A feature extraction method and a registration method of blood vessel structure in a retinal OCT image

Abstract

The application discloses a feature extraction method and a registration method of blood vessel structure in a retinal OCT image, the feature extraction method is based on the retinal OCT image to extract a first feature reflecting 3D features of the blood vessel structure, a second feature reflecting intersection features of the blood vessel structure and a third feature reflecting SURF features of the blood vessel structure, and the extracted features are fused, so that the obtained features can have more comprehensive blood vessel structure information, laying a foundation for subsequent registration and other operations. In the application, the feature extraction method designs an extraction step of the second feature extraction for reflecting the intersection features of the blood vessel structure, extracts blood vessel intersections through a first feature extraction matrix to obtain screened blood vessel intersections, and then excludes pseudo feature points from the screened blood vessel intersections through a second feature extraction matrix to obtain twice-screened blood vessel intersections, so that the blood vessel intersections can be avoided to be missed or wrong.

Description

[0001] Technology Neighborhood

[0002] This invention belongs to the field of medical image processing technology, and particularly relates to a method for feature extraction and registration of vascular structures in retinal OCT images. Background Technology

[0003] Optical coherence tomography (OCT), a non-invasive, high-resolution imaging technique, is widely used in ophthalmology and other medical fields. It can produce high-resolution cross-sectional images of each layer of the retina. These images are used to diagnose and monitor various eye diseases, such as macular degeneration, glaucoma, retinal detachment, and retinal vascular occlusion, playing a crucial role in ophthalmic diagnosis. An OCT enface image is an image projected from top to bottom onto a two-dimensional plane along the Z-axis from three-dimensional OCT volumetric data of the retina. This projection allows for visualization and analysis of retinal images at a fixed depth or volume, enabling physicians to visually observe and analyze retinal structures.

[0004] To achieve accurate monitoring of various ophthalmic diseases, it is necessary to register and align OCT enface images from adjacent time points. This process involves spatially aligning (i.e., registering) enface images acquired at different time points using image processing techniques to ensure comparability between images. Registered images can visually demonstrate changes in the lesion area, including neovascularization, atrophy, and improvement or deterioration of retinal structure. This is crucial for evaluating treatment effectiveness and guiding the development of subsequent treatment strategies. However, eye movements and interference from the lesions themselves pose significant challenges to registration, hindering subsequent observation and research of the lesions. Therefore, it is essential to develop a technique capable of accurately registering OCT enface images from adjacent time series.

[0005] There are many methods for OCT enface image registration, which can be mainly divided into two categories: feature-based methods and intensity-based methods. Feature-based registration methods extract key feature points from two images and then establish the correspondence between the images by matching these feature points. Intensity-based methods directly compare the pixel intensities of the images and find the optimal registration parameters by optimizing an objective function (such as minimizing the mean square error between the two images). Since intensity-based methods do not require feature extraction and directly calculate pixel intensities, they have a wider range of applicability. However, these methods are more sensitive to noise and have poor registration accuracy in complex images. Considering that OCT enfaces contain speckle noise, feature-based image registration methods are very effective in processing OCT enface images.

[0006] Among feature-based methods, the most commonly used is the registration method based on vascular structure feature points. This method extracts vascular structures from retinal images, uses these structures for feature description, matches the features of two images, estimates a transformation model, and ultimately achieves image registration. However, this type of method is highly dependent on vascular features, and the presence of lesions often damages and affects vascular structures, thus impacting the matching of vascular features. Furthermore, if there are many features in the image, it will also increase computational complexity, affecting computational efficiency and cost, and causing inconvenience to clinical diagnosis. Currently, no literature has proposed a method that can comprehensively consider and balance the issues of accuracy and time complexity in image registration.

[0007] Therefore, how to propose a feature extraction and registration method for vascular structures in retinal OCT images that comprehensively considers and balances the accuracy and time complexity of image registration is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] To address the problems existing in the prior art, this invention proposes a feature extraction method and registration method for vascular structures in retinal OCT images.

