OCTA projection image-based arteriovenous centricity calculation method

Abstract

The application discloses an arteriovenous centricity calculation method based on an OCTA projection image, and comprises the following steps: carrying out tissue layer segmentation on an input frequency domain optical coherence tomography (SD-OCT) image, positioning an inner limiting membrane (ILM) upper boundary and an outer reticular layer (OPL) lower boundary, generating an OCTA projection image according to the ILM layer upper boundary and the OPL layer lower boundary, segmenting blood vessels of the OCTA projection image by using an image projection network (IPN) model, then distinguishing arteriovenous blood vessels according to the blood vessel thickness and light and dark characteristics on a color fundus image, then adopting a semi-automatic registration method to register the OCTA projection image and the color fundus image, drawing arteriovenous blood vessel images on a blood vessel segmentation result image according to the blood vessel continuity, finally artificially grading the blood vessels, and calculating the blood vessel centricity of each blood vessel according to a blood vessel centricity formula. Experimental results show that the arteriovenous blood vessel centricity calculated by the application has a statistically significant difference.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, particularly the field of quantitative analysis of retinal vessels, and especially relates to a method for calculating arteriovenous centripetal force based on OCTA projection images. Background Technology

[0002] OCTA is a non-invasive fundus imaging technique that can identify retinal blood signals at high resolution and is currently widely used for vascular quantification. Artery-vein (AV) classification can provide a reliable basis for the diagnosis of retinal diseases such as diabetes. Retinal vessel width is highly correlated with the occurrence or risk of many diseases, and the retinal arteriole-venule ratio has become an important indicator of disease. Therefore, quantitative analysis of retinal arteries and veins is helpful for the research of retinal biomarkers.

[0003] Retinal vessel tortuosity has been recognized as one of the early indicators of many vascular and non-vascular diseases. With the development of retinal vessel segmentation technology, tortuosity has been widely used as a quantitative indicator of retinal vessels. Although various methods exist for quantifying tortuosity, limitations remain, such as the lack of a unified large-scale tortuosity dataset and the lack of the most effective tortuosity for a specific disease. All tortuosity indices are unaffected by translation, rotation, and scaling; the position and orientation of the vessels do not affect the tortuosity.

[0004] Furthermore, existing tortuosity indices only focus on local AV characteristics and there are no vascular morphology indices that take into account global information, and current research is concentrated on color fundus images. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the prior art by providing a method for calculating arteriovenous centripetal force based on OCTA projection images. The method includes the following steps:

[0006] Step 1: Acquire color fundus images and 3D SD-OCT and OCTA retinal images;

[0007] Step 2: Segment the SD-OCT retinal image into ILM and OPL layers, and simultaneously obtain the center position of the avascular zone FAZ in the fovea.

[0008] Step 3: Determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer;

[0009] Step 4: Obtain two-dimensional vascular images based on three-dimensional SD-OCT and OCTA retinal images;

[0010] Step 5: Register the OCTA projection image with the color fundus image;

[0011] Step 6: On the color fundus image, distinguish arteries and veins according to the thickness and brightness of blood vessels, and draw the gold standard for arteriovenous segmentation of blood vessel images based on the continuity of blood vessels.

[0012] Step 7: Obtain the arteriovenous skeleton image by skeletalizing the arteriovenous image;

[0013] Step 8: Perform vascular grading on the arteriovenous skeleton image to obtain a vascular grading image;

[0014] Step 9: Obtain the location of each blood vessel based on the vascular grading image, and take the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point.

[0015] Step 10: According to the formula for calculating the tortuosity of blood vessels, each blood vessel is traversed from the starting point to the ending point to calculate the centripetal degree of the blood vessel.

[0016] Step 11: Compare the centripetal force of arteries and veins of the same level.

