Methods and systems for cloud-based automated quantitative assessment of retinal microvasculature using optical coherence tomographic angiography images

Abstract

Systems and methods are provided for the automated quantitative assessment of the retinal microvascular system and the quantification of microvascular properties for the diagnosis and treatment of retinal and extraretinal diseases.

Description

Technical Field

[0001] This invention relates to a method and system for cloud-based automated quantitative assessment of the retinal microvascular system using optical coherence tomography angiography images. Background Technology

[0002] Retinal diseases in humans can manifest as various physiological or pathological conditions, including diabetes, which can lead to diabetic retinopathy (DR), retinal vein occlusion (RVO), age-related macular degeneration (AMD), and more. Recently, increased life expectancy and a sedentary, stressful lifestyle have led to a rapid increase in the number of people suffering from these vision-threatening conditions. There is an urgent need for large-scale improvements in the ways that screen, diagnose, and treat these diseases.

[0003] Retinal imaging is the only method for directly examining blood vessels throughout the body and their pathological changes. Retinal imaging not only reflects retinal vascular diseases but also indicates the risk of systemic diseases, including cardiovascular disease and hypertension. Most retinal diseases cause abnormalities in the eye's microvascular system, including diabetic retinopathy (DR), retinopathy of prematurity (RVO), and atherosclerosis (AMD). Furthermore, retinal characteristics provide important information about the risk of developing vascular diseases. Systemic diseases such as cardiovascular disease, hypertension, and atherosclerosis themselves manifest as changes in retinal microvessels. Therefore, the measurement of the retinal microvascular system plays a crucial role in the management of vascular diseases.

[0004] Optical coherence tomography angiography (OCTA) has recently been introduced for imaging the microvascular network in the human eye. Compared to traditional dye-based angiography, OCTA offers a safer, faster, and more cost-effective non-invasive method for diagnosing and monitoring retinal vascular abnormalities. OCTA has the ability to visualize superficial, deep, and avascular retinal images, as well as the microvascular system of the choroidal capillary layer and choroid.

[0005] Typically, patients regularly visit an ophthalmologist or other eye care professional for OCTA measurements. The professional can examine the images and rationally compare them to other retinal images they have observed in the past. While this experience-based review is effective in detecting overall abnormalities and large changes, it can be ineffective in observing minute changes in a patient's retina. The assessment is also subjective, and its accuracy heavily relies on the clinician's prior experience, making diagnosis susceptible to human error. Inexperienced ophthalmologists often consult external specialists for interpretation of laboratory results and medical images to improve their diagnostic accuracy. On the other hand, due to the limited descriptive capabilities of the built-in software of OCTA machines, most patients do not have access to renowned ophthalmologists and cannot obtain sufficient information directly from their OCTA reports. When patients become concerned about their eye health as a result, they may have to seek further advice from external specialists, which can be time-consuming and costly. Therefore, current retinal analysis systems are subjective, time-consuming, error-prone, and require significant human input from healthcare professionals. Furthermore, comprehensive biomarkers based on quantitative OCTA images for reliable monitoring of vascular disease are lacking.

[0006] Advances in computer-aided image processing and analysis techniques are crucial for making imaging-based disease diagnosis scalable, cost-effective, and reproducible. These advancements will directly lead to effective patient triage, resulting in earlier diagnosis, timely treatment, and improved quality of life. Summary of the Invention

[0007] Systems and methods for automated quantitative assessment of the retinal microvascular system and generation of retinal digital maps are provided for the diagnosis and treatment of retinal and extraretinal diseases. The systems and methods of the present invention provide computational retinal microvascular biomarkers (CRMBs) that can be used to quantify the state of retinal microvascular disease. Advantageously, the systems and methods of the present invention can be integrated into cloud computing platforms for cloud-based analytics that provide broad accessibility to patients and healthcare providers and enable the use of machine learning to identify novel CRMBs. Attached Figure Description

[0008] Figure 1 A schematic diagram of an example architecture for a cloud platform used in a computerized automated optical coherence tomography angiography (OCTA) image analysis system is shown.

[0009] Figure 2 A schematic diagram of a process for extracting computational retinal microvascular biomarkers (CRMB) from OCTA images according to an embodiment of the present technology is shown.

[0010] Figure 3A Representative raw OCTA images of both deep and superficial capillary plexuses are shown. Figure 3B Representative OCTA scan images of both deep and superficial capillary plexuses are shown.

