A method and system for registering fundus angiogram images

Abstract

The present application relates to medical image registration technology, disclose a kind of for fundus angiogram image registration method and system, its extraction blood vessel tree structure method includes, extraction center line and division branch point;The method for extracting tree structure feature;According to tree structure, the method for segmenting and extracting curve segment feature of blood vessel curve;After comparing tree structure feature similarity, determine branch point, then compare the similarity of adjacent curve feature registration method.The present application extracts fundus blood vessel tree, and according to branch point, the feature of branch point and blood vessel segment is summarized, the part of high similarity in different images is found, the registration between two images with different fields of view is realized, and the scene demand is satisfied.

Description

TECHNICAL FIELD

[0001] The present application relates to medical image registration technology, and in particular to a method and system for fundus angiography image registration. BACKGROUND

[0002] With the development of medical technology, the types of image examination in clinical scenarios are increasing, and medical workers have an increasing demand for automated image processing. In ophthalmic clinics, fundus angiography is one of the important examinations for checking abnormalities such as blood vessel leakage and blockage, and is also an important reference for fundus laser surgery. However, the fundus vascular structure is complex, and the images in the examination report often have a dispersed visual field and large differences, making it difficult for doctors to provide a global view. Therefore, there is a high demand for image registration and stitching, but there is currently no reliable fundus angiography image registration method.

[0003] In similar scenarios, such as prior art 1: patent CN109767459B provides a fundus blood vessel method based on deep learning to generate a deformation field, but the visual field of the fundus image is similar, and the required deformation field is mostly scaling and rotation, lacking visual field matching;

[0004] Prior art 2: patent CN102548478B uses features based on vascular calcification to register different images in the X-ray vascular intervention scenario, but fundus vascular lesions are mainly leakage, and over time, contrast agent leakage can cause a large difference in the texture features of the same lesion;

[0005] Prior art 3: patent CN112508881A uses a registration method based on mutual information and image pyramids to register images, but the research object of this method is head radiographic images, and there is no large difference in visual field. Therefore, the mutual information of different visual fields is greatly different, and therefore is not suitable for the fundus angiography scenario. SUMMARY

[0006] The present application provides a method and system for fundus angiography image registration to solve the problem of reducing the quality of the projection picture by the trapezoidal correction method in the prior art.

[0007] To solve the above technical problems, the present application solves the problems by the following technical solutions:

[0008] A method for fundus angiography image registration, the method comprising:

[0009] Construction of a fundus perfusion vessel segmentation model M';

[0010] Extraction of the blood vessel centerline, the blood vessel centerline l is extracted from the fundus perfusion vessel segmentation model M' i ;

[0011] Tree structure feature extraction, through the blood vessel center line l i Extracting the multi-branch tree structure feature;

[0012] Obtaining the curve feature, intercepting the blood vessel center curve l i The connecting line coordinates between any two adjacent branch points of the curve, and the tangent and inverse tangent are calculated one by one to obtain the segmented curve feature;

[0013] Similarity comparison registration, for any two images I1 and I2 in the fundus angiogram, the tree structure feature of each pair of branch points and the curve feature of the adjacent blood vessel segment of each connected region are compared for similarity comparison registration;

[0014] Image registration, for the registrable curve segment and The connecting lines of the branch nodes at both ends are and The included angle α of the connecting lines is calculated counterclockwise; if the tangent and inverse tangent of the two curve segments are registrable, α = α + 180; for images I1 and I2, the two are horizontally superimposed with and as the center, and the image registration is completed by counterclockwise rotation of α degrees.

[0015] As preferred, the method for constructing the fundus perfusion blood vessel segmentation model M' comprises:

[0016] Construction of the blood vessel segmentation model M, a two-dimensional UNet structure is used to construct the blood vessel segmentation model M;

[0017] Obtaining the training set T, N fundus angiograms are collected and pixel-level segmentation annotation of blood vessels is performed by professional ophthalmologists to form the training set T;

[0018] Obtaining the fundus perfusion blood vessel segmentation model M', the blood vessel segmentation model M is trained using the training set T to obtain the trained model M'.