[0009] The technical solution of the present invention is as follows:

[0010] A method for feature extraction of vascular structures in retinal OCT images includes:

[0011] The first feature is extracted from the retinal OCT image to represent the 3D features of the vascular structure.

[0012] The retinal OCT image is converted into a vascular skeleton image to obtain a vascular skeleton image;

[0013] The vascular skeleton image is subjected to a second feature extraction to reflect the intersection features of the vascular structure and a third feature extraction to reflect the SURF features of the vascular structure, thus obtaining the second and third features respectively.

[0014] The first feature, the second feature, and the third feature are fused together with respect to the feature coordinates to form a combined feature, which is used as the feature extraction result.

[0015] Furthermore, the specific steps for extracting the first feature from the retinal OCT image to represent the 3D features of the vascular structure and obtaining the first feature include:

[0016] 1) Denoise the B-scan image corresponding to the retinal OCT image;

[0017] 2) For each B-scan image after denoising, calculate the median, mean, and variance of each cylinder;

[0018] 3) Based on the calculated median, mean, and variance, a pixel importance evaluation model for retinal OCT images is constructed, and based on the pixel importance evaluation model for retinal OCT images, the intensity difference of cylindrical pixels in the retinal OCT images is calculated.

[0019] 4) Compare the intensity difference of the cylindrical pixels with a set threshold related to the median value, filter out pixels that are greater than or equal to the set threshold, and use the filtered pixels as the first feature that reflects the 3D features of the blood vessel structure.

[0020] Furthermore, the expression for the pixel importance evaluation model is:

[0021]

[0022] In the formula, f(x) (s,i) M represents the importance evaluation result of the pixels in the i-th column of the s-th B-scan image; (s,i) σ is the pixel mean of the i-th column of the s-th B-scan image; (s,i) Let be the pixel variance of the i-th column of the s-th B-scan image; e is a mathematical constant;

[0023] The formula for calculating the intensity difference of the cylindrical pixels is:

[0024] Sub (s,i,j) =|I (s,i,j) ×(1-f(x) (s,i) )|

[0025] In the formula, Sub (s,i,j) Let I be the pixel grayscale difference estimation function value of the i-th column and j-th row of the s-th B-scan image; (s,i,j) Let be the pixel value of the i-th column and j-th row of the s-th B-scan image.

[0026] Furthermore, the specific steps for converting the retinal OCT image into a vascular skeleton image to obtain a vascular skeleton image include:

[0027] 1) Project the retinal OCT image along the Z-axis to obtain an OCT face image;

[0028] 2) Then, extract the vascular skeleton image based on the OCT enface image.

[0029] Furthermore, the specific steps for extracting the second feature to reflect the intersection features of vascular structures include:

[0030] 1) Identification of intersections of vascular structures: Neighborhood graph analysis is performed on the vascular skeleton image to obtain potential intersections of vascular structures;

[0031] 2) Extraction of vascular structure intersections: Based on the pre-constructed first feature extraction matrix, vascular structure intersections are extracted from potential vascular intersections to obtain the vascular intersections after initial screening.

[0032] The refined processing of vascular structure intersections involves eliminating pseudo-features from the initially screened vascular structure intersections based on a pre-constructed second feature extraction matrix, resulting in vascular intersections after secondary screening.

[0033] The marking and application of vascular structure intersections: The vascular intersections after secondary screening are marked as a second feature extraction that reflects the characteristics of vascular structure intersections.

[0034] Furthermore, the expression for the first feature extraction matrix T1 is:

[0035]

[0036] The specific extraction method for pre-constructed first feature extraction matrix to extract potential vascular intersections and obtain the initially screened vascular intersections is as follows: The first feature extraction matrix is ​​used to sum the values ​​of eight neighborhoods for each pixel in the OCT enface image. Pixels whose summation result is greater than a first set value are selected as potential vascular intersections. The expression for the summation is:

[0037] g r1 (x,y)=f(x-1,y-1)+f(x-1,y)+f(x-1,y+1)+f(x,y-1)+f(x,y)

[0038] +f(x,y+1)+f(x+1,y-1)+f(x+1,y)+f(x+1,y+1)

[0039] In the formula, g r1 (x,y) is the sum of the 8 neighborhood values ​​corresponding to the pixel with coordinates (x,y) in the OCT enface image; f(x-1,y-1), f(x-1,y), f(x-1,y+1), f(x,y-1), f(x,y), f(x,y+1), f(x+1,y-1), f(x+1,y+1), and f(x+1,y+1) are the pixel with coordinates (x-1,y-1), (x-1,y), (x-1,y+1), (x,y-1), (x,y), (x,y+1), (x+1,y-1), (x+1,y), and (x+1,y+1) in the OCT enface image, respectively.