[0017] Further, step 2 involves segmenting the SD-OCT retinal image into ILM and OPL layers, and simultaneously obtaining the center position of the foveal avascular zone (FAZ). Specifically:

[0018] The retinal tissue layer segmentation algorithm of OCTExplorer software was used to segment the SD-OCT retinal image into ILM and OPL layers, and the center position of the foveal avascular zone (FAZ) was obtained. Specifically, the retinal layer segmentation result and the center position of FAZ were obtained by using the Open→Folder→Executables→Segmentation of 10 Retinal Layers (Macular OCT) function in OCTExplorer software. In the layer segmentation result, the first layer is the ILM layer and the sixth layer is the OPL layer.

[0019] Furthermore, in step 3, the OCTA projection image is determined based on the upper boundary of the ILM layer and the lower boundary of the OPL layer, using the following formula:

[0020]

[0021] In the formula, MaxProjection(x,y) represents the OCTA projected image, x represents the width in the A-scan direction, y represents the sequence number in the B-scan direction, img(x,i,y) is the 3D OCTA image data, ILM(x,y) represents the ILM layer position matrix, and OPL(x,y) represents the OPL layer position matrix.

[0022] Further, step 4 applies the IPN model to segment the three-dimensional SD-OCT and OCTA retinal images to obtain two-dimensional vascular images. Specifically, the model input is the three-dimensional SD-OCT and OCTA retinal images, and the output is the vascular segmentation image.

[0023] Furthermore, step 5 utilizes the semi-automatic registration function of MATLAB software to register the OCTA projection image and the color fundus image, specifically as follows:

[0024] Semi-automatic registration can be achieved by interactively selecting reference points using the cpselect function in MATLAB software. The source image is set to an OCTA projection image, and the target image is set to a color fundus image. Registration can be completed by selecting more than five sets of paired points.

[0025] Furthermore, step 6, which involves distinguishing arteries and veins on a color fundus image based on the thickness and brightness of blood vessels, and drawing the gold standard for arteriovenous segmentation of the blood vessel image based on the continuity of the blood vessels, specifically includes:

[0026] Step 6-1: Differentiate arteries and veins on color fundus images according to the following rules:

[0027] 1) Arteries are brighter in color than veins;

[0028] 2) The artery is thinner than the adjacent vein;

[0029] 3) The central reflex of arteries is relatively wide, while the central reflex of veins is relatively small;

[0030] 4) Arteries and veins alternate near the optic disc before branching, which means that near the optic disc, an artery is usually adjacent to two veins, and vice versa;

[0031] Step 6-2: The arteries and veins extend from the optic disc area to the macula. Based on the continuity of the blood vessels, the artery and vein classification of the OCTA projection image is determined, and the gold standard for arteriovenous segmentation of the blood vessel image is drawn.

[0032] Further, in step 7, the arteriovenous image is skeletonized using the bwmorph function of MATLAB software to obtain the arteriovenous skeleton image, where the parameter operation of the bwmorph function is set to 'skel' and n is set to Inf.

[0033] Further, step 8 involves grading the arteriovenous skeleton image to obtain a vascular grading image. The specific process includes:

[0034] The arteriovenous skeleton images were graded into primary vessels, secondary vessels, and other vessels. The grading is explained as follows: 1) Primary: large vessels that run directly from the optic disc area; 2) Secondary: branch vessels of primary vessels; 3) Other: all vessels other than primary and secondary vessels.

[0035] The level of branch vessels is further determined based on the extension trend and thickness of the branches: if the extension trend of the vessels after the bifurcation point is consistent and the change in vessel thickness is less than the preset threshold, then the vessels before and after the bifurcation point are classified into the same level.

[0036] Further, step 9 specifically involves: applying a depth-first search method to the vascular grading image to obtain the location of each vessel, and using the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point. The specific process includes:

[0037] Step 9-1: Use MATLAB's bwlabel function to label connected components;

[0038] Step 9-2: Apply depth-first search to each connected component to obtain the arteriovenous vascular skeleton sequence, calculate the distance from the two endpoints of the sequence to the FAZ, and take the point farthest from the center of the FAZ as the starting point and the point closest to the center of the FAZ as the ending point.

