A blood vessel proliferation quantification grading method and device, electronic equipment and storage medium

Abstract

The application provides a blood vessel proliferation quantification grading method and device, electronic equipment and a storage medium, applied to the technical field of image processing, and provides double-view image data for subsequent steps by acquiring a stereoscopic image of a lesion area; and a three-dimensional point cloud model is constructed by using a parallax map, a corresponding second blood vessel area is obtained by positioning a first blood vessel area recognized in a two-dimensional image to a three-dimensional space, a three-dimensional geometric feature of the blood vessel is extracted by calculating a blood vessel geometric parameter, and necessary parameters are provided for more accurate blood vessel quantification analysis. The three-dimensional density value calculated by using the three-dimensional geometric parameter realizes objective and quantitative grading of the degree of blood vessel proliferation, overcomes the defects of subjectivity and lack of physical distance guidance of the traditional method, and improves the accuracy and reliability of blood vessel proliferation evaluation.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method, device, electronic device and storage medium for quantitative grading of vascular proliferation. Background Art

[0002] With the rapid advancement and maturation of science and technology, minimally invasive surgery has become an important surgical method in hospitals. Laparoscopy is a key tool in gynecological examinations. Doctors can explore the abdominal cavity and observe the patient's intracavitary details in detail.

[0003] In existing gynecological surgical exploration, people only focus on anatomical changes or tumor characteristics, but not on blood vessels. There are no relevant indicators or corresponding standards. However, everyone recognizes that vascular changes have clinical significance.

[0004] During gynecological laparoscopy, the blood vessels observed primarily on the surfaces of pelvic organs, such as the peritoneum, bladder, uterus, and fallopian tubes, are often obscured. However, the distribution, density, morphology, color, and fullness of these vessels remain largely unnoticed, primarily due to a lack of objective assessment methods and means, let alone standards. Vascular hyperplasia is clearly associated with inflammation. This results in some nonspecific inflammation remaining undiagnosed. Even after surgery, these conditions can be overlooked and omitted due to the lack of objective indicators, compromising both understanding and treatment of the disease.

[0005] With the development of image recognition technology, the ability to calculate and compare various microscopic vascular features, combined with clinical data, to design and infer diagnostic models has the potential to become a new diagnostic method and tool. In particular, key features such as vascular density, as a tool for understanding disease, can help clinicians achieve clearer diagnoses and more targeted treatments.

[0006] However, traditional methods of capturing images with 2D endoscopes and calculating vascular density at the pixel level lack guidance from actual physical distances. This can easily lead to incorrect judgments about vascular density due to the use of handheld endoscopes at both near and far distances, which can adversely affect disease diagnosis. Furthermore, traditional technologies lack intraoperative vascular quantitative grading systems based on three-dimensional spatial information, particularly solutions that integrate vascular physical dimensions, dynamic blood flow parameters, and clinical grading standards.

[0007] Therefore, the existing technology lacks a solution that can overcome the lack of real physical distance guidance and use binocular ranging under 3D laparoscopy to more accurately obtain vascular density and other characteristics to judge the outcome of the lesion. Summary of the Invention

[0008] In view of the above-mentioned shortcomings of the existing technology, the present application provides a method, device, electronic device and storage medium for quantitative grading of vascular proliferation, which are applied to the field of image processing technology. This method can overcome the problem of lack of real physical distance guidance and introduce a binocular ranging solution based on 3D laparoscopy, which has the beneficial effect of more accurately obtaining vascular density and other vascular geometric parameters and judging the results of lesions.

[0009] In a first aspect, a method for quantitatively grading angiogenesis comprises:

[0010] S1: Acquire a stereoscopic image of the lesion area, wherein the stereoscopic image includes at least left-eye image information and right-eye image information;

[0011] S2: performing stereo correction on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and constructing a three-dimensional point cloud model based on the disparity map;

[0012] S3: Segmenting a first blood vessel region according to the corrected left-eye image information, and converting the first blood vessel region into a corresponding second blood vessel region in the three-dimensional point cloud model;

[0013] S4: Calculating blood vessel geometric parameters of the second blood vessel region, where the blood vessel geometric parameters include at least a blood vessel centerline and a blood vessel center point;

[0014] S5: Calculating the blood vessel width according to the blood vessel center point, calculating the three-dimensional density value of the blood vessel according to the blood vessel centerline and the blood vessel width, and grading the blood vessel according to the three-dimensional density value.

[0015] This application provides a method for quantitatively grading vascular proliferation. By acquiring stereoscopic images of the lesion area and providing dual-perspective image data for subsequent steps, a three-dimensional point cloud model is constructed using a disparity map, achieving conversion from a two-dimensional image to a three-dimensional space, providing three-dimensional spatial information for subsequent vascular analysis. The technical contribution of this step lies in obtaining three-dimensional structural information of the lesion area through stereo vision technology, overcoming the limitation of traditional two-dimensional images lacking depth information. By locating the first vascular region identified in the two-dimensional image in three-dimensional space to obtain the corresponding second vascular region, accurate conversion of the first vascular region from two-dimensional image space to three-dimensional point cloud space is achieved, laying the foundation for subsequent calculation of vascular geometric parameters. By calculating the vascular geometric parameters, the geometric features of the vessels in three-dimensional space are extracted, providing the necessary parameters for more accurate vascular quantitative analysis. The three-dimensional density values ​​calculated using the three-dimensional geometric parameters enable objective and quantitative grading of the degree of vascular proliferation, overcoming the subjectivity and lack of physical distance guidance of traditional methods, and improving the accuracy and reliability of vascular proliferation assessment.

[0016] Furthermore, step S2 includes:

[0017] S21: Acquire calibration parameters according to the left-eye image information and the right-eye image information;

[0018] S22: performing epipolar correction on the left-eye image information and the right-eye image information according to the calibration parameters to eliminate lens distortion;

[0019] S23: Calculating a disparity map between the left-eye image information and the right-eye image information after eliminating lens distortion;

[0020] S24: Constructing the three-dimensional point cloud model according to the disparity map.