[0040] Furthermore, the expression for the second feature extraction matrix T2 is:

[0041]

[0042] The specific method for eliminating false feature points from the initially screened vascular intersections based on the pre-constructed second feature extraction matrix to obtain the secondary screening vascular intersections includes: summing the values ​​of the eight neighborhoods of the initially screened vascular intersections using the first feature extraction matrix T2, and selecting pixels whose summation result is greater than or equal to a second set value or equal to a third set value as potential secondary screening vascular intersections; wherein, the expression for the summation is:

[0043] g r2 (x,y)=f(x,y-1)+f(x-1,y)+f(x,y)+f(x+1,y)+f(x,y+1)

[0044] In the formula, g r2 (x,y) is the sum of the 5 neighborhood values ​​corresponding to the pixel with coordinates (x,y) in the OCT enface image; f(x,y-1), f(x-1,y), f(x,y), f(x+1,y) and f(x,y+1) are the pixel with coordinates (x,y-1), (x-1,y), (x,y), (x+1,y) and (x,y+1) in the OCT enface image, respectively.

[0045] Furthermore, the specific steps for extracting the third feature to represent the SURF characteristics of vascular structures include:

[0046] The SURF algorithm was used to extract the third feature to represent the SURF features of vascular structure, and the extraction result was used as the third feature.

[0047] A method for registering vascular structures in retinal OCT images includes:

[0048] Acquire two adjacent retinal OCT images;

[0049] The feature extraction method for vascular structures in retinal OCT images of the present invention is used to extract combined features from two adjacent time-series retinal OCT images respectively, resulting in two feature extraction results for the corresponding two adjacent time-series retinal OCT images;

[0050] The two adjacent time-series retinal OCT images are converted into vascular skeleton images to obtain corresponding vascular skeleton images. Then, the vascular skeleton features of each vascular skeleton image are extracted and the coordinates are simplified to obtain two vascular skeleton features corresponding to the two adjacent time-series retinal OCT images.

[0051] Based on the feature extraction results and vascular skeleton features of two adjacent time-series retinal OCT images, vascular structure registration is performed on the two adjacent time-series retinal OCT images.

[0052] Furthermore, the specific method for registering the vascular structures of the two adjacent retinal OCT images based on the feature extraction results and vascular skeleton features corresponding to the two adjacent retinal OCT images includes:

[0053] Based on the feature extraction results and vascular skeleton features of two adjacent time-series retinal OCT images, the CPD algorithm is used to register the vascular structures of the two adjacent time-series retinal OCT images.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] This invention proposes a feature extraction method for vascular structures in retinal OCT images. This method extracts a first feature that reflects the 3D features of vascular structures, a second feature that reflects the intersection features of vascular structures, and a third feature that reflects the SURF features of vascular structures based on retinal OCT images. The extracted features are then fused, resulting in features that have more comprehensive vascular structure information, laying the foundation for subsequent registration and other operations.

[0056] The feature extraction method of this invention designs a pixel importance evaluation model and selects the first feature that reflects the 3D features of the vascular structure based on the intensity difference of cylindrical pixels. This first feature is used to supplement the 2D features such as the intersection features and SURF features of the vascular structure, and to extract the vascular structure features in multiple dimensions.

[0057] The feature extraction method of the present invention is designed with a second feature extraction step to reflect the characteristics of the intersection of vascular structures. The first feature extraction matrix is ​​used to extract potential vascular intersections to obtain the vascular intersections after initial screening. Then, the second feature extraction matrix is ​​used to eliminate false feature points of the vascular intersections after initial screening to obtain the vascular intersections after secondary screening. This can avoid the omission of vascular intersections and make the extraction of the second feature more accurate.