[0011] Figure 4 An illustration shows an algorithm for extracting and generating a digital vascular system map (DVM) from a calibrated OCTA image.

[0012] Figure 5A An illustration shows the process of applying the Early Treatment for Diabetic Retinopathy (ETDRS) grid to the DVM. Figure 5B The segment is shown as defined by the ETDRS grid in the left and right eyes.

[0013] Figure 6 Image A shows a digital vascular map (DVM) generated using OCTA images. Figure 6 Image B shows a thresholded image of the DVM image. Figure 6 Image C shows a thresholded subcentral concave image of the DVM image. Figure 6 Image D shows a ridge-filtered image of a DVM image. Figure 6 Image E shows a thresholded ridge filter image of a DVM image. Figure 6 Image F shows the object with small blood vessels removed. Figure 6 Image E. Figure 6 Image G shows a depiction of retinal vessels. Figure 6 Image F. Figure 6 Image H shows skeletalized blood vessels. Figure 6 Image F. Figure 6 Image I shows the main vascular segments with separation. Figure 6 Image H.

[0014] Figure 7 A diagram illustrating the process for calculating the vascular dispersion index of DVM is shown.

[0015] Figure 8 A diagram illustrating an algorithm for detecting endpoints and branch points in a skeletonized DVM is shown.

[0016] Figure 9A An image of the DVM with the calculated vascular dispersion index marked on each vessel is shown. Figure 9B An image of a DVM in which the calculated vessel diameter index is marked for each of the depicted vessels is shown. Figure 9C An image of the DVM, in which the calculated vascular tortuosity index is marked for each isolated vascular segment, is shown.

[0017] Figure 10 Image A shows a digital vascular map (DVM) of the retina. Figure 10 Image B shows the masked and denoised DVM. Figure 10 Image C shows thresholded and closed DVM. Figure 10 Image D shows the bridged DVM. Figure 10 Image E shows a DVM with polygonal endpoints of a central concave avascular region. Figure 10 Image F shows the DVM in which the avascular region of the fovea is marked.

[0018] Figure 11 A visualization of the distribution and pairwise relationships among different CRMBs in samples from healthy subjects and patients with retinal vein occlusion is shown. Detailed Implementation

[0019] Systems and methods are provided for the automated assessment of retinal microvascular architecture, the quantification of changes in the retinal microvascular system associated with retinal or extraretinal diseases, disease treatment, and the quantitative assessment of treatment success based on the reversal of disease-related changes in the retinal microvascular system. Using the systems and methods of the present invention, a method is also provided for establishing comprehensive computational retinal microvascular system biomarkers (CRMBs) through a knowledge-driven, computerized, automated analysis system based on fractal analysis using OCTA images. The systems and methods of the present invention employ cloud computing networks with distributed input devices and cloud-based analysis platforms for the automated analysis of OCTA images, the quantification of ocular diseases based on CRMBs established using the systems and methods of the present invention, and the treatment of corresponding ocular diseases. Cloud computing refers to the ability to access computing resources via the Internet for the purposes of data storage, aggregation, synthesis, and retrieval, and to utilize computational algorithms and software packages on data.

[0020] In a preferred embodiment, the present invention provides a computerized automated OCTA image analysis system for developing and quantifying computational retinal microvascular system biomarkers (CRMBs) based on an OCTA image database and for comprehensively and readily availablely assessing and quantifying retinal characteristics. Using the system of the present invention, clinicians and patients can obtain and evaluate data reports within minutes, enabling comprehensive diagnosis and treatment in clinical settings or outpatient visits. Rapid and comprehensive analysis also allows for seamless re-imaging when conclusive results cannot be obtained using the initial OCTA report.

[0021] Advantageously, the system and the quantitative retinal microvascular system biomarkers generated using the system and method of the present invention can also be used for effective downstream research.

[0022] In some embodiments, the cloud database storage of the system of the present invention originates from images of thousands of patients, which are either uploaded by the patients themselves, collected as part of clinical studies for disease or drug discovery and development, or generated by routine clinical workflows (where a group of patients is analyzed in batch mode).

[0023] In a preferred embodiment, the data obtained using the system of the present invention is provided to an e-cloud deep learning system configured to evaluate and identify early biomarkers of retinal diseases (including but not limited to retinal vascular diseases) using machine learning techniques.

[0024] By utilizing computational microvascular biomarkers generated using the methods and systems of this invention, treatment plans can be designed that are associated with the presence of retinal microvascular biomarkers and / or their potential medical conditions, and patients can be treated and treatment efficacy can be evaluated using the CRMB of this invention.