[0021] This application provides a method for quantitatively grading angiogenesis. Calibration parameters are obtained, including camera intrinsic and extrinsic parameters, as well as distortion parameters. Using these calibration parameters, epipolar correction is performed on the left and right image information, effectively eliminating or reducing lens distortion and improving image quality. By calculating the disparity map between the left and right image information, the resulting disparity map is more accurate because the input images have already undergone distortion correction. Finally, a three-dimensional point cloud model is constructed using this high-precision disparity map, ensuring its accuracy and reliability.

[0022] Furthermore, step S24 includes:

[0023] S241: Obtaining a disparity value of each pixel in the left-eye image information and the right-eye image information according to the disparity map;

[0024] S242: Obtain the focal length, baseline distance, and noise compensation item of the camera;

[0025] S243: Calculating a depth value of each pixel according to the focal length, the baseline distance, the noise compensation item, and the disparity value;

[0026] S244: Convert the depth value into a depth map, and construct the three-dimensional point cloud model according to the depth map.

[0027] The present application provides a method for quantitative grading of vascular proliferation, which obtains the disparity value of each pixel through the disparity map; the disparity value reflects the position difference of the pixel in the left and right eye images, and can be used to calculate the depth value; by obtaining the focal length, baseline distance and noise compensation term of the camera; the focal length and baseline distance are important parameters of the stereo vision system, which determine the proportional relationship of the depth calculation; the noise compensation term takes into account the errors that may exist in practical applications and improves the accuracy of the depth calculation. The depth value of each pixel is obtained by calculation and converted into a depth map, so that a three-dimensional point cloud model can be constructed based on the depth map. This method provides a complete set of steps for constructing a three-dimensional point cloud model with strong operability, making it possible to construct an accurate three-dimensional point cloud model based on the disparity map, laying the foundation for the subsequent three-dimensional analysis of the vascular area.

[0028] Furthermore, step S3 includes:

[0029] S31: performing CLAHE contrast enhancement on the corrected left-eye image information;

[0030] S32: performing semantic segmentation on the enhanced left-eye image information to extract a binary mask of a blood vessel;

[0031] S33: performing a morphological opening operation on the binary mask of the blood vessel to obtain a pixel coordinate set of the first blood vessel region;

[0032] S34: Mapping the pixel coordinate set of the first blood vessel region to the three-dimensional point cloud model to obtain the second blood vessel region.

[0033] This application provides a method for quantitatively grading vascular proliferation. First, CLAHE contrast enhancement is performed on the corrected left-eye image information, which can effectively improve the local contrast of the image, making the blood vessels clearer in the image and overcoming the problem of insufficient contrast in the original image. Next, semantic segmentation technology is used to process the enhanced left-eye image information to accurately identify the blood vessel regions in the image and generate a binary mask of the blood vessels. Then, a morphological opening operation is performed on the binary mask of the blood vessels to eliminate small noise points and burrs in the image, smooth the blood vessel contours, further optimize the blood vessel segmentation results, and obtain a more accurate set of pixel coordinates of the first blood vessel region. Finally, the pixel coordinate set of the first blood vessel region obtained through the above processing is mapped to a three-dimensional point cloud model to obtain a second blood vessel region corresponding to the first blood vessel region, achieving the purpose of accurately converting the two-dimensional image segmentation results into three-dimensional space. Through the above series of steps, this technical solution can more accurately and reliably segment and extract blood vessel regions, laying the foundation for subsequent calculation of blood vessel geometric parameters and blood vessel grading, and effectively improving the accuracy and reliability of quantitative grading of vascular proliferation.

[0034] Furthermore, step S34 includes:

[0035] S341: performing positioning calculation on the pixel coordinate set of the first blood vessel region based on a normalized cross-correlation algorithm to obtain two-dimensional blood vessel pixel coordinates of the pixel coordinate set of the first blood vessel region;

[0036] S342: Performing a three-dimensional matrix transformation on the two-dimensional blood vessel pixel coordinates to obtain a three-dimensional blood vessel point cloud subset, wherein the position corresponding to the three-dimensional blood vessel point cloud subset in the three-dimensional point cloud model is a second blood vessel region.

[0037] Furthermore, step S4 includes:

[0038] S41: performing DBSCAN clustering calculation on the binary mask to generate a plurality of independent blood vessel point cloud clusters, where the independent blood vessel point cloud clusters are the blood vessel centerlines of each blood vessel in the second blood vessel region;

[0039] S42: Calculate the three-dimensional centroid of each of the independent blood vessel point cloud clusters to obtain the blood vessel center point of each blood vessel in the second blood vessel region.

[0040] Furthermore, step S5 includes:

[0041] S51: arbitrarily selecting a cubic region in the three-dimensional point cloud model, where the cubic region contains one or more blood vessels;

[0042] S52: performing cylindrical fitting on the center point of the blood vessel to obtain the blood vessel width, and calculating the three-dimensional density value of the blood vessel in the cubic area according to the blood vessel centerline and the blood vessel width;

[0043] S53: Classifying the blood vessels according to the three-dimensional density values.

[0044] In a second aspect, a device for quantitatively grading vascular proliferation is provided, which is used in any of the steps of the above-mentioned methods, and comprises:

[0045] Image acquisition module: used to acquire a stereoscopic image of the lesion area, the stereoscopic image including at least left-eye image information and right-eye image information;

[0046] Point cloud construction module: used to perform stereo correction on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and construct a three-dimensional point cloud model based on the disparity map;

[0047] A blood vessel segmentation module is configured to segment a first blood vessel region according to the corrected left-eye image information, and convert the first blood vessel region into a corresponding second blood vessel region in the three-dimensional point cloud model;

[0048] A parameter calculation module: configured to calculate the vascular geometric parameters of the second vascular region, wherein the vascular geometric parameters include at least a vascular centerline and a vascular center point;

[0049] Blood vessel grading module: used to calculate the blood vessel width according to the blood vessel center point, calculate the three-dimensional density value of the blood vessel according to the blood vessel centerline and the blood vessel width, and grade the blood vessel according to the three-dimensional density value.

[0050] In a third aspect, the present application provides an electronic device comprising a processor and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in any of the above methods are executed.

[0051] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which runs the steps of any of the above methods when executed by a processor.