[0058] The feature extraction method of this invention uses Douglas-Peucker to simplify the coordinates of blood vessel points, which meets the requirement of rapid feature extraction and is especially suitable for clinical registration and other operations.

[0059] This invention proposes a registration method for vascular structures in retinal OCT images. This method uses the feature extraction method of this invention, which can optimize accuracy while ensuring registration speed. Attached Figure Description

[0060] Figure 1This is a flowchart illustrating the feature extraction method for vascular structures in retinal OCT images in this embodiment;

[0061] Figure 2 This is a flowchart illustrating the registration method for vascular structures in retinal OCT images in this embodiment;

[0062] Figure 3(a) is one of the projection diagrams in the embodiment;

[0063] Figure 3(b) is one of the projection diagrams in the embodiment;

[0064] Figure 4 This is a schematic diagram of 3D feature points in the embodiment;

[0065] Figure 5 This is a schematic diagram of blood vessel intersection identification in the embodiment;

[0066] Figure 6 This is a schematic diagram of pseudo-feature points in the embodiment;

[0067] Figure 7 This is a schematic diagram of the markings at the intersection of blood vessels in the embodiment;

[0068] Figure 8 This is a simplified schematic diagram of blood vessel coordinates in the embodiment;

[0069] Figure 9 This is a schematic diagram of the image registration results in the embodiment. Detailed Implementation

[0070] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0071] Example 1:

[0072] This invention provides a method for feature extraction of vascular structures in retinal OCT images, such as... Figure 1 and Figure 2 As shown, it includes:

[0073] S1. Extract the first feature from the retinal OCT image to represent the 3D features of the vascular structure, and obtain the first feature;

[0074] S2. Convert the retinal OCT image into a vascular skeleton image to obtain a vascular skeleton image;

[0075] S3. Extract the second feature from the vascular skeleton image to reflect the intersection features of the vascular structure and extract the third feature to reflect the SURF features of the vascular structure, thus obtaining the second and third features respectively.

[0076] S4. The first feature, second feature, and third feature are fused with respect to the feature coordinates to form a combined feature, which is used as the feature extraction result.

[0077] Furthermore, the first feature extraction, which is used to represent the 3D features of the vascular structure, is performed on the retinal OCT image. The specific steps to obtain the first feature include:

[0078] 1) Denoising is performed on each B-scan image corresponding to the retinal OCT image because retinal OCT images may generate noise that is difficult to remove during the imaging process, especially multiplicative noise. Specifically, the BM3D algorithm can be used for denoising to effectively remove this noise and improve the accuracy of subsequent feature extraction;

[0079] 2) For each B-scan image after denoising, i.e. each slice, project it from top to bottom along the Z-axis, as shown in Figure 3(a) and Figure 3(b), and calculate the median, mean and variance of each cylinder (Z direction);

[0080] 3) Based on the calculated median, mean, and variance, a pixel importance assessment model for retinal OCT images is constructed, and based on the pixel importance assessment model for retinal OCT images, the intensity difference of cylindrical pixels in retinal OCT images is calculated.

[0081] 4) Compare the intensity differences of the cylindrical pixels with a set threshold related to the median, and filter out pixels that are greater than or equal to the set threshold (i.e., pixels with significant changes), such as... Figure 4 As shown, the selected pixels are used as the first feature to represent the 3D features of the blood vessel structure, and a first feature point set can be further formed.

