels M in the denoised binarized vascular image to obtain the average vascular density.

6. The method according to claim 5, characterized in that, Calculating the average blood vessel diameter includes: The total number of pixels with a value of 1 in the vascular skeleton image is counted as L, and the ratio of the total number of pixels A to the total number of pixels L is calculated as R1. Multiply the ratio R1 by the scale factor to obtain the average blood vessel diameter; The scale factor represents the number of micrometers corresponding to one pixel.

7. The method according to claim 6, characterized in that, Calculating the proportion of blood vessel lengths of different diameters includes: Calculate the maximum Euclidean distance from each skeleton pixel in the vascular skeleton image to the nearest background pixel; wherein, the background pixel is the pixel with a pixel value of 0 in the vascular skeleton image; The local blood vessel diameter is calculated based on the maximum Euclidean distance, and the local blood vessel diameters are grouped according to preset diameter intervals to obtain multiple diameter ranges; the formula for calculating the local blood vessel diameter is: d=2×d max ×c L ; Where d is the diameter of the local blood vessel, d max For the maximum Euclidean distance, c L The scale factor is the aforementioned scale coefficient; In different pipe diameter ranges, the following formula is used to calculate the value per 1 mm. 2 The percentage of blood vessel lengths of different diameters, R2: Where A is the total number of pixels with a value of 1 in the denoised binarized vascular image, L is the total number of pixels with a value of 1 in the vascular skeleton image, and M is the total number of pixels in the denoised binarized vascular image.

8. The method according to claim 1, characterized in that, S6 includes: The ordinate of the dermal-epidermal boundary coordinate set is taken as the depth reference zero point z0; In the denoised binarized vascular image, the number of vascular pixels V within the depth layer from z0 to z0+Δz1 is counted. shallow and the number of blood vessel pixels V in the depth layer from z0+Δz1 to z0+Δz2 deep Where Δz1 and Δz2 are preset depth values; The density ratio R3 of superficial blood vessels to deep blood vessels is calculated using the following formula: Among them, M shallow M represents the total number of pixels in the superficial region of the denoised binarized vascular image. deep The total number of pixels in the deep region of the denoised binarized vascular image.

9. The method according to claim 1, characterized in that, S7 includes: In the denoised binary vascular image, non-continuous blood vessels are outlined with rectangular boxes and are considered as non-continuous blood vessels. If the target condition is met, the non-continuous blood vessels are determined to be roughly circular. The ratio of the sum of the circular pixels to the total number of pixels in the denoised binarized blood vessel image is taken as the vertical blood vessel proportion. The target conditions are: the difference between the length and width of the rectangle does not exceed δ1, the area ratio of the blood vessel in the rectangle is greater than δ2, and the proportion of the non-continuous blood vessel in the denoised binary blood vessel image is greater than δ3, where δ1, δ2, and δ3 are preset parameters.

10. A system for extracting port-wine stain lesion features based on OCTA technology, applied to the method according to any one of claims 1 to 9, characterized in that, The system includes: The image acquisition module is used to scan the target tissue using OCTA technology based on the OMAG algorithm, and to acquire a three-dimensional tomographic image containing tissue depth and structural information as well as a three-dimensional projection image containing blood flow information. The boundary recognition module is used to identify the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set based on the three-dimensional tomographic image. The thickness calculation module is used to calculate the epidermal layer thickness based on the epidermal-air boundary coordinate set and the dermal-epidermal boundary coordinate set, and to obtain an epidermal layer thickness distribution map. The image processing module is used to perform adaptive threshold segmentation and median filtering on the three-dimensional projection image to generate a denoised binarized blood vessel image, and to perform skeletonization processing on the denoised binarized blood vessel image to generate a blood vessel skeleton map with a single pixel width. The vascular feature calculation module is used to calculate the average vascular density based on the denoised binarized vascular image, and to calculate the average vascular diameter and the proportion of vascular lengths of different diameters based on the denoised binarized vascular image and the vascular skeleton map. The vascular layering module is used to divide the vascular layering region using the dermal-epidermal boundary coordinate set as the depth reference plane, and to calculate the ratio of superficial vascular density to deep vascular density based on the vascular layering region. The vertical blood vessel recognition module is used to identify vertical blood vessels and calculate the proportion of vertical blood vessels based on the denoised binarized blood vessel image. The feature integration module is used to use the epidermal thickness distribution map, the average blood vessel density, the average blood vessel diameter, the proportion of blood vessel length with different diameters, the ratio of superficial blood vessel density to deep blood vessel density, and the proportion of vertical blood vessels as features of port-wine stain lesions.

Images