[0039] Furthermore, in step 10, the centripetal force of each blood vessel is calculated by traversing from the starting point to the ending point according to the formula for calculating blood vessel tortuosity. The formula used is as follows:

[0040]

[0041] d n =||CP n ||2

[0042]

[0043] Where VC is the centripetal force of the blood vessel, C is the center coordinate of the FAZ, and P n Let be the nth point in the vascular skeleton sequence, l be the length of the vascular skeleton, and s be the step size.

[0044] Compared with the prior art, the significant advantages of this invention are:

[0045] (1) This invention presents a method for calculating arteriovenous centripetal direction based on OCTA projection images. Given the current focus on color fundus images, morphological analysis is performed on OCTA projection images. Compared to existing methods that only consider the curvature of retinal vessels (AVs), this proposed method comprehensively considers global information about retinal vessels, resulting in better resolution and stability.

[0046] (2) Experimental results on an annotated database showed that there was a statistically significant difference in VC between arteries and veins, and the difference was greater than that of tortuosity.

[0047] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0048] Figure 1 This is a flowchart of the method for calculating arteriovenous centripetal direction based on OCTA projection images according to the present invention.

[0049] Figure 2 This is a flowchart of the gold standard for arteriovenous segmentation in drawing vascular images.

[0050] Figure 3 This is a flowchart of vascular grading.

[0051] Figure 4 It is an OCTA projection image.

[0052] Figure 5 This is an OCTA vessel segmentation image.

[0053] Figure 6 It is a registration image between the OCTA projection image and the color fundus image.

[0054] Figure 7 This is a schematic diagram of the gold standard for arteriovenous circulation.

[0055] Figure 8 It is an image of arteriovenous skeletonization. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] In one embodiment, combined Figure 1 This invention proposes a method for calculating arteriovenous centripetal force based on OCTA projection images, the method comprising the following steps:

[0058] Step 1: Acquire color fundus images and 3D SD-OCT and OCTA retinal images.

[0059] Step 2: Use the retinal tissue layer segmentation algorithm of OCTExplorer software to segment the SD-OCT retinal image into ILM and OPL layers, and obtain the center position of the foveal avascular zone (FAZ). Specifically, use the Open→Folder→Executables→Segmentation of 10 Retinal Layers (Macular OCT) function in OCTExplorer software to obtain the retinal layer segmentation results and the center position of FAZ. In the layer segmentation results, the first layer is the ILM layer and the sixth layer is the OPL layer.

[0060] Step 3: Determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer; the formula used is:

[0061]

[0062] In the formula, MaxProjection(x,y) represents the OCTA projected image, x represents the width in the A-scan direction, y represents the sequence number in the B-scan direction, img(x,i,y) is the 3D OCTA image data, ILM(x,y) represents the ILM layer position matrix, and OPL(x,y) represents the OPL layer position matrix.

[0063] Step 4: Apply the IPN model to segment the 3D SD-OCT and OCTA retinal images to obtain 2D vascular images. Specifically, the model input is the 3D SD-OCT and OCTA retinal images, and the output is the vascular segmentation image.

[0064] Step 5: Register the OCTA projection image and the color fundus image using the semi-automatic registration function in MATLAB software. Specifically:

[0065] Semi-automatic registration can be achieved by interactively selecting reference points using the cpselect function in MATLAB software. The source image is set to an OCTA projection image, and the target image is set to a color fundus image. Registration can be completed by selecting more than five sets of paired points.