[0082] The threshold value related to the median can be: α×K (s,i) In the formula, α is the threshold coefficient, typically taken as 0.0001; K (s,i) It is the median value of the pixels in the i-th column of the s-th B-scan image;

[0083] Furthermore, the expression for the pixel importance evaluation model is:

[0084]

[0085] In the formula, f(x) (s,i) M represents the importance evaluation result of the pixels (also called voxels) in the i-th column of the s-th B-scan image; (s,i) σ is the pixel mean of the i-th column of the s-th B-scan image; (s,i) Let be the pixel variance of the i-th column of the s-th B-scan image, i∈[1,n]; e is a mathematical constant;

[0086] The formula for calculating the intensity difference of cylindrical pixels is:

[0087] Sub (s,i,j) =|I (s,i,j) ×(1-f(x) (s,i) )

[0088] In the formula, Sub (s,i,j) Let I be the pixel grayscale difference estimation function value of the i-th column and j-th row of the s-th B-scan image; (s,i,j) Let be the pixel value in the i-th column and j-th row of the s-th B-scan image;

[0089] Furthermore, the specific steps for converting retinal OCT images into vascular skeleton images include:

[0090] 1) Project the retinal OCT image along the Z-axis to obtain the OCT enface image;

[0091] 2) Then, extract the vascular skeleton image based on the OCT enface image;

[0092] 3) The coordinates of the vascular skeleton points in the extracted vascular skeleton image can be simplified using the Douglas-Peucker method.

[0093] Furthermore, the specific steps for extracting the second feature to represent the intersection features of vascular structures include:

[0094] 1) Identification of intersections of vascular structures: Neighborhood graph analysis is performed on the vascular skeleton image to obtain potential intersections of vascular structures, such as... Figure 5 As shown;

[0095] 2) Extraction of vascular structure intersections: Based on the pre-constructed first feature extraction matrix, vascular structure intersections are extracted from potential vascular intersections to obtain the vascular intersections after initial screening.

[0096] The refined processing of vascular structure intersections involves eliminating false feature points from the initially screened vascular structure intersections based on a pre-constructed second feature extraction matrix, resulting in vascular intersections after a second round of screening. Figure 6 As shown;

[0097] The marking and application of vascular structure intersections: After secondary screening, the vascular intersections are marked as a second feature extraction, reflecting the characteristics of vascular structure intersections, which can further form a second feature point set. After extracting the vascular intersections, the pixels of the entire binary image are replaced with 0s and 1s. The background of the vascular image becomes white (grayscale value 1), and the vascular pixels become black (grayscale value 0). The extracted intersections are marked to facilitate more intuitive observation of the image, such as... Figure 7 As shown.

[0098] Furthermore, the expression for the first feature extraction matrix T1 is:

[0099]

[0100] Based on the pre-constructed first feature extraction matrix, potential vascular intersections are extracted. The specific extraction method for the initially screened vascular intersections is as follows: the first feature extraction matrix is ​​used to sum the values ​​of 8 neighborhoods for each pixel in the OCT enface image, and pixels whose summation result is greater than a first set value (3 in this example) are selected as potential vascular intersections. The expression for the summation is:

[0101] g r1 (x,y)=f(x-1,y-1)+f(x-1,y)+f(x-1,y+1)+f(x,y-1)+f(x,y)

[0102] +f(x,y+1)+f(x+1,y-1)+f(x+1,y)+f(x+1,y+1)

[0103] In the formula, g r1 (x,y) is the sum of the 8 neighborhood values ​​corresponding to the pixel with coordinates (x,y) in the OCT enface image; f(x-1,y-1), f(x-1,y), f(x-1,y+1), f(x,y-1), f(x,y), f(x,y+1), f(x+1,y-1), f(x+1,y+1), and f(x+1,y+1) are the pixel with coordinates (x-1,y-1), (x-1,y), (x-1,y+1), (x,y-1), (x,y), (x,y+1), (x+1,y-1), (x+1,y), and (x+1,y+1) in the OCT enface image, respectively.

[0104] Furthermore, the expression for the second feature extraction matrix T2 is:

[0105]

[0106] Based on a pre-constructed second feature extraction matrix, the method for eliminating false feature points from the initially screened vascular intersections to obtain the secondary screening vascular intersections includes: summing the values ​​of the eight neighborhoods of the initially screened vascular intersections using the first feature extraction matrix T2, and selecting pixels whose summation is greater than or equal to a second set value or equal to a third set value as potential secondary screening vascular intersections; wherein, the expression for the summation is:

[0107] g r2(x,y)=f(x,y-1)+f(x-1,y)+f(x,y)+f(x+1,y)+f(x,y+1)

[0108] In the formula, g r2 (x,y) is the sum of the 5 neighborhood values ​​corresponding to the pixel with coordinates (x,y) in the OCT enface image; f(x,y-1), f(x-1,y), f(x,y), f(x+1,y) and f(x,y+1) are the pixel with coordinates (x,y-1), (x-1,y), (x,y), (x+1,y) and (x,y+1) in the OCT enface image, respectively.