[0052] Beneficial Effects: The method, device, electronic device, and storage medium proposed in this application acquire stereoscopic images of the lesion area, providing dual-perspective image data for subsequent steps. A 3D point cloud model is constructed from the disparity map, achieving the conversion from 2D images to 3D space and providing 3D spatial information for subsequent vascular analysis. The technical contribution of this step lies in obtaining 3D structural information of the lesion area through stereo vision technology, overcoming the limitation of traditional 2D images lacking depth information. By locating the first vascular region identified in the 2D image in 3D space and obtaining the corresponding second vascular region, the first vascular region is accurately converted from 2D image space to 3D point cloud space, laying the foundation for subsequent calculation of vascular geometric parameters. By calculating the vascular geometric parameters, the geometric features of the vessels in 3D space are extracted, providing the necessary parameters for more accurate vascular quantitative analysis. The 3D density values ​​calculated using the 3D geometric parameters enable objective and quantitative grading of the degree of vascular proliferation, overcoming the subjectivity and lack of physical distance guidance of traditional methods, and improving the accuracy and reliability of vascular proliferation assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method for quantitative grading of angiogenesis proposed in this application.

[0054] Figure 2 This is a structural diagram of a device for quantitatively grading angiogenesis proposed in this application.

[0055] Figure 3 This is a structural diagram of an electronic device proposed in this application.

[0056] Explanation of reference numerals: 201, image acquisition module; 202, point cloud construction module; 203, blood vessel segmentation module; 204, parameter calculation module; 205, blood vessel classification module; 301, processor; 302, memory; 303, communication bus; 3, electronic equipment. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and marked in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work fall within the scope of protection of the present application.

[0058] It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0059] The following disclosure provides a number of different implementations or examples for achieving the objectives of the present invention, addressing the problems of conventional techniques lacking guidance based on real physical distances and a lack of an intraoperative vascular quantitative grading system based on three-dimensional spatial information, particularly a solution that integrates vascular physical dimensions, dynamic blood flow parameters, and clinical grading criteria. Therefore, this application proposes a method, apparatus, electronic device, and storage medium for quantitative grading of vascular proliferation, as follows:

[0060] Please refer to Figure 1 In a first aspect, a method for quantitatively grading angiogenesis comprises:

[0061] S1: Acquire a stereoscopic image of the lesion area, where the stereoscopic image includes at least left-eye image information and right-eye image information;

[0062] S2: Perform stereo correction on the left and right image information to obtain a disparity map between the two images, and construct a 3D point cloud model based on the disparity map;

[0063] S3: Segmenting the first blood vessel region according to the corrected left eye image information, and converting the blood vessel region into a corresponding second blood vessel region in the three-dimensional point cloud model;

[0064] S4: Calculating blood vessel geometric parameters of the second blood vessel region, where the blood vessel geometric parameters include at least: a blood vessel centerline and a blood vessel center point;

[0065] S5: Calculate the vessel width based on the vessel center point, calculate the three-dimensional density value of the vessel based on the vessel centerline and the vessel width, and grade the vessel based on the three-dimensional density value.

[0066] In step S1, the acquisition of stereo images can be accomplished using a binocular vision system equipped with two cameras, one for capturing left-eye image information and the other for capturing right-eye image information of the lesion area. The two cameras are configured to simulate the perspective of the human eye, capturing images of the same lesion area from different angles, thereby obtaining a stereo image pair containing parallax information. The binocular vision system can be a 3D laparoscope with a resolution of 1920×1080@30fps.

[0067] In step S2, stereo correction is an image preprocessing technique used to eliminate or reduce geometric image distortion caused by camera lens distortion and installation errors. Epipolar correction is one possible stereo correction method. This method aligns the scan lines of the left and right images, simplifying subsequent disparity calculations. Disparity maps can be generated using stereo matching algorithms such as block matching or optical flow. These algorithms aim to find the positional deviation between corresponding pixels in the left and right images, i.e., the disparity value.

[0068] The construction of the three-dimensional point cloud model is based on the disparity map and camera calibration parameters. The three-dimensional coordinates of each pixel can be calculated through the disparity value and calibration parameters. The three-dimensional coordinate set of all pixels constitutes the three-dimensional point cloud model, which is a discrete representation of the three-dimensional structure of the lesion area.

[0069] In step S3, the first vascular region can be segmented using a variety of image segmentation algorithms, such as threshold segmentation, region growing, edge detection, or semantic segmentation. Semantic segmentation is a preferred image segmentation method that can identify vascular regions at the pixel level. The conversion of the vascular region from a 2D image to a 3D point cloud model can be achieved through a coordinate mapping relationship. Specifically, based on the camera imaging model and calibration parameters, the 2D image pixel coordinates are converted to 3D point cloud coordinates, thereby obtaining the second vascular region in 3D space.

[0070] In step S4, the calculation of the blood vessel geometric parameters can be achieved with the help of point cloud processing and analysis technology, the extraction of the blood vessel centerline can be achieved through a skeleton extraction algorithm or a topological analysis method, and the blood vessel center point can be the average coordinates or centroid coordinates of the points on the blood vessel centerline.

[0071] In step S5, the vessel width can be calculated by cylindrical fitting of the vessel center point. The calculated 3D vessel density value can be defined as the length or volume of the vessel per unit volume. The cubic region can be a regular region manually set in the 3D point cloud model or an irregular region adaptively determined based on the anatomical structure or pathological characteristics of the lesion. The development of vascular grading standards can refer to existing vascular proliferation grading standards or clinical guidelines, or new grading standards can be established through statistical analysis and experimental verification. The grading results can be used to assess the degree of vascular proliferation and the severity of the lesion.

[0072] Specifically, the method for quantitatively grading vascular proliferation works as follows: First, a stereoscopic image of the lesion area is acquired using a binocular vision system, providing the data foundation for subsequent 3D reconstruction and analysis. Then, the stereoscopic image is rectified and disparity calculated to eliminate image distortion, resulting in a disparity map. Based on this disparity map, a 3D point cloud model of the lesion area is constructed, achieving the conversion from 2D image to 3D space. Next, the vascular region is segmented from the rectified left-eye image, and the segmentation results are mapped onto the 3D point cloud model to obtain vascular point cloud data in 3D space. Based on this, the geometric parameters of the vessels, including centerline, center point, and width, are calculated. These parameters quantitatively describe the vascular morphology and size. Finally, the 3D density of the vessels is calculated based on these geometric parameters, and the degree of vascular proliferation is graded based on this density value, achieving a quantitative assessment of vascular proliferation. Through these steps, this method can accurately quantify the degree of vascular proliferation in 3D space, overcoming the limitation of traditional 2D methods that lack depth information. This method improves the objectivity and accuracy of vascular proliferation assessment, providing a more reliable basis for clinical diagnosis and treatment.