[0066] Step 6: On the color fundus image, arteries and veins are distinguished based on the thickness and brightness of blood vessels, and the gold standard for arteriovenous segmentation of the blood vessel image is drawn based on the continuity of the blood vessels; combined with... Figure 2 The specific process includes:

[0067] Step 6-1: Differentiate arteries and veins on color fundus images according to the following rules:

[0068] 1) Arteries are brighter in color than veins;

[0069] 2) The artery is thinner than the adjacent vein;

[0070] 3) The central reflex of arteries is relatively wide, while the central reflex of veins is relatively small;

[0071] 4) Arteries and veins alternate near the optic disc before branching, which means that near the optic disc, an artery is usually adjacent to two veins, and vice versa;

[0072] Step 6-2: The arteries and veins extend from the optic disc area to the macula. Based on the continuity of the blood vessels, the artery and vein classification of the OCTA projection image is determined, and the gold standard for arteriovenous segmentation of the blood vessel image is drawn.

[0073] Step 7: Use the bwmorph function in MATLAB software to skeletonize the arteriovenous image to obtain the arteriovenous skeleton image, where the parameters operation and n of the bwmorph function are set to 'skel' and Inf.

[0074] Step 8: Perform vascular grading on the arteriovenous skeleton image to obtain a vascular grading image, and combine it with... Figure 3 The specific process includes:

[0075] The arteriovenous skeleton images were graded into primary vessels, secondary vessels, and other vessels. The grading is explained as follows: 1) Primary: large vessels that run directly from the optic disc area; 2) Secondary: branch vessels of primary vessels; 3) Other: all vessels other than primary and secondary vessels.

[0076] The level of branch vessels is further determined based on the extension trend and thickness of the branches: if the extension trend of the vessels after the bifurcation point is consistent and the change in vessel thickness is less than the preset threshold, then the vessels before and after the bifurcation point are classified into the same level.

[0077] Step 9: Apply a depth-first search method to the vascular grading image to obtain the location of each vessel, and use the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point. The specific process includes:

[0078] Step 9-1: Use MATLAB's bwlabel function to label connected components;

[0079] Step 9-2: Apply depth-first search to each connected component to obtain the arteriovenous vascular skeleton sequence, calculate the distance from the two endpoints of the sequence to the FAZ, and take the point farthest from the center of the FAZ as the starting point and the point closest to the center of the FAZ as the ending point.

[0080] Step 10: Based on the formula for calculating blood vessel tortuosity, traverse each blood vessel from the starting point to the ending point to calculate the centripetal degree of the blood vessel. The formula used is as follows:

[0081]

[0082] d n =||CP n ||2

[0083]

[0084] Where VC is the centripetal force of the blood vessel, C is the center coordinate of the FAZ, and P n Let be the nth point in the vascular skeleton sequence, l be the length of the vascular skeleton, and s be the step size.

[0085] Step 11: Compare the centripetal force of arteries and veins of the same level.

[0086] In one embodiment, an arteriovenous centripetal calculation system based on OCTA projection images is provided, the system comprising:

[0087] The image acquisition module is used to acquire color fundus images and three-dimensional SD-OCT and OCTA retinal images;

[0088] The image segmentation module is used to segment SD-OCT retinal images into ILM and OPL layers, and simultaneously obtain the center position of the foveal avascular zone (FAZ).

[0089] The image recognition module is used to determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer;

[0090] The image extraction module is used to acquire two-dimensional vascular images based on three-dimensional SD-OCT and OCTA retinal images;

[0091] The image registration module is used to register OCTA projected images with color fundus images;

[0092] The arteriovenous segmentation module is used to distinguish arteries and veins on color fundus images based on the thickness and brightness of blood vessels, and to draw the gold standard for arteriovenous segmentation of blood vessel images based on the continuity of blood vessels.

[0093] The image skeletonization module is used to skeletonize arteriovenous images to obtain arteriovenous skeleton images;

[0094] The vascular grading module is used to grade the arteriovenous skeleton image to obtain a vascular grading image.

[0095] The vessel location determination module is used to obtain the location of each vessel based on the vascular grading image, and to take the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point.