[0019] As preferred, the blood vessel segmentation result S is obtained by passing the blood vessel angiogram I through the blood vessel segmentation model M' to obtain the blood vessel segmentation result S;

[0020] Extracting the connected region, calculating the connected region, and extracting each connected region set S' = {s0,…,st}; t};

[0021] Labeling each connected region s i ,i∈{0,…,t} to obtain the labeled point p;

[0022] Connect all the marked points p in order of adjacency and smooth by spline interpolation to obtain a connected region s i centerline l i .

[0023] As preferred, mark each connected region s i , i∈{0,…,t} to obtain marked points p; and perform the following steps:

[0024] S1. Select the nearest place to the image edge as the entrance, and measure the pixel width w of the entrance, the center point coordinate of the entrance being p=(x,y);

[0025] S2. Draw an intersection circle O with p as the center and k·w / 2 as the radius (k>1);

[0026] S3. Calculate the set C of the intersection of the boundary of the intersection circle O and the segmentation result s i , and count the number m of the intersection results: if m>1, p is a branch node; if m=1, p is a non-branch node; if m=0, p is a terminal point;

[0027] S4. Select an unmarked intersection c∈C, extract the intersection points on both sides of the blood vessel segmentation result, and calculate the distance w′ between the two intersection points and the center point p′=(x′,y′);

[0028] S5. Take p′ as p and w′ as w, repeat steps a to d until there is no unmarked intersection, and end.

[0029] As preferred, the extraction of the tree structure features is performed for each centerline l i , and a multi-branch tree structure is extracted, taking the direction of the upper left corner of the image as the starting point, dividing all the centerline points into a branch point set B={b i , i∈(0,1,…)} and a non-branch point set E={e i , i∈(0,1,…)}, and extracting the tree structure features according to the following steps:

[0030] Calculate the angle degree θ between every two adjacent edges e1 and e2 of each branch point b where ∠e 1i be 2i represents the angle degree between the line connecting the i-th point on e 1i and e 2i and the branch point b;

[0031] Express the tree structure features of the branch point b in degrees m is the branch number;

[0032] Collect the tree structure features of the branch point set on a centerline region to obtain the tree structure features of the blood vessel

[0033] By combining the tree structure features of all connected regions, we obtain the tree structure feature set in image I.

[0034] Preferably, the curve features are obtained by taking the coordinates of the connecting line between any two adjacent branch points of the vascular center curve and calculating the tangent for each branch as the piecewise curve features. The specific steps include:

[0035] Extract the center line s of the blood vessels in image I i Two adjacent branch points b p and b q The sequence of segmented vascular curves between the segments v = {(x t ,y t Let t∈{0,1,…}}, and let the starting coordinates be b. p The coordinates of the endpoint are b. q ;

[0036] Except for the starting and ending coordinates, calculate the tangential tan and inverse tangential retan along the coordinate sequence v one by one. For (x... t ,y t The calculation method is as follows:

[0037]

[0038]

[0039] Where norm is the vector normalization. (x t ,y t The curve characteristics are as follows:

[0040] For a sequence of curve segments v, its piecewise curve characteristics are as follows:

[0041] For blood vessel s, its curve characteristics are as follows:

[0042] Preferably, similarity comparison registration is performed on two images I1 and I2 after extracting the tree structure features and curve features of each connected region. The specific steps are as follows:

[0043] Step 1: Select one unmatched connected component in each of images I1 and I2. and Extract its tree structure features and curve features, and combine the branch points of the two blood vessels in pairs;

[0044] Step 2, calculate all branch point combinations The tree structure similarity is calculated by Hausdorff distance.

[0045] Step 3, set the tree structure similarity threshold value epsilon, select two branch points with the maximum similarity and greater than epsilon as the starting branch points, and enter step 4; if there is no branch point combination with the similarity greater than epsilon, return to step 1 to select the unmatched blood vessel combination to calculate the tree structure similarity. If there is no unmatched blood vessel combination, it indicates that the images I1 and I2 have no overlapping area and cannot be registered.