[0025] In some embodiments, the quantified CRMB is used to quantify the severity of extraretinal disease symptoms. In some embodiments, the quantified CRMB is used to quantify the severity of retinal disease symptoms. In some embodiments, the quantified CRMB is used to quantify both extraretinal disease symptoms and retinal disease symptoms.

[0026] In some embodiments, the system of the present invention is integrated into the interface of an OCTA machine, and users can view the results of real-time quantification of disease biomarkers after capturing OCTA images in an ophthalmology clinic.

[0027] In some embodiments, the system of the present invention provides machine learning techniques to develop quantitative parameters for various extraretinal and retinal disease symptoms. In specific embodiments, the machine learning techniques include, but are not limited to, deep learning and gradient boosting decision trees. Advantageously, the provided system and method of the present invention are used to validate and evaluate novel objective microvascular system biomarkers for eye diseases, thereby facilitating the diagnosis, monitoring, and treatment of the corresponding eye diseases.

[0028] In preferred embodiments, these systems and methods quantify retinal microvascular parameters, including but not limited to fractal dimension index, foveal avascular area, and blood flow index. In another preferred embodiment, these systems and methods quantify retinal microvascular parameters including retinal vascular geometry, including but not limited to vascular dispersion, vascular diameter, and vascular tortuosity.

[0029] In another embodiment, machine learning techniques, including but not limited to e-cloud deep learning, are provided to identify and quantify new CRMBs.

[0030] In certain embodiments, the system of the present invention acquires OCTA image data. For example, in some embodiments of cloud-based operation of the system of the present invention, OCTA images of the patient are generated using, for example, OPTO VUE, Cirrus 5000, and / or SPECTRALIS OCT2 OCTA machines at a medical center, including, but not limited to, two consecutive 3×3mm, frontal OCTA images of superficial and deep capillary plexuses. In other embodiments, scanned images are generated from the output of the OCTA machine. Advantageously, data input can be performed by both the ophthalmologist user and the patient, both of whom can access the cloud platform and scan OCTA reports from hospital visits to upload to the system of the present invention. OCTA images are uploaded to the cloud via an Internet application programming interface (API). In some embodiments, OCTA images are uploaded as part of research and development studies and / or clinical studies for disease and / or drug development. In some embodiments, OCTA images are uploaded in batch mode during routine clinical workflows or by the patient themselves.

[0031] In some embodiments, the uploaded OCTA images are anonymized by removing certain types of metadata and stored in a cloud with server-side encryption. Advantageously, cloud infrastructures using the system of this invention can accommodate the management and analysis of large-scale imaging data.

[0032] In some embodiments, the system of the present invention is configured to calibrate the original OCTA image so that the image is input in a predetermined orientation. For example, when scanning an original OCTA image from the printout of an OCTA machine and the image is a portrait view, the system of the present invention rotates the image by 90 degrees.

[0033] In some embodiments, the scanned image is converted from RGB(R,G,B) to Lab color space(L,a,B) and is equally divided into left and right halves.

[0034] In a particular embodiment, pixels in each half of the image are analyzed, and more are included in the left half of the OCTA image when rotated clockwise. An image of pixels.

[0035] In other embodiments, counterclockwise rotation includes less [data / texture] in the left half of the OCTA image. The image consists of pixels. After a corresponding rotation, the scanned image is presented in landscape view format and with a predetermined orientation.

[0036] In a preferred embodiment, the image after being rotated to the predetermined orientation is further rotated by a small angle so that the reference lines are strictly horizontal and vertical, respectively.

[0037] In a preferred embodiment, the image rotated to the predetermined orientation is further rotated by the required degree (ranging from -90 degrees to 90 degrees) so that the reference lines are strictly horizontal and vertical, respectively.

[0038] In some embodiments, the scanned image is converted to grayscale, Gaussian blurred to reduce noise, thresholded, and edge-processed using a Canny edge detector.

[0039] In a preferred embodiment, the rotation angle is determined by detecting the orientation of reference lines in the edge-processed image using Hough line transform.

[0040] In another embodiment, following the calibration process described above, a digital vascular map (DVM) is generated from the OCTA image. For this purpose, the OCTA image is first converted to grayscale and thresholded using Otsu's binarization. All vertical and horizontal line segments in the resulting image are identified using Hough line transform, and the detected line segments are combined into a new image. Contours in the new image are identified and classified; these contours correspond to rectangles in the original OCTA image.