[0096] The vessel centripetality calculation module is used to calculate the vessel centripetality by traversing each vessel from the starting point to the ending point according to the vessel tortuosity calculation formula.

[0097] The comparison module is used to compare the centripetal force of arteries and veins of the same level.

[0098] Specific limitations regarding the arteriovenous centripetal calculation system based on OCTA projection images can be found in the limitations of the arteriovenous centripetal calculation method based on OCTA projection images mentioned above, and will not be repeated here. Each module in the aforementioned arteriovenous centripetal calculation system based on OCTA projection images can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0099] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0100] Step 1: Acquire color fundus images and 3D SD-OCT and OCTA retinal images;

[0101] Step 2: Segment the SD-OCT retinal image into ILM and OPL layers, and simultaneously obtain the center position of the avascular zone FAZ in the fovea.

[0102] Step 3: Determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer;

[0103] Step 4: Obtain two-dimensional vascular images based on three-dimensional SD-OCT and OCTA retinal images;

[0104] Step 5: Register the OCTA projection image with the color fundus image;

[0105] Step 6: On the color fundus image, distinguish arteries and veins according to the thickness and brightness of blood vessels, and draw the gold standard for arteriovenous segmentation of blood vessel images based on the continuity of blood vessels.

[0106] Step 7: Obtain the arteriovenous skeleton image by skeletalizing the arteriovenous image;

[0107] Step 8: Perform vascular grading on the arteriovenous skeleton image to obtain a vascular grading image;

[0108] Step 9: Obtain the location of each blood vessel based on the vascular grading image, and take the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point.

[0109] Step 10: According to the formula for calculating the tortuosity of blood vessels, each blood vessel is traversed from the starting point to the ending point to calculate the centripetal degree of the blood vessel.

[0110] Step 11: Compare the centripetal force of arteries and veins of the same level.

[0111] For specific limitations on each step, please refer to the limitations on the method for calculating arteriovenous centripetality based on OCTA projection images above, which will not be repeated here.

[0112] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0113] Step 1: Acquire color fundus images and 3D SD-OCT and OCTA retinal images;

[0114] Step 2: Segment the SD-OCT retinal image into ILM and OPL layers, and simultaneously obtain the center position of the avascular zone FAZ in the fovea.

[0115] Step 3: Determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer;

[0116] Step 4: Obtain two-dimensional vascular images based on three-dimensional SD-OCT and OCTA retinal images;

[0117] Step 5: Register the OCTA projection image with the color fundus image;

[0118] Step 6: On the color fundus image, distinguish arteries and veins according to the thickness and brightness of blood vessels, and draw the gold standard for arteriovenous segmentation of blood vessel images based on the continuity of blood vessels.

[0119] Step 7: Obtain the arteriovenous skeleton image by skeletalizing the arteriovenous image;

[0120] Step 8: Perform vascular grading on the arteriovenous skeleton image to obtain a vascular grading image;

[0121] Step 9: Obtain the location of each blood vessel based on the vascular grading image, and take the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point.

[0122] Step 10: According to the formula for calculating the tortuosity of blood vessels, each blood vessel is traversed from the starting point to the ending point to calculate the centripetal degree of the blood vessel.

[0123] Step 11: Compare the centripetal force of arteries and veins of the same level.

[0124] For specific limitations on each step, please refer to the limitations on the method for calculating arteriovenous centripetality based on OCTA projection images mentioned above, which will not be repeated here.

[0125] This invention uses color fundus images, three-dimensional SD-OCT and OCTA retinal images as input, and employs image processing techniques to automatically register images under different modalities.