[0046] Step 4, set the curve feature similarity threshold value theta. For the branch node and select one adjacent branch node to obtain the curve features and calculate the average Euclidean distance omega of the tangent direction of and the tangent direction of If omega is less than or equal to theta, the two curve segments can be registered and enter the image registration;

[0047] If omega is greater than theta, calculate the average Euclidean distance omega' of the tangent direction of and the inverse tangent direction of If omega' is less than or equal to theta, the two curve segments can be registered and enter the image registration; if omega' is greater than theta, the two curve segments cannot be registered and enter step 5.

[0048] Step 5, select another curve segment feature of and calculate the curve feature similarity, and repeat step 4; if does not exist, select and calculate the similarity, and repeat step 4; if also does not exist, the two curve segments cannot be registered.

[0049] As a preferred, it also includes the determination of unregistered, traversing all branch node combinations with the tree structure similarity less than the threshold value epsilon, repeating step 4; if there is no branch node combination that has not been calculated and meets the condition of being less than the threshold value epsilon, the two connected regions are marked as matched;

[0050] Traverse the unmatched connected region combination, repeat step 1; if there is no unmatched connected region combination, the images I1 and I2 have no overlapping area and cannot be registered.

[0051] In order to solve the above technical problems, the application also provides a system for fundus angiography image registration, which is composed of the method for fundus angiography image registration.

[0052] To solve the above technical problems, the application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the fundus angiography image registration method.

[0053] The application has the following technical effects:

[0054] The application extracts fundus vascular trees, segments the blood vessels according to branch points, collects the characteristics of the branch points and the blood vessel segments, finds the parts with high similarity in different images, realizes the registration between two images with different fields of view, and meets the scene requirements. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a flowchart of the application;

[0056] Figure 2-1 is a first original image of the application, Figure 2-2 is a second original image of the application;

[0057] Figure 3-1 is a blood vessel segmentation and selected connected region image corresponding to the first original image of the application; Figure 3-2 is a blood vessel segmentation and selected connected region image corresponding to the second original image of the application;

[0058] Figure 4-1 is a connected region center line extraction image corresponding to the first image of the application; Figure 4-2 is a connected region center line extraction image corresponding to the second image of the application;

[0059] Figure 5 is an image registration result image of the application. DETAILED DESCRIPTION

[0060] The application will be further described in detail below in combination with the drawings and embodiments.

[0061] Embodiment 1

[0062] A method for fundus angiography image registration, the method comprising:

[0063] Construction of a fundus perfusion blood vessel segmentation model M'; Extraction of a blood vessel center line l i Extraction of a tree structure feature through the blood vessel center line l i Extraction of a multi-branch tree structure feature;

[0064] Acquisition of a curve feature by intercepting the blood vessel center curve l iThe coordinates of the connecting line between any two adjacent branch points are obtained, and the tangential and antitangential directions are calculated one by one to obtain the piecewise curve characteristics;

[0065] Similarity comparison registration involves comparing the tree structure features of each pair of branch points in every connected region and the curve features of adjacent vessel segments for any two images I1 and I2 in fundus angiography. Image registration is then performed for registrable curve segments. and The lines connecting the branch nodes at both ends are respectively and Calculate the counterclockwise angle α between the two curve segments; if the two curve segments Tangent and Inverse tangential registration is possible, α = α + 180; for images I1 and I2, with and Centered on the image, the two images are horizontally superimposed and rotated counterclockwise by α degrees to complete image registration.

[0066] The methods for constructing the fundus perfusion vessel segmentation model M′ include:

[0067] The blood vessel segmentation model M is constructed using a two-dimensional UNet structure.

[0068] The training set T is obtained by collecting N fundus angiography images and having professional ophthalmologists perform pixel-level segmentation and annotation of the blood vessels to form the training set T;

[0069] To obtain the fundus perfusion vessel segmentation model M′, the training set T is used to train the vessel segmentation model M, resulting in the trained model M′.