[0109] Furthermore, the specific steps for extracting the third feature to represent the SURF characteristics of vascular structures include:

[0110] The SURF algorithm is used to extract the third feature to represent the SURF features of the vascular structure. The extraction result is used as the third feature, and a third feature point set can be further formed.

[0111] Example 2:

[0112] The present invention provides a method for registering vascular structures in retinal OCT images, such as... Figure 2 As shown, it includes:

[0113] Acquire two adjacent time-series retinal OCT images;

[0114] The feature extraction method for vascular structures in retinal OCT images of the present invention is used to extract combined features from two adjacent time-series retinal OCT images to obtain two feature extraction results for the corresponding two adjacent time-series retinal OCT images.

[0115] Two adjacent retinal OCT images are transformed into vascular skeleton images, respectively, to obtain corresponding vascular skeleton images. Then, vascular skeleton features are extracted and coordinates are simplified for each vascular skeleton image, resulting in two vascular skeleton features corresponding to the two adjacent retinal OCT images. Further, a vascular skeleton feature point set can be formed, as shown in the following figure. Figure 8 As shown;

[0116] Based on the feature extraction results and vascular skeleton features of two adjacent retinal OCT images, vascular structure registration is performed between the two adjacent retinal OCT images.

[0117] Furthermore, based on the feature extraction results and vascular skeleton features of two adjacent retinal OCT images, the specific method for registering vascular structures in two adjacent retinal OCT images includes:

[0118] Based on the feature extraction results and vascular skeleton features of two adjacent retinal OCT images, the CPD algorithm is used to register the vascular structures of the two adjacent retinal OCT images. The results are as follows: Figure 9 As shown.

[0119] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for feature extraction of vascular structures in retinal OCT images, characterized in that, include: The first feature is extracted from the retinal OCT image to represent the 3D features of the vascular structure. The retinal OCT image is converted into a vascular skeleton image to obtain a vascular skeleton image; The vascular skeleton image is subjected to a second feature extraction to reflect the intersection features of the vascular structure and a third feature extraction to reflect the SURF features of the vascular structure, thus obtaining the second and third features respectively. The first feature, the second feature, and the third feature are fused with respect to the feature coordinates to form a combined feature, which is used as the feature extraction result. The specific steps for extracting the first feature from the retinal OCT image to represent the 3D features of the vascular structure and obtaining the first feature include: 1) Denoise the retinal OCT image for each B-scan image; 2) For each B-scan image after denoising, calculate the median, mean, and variance of each cylinder; 3) Based on the calculated median, mean, and variance, a pixel importance evaluation model for retinal OCT images is constructed, and based on the pixel importance evaluation model for retinal OCT images, the intensity difference of cylindrical pixels in the retinal OCT images is calculated. 4) Compare the intensity difference of the cylindrical pixels with a set threshold related to the median value, filter out pixels that are greater than or equal to the set threshold, and use the filtered pixels as the first feature that reflects the 3D features of the blood vessel structure. The specific steps for converting the retinal OCT image into a vascular skeleton image to obtain a vascular skeleton image include: 1) Project the retinal OCT image along the Z-axis to obtain an OCT enface image; 2) Then, extract the vascular skeleton image based on the OCT enface image.

2. The method for feature extraction of vascular structures in retinal OCT images according to claim 1, characterized in that, The expression for the pixel importance evaluation model is: ， In the formula, The importance evaluation result of the pixel in the i-th column in the s-th B-scan image; Let be the average pixel value of the i-th column of the s-th B-scan image; Let be the pixel variance of the i-th column of the s-th B-scan image; It is a mathematical constant; The formula for calculating the intensity difference of the cylindrical pixels is: ， In the formula, Let be the pixel grayscale difference estimation function value of the i-th column and j-th row of the s-th B-scan image; Let be the pixel value of the i-th column and j-th row of the s-th B-scan image.