[0126] As a specific example, in one embodiment, the invention is further verified and illustrated. The method for calculating arteriovenous centripetal direction based on OCTA projection images includes the following: the acquired color fundus image is 2736×1824 pixels, and the three-dimensional SD-OCT and OCTA retinal images are 640×400×400 pixels. First, the three-dimensional SD-OCT retinal image is segmented into ILM and OPL layers using OCTExplorer software. Then, based on the boundaries of the ILM and OPL layers, the maximum value of the grayscale value between the two boundaries is removed, thereby generating... Figure 4 The OCTA projection image is shown. The IPN model is applied to segment the 3D SD-OCT and OCTA retinal images to obtain the vascular image projected by the OCTA, as shown. Figure 5 As shown. The cpselect function provided by MATLAB software is used to semi-automatically register the OCTA projection image with the color fundus image, as follows. Figure 6 As shown. Arteries and veins are distinguished on color fundus images based on the brightness and thickness of blood vessels. Then, based on the continuity of blood vessels, the macula region within the OCTA image is segmented into arteries and veins, obtaining the gold standard for arteriovenous segmentation, as shown. Figure 7 As shown. The `bwlabel` function provided by MATLAB software is used to perform skeletonization processing on arteriovenous images, such as... Figure 8 As shown in the figure. Then, the vessels were graded based on their orientation and the thickness characteristics before and after branching points. Each connected region represents a vessel. The position and sequence of each vessel can be obtained from the connected regions. Then, based on the distances from the two endpoints of the sequence to the center of the FAZ, the farther endpoint is taken as the starting point and the closer endpoint as the ending point. Finally, the centripetal force is calculated by traversing from the starting point to the ending point using the centripetal force formula. For arteries and veins of the same grade, the centripetal force is compared. The specific centripetal force values ​​calculated for different grades of vessels in this experiment are shown in Table 1. The table shows that there are statistically significant differences in the centripetal force between arteries and veins.

[0127] Table 1. Quantitative comparison of concentricity of different grades of blood vessels (mean ± standard deviation).

[0128]

[0129] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A method for calculating arteriovenous centripetal force based on OCTA projection images, characterized in that, The method includes the following steps: Step 1: Acquire color fundus images and 3D SD-OCT and OCTA retinal images; Step 2: Segment the SD-OCT retinal image into ILM and OPL layers, and simultaneously obtain the center position of the avascular zone FAZ in the fovea. Step 3: Determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer; Step 4: Obtain two-dimensional vascular images based on three-dimensional SD-OCT and OCTA retinal images; Step 5: Register the OCTA projection image with the color fundus image; Step 6: On the color fundus image, distinguish arteries and veins according to the thickness and brightness of blood vessels, and draw the gold standard for arteriovenous segmentation of blood vessel images based on the continuity of blood vessels. Step 7: Obtain the arteriovenous skeleton image by skeletalizing the arteriovenous image; Step 8: Perform vascular grading on the arteriovenous skeleton image to obtain a vascular grading image; Step 9: Obtain the location of each blood vessel based on the vascular grading image, and take the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point. Step 10: According to the formula for calculating the tortuosity of blood vessels, each blood vessel is traversed from the starting point to the ending point to calculate the centripetal degree of the blood vessel. Step 11: For arteries and veins of the same level, compare the centripetal force of the vessels; Step 10 describes calculating the centripetal force of each blood vessel by traversing it from the starting point to the ending point using the blood vessel tortuosity calculation formula. The formula used is as follows: in, Where C represents the centripetal force of the blood vessels, and C is the center coordinate of the FAZ. Let be the nth point in the vascular skeleton sequence, l be the length of the vascular skeleton, and s be the step size.

2. The method for calculating arteriovenous centripetal direction based on OCTA projection images according to claim 1, characterized in that, Step 3 describes determining the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer, using the following formula: In the formula, Represents the OCTA projection image. Indicates the width in the A-scan direction. This represents the number of sequences in the B-scan direction. For 3D OCTA image data, Represents the ILM layer location matrix. This represents the OPL layer position matrix.

3. The method for calculating arteriovenous centripetal direction based on OCTA projection images according to claim 1, characterized in that, Step 4 applies the IPN model to segment the 3D SD-OCT and OCTA retinal images to obtain 2D vascular images. Specifically, the model input is the 3D SD-OCT and OCTA retinal images, and the output is the vascular segmentation image.