[0070] The blood vessel segmentation result S is obtained by passing the angiography image I through the blood vessel segmentation model M′.

[0071] Extraction of connected components: Calculate connected components and extract the set of each connected component S′={s0,…,s t};

[0072] For each connected region s i Mark the points p in the i∈{0,…,t} groups;

[0073] Connect all the marked points p in order of adjacency, and then smooth the connection using spline interpolation to obtain the connected region s. i centerline l i .

[0074] For each connected region s i Mark the points p in the i∈{0,…,t} groups;

[0075] The following steps are performed:

[0076] S1. Select the nearest point to the edge of the connected region as the entrance, and measure the pixel width w of the entrance. The center point of the entrance is p = (x, y);

[0077] S2. Draw the intersection circle O with p as the center and k·w / 2 as the radius (k > 1);

[0078] S3. Calculate the intersection set C of the boundary of the intersection circle O and the segmentation result s i , and count the number of intersections m: if m > 1, p is a branch node; if m = 1, p is a non-branch node; if m = 0, p is an end point;

[0079] S4. Select an unmarked intersection c ∈ C, extract the intersection points on both sides of the blood vessel segmentation result, and calculate the distance w' between the two intersection points and the center point p' = (x', y');

[0080] S5. Repeat steps a to d with p' as p and w' as w until there is no unmarked intersection, and end.

[0081] The extraction of the tree structure features is performed for each center line l i , and a multi-branch tree structure is extracted. The image upper left corner direction is taken as the starting point, and all center line points are divided into a branch point set B = {b i , i ∈ (0, 1, …)} and a non-branch point set E = {e i , i ∈ (0, 1, …)}. The tree structure features are extracted according to the following steps:

[0082] The angle degree θ between each two adjacent edges e1 and e2 at each branch point b is calculated where ∠e 1i be 2i represents the angle degree between the line connecting the i-th point on e 1i and e 2i and the branch point b.

[0083] The tree structure features of the branch point b are expressed in degrees m is the number of branches;

[0084] The tree structure features of the branch point set in a center line region are collected to obtain the tree structure features of the blood vessel

[0085] The tree structure features of all connected regions are collected to obtain the tree structure feature set in the image I

[0086] The curve features are obtained by extracting the coordinates of the connecting line between any two adjacent branch points of the vascular center curve and calculating the tangent for each point as the piecewise curve feature. Specific steps include:

[0087] Extract the center line s of the blood vessels in image I i Two adjacent branch points b p and b q The sequence of segmented vascular curves between the segments v = {(x t ,y t Let t∈{0,1,…}}, and let the starting coordinates be b. p The coordinates of the endpoint are b. q ;

[0088] Except for the starting and ending coordinates, calculate the tangential tan and inverse tangential retan along the coordinate sequence v one by one. For (x... t ,y t The calculation method is as follows:

[0089]

[0090]

[0091] Where norm is the vector normalization. (x t ,y t The curve characteristics are as follows:

[0092] For a sequence of curve segments v, its piecewise curve characteristics are as follows:

[0093] For blood vessel s, its curve characteristics are as follows:

[0094] Similarity comparison registration is performed on two images I1 and I2. After extracting the tree structure features and curve features of each connected region, similarity comparison registration is performed. The specific steps are as follows:

[0095] Step 1: Select one unmatched connected component in each of images I1 and I2. and Extract its tree structure features and curve features, and combine the branch points of the two blood vessels in pairs;

[0096] Step 2, calculate all branch point combinations The similarity of tree structure features is calculated using the Hausdorff distance.

[0097] Step 3, set the tree structure similarity threshold ε, select two branch points with the maximum similarity and greater than ε as the starting branch points, and enter step 4; if there is no branch point combination with a similarity greater than ε, return to step 1 to select the unmatched blood vessel combination to calculate the tree structure similarity. If there is no unmatched blood vessel combination, it indicates that the images I1 and I2 have no overlapping area and cannot be registered.