3. The method for feature extraction of vascular structures in retinal OCT images according to claim 1, characterized in that, The specific steps for extracting the second feature to reflect the intersection features of vascular structures include: 1) Identification of intersections of vascular structures: Neighborhood graph analysis is performed on the vascular skeleton image to obtain potential intersections of vascular structures; 2) Extraction of vascular structure intersections: Based on the pre-constructed first feature extraction matrix, vascular structure intersections are extracted from potential vascular intersections to obtain the vascular intersections after initial screening. The refined processing of vascular structure intersections involves eliminating pseudo-features from the initially screened vascular structure intersections based on a pre-constructed second feature extraction matrix, resulting in vascular intersections after secondary screening. The marking and application of vascular structure intersections: The vascular intersections after secondary screening are marked as a second feature extraction that reflects the characteristics of vascular structure intersections.

4. The method for feature extraction of vascular structures in retinal OCT images according to claim 3, characterized in that, The first feature extraction matrix The expression is: ， The specific extraction method for pre-constructed first feature extraction matrix to extract potential vascular intersections and obtain the initially screened vascular intersections is as follows: The first feature extraction matrix is ​​used to sum the values ​​of eight neighborhoods for each pixel in the OCT enface image. Pixels whose summation result is greater than a first set value are selected as potential vascular intersections. The expression for the summation is: ， In the formula, The coordinates in the OCT enface image are The summation of the 8 neighborhood values ​​corresponding to the pixel; , , , , , , , and The coordinates in the OCT enface image are respectively , , , , , , , and The pixels.

5. The method for feature extraction of vascular structures in retinal OCT images according to claim 3, characterized in that, Second feature extraction matrix The expression is: ， The specific method for eliminating false feature points from the initially screened vascular structure intersections based on a pre-constructed second feature extraction matrix to obtain the second-screened vascular intersections includes: using the first feature extraction matrix... The values ​​of the five neighborhoods of the initially screened vessel intersections are summed. Pixels whose summation is greater than or equal to a second or third preset value are selected as potential vessel intersections for secondary screening. The expression for the summation is: ， In the formula, The coordinates in the OCT enface image are The summation of the 5 neighborhood values ​​corresponding to the pixel; , , , and The coordinates in the OCT enface image are respectively , , , and The pixels.

6. The method for feature extraction of vascular structures in retinal OCT images according to claim 1, characterized in that, The specific steps for extracting the third feature to represent the SURF characteristics of vascular structures include: The SURF algorithm was used to extract the third feature to represent the SURF features of vascular structure, and the extraction result was used as the third feature.

7. A method for registering vascular structures in retinal OCT images, characterized in that, include: Acquire two adjacent retinal OCT images; The feature extraction method for vascular structures in retinal OCT images according to any one of claims 1 to 6 is used to extract combined features from the two adjacent time-series retinal OCT images respectively, to obtain two feature extraction results for the corresponding two adjacent time-series retinal OCT images; The two adjacent time-series retinal OCT images are converted into vascular skeleton images to obtain corresponding vascular skeleton images. Then, the vascular skeleton features of each vascular skeleton image are extracted and the coordinates are simplified to obtain two vascular skeleton features corresponding to the two adjacent time-series retinal OCT images. Based on the feature extraction results and vascular skeleton features of two adjacent time-series retinal OCT images, vascular structure registration is performed on the two adjacent time-series retinal OCT images.

8. The registration method for vascular structures in retinal OCT images according to claim 7, characterized in that, The specific method for registering the vascular structure of two adjacent retinal OCT images based on the feature extraction results and vascular skeleton features of the two adjacent time-series retinal OCT images includes: Based on the feature extraction results and vascular skeleton features of two adjacent time-series retinal OCT images, the CPD algorithm is used to register the vascular structures of the two adjacent time-series retinal OCT images.

Images