4. The method for calculating arteriovenous centripetal direction based on OCTA projection images according to claim 1, characterized in that, Step 6 describes distinguishing arteries and veins on a color fundus image based on the thickness and brightness of blood vessels, and drawing the gold standard for arteriovenous segmentation of the vascular image based on the continuity of the blood vessels. The specific process includes: Step 6-1: Differentiate arteries and veins on color fundus images according to the following rules: 1) Arteries are brighter in color than veins; 2) Arteries are thinner than adjacent veins; 3) The central reflex of arteries is relatively wide, while the central reflex of veins is relatively small; 4) Arteries and veins alternate near the optic disc before branching, which means that near the optic disc, an artery is usually adjacent to two veins, and vice versa; Step 6-2: The arteries and veins extend from the optic disc area to the macula. Based on the continuity of the blood vessels, the artery and vein classification of the OCTA projection image is determined, and the gold standard for arteriovenous segmentation of the blood vessel image is drawn.

5. The method for calculating arteriovenous centripetal direction based on OCTA projection images according to claim 1, characterized in that, Step 8 involves grading the arteriovenous skeleton image to obtain a vascular grading image. The specific process includes: The arteriovenous skeleton images were graded into primary vessels, secondary vessels, and other vessels. The grading is explained as follows: 1) Primary: Large vessels that run directly from the optic disc area; 2) Secondary: Branch vessels of primary vessels; 3) Other: All vessels other than primary and secondary vessels. The level of branch vessels is further determined based on the extension trend and thickness of the branches: if the extension trend of the vessels after the bifurcation point is consistent and the change in vessel thickness is less than the preset threshold, then the vessels before and after the bifurcation point are classified into the same level.

6. The method for calculating arteriovenous centripetal direction based on OCTA projection images according to claim 1, characterized in that, Step 9 specifically involves applying a depth-first search method to the vascular grading image to obtain the location of each vessel, and using the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point. The specific process includes: Step 9-1, mark the connected components; Step 9-2: Apply depth-first search to each connected component to obtain the arteriovenous vascular skeleton sequence, calculate the distance from the two endpoints of the sequence to the FAZ, and take the point farthest from the center of the FAZ as the starting point and the point closest to the center of the FAZ as the ending point.

7. A system for calculating arteriovenous centripetal direction based on OCTA projection images, implemented according to the method described in any one of claims 1 to 6, characterized in that, The system includes: The image acquisition module is used to acquire color fundus images and three-dimensional SD-OCT and OCTA retinal images; The image segmentation module is used to segment SD-OCT retinal images into ILM and OPL layers, and simultaneously obtain the center position of the foveal avascular zone (FAZ). The image recognition module is used to determine the OCTA projection image based on the upper boundary of the ILM layer and the lower boundary of the OPL layer; The image extraction module is used to acquire two-dimensional vascular images based on three-dimensional SD-OCT and OCTA retinal images; The image registration module is used to register OCTA projected images with color fundus images; The arteriovenous segmentation module is used to distinguish arteries and veins on color fundus images based on the thickness and brightness of blood vessels, and to draw the gold standard for arteriovenous segmentation of blood vessel images based on the continuity of blood vessels. The image skeletonization module is used to skeletonize arteriovenous images to obtain arteriovenous skeleton images; The vascular grading module is used to grade the arteriovenous skeleton image to obtain a vascular grading image. The vessel location determination module is used to obtain the location of each vessel based on the vascular grading image, and takes the point farthest from the FAZ center as the starting point and the point closest to the FAZ center as the ending point. The vessel centripetality calculation module is used to calculate the vessel centripetality by traversing each vessel from the starting point to the ending point according to the vessel tortuosity calculation formula. The comparison module is used to compare the centripetal force of arteries and veins of the same level.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images