[0098] Step 4, set the curve feature similarity threshold θ. For the branch node and select one adjacent branch node to obtain the curve features and of the intermediate blood vessel segment curve , respectively. Calculate the average Euclidean distance ω of the tangent direction of and the tangent direction of . If ω≤θ, the two curve segments can be registered, and enter the image registration;

[0099] If ω>θ, calculate the average Euclidean distance ω' of the tangent direction of and the inverse tangent direction of . If ω'≤θ, the two curve segments can be registered, and enter the image registration; if ω'>θ, the two curve segments cannot be registered, and enter step 5.

[0100] Step 5, select another curve segment feature of and calculate the curve feature similarity, repeat step 4; if does not exist, select and to calculate the similarity, repeat step 4; if also does not exist, the two curve segments cannot be registered.

[0101] Example 2

[0102] Based on example 1, the present embodiment has two existing fundus angiography images Figure 2-1 and 2-2 , and the overlapping area is located and the images are registered according to the steps of the present application.

[0103] Construct a fundus perfusion blood vessel segmentation model.

[0104] Use a two-dimensional UNet structure to construct a blood vessel segmentation model M;

[0105] Collect 500 fundus angiography images and perform pixel-level segmentation annotation of blood vessels by professional ophthalmologists to form a training set T;

[0106] Use the training set T to train the blood vessel segmentation model M to obtain the trained model M'.

[0107] Based on the segmentation result, a blood vessel centerline is extracted.

[0108] Angiography image Figure 2-1 and 2-2 The blood vessel segmentation model M' is passed through to obtain a blood vessel segmentation result S;

[0109] The connected regions are calculated and extracted Figure 2-1 and 2-2 The connected region set of FIG. 2

[0110] For the connected region Figure 2-1 , the following steps are performed: (as shown in Figure 3-1 ) and Figure 2-1 the connected region (as shown in Figure 3-1 ):

[0111] a. Select the nearest edge of the connected region as the entrance, and measure the pixel width w of the entrance. The center point coordinate of the entrance is p = (x, y);

[0112] b. Draw the intersection circle O with p as the center and 1.05·w / 2 as the radius;

[0113] c. Calculate the intersection set C of the intersection circle O boundary and the segmentation result s i , and count the number m of the intersection results: if m > 1, p is a branch node; if m = 1, p is a non-branch node; if m = 0, p is a terminal point;

[0114] d. Select an unmarked intersection c ∈ C, extract the intersection points on both sides of the blood vessel segmentation result, and calculate the distance w' between the two intersection points and the center point p' = (x', y');

[0115] e. Repeat steps a to d with p' as p and w' as w until there is no unmarked intersection, and end the step

[0116] Connect all the marked points p in order according to the adjacent relationship, and after spline interpolation smoothing, the centerline of the connected region is obtained and the centerline of the connected region is obtained Figure 4-1 as shown in 4-2 .

[0117] For the centerline in Figure 4-1 and 4-2 , a multi-branch tree structure is extracted, and the tree structure features of all centerlines are extracted from the left upper corner direction of the image as the starting point;

[0118] For the centerline inFigure 4-1 and 4-2 The centerline multi-branch tree structure is used to calculate curve characteristics by taking the curve segment in adjacent branch points as the unit.

[0119] Set the tree similarity threshold ε = 2.0 and compare the branch points. ( Figure 4-1 Tree structure features (indicated by the white arrow on the left) and branch points ( Figure 4-2 Tree structure features (indicated by the white arrow on the left) According to the Hausdorff distance formula, the similarity of tree structure features of branch combinations is 0.6 < ε, that is... It appears to be the same branch point;

[0120] Set the curve feature similarity threshold θ = 0.35, and calculate the curve segment. and Based on the curve feature similarity and the mean Euclidean distance, the similarity ω = 0.1744 < θ, indicating that the two curve segments can be registered.