[0073] In some specific embodiments, a 3D laparoscope dual camera is used to synchronously capture stereoscopic images of the lesion area, the optical axes of the two cameras are parallel and the baseline distance is 10 mm, and the image resolution is 1920x1080@30fps.

[0074] The Bouguet calibration method is used to calibrate the camera to obtain the intrinsic parameter matrix, distortion parameters and extrinsic parameter matrix. The distortion parameters include radial distortion coefficient and tangential distortion, and the extrinsic parameter matrix includes rotation matrix and translation matrix. The left and right eye images are corrected using the epipolar correction algorithm. The horizontal parallax range of the corrected image is between 0 and 200 pixels.

[0075] The Semi-Global Block Matching (SGBM) algorithm is used to calculate the disparity map. The disparity search range is 1-256 pixels, and the penalty parameters P1=8 and P2=32.

[0076] Based on the disparity map and calibration parameters, the three-dimensional coordinates of each pixel are calculated using a triangulation method to construct a three-dimensional point cloud model of the lesion area with a point cloud density of 1000 points per cubic millimeter.

[0077] To segment the first vessel region, the left image is first contrast-enhanced using CLAHE. A pre-trained deep learning model (e.g., U-Net) is then used to perform semantic segmentation of the vessels, generating a binary mask. This mask is then morphologically opened using a 3x3 elliptical kernel to remove small noise and artifacts, yielding the pixel coordinates of the first vessel region. The normalized cross-correlation algorithm is then used to search for the corresponding pixel coordinates of the first vessel region in the right image, converting the 2D image to a 3D point cloud model to obtain the second vessel region.

[0078] The point cloud data for the second vessel region was then clustered using DBSCAN with a cluster radius of 0.5 mm and a minimum point count of 5. This generated multiple independent vessel point cloud clusters, each representing the centerline of a single vessel. The 3D centroid of each cluster was calculated as the vessel center. A cylindrical fit was performed on each cluster to determine the vessel radius, with the vessel width being twice the radius.

[0079] Finally, a 10x10x10mm cube region can be manually selected in the 3D point cloud model to calculate the total volume of all blood vessels within the region. The 3D vascular density is defined as the ratio of the total vascular volume to the volume of the cube region. Based on the 3D density values, the degree of vascular proliferation is divided into four grades: Grade I (density value ρ∈[0, 0.2]), Grade II (density value ρ∈[0.2, 0.4]), Grade III (density value ρ∈[0.4, 0.6]), and Grade IV (density value ρ≥0.6). This grading standard can be adjusted and optimized based on clinical pathological results. This allows for a quantitative grading assessment of the degree of vascular proliferation in pelvic lesions in gynecological patients.

[0080] Furthermore, step S2 includes:

[0081] S21: Obtaining calibration parameters based on the left eye image information and the right eye image information;

[0082] S22: performing epipolar correction on the left eye image information and the right eye image information according to the calibration parameters to eliminate lens distortion;

[0083] S23: Calculating a disparity map between the left-eye image information and the right-eye image information after eliminating lens distortion;

[0084] S24: Construct a 3D point cloud model based on the disparity map.

[0085] In step S21, the calibration parameters can be obtained through a standard camera calibration process. For example, using the checkerboard calibration method, multiple images of a calibration plate are taken at different angles. Through corner detection and parameter optimization, the camera intrinsic parameter matrix, distortion coefficients, and extrinsic parameter matrices of the left and right cameras are calculated. The distortion parameters include radial distortion coefficients and tangential distortion, and the extrinsic parameter matrix includes a rotation matrix and a translation matrix.

[0086] In step S22, epipolar correction is performed by calibrating parameters to project the left and right image information onto the same plane, aligning the pixel rows of the two images. This constrains the subsequent disparity search to the horizontal direction, improving search efficiency and accuracy and eliminating the effects of lens distortion.

[0087] In step S23, the disparity map can be calculated using the Semi-Global Block Matching (SGBM) algorithm. For the corrected left-eye image information and the right-eye image information, matching pixels are searched in the horizontal direction, and the disparity value of the pixel is calculated. The disparity value is inversely proportional to the distance between the pixels.

[0088] In step S24, the three-dimensional point cloud model is constructed by using the disparity map and camera parameters and the principle of triangulation to calculate the coordinates of each pixel in the three-dimensional space. These three-dimensional coordinate sets constitute the three-dimensional point cloud model.

[0089] Furthermore, step S24 includes:

[0090] S241: Obtaining a disparity value of each pixel in the left-eye image information and the right-eye image information according to the disparity map;

[0091] S242: Obtain the focal length, baseline distance, and noise compensation item of the camera;

[0092] S243: Calculate the depth value of each pixel according to the focal length, baseline distance, noise compensation item, and disparity value;

[0093] S244: Convert the depth value into a depth map, and construct a three-dimensional point cloud model according to the depth map.

[0094] In step S241, each pixel value in the disparity map represents the horizontal displacement between the corresponding pixel points of the left and right images, and this displacement is the disparity value. The disparity value can be obtained by directly reading the pixel values ​​of the disparity map.

[0095] In step S242 , the focal length and baseline distance are intrinsic parameters of the stereo vision system and can be pre-determined through camera calibration. The noise compensation term is introduced to improve the accuracy of depth calculation and may be a fixed value obtained through experimentation or empirical estimation to correct for system errors.

[0096] In step S243, the depth value is calculated based on the triangulation principle. The specific formula is:

[0097] Depth value = (focal length * baseline distance) / parallax value + noise compensation item.