[0041] In a preferred embodiment, a top-left rectangle with a width / height greater than 0.27 times the width of the original OCTA image is extracted. In embodiments where the top-left rectangle does not have equal length and width, the rectangle is cropped into a square, wherein the cropping decision is determined by performing Harris angle detection on a padded, Gaussian blurred, and thresholded version of the image. The resulting square is a DVM extracted from the calibrated OCTA image.

[0042] In some embodiments, the systems and methods of the present invention apply an Early Treatment for Diabetic Retinopathy (ETDRS) mesh to the generated DVM as described above. In specific embodiments, these systems and methods convert the DVM to the Lab color space (L, a, b) and replace the pixels with white pixels. The pixels are selected to retain only the color components while removing most of the vascular signals. The resulting image is then converted to grayscale and skeletonized, and a reference circle in the ETDRS mesh is identified using a Hough circle transform. In a particular embodiment, the inner reference circle C 内 Having a center (x) 内 ,y 内 ) and r 内 The radius, and the outer reference circle C 外 Having a center (x) 外 ,y 外 )=(x 内 ,y 内 ) and r 外 The radius. In a preferred embodiment, C 内 The area within roughly corresponds to the central concave region, while the secondary central concave region refers to C.内 and C 外 The annular region between them.

[0043] In specific embodiments, the parafovea is then divided into superior, inferior, nasal, and temporal segments. In some embodiments, these systems and methods use recording parameters of a reference circle in the ETDRS mesh to calculate the segment boundaries for the right eye (OD) or left eye (OS), respectively. In some embodiments, with the coordinates of the upper left corner of the DVM being (0,0), the endpoints of the line segments dividing the superior and nasal segments of the right eye are: and

[0044] In some embodiments, the systems and methods of the present invention calculate the fractal dimension index (FD) to quantify the complexity of retinal microvascular structures in both superficial capillary plexuses (SCP) and deep capillary plexuses (DCP). The generated DVM is first binarized by adaptive thresholding, and the FD is calculated by applying a box-counting algorithm to the resulting graph (Lemmens et al., 2020).

[0045] In some embodiments, these systems and methods automatically depict retinal vessels in the SCP. In a preferred embodiment, these systems and methods binarize the original DVM using adaptive global thresholding to produce a thresholded DVM image. In a preferred embodiment, only the foveal and subfoveal regions are preserved according to the ETDRS grid applied as described above. In some embodiments, the systems and methods of the present invention use a Sato filter to detect continuous ridges corresponding to vessels in the resulting DVM. Subsequently, the ridge-filtered DVM image is binarized using adaptive thresholding based on the median of the subfoveal pixel intensity. In another embodiment, these systems and methods remove small objects from the binarized DVM image, which correspond to insignificant isolated vessel branches or noise. In yet another embodiment, these systems and methods identify contours in the resulting DVM, where each contour corresponds to a depicted retinal vessel.

[0046] In some embodiments, the capillary perfusion density index (PDC) is calculated for both DCP and SCP based on the generated DVM and its accompanying ETDRS mesh. First, the image is converted to grayscale, where pixel intensity values ​​range from 0 (black) to 255 (white). Next, the PDC is calculated as the average intensity of all undepicted vascular pixels in the considered segment of the image. Because brighter pixels (with larger intensity values) typically correspond to vascular structures in OCTA images, an image with a larger calculated PDC has denser capillaries.

[0047] In some embodiments, the large vessel perfusion density index (PDL) is calculated based on the DVM generated in the SCP and its ETDRS mesh. For a segment of the graph under consideration, the PDL is calculated as the proportion of pixels belonging to the depicted vessels or larger vessels. A larger calculated PDL indicates a denser network of large vessels.

[0048] In a preferred embodiment, these systems and methods calculate the vascular dispersion index (VDisp). Because in a healthy eye, parafoveal vessels are more centripetal towards the center of the fovea, VDisp refers to the degree of centripetal tendency of parafoveal vessels. The larger the VDisp, the less centripetal the vessels have on average.

[0049] In some embodiments, these systems and methods, after removing small objects, fill and crop a thresholded, ridge-filtered DVM image to center the subfoveal region. In another embodiment, for each depicted retinal vessel V i (i = 1, ..., N, where N is the total number of blood vessels depicted), these systems and methods calculate the relationship between the region V and the region V. i The same second-moment ellipse E is covered. In some embodiments, if V i If it is completely centripetal, then it is equally oriented to connect E. i The line segment whose center of mass and the center of the central concave region.