[0121] calculate Connecting segments and The included angle α of the connecting line segments is 0.81 degrees. and To complete registration, rotate clockwise α to ensure alignment. Figure 5 As shown.

[0122] Example 3

[0123] Based on Example 1, Figure 1 This embodiment also includes determining the unregistered cases by traversing all branch node combinations with tree structure feature similarity less than the threshold ε and repeating step 4; if there are no uncalculated branch node combinations that satisfy the condition of less than the threshold ε, the two connected regions to which they belong are marked as matched;

[0124] Traverse unmatched connected component combinations and repeat step 1; if there are no unmatched connected component combinations, images I1 and I2 do not have overlapping regions and cannot be registered.

[0125] Example 4

[0126] Based on Example 1, this example is a system for registering fundus angiography images, which is a system constructed using the method for registering fundus angiography images described above.

[0127] Example 5

[0128] Based on the embodiment 1, this embodiment is a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the fundus angiography image registration method.

Claims

1. A method for image registration in fundus angiography, the method comprising: Construction of a segmentation model M' for fundus perfusion vessels; The extraction of the vascular centerline was performed using a fundus perfusion vessel segmentation model M'. i ; Tree structure feature extraction, through the blood vessel centerline l i Extracting multi-way tree structural features; Obtaining curve features, extracting the vessel centerline l i The coordinates of the connecting line between any two adjacent branch points are obtained, and the tangential and antitangential directions are calculated one by one to obtain the piecewise curve characteristics; Similarity comparison registration is performed on any two images I1 and I2 in fundus angiography images. The tree structure features of each pair of branch points in each connected region and the curve features of adjacent vascular segments are compared and registered. Image registration, for registerable curve segments and The lines connecting the branch nodes at both ends are respectively and Calculate the counterclockwise angle α between the two curve segments; if the two curve segments Tangent and Inverse tangential registration is possible, α = α + 180; for images I1 and I2, with and Centered on the image, the two images are horizontally superimposed and rotated counterclockwise by α degrees to complete image registration.

2. The method for image registration in fundus angiography according to claim 1, characterized in that, The methods for constructing the fundus perfusion vessel segmentation model M' include: The blood vessel segmentation model M is constructed using a two-dimensional UNet structure. The training set T is obtained by collecting N fundus angiography images and having professional ophthalmologists perform pixel-level segmentation and annotation of the blood vessels to form the training set T; To obtain the fundus perfusion vessel segmentation model M', the training set T is used to train the vessel segmentation model M, resulting in the trained model M'.

3. The method for image registration in fundus angiography according to claim 1, characterized in that, The blood vessel segmentation result S is obtained by passing the angiography image I through the blood vessel segmentation model M′. Extraction of connected components: Calculate connected components and extract the set of each connected component S' = {s0, ..., s...} t }; For each connected region s i Mark the points p in the i∈{0,…,t} groups; Connect all the marked points p in order of adjacency, and then smooth the connection using spline interpolation to obtain the connected region s. i centerline l i .

4. The method for image registration in fundus angiography according to claim 3, characterized in that, For each connected region s i Mark the points p in the i∈{0,…,t} groups; Perform the following steps: S1. Select the point in the connected region that is closest to the edge of the image as the entry point, and measure the pixel width w of the entry point. The coordinates of the center point of the entry point are p = (x, y). S2. With p as the center and k·w / 2 as the radius, and k>1, draw the intersection circle O; S3. Calculate the boundary of the intersection circle O and the segmentation result s. i Let C be the set of overlapping points, and count the number m of overlapping results: if m>1, p is a branch node; if m=1, p is a non-branch node; if m=0, p is the endpoint. S4. Select an unmarked overlapping point c∈C, extract its intersection points with the two sides of the blood vessel segmentation result, and calculate the distance w' between the two intersection points and the center point p'=(x′,y′); S5. Using p' as p and w' as w, repeat steps S1 to S4 until there are no more unmarked overlapping areas, then end.