[0098] In step S244, the depth map is an image in which the grayscale value of each pixel represents the depth value corresponding to the pixel. The depth map can be obtained by arranging the depth value of each pixel calculated in step S243 into a matrix form. The three-dimensional point cloud model is a collection of points in space, and each point is represented by a three-dimensional coordinate. The specific method of constructing a three-dimensional point cloud model can be that for each pixel in the depth map, according to the coordinates and depth value of the pixel, combined with the internal parameters of the camera, the coordinates of the pixel in the three-dimensional space are calculated, and the set of three-dimensional coordinates of all pixels constitutes the three-dimensional point cloud model. Alternatively, a three-dimensional point cloud model can be constructed by a Poisson reconstruction algorithm based on the depth map and RGB color mapping.

[0099] Furthermore, step S3 includes:

[0100] S31: performing CLAHE contrast enhancement on the corrected left eye image information;

[0101] S32: performing semantic segmentation on the enhanced left-eye image information and extracting the binary mask of the blood vessels;

[0102] S33: performing a morphological opening operation on the binary mask of the blood vessel to obtain a pixel coordinate set of the first blood vessel region;

[0103] S34: Mapping the pixel coordinate set of the first blood vessel region to the three-dimensional point cloud model to obtain a second blood vessel region.

[0104] Among them, CLAHE contrast enhancement improves the local contrast of the image by analyzing the histogram of the local area of ​​the image and redistributing the pixel intensity.

[0105] Semantic segmentation uses a deep learning model, such as U-Net, to perform pixel-level classification on the enhanced image and generate accurate binary blood vessel masks. The specific U-Net encoder depth is 5.

[0106] The morphological opening operation can eliminate small noise points and burrs in the image, smooth the blood vessel contour, further optimize the blood vessel segmentation results, and obtain the pixel coordinate set of the first blood vessel region more accurately.

[0107] The morphological opening operation includes erosion and dilation operations. By performing erosion followed by dilation, noise points are eliminated and the edges of blood vessels are smoothed.

[0108] The mapping of the pixel coordinate set is achieved through camera calibration parameters and a disparity map. The pixel coordinates of the first blood vessel region are converted into three-dimensional point cloud coordinates, thereby obtaining the second blood vessel region.

[0109] In some specific embodiments, the CLAHE contrast enhancement algorithm is used to enhance the corrected left-eye image information. For example, the parameters of the CLAHE algorithm, such as tile size and contrast limiter, can be adjusted according to the actual image quality and vascular characteristics to achieve the optimal enhancement effect. As a preferred embodiment, the tile size is set to 8x8 pixels and the contrast limiter is set to 2. The semantic segmentation model adopts a U-Net network structure and is trained using a medical image dataset containing vascular images. The trained model can accurately identify and segment vascular regions. As a preferred embodiment, the encoder part of the U-Net network uses ResNet34 as the backbone network, and the decoder part uses deconvolution and upsampling operations to gradually restore the image resolution. The loss function uses the cross-entropy loss function. The morphological opening operation uses a 3x3 elliptical structuring element, and the number of iterations is set to 1 to eliminate noise in the binary mask of the vascular vessel and smooth the vascular contour. As a preferred embodiment, the erosion operation and the dilation operation both use a 3x33x3 elliptical structuring element. The erosion operation is first performed to remove noise, and then the dilation operation is performed to restore the original size of the vascular region.

[0110] The coordinate mapping process is implemented through the following steps: First, the camera calibration parameters, including the intrinsic parameter matrix, extrinsic parameter matrix, and distortion parameters, are obtained. Then, the 2D coordinates of each pixel are calculated based on the pixel coordinates and disparity map of the left eye image. Finally, each pixel coordinate in the pixel coordinate set of the first vessel region is converted into 3D spatial coordinates to obtain a 3D point cloud subset of the second vessel region.

[0111] Specifically, the pixel coordinate set is: ,in, is the pixel coordinate set, is the two-dimensional coordinate of each pixel, is the binary mask of blood vessels.

[0112] Furthermore, step S34 includes:

[0113] S341: performing positioning calculation on the pixel coordinate set of the first blood vessel region based on a normalized cross-correlation algorithm to obtain two-dimensional blood vessel pixel coordinates of the pixel coordinate set of the first blood vessel region;

[0114] S342: Performing a three-dimensional matrix transformation on the two-dimensional blood vessel pixel coordinates to obtain a three-dimensional blood vessel point cloud subset, wherein the position corresponding to the three-dimensional blood vessel point cloud subset in the three-dimensional point cloud model is the second blood vessel region.

[0115] In step S341, a normalized cross-correlation algorithm is used to locate the pixel coordinates of the first blood vessel region. Specifically, the algorithm uses the pixel coordinates of the first blood vessel region in the left image as a template and searches for an area in the right image that best matches the template. By calculating the normalized cross-correlation coefficient between the template and the search area, the optimal matching location is determined. This optimal matching location is the two-dimensional blood vessel pixel coordinates of the pixel coordinates of the first blood vessel region. This allows the precise location of the first blood vessel region in the left and right images.

[0116] In step S342, a three-dimensional matrix transformation converts the two-dimensional vessel pixel coordinates into three-dimensional space. Specifically, camera calibration parameters, such as the camera intrinsic parameter matrix, extrinsic parameter matrix, and distortion parameters, are used to construct a three-dimensional matrix transformation relationship. Using this three-dimensional matrix transformation relationship, the two-dimensional vessel pixel coordinates can be projected onto the three-dimensional point cloud model to obtain corresponding three-dimensional coordinates. These three-dimensional coordinates constitute a three-dimensional vessel point cloud subset, which is the second vessel region. Through precise positioning using the normalized cross-correlation algorithm and accurate conversion using the three-dimensional matrix transformation, the vessel region in the two-dimensional image can be accurately mapped onto the three-dimensional point cloud model.

[0117] Specifically, to address the existing problem of vascular density calculation lacking guidance from actual physical distances, the technical solution of the present application utilizes a normalized cross-correlation algorithm in step S34 to precisely locate the position of the 2D vascular region within the 3D point cloud model and performs a 3D matrix transformation, thereby obtaining a precise 3D vascular point cloud subset, namely the second vascular region. In this way, the vascular region segmented from the 2D image can be accurately converted to 3D space, overcoming the problem of insufficient positioning accuracy that may result from simple direct mapping. The resulting second vascular region more accurately reflects the true distribution of blood vessels in 3D space, providing a more reliable data foundation for the subsequent calculation of vascular geometric parameters and vascular grading. Compared to methods that simply map 2D pixel coordinates to 3D point cloud models, the technical solution of the present application improves the accuracy and reliability of the conversion from 2D image vascular regions to 3D point cloud model vascular regions, thereby providing more accurate basic data for the subsequent quantitative grading of vascular proliferation.