[0050] According to the method of the present invention, VDisp is defined as E i spindle and connection E i The center of mass and C 内 The average of all angles between the line segments at the center, corresponding to the central fovea identified when the ETDRS grid is applied to the extracted and calibrated DVM image. In some embodiments, the systems and methods of the present invention calculate multiple vascular dispersion indices for individual vessels in the DVM image. In some embodiments, these systems and methods calculate the total VDisp of the DVM image, which is the average of all individual vascular dispersion indices.

[0051] In certain embodiments, these systems and methods separate primary and secondary retinal vessel segments. In certain embodiments, these systems and methods utilize the depicted retinal vessels to skeletonize a vascular system map. In other embodiments, these systems and methods detect endpoints and branch points in a skeletonized DVM image using an algorithm comprising the steps of: traversing each positive pixel in the DVM image and counting the number (n) of positive pixels in eight neighboring segments of the pixel; calculating the number of pixels containing a single positive pixel in its eight neighboring segments and applying endpoint labels to the pixels; calculating the number of pixels containing three or more positive pixels in its eight neighboring segments and applying branch point labels to the pixels; calculating the pixels including branch point labels and which pixels with branch point labels are not adjacent, and applying true branch point labels to the pixels.

[0052] In some embodiments, these systems and methods mask marked branch points from a DVM image and compute all contours in the masked image, where each contour corresponds to a vessel segment. In other embodiments, these systems and methods represent the total number of branch points as N. bP Then the total number of blood vessel segments N seg Equals 1 + 2N bP In another embodiment, these systems and methods represent the length of a geodesic segment as L. seg And the geodesic length of its main blood vessel is denoted as L. ves .

[0053] if If the segment is not marked as a minor vessel segment, it is marked as a major vessel segment.

[0054] In some embodiments, the system and method of the present invention calculate the vessel diameter index (VDiam). VDiam is defined as the average ratio between the pixel area and the geodesic length of all depicted retinal vessels. In the image, S represents the side length of the DVM, N represents the number of depicted vessels, and VDiam represents the number of depicted vessels. i The formula for VDiam, representing the i-th depicted blood vessel, is:

[0055]

[0056] In some embodiments, the total VDiam of the DVM image is calculated, and it is the average of the individual VDiams of all markers in the image divided by S.

[0057] In some embodiments, the system and method of the present invention calculate the vascular tortuosity index (VT). After calculating the endpoints and branch points, VT is quantified as the average ratio between the geodesic length and Euclidean length of the separated vascular segments in the DVM image. M1 represents the number of primary vascular segments, M2 represents the number of secondary vascular segments, and S... i 主要 Represented as the i-th major vascular segment and S i 次要 Let VT be the i-th minor vascular segment. The formula for VT is:

[0058]

[0059] Where a = 0.8 is the weighting factor. In this VT calculation, minor vessel segments have lower importance because they typically correspond to secondary vessels. If the vessels in the DVM image are more tortuous and distorted on average, the image will have a larger VT. In some embodiments, individual VTs are displayed in the DVM image. In some embodiments, the total VT of the DVM image is displayed, and it is a weighted average of all individual VTs marked in the DVM.

[0060] In some embodiments, these systems and methods calibrate the FAZ in a DVM image, which is defined as a non-capillary region within the innermost ring of the subcentral foveal capillary plexus. In a particular embodiment, reference circles and lines representing the ETDRS grid in the DVM are masked using median-filtered versions of their corresponding regions in the image. In some embodiments, the systems and methods apply median filtering to suppress noise. In a preferred embodiment, the masked and denoised image is adaptively thresholded and closed (dilated, subsequently eroded), and small objects are removed. In another embodiment, these systems and methods apply bridging, i.e., if a 0-value pixel has two non-contiguous non-zero neighboring pixels, the 0-value pixel is set to 1 and the resulting image is thinned. In yet another embodiment, these systems and methods remove small isolated objects from the image to improve the robustness of the FAZ identification method.