5. The method for image registration in fundus angiography according to claim 1, characterized in that, Extraction of tree structure features For each centerline l i Extract the multi-branch tree structure, starting from the top left corner of the image, and divide all centerline points into a branch point set B = {b}. i The set of non-branching points E = {e, i∈(0,1,…)} i For each i ∈ (0, 1, ...), extract the tree structure features as follows: Calculate the angle between every two adjacent sides e1 and e2 at each branch point b. Where ∠e 1i be 2i Represents e 1i and e 2i The angle formed by the lines connecting the i-th point and the branch point b; Using degree to represent the tree structure characteristics of branch point b m is the number of branches; The tree structure features of the blood vessel are obtained by combining the tree structure features of the branch point set on a central axis region. By combining the tree structure features of all connected regions, we obtain the tree structure feature set in image I.

6. The method for image registration in fundus angiography according to claim 1, characterized in that, The curve features are obtained by extracting the coordinates of the connecting line between any two adjacent branch points of the vascular center curve and calculating the tangent for each point as the piecewise curve feature. Specific steps include: Extract the center line of the blood vessels in image I. i Two adjacent branch points b p and b q The sequence of segmented vascular curves between the segments v = {(x t ,y t Let t∈{0,1,…}}, and let the starting coordinates be b. p The coordinates of the endpoint are b. q ; Except for the starting and ending coordinates, calculate the tangential tan and inverse tangential retan along the coordinate sequence v one by one. For (x... t ,y t The calculation method is as follows: Where norm is the vector normalization, (x t ,y t The curve characteristics are as follows: For a sequence of curve segments v, its piecewise curve characteristics are as follows: For blood vessel s, its curve characteristics are as follows:

7. The method for image registration in fundus angiography according to claim 1, characterized in that, Similarity comparison registration is performed on two images I1 and I2. After extracting the tree structure features and curve features of each connected region, similarity comparison registration is performed. The specific steps are as follows: Step 1: Select one unmatched connected component in each of images I1 and I2. and Extract its tree structure features and curve features, and combine the branch points of the two blood vessels in pairs; Step 2, calculate all branch point combinations The similarity of tree structure features is calculated using the Hausdorff distance. Step 3: Set a tree structure similarity threshold ε. Select the two branch points with the highest similarity that are greater than the similarity threshold ε as the starting branch points and proceed to Step 4. If there are no branch point combinations with similarity greater than ε, return to Step 1 and select unmatched blood vessel combinations to calculate the tree structure similarity. If there are no unmatched blood vessel combinations, it means that images i1 and I2 have no overlapping areas and cannot be registered. Step 4: Set the curve feature similarity threshold θ. For branch nodes... and Select an adjacent branch node to obtain the curve features of the intermediate blood vessel segment curve. and calculate tangential and The average Euclidean distance ω in the tangential direction. If ω ≤ θ, then the two curve segments can be registered and included in the image registration process; If ω>θ, tangential and The average Euclidean distance ω' is calculated in the inverse tangential direction. If ω'≤θ, the two curve segments can be registered and proceed to image registration; if ω′>θ, the two curve segments cannot be registered and proceed to step 5. Step 5, select Another feature of the curve segment and Calculate curve feature similarity and repeat step 4; if If it does not exist, then select. and Calculate similarity and repeat step 4; if If neither exists, then the two curve segments cannot be registered.

8. The method for image registration in fundus angiography according to claim 1, characterized in that, It also includes determining the unregistered cases, traversing all tree structure feature similarity combinations less than the threshold ε, and repeating step 4; if there is no uncalculated branch node combination that satisfies the condition of less than the threshold ε, the two connected regions to which they belong are marked as matched; Traverse unmatched connected component combinations and repeat step 1; if there are no unmatched connected component combinations, images I1 and I2 do not have overlapping regions and cannot be registered.

9. A system for image registration in fundus angiography, characterized in that, The system is constructed using the method for retinal angiography image registration as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of a method for registering fundus angiography images as described in any one of claims 1 to 8.

Images