[0118] Specifically, the calculation method of the 3D blood vessel point cloud subset is:

[0119]

[0120] in, is a subset of the 3D blood vessel point cloud, is the three-dimensional space coordinate of each pixel, is the two-dimensional coordinate of each pixel, is the coordinate transformation matrix.

[0121] Furthermore, step S4 includes:

[0122] S41: performing DBSCAN clustering calculation on the binary mask to generate multiple independent blood vessel point cloud clusters, where the independent blood vessel point cloud cluster is the blood vessel centerline of each blood vessel in the second blood vessel region;

[0123] S42: Calculate the three-dimensional centroid of each independent blood vessel point cloud cluster to obtain the blood vessel center point of each blood vessel in the second blood vessel region.

[0124] Among them, in step S41, the DBSCAN clustering algorithm, as a density-based clustering method, can effectively identify the vascular region in the binary mask. Specifically, the DBSCAN algorithm divides the pixels in the binary mask into different clusters by setting two parameters: the neighborhood radius and the minimum number of neighborhood points. In the application scenario of quantitative grading of vascular proliferation, each cluster in the binary mask can be considered as an independent blood vessel. Therefore, the multiple independent vascular point cloud clusters generated by the DBSCAN clustering calculation can be defined as the vascular centerline of each blood vessel in the second vascular region.

[0125] Specifically, DBSCAN clustering is performed on the binary mask with eps = 0.1 mm to generate N independent blood vessel point cloud clusters: , ... , where for each independent blood vessel point cloud cluster , k=1, 2, 3...N.

[0126] In step S42, for each independent blood vessel point cloud cluster generated in step S41 The 3D centroid is obtained by calculating the average 3D coordinates of all pixels within the cluster. This 3D centroid represents the center position of the corresponding vessel point cloud cluster in 3D space. Therefore, the 3D centroid of each independent vessel point cloud cluster can be defined as the vessel center point of each vessel in the second vessel region.

[0127] The specific calculation formula for the 3D centroid of each independent blood vessel point cloud cluster is:

[0128] ,in is each pixel in the cluster, The three-dimensional coordinates of , , , It is the center point of the blood vessel.

[0129] In some specific implementations, in step S41 , specific parameters of the DBSCAN clustering algorithm may be set as follows: the neighborhood radius epsilon is 3 pixels, and the minimum number of neighborhood points min_samples is 5.

[0130] Furthermore, step S5 includes:

[0131] S51: arbitrarily selecting a cubic region in the three-dimensional point cloud model, where the cubic region contains one or more blood vessels;

[0132] S52: performing cylindrical fitting on the center point of the blood vessel to obtain the blood vessel width, and calculating the three-dimensional density value of the blood vessel in the cube area based on the blood vessel centerline and the blood vessel width;

[0133] S53: Classify blood vessels according to three-dimensional density values.

[0134] The cubic region used as the spatial range for calculating vascular density values ​​can be selected manually or automatically. Manual selection involves defining a cubic region within the 3D point cloud model through an interactive interface. This cubic region must encompass one or more vessels to be analyzed. Automatic selection involves pre-setting the size of the cubic region and then sliding it within the 3D point cloud model at a specific step size, or automatically generating a cubic region centered around the center point of a vessel.

[0135] Cylinder fitting is performed on each independent vessel point cloud cluster. Blood vessels typically appear as approximately cylindrical structures in three-dimensional space. Cylinder fitting is a method for estimating cylinder parameters using point cloud data. Specifically, by performing cylinder fitting on each independent vessel point cloud cluster, the cylinder parameters that best fit the point cloud cluster, such as axis direction and radius, can be obtained. The radius parameter obtained by cylinder fitting can be used to characterize the vessel width (i.e., vessel diameter). Therefore, the cylinder radius obtained by cylinder fitting for each independent vessel point cloud cluster can be defined as the vessel width of each vessel in the second vessel region. In some embodiments, the cylinder fitting algorithm can utilize the RANSAC algorithm, with the number of iterations set to 100 and the inlier threshold set to 1 pixel. For example, for an independent vessel point cloud cluster, DBSCAN clustering and cylinder fitting can be used to calculate the vessel centerline as a series of 3D point coordinates, the vessel center point as a single 3D coordinate value, and the vessel width as 0.5 mm. By setting these specific parameters and selecting the algorithm, accurate calculation of vessel width can be achieved.

[0136] Specifically, in step S53, the three-dimensional density value may be calculated based on the Monte Carlo integration method:

[0137] ,in, is the number of blood vessels in the cubic area, is the diameter of a single blood vessel, is the centerline length of a single blood vessel, is the volume of the cube region, is the serial number of the blood vessel, k=1, 2, 3...N.

[0138] For grading blood vessels according to three-dimensional density values, the mapping relationship between density values ​​and blood vessel grades can be pre-set. For example, the density value ρ∈[0, 0.2) can be set as grade I, the density value ρ∈[0.2, 0.4) can be set as grade II, the density value ρ∈[0.4, 0.6) can be set as grade III, and the density value ρ≥0.6 can be set as grade IV. Then, the corresponding blood vessel grade can be found based on the calculated three-dimensional density value.

[0139] In summary, the present application fundamentally overcomes the limitations of traditional 2D solutions through a 3D laparoscopic binocular stereo vision system: it directly obtains the true physical size of blood vessels (millimeter-level accuracy) based on physical ranging (binocular baseline distance + triangulation), replacing the traditional 2D laparoscopic estimation method that relies on pixel distance, eliminating the magnification error caused by changes in lens distance; and calculates spatial vascular density through three-dimensional point cloud reconstruction and Monte Carlo integration method, solving the problem of density underestimation caused by vascular overlap and viewing angle tilt in the 2D projection method, which has the advantage of improving the accuracy and reliability of vascular proliferation assessment.

[0140] Please refer to Figure 2 A device for quantitatively grading angiogenesis, used in any of the steps of the above methods, comprising:

[0141] Image acquisition module 201: used to acquire a stereoscopic image of the lesion area, the stereoscopic image including at least left-eye image information and right-eye image information;

[0142] Point cloud construction module 202: used to perform stereo correction on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and construct a 3D point cloud model based on the disparity map;

[0143] Blood vessel segmentation module 203: configured to segment a first blood vessel region according to the corrected left eye image information, and convert the first blood vessel region into a corresponding second blood vessel region in a three-dimensional point cloud model;

[0144] Parameter calculation module 204: used to calculate blood vessel geometric parameters of the second blood vessel region, the blood vessel geometric parameters at least including: blood vessel centerline and blood vessel center point;

[0145] The blood vessel grading module 205 is used to calculate the blood vessel width according to the blood vessel center point, calculate the three-dimensional density value of the blood vessel according to the blood vessel centerline and the blood vessel width, and grade the blood vessels according to the three-dimensional density value.

[0146] Among them, the image acquisition module 201 can specifically be a binocular camera system, which is placed in parallel to simulate human eye vision, and synchronously collects left-eye image information and right-eye image information of the lesion area from different angles to provide data support for subsequent stereoscopic vision processing.

[0147] Point cloud construction module 202, in which stereo correction specifically uses the Bouguet algorithm, which corrects the distortion of the original image through calibration parameters and aligns the two images so that the pixels of the two images are in the same row, simplifying subsequent disparity calculations. The disparity map is obtained by calculating the disparity values ​​of the pixels through a semi-global matching algorithm. The 3D point cloud model is constructed by calculating the 3D spatial coordinates of each pixel in the disparity map using the disparity value and pre-calibrated camera parameters, thereby constructing a 3D point cloud model of the lesion area.

[0148] The vascular segmentation module 203 performs CLAHE contrast enhancement on the corrected left-eye image information. CLAHE contrast enhancement can improve the local contrast of the image, making the vascular structure clearer. Semantic segmentation specifically utilizes the U-Net network, a deep learning network that can effectively segment vascular regions in the image and obtain a binary mask of the vessels. A morphological opening operation is used to remove noise and small connected regions from the binary mask of the vessels, obtaining a more accurate set of pixel coordinates for the first vascular region. The vascular region is converted to a 3D point cloud model, specifically using a normalized cross-correlation algorithm to accurately locate the pixel coordinates of the first vascular region, obtaining 2D vascular pixel coordinates. These 2D vascular pixel coordinates are then converted to a 3D point cloud space through a 3D matrix transformation, obtaining a 3D vascular point cloud subset, which serves as the second vascular region.

[0149] Parameter calculation module 204, DBSCAN clustering calculation, is used to cluster the 3D vessel point cloud subset of the second vessel region, grouping point clouds belonging to the same vessel into a cluster to form multiple independent vessel point cloud clusters. Each independent vessel point cloud cluster represents the centerline of a single vessel. The vessel center point is obtained by calculating the 3D centroid of each independent vessel point cloud cluster.

[0150] The vascular grading module 205 selects the cube region, allowing for manual or automatic selection of the region containing the blood vessels. A least-squares method is used to perform cylindrical fitting on each independent vascular point cloud cluster, obtaining the parameters of the cylinder that best approximates the vessel shape. The cylinder diameter is the vessel width. This is calculated by calculating the ratio of the sum of the volumes of all vessels within the cube region to the volume of the cube region. The vessel volume is calculated as the ratio of the vessel centerline length to the square of the vessel width. For vascular grading, a predefined mapping relationship between 3D density values ​​and vascular proliferation grades is used. The vascular proliferation grade is determined by searching a mapping table based on the calculated 3D density values.

[0151] Specifically, the device for quantitatively grading vascular proliferation proposed in this application, during operation, first obtains a stereoscopic image of the lesion area by the image acquisition module 201. The stereoscopic image contains left and right image information, providing raw data for subsequent processing. Then, the point cloud construction module 202 receives the stereoscopic image and performs stereoscopic correction on the left and right image information to eliminate image distortion and align the images. It then calculates a disparity map of the corrected image, which reflects the depth information of the image pixels. Based on the disparity map, a three-dimensional point cloud model of the lesion area is constructed, achieving the conversion from a two-dimensional image to a three-dimensional space. Next, the vessel segmentation module 203 receives the corrected left image information and segments a first vessel region from the left image information. The first vessel region is a collection of pixels representing the vessel in the two-dimensional image. The first vessel region is converted into a corresponding second vessel region in the three-dimensional point cloud model. The second vessel region is a collection of point clouds representing the vessel in three-dimensional space, achieving three-dimensional positioning of the vessel region. The parameter calculation module 204 calculates the geometric parameters of the vessels in the second vessel region, obtaining key vessel parameters such as the vessel centerline and center point, providing a basis for subsequent density calculation and grading. Finally, the vessel grading module 205 calculates the vessel width based on the vessel center point, and then calculates the three-dimensional density value of the vessel based on the vessel centerline and the vessel width. The three-dimensional density value comprehensively reflects the number and thickness of the vessels, and the vessels are graded based on the three-dimensional density value to achieve a quantitative assessment of vascular proliferation.

[0152] Through the collaborative work of the above modules, the device proposed in this application can automatically, efficiently, and accurately perform quantitative grading of vascular proliferation. Compared with traditional methods, the device proposed in this application uses a three-dimensional point cloud model for vascular analysis, overcoming the lack of depth information in traditional two-dimensional image analysis methods. It can more accurately obtain characteristic parameters such as vascular density, thereby more objectively and accurately assessing the degree of vascular proliferation.

[0153] Please refer to Figure 3 , Figure 3This is a structural diagram of an electronic device provided in an embodiment of the present application. The present application provides an electronic device 3, including: a processor 301 and a memory 302. The processor 301 and the memory 302 are interconnected and communicate with each other via a communication bus 303 and / or other forms of connection mechanisms (not shown). The memory 302 stores computer-readable instructions executable by the processor 301. When the electronic device is running, the processor 301 executes the computer-readable instructions to execute the method in any optional implementation of the above embodiment to achieve the following functions: obtaining a stereoscopic image of the lesion area, the stereoscopic image at least The method includes left-eye image information and right-eye image information; stereo correction is performed on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and a three-dimensional point cloud model is constructed based on the disparity map; a first blood vessel region is segmented based on the corrected left-eye image information, and the first blood vessel region is converted into a corresponding second blood vessel region in the three-dimensional point cloud model; blood vessel geometric parameters of the second blood vessel region are calculated, and the blood vessel geometric parameters include at least: a blood vessel centerline and a blood vessel center point; the blood vessel width is calculated based on the blood vessel center point, a three-dimensional density value of the blood vessel is calculated based on the blood vessel centerline and the blood vessel width, and the blood vessels are graded based on the three-dimensional density value.

[0154] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method of any optional implementation of the above embodiment is executed to achieve the following functions: obtaining a stereoscopic image of a lesion area, the stereoscopic image including at least left-eye image information and right-eye image information; performing stereoscopic correction on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and constructing a three-dimensional point cloud model based on the disparity map; segmenting a first blood vessel region based on the corrected left-eye image information, and converting the first blood vessel region into a corresponding second blood vessel region in the three-dimensional point cloud model; calculating blood vessel geometric parameters of the second blood vessel region, the blood vessel geometric parameters including at least: a blood vessel centerline and a blood vessel center point; calculating a blood vessel width based on the blood vessel center point, calculating a three-dimensional density value of the blood vessel based on the blood vessel centerline and the blood vessel width, and grading the blood vessels based on the three-dimensional density value.

[0155] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0156] In addition, the units described as separate components 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 units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0157] Furthermore, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0158] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0159] The foregoing is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Persons skilled in the art will readily appreciate that the present application may be modified and altered in various ways. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for quantitatively grading angiogenesis, characterized in that: The method comprises: S1: Acquire a stereoscopic image of the lesion area, wherein the stereoscopic image includes at least left-eye image information and right-eye image information; S2: performing stereo correction on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and constructing a three-dimensional point cloud model based on the disparity map; S3: Segmenting a first blood vessel region according to the corrected left-eye image information, and converting the first blood vessel region into a corresponding second blood vessel region in the three-dimensional point cloud model; S4: Calculating blood vessel geometric parameters of the second blood vessel region, where the blood vessel geometric parameters include at least a blood vessel centerline and a blood vessel center point; S5: calculating the blood vessel width according to the blood vessel center point, calculating a three-dimensional density value of the blood vessel according to the blood vessel centerline and the blood vessel width, and grading the blood vessel according to the three-dimensional density value; Step S3 includes: S31: performing CLAHE contrast enhancement on the corrected left-eye image information; S32: performing semantic segmentation on the enhanced left-eye image information to extract a binary mask of a blood vessel; S33: performing a morphological opening operation on the binary mask of the blood vessel to obtain a pixel coordinate set of the first blood vessel region; S34: mapping the pixel coordinate set of the first blood vessel region to the three-dimensional point cloud model to obtain the second blood vessel region; Step S34 includes: S341: performing positioning calculation on the pixel coordinate set of the first blood vessel region based on a normalized cross-correlation algorithm to obtain two-dimensional blood vessel pixel coordinates of the pixel coordinate set of the first blood vessel region; S342: performing a three-dimensional matrix transformation on the two-dimensional blood vessel pixel coordinates to obtain a three-dimensional blood vessel point cloud subset, wherein the position corresponding to the three-dimensional blood vessel point cloud subset in the three-dimensional point cloud model is a second blood vessel region; Step S4 includes: S41: performing DBSCAN clustering calculation on the binary mask to generate a plurality of independent blood vessel point cloud clusters, where the independent blood vessel point cloud clusters are the blood vessel centerlines of each blood vessel in the second blood vessel region; S42: Calculate the three-dimensional centroid of each of the independent blood vessel point cloud clusters to obtain the blood vessel center point of each blood vessel in the second blood vessel region.

2. A method for quantitative grading of angiogenesis according to claim 1, characterized in that: Step S2 includes: S21: Acquire calibration parameters according to the left-eye image information and the right-eye image information; S22: performing epipolar correction on the left-eye image information and the right-eye image information according to the calibration parameters to eliminate lens distortion; S23: Calculating a disparity map between the left-eye image information and the right-eye image information after eliminating lens distortion; S24: Constructing the three-dimensional point cloud model according to the disparity map.

3. A method for quantitative grading of angiogenesis according to claim 2, characterized in that: Step S24 includes: S241: Obtaining a disparity value of each pixel in the left-eye image information and the right-eye image information according to the disparity map; S242: Obtain the focal length, baseline distance, and noise compensation item of the camera; S243: Calculating a depth value of each pixel according to the focal length, the baseline distance, the noise compensation item, and the disparity value; S244: Convert the depth value into a depth map, and construct the three-dimensional point cloud model according to the depth map.

4. A method for quantitative grading of angiogenesis according to claim 1, characterized in that: Step S5 includes: S51: arbitrarily selecting a cubic region in the three-dimensional point cloud model, where the cubic region contains one or more blood vessels; S52: performing cylindrical fitting on the center point of the blood vessel to obtain the blood vessel width, and calculating the three-dimensional density value of the blood vessel in the cubic area according to the blood vessel centerline and the blood vessel width; S53: Classifying the blood vessels according to the three-dimensional density values.

5. A device for quantitatively grading vascular proliferation, characterized in that: Applied to the steps of the method according to any one of claims 1 to 4 above, the device comprises: Image acquisition module: used to acquire a stereoscopic image of the lesion area, the stereoscopic image including at least left-eye image information and right-eye image information; Point cloud construction module: used to perform stereo correction on the left-eye image information and the right-eye image information to obtain a disparity map between the two images, and construct a three-dimensional point cloud model based on the disparity map; A blood vessel segmentation module is configured to segment a first blood vessel region according to the corrected left-eye image information, and convert the first blood vessel region into a corresponding second blood vessel region in the three-dimensional point cloud model; A parameter calculation module: configured to calculate the vascular geometric parameters of the second vascular region, wherein the vascular geometric parameters include at least a vascular centerline and a vascular center point; Blood vessel grading module: used to calculate the blood vessel width according to the blood vessel center point, calculate the three-dimensional density value of the blood vessel according to the blood vessel centerline and the blood vessel width, and grade the blood vessel according to the three-dimensional density value.

6. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 4 are executed.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are executed.

Images