[0061] In some embodiments, these systems and methods calculate the largest polygon in the foveal avascular region, i.e., a polygon that does not cover any of the positive vascular signals in the resulting image. In some embodiments, (0,0) is indicated to correspond to the upper left corner of the DVM, and the center of the foveal region is (x... 内 ,y 内 And the side length of the DVM is s. In some embodiments, for Θ = 0, 3, 6, 9, ..., 357 (i.e., a total of 120 Θs), consider starting from (x 内 ,y 内 ) to (x 内 +s·sinθ,y 内The FAZ is defined as all pixels within the image on a directed line segment (n = s·cosθ). In some embodiments, the first pixel with a positive signal (i.e., part of a blood vessel) is recorded as one of the endpoints of the FAZ. If all pixels within the image on the directed line segment have a value of zero, this particular θ is skipped. If fewer than three searches return a positive signal, an error is reported. In a particular embodiment, the polygon is defined by connecting all n = 120 endpoints in the order they are found, and is an approximate FAZ. In other embodiments, the polygon is defined to calibrate the FAZ by connecting fewer than 120 endpoints when fewer than all endpoints are used in some searches where no positive pixels are found.

[0062] In some embodiments, the systems and methods of the present invention calculate the FAZ area index (FAI). In some embodiments, these systems and methods label the vertices of the FAZ polygon as (a1, b1), (a2, b2), ..., (a3, b4) in a clockwise order. n ,b n In specific embodiments, these systems and methods use the shoelace formula to calculate the area of ​​a polygon:

[0063]

[0064] In another embodiment, these systems and methods define the FAI as the area of ​​the FAZ polygon divided by the pixel area of ​​the DVM. Advantageously, if the DVM is always focused on a 3×3 region, the FAI is comparable between images of different sizes.

[0065] In some embodiments, the systems and methods of the present invention calculate the FAZ perimeter index (FPI). In some embodiments, these systems and methods calculate the perimeter of the FAZ polygon using the following formula:

[0066]

[0067] In some embodiments, FPI is defined as the perimeter of the FAZ polygon divided by the side length of the DVM (in pixels). Advantageously, if the DVM is always focused on a 3×3 region, the FPI is comparable between images of different sizes.

[0068] In some embodiments, the system and method of the present invention calculate the FAZ non-circularity index (FACI). In a particular embodiment, the FACI quantifies the irregularity of the shape of the calibrated FAZ and is calculated using the following formula:

[0069]

[0070] Advantageously, when FACI equals one, FAZ is a perfect circle. In some embodiments, when the FACI value is greater than one, FAZ has a more irregular shape.

[0071] In a preferred embodiment, the system of the present invention employs a learning algorithm and training data generated using a cloud-based OCTA analysis system to derive model parameters, thereby developing a CRMB for the quantified specific disease symptoms.

[0072] In some embodiments, the present invention provides a method for automated quantitative assessment of retinal microvascular systems and generation of retinal digital vascular maps, comprising:

[0073] Recording optical coherence tomography (OCTA) image data;

[0074] Preprocessing OCTA image data; and

[0075] Generate a digital vascular map (DVM) of the retina.

[0076] In some embodiments, the preprocessing steps of the method further include applying an early treatment study grid for diabetic retinopathy to a DVM; calculating the fractal dimension index; and calibrating retinal vessels.

[0077] In another embodiment, the preprocessing steps of the method further include calculating the vascular dispersion index; separating the main vascular segments and secondary vascular segments; calculating the vascular diameter index; calculating the vascular tortuosity index; calibrating the avascular region of the central fossa; calculating the area index of the avascular region of the central fossa; and calculating the non-circularity index of the avascular region of the central fossa.

[0078] In a preferred embodiment, the present invention provides a method for extracting computational retinal microvascular biomarkers (CRMBs) from a DVM generated using the method and system of the present invention.

[0079] In some embodiments, CRMB is a fractal dimension index, a vascular dispersion index, a capillary perfusion density index, a large vessel perfusion density index, a vessel diameter index, a vascular tortuosity index, a central foveal avascular area index, a central foveal avascular perimeter index, and / or a central foveal avascular non-circularity index. In a preferred embodiment, CRMB is a vascular dispersion index or a vascular tortuosity index.

[0080] In a specific embodiment, the present invention provides an automated quantification method for computational retinal microvascular biomarkers (CRMBs) of superficial retinal capillary plexuses from a subject suspected of having retinal disease, and a method for treating the subject if the CRMB quantification indicates that the subject has retinal disease, the method comprising:

[0081] Recording optical coherence tomography (OCTA) image data;

[0082] Preprocess OCTA image data;

[0083] Generate digital vascular maps (DVMs) of the superficial capillary plexus in the retina;

[0084] Provides retinal DVM for normal subjects;

[0085] Quantify at least one CRMB in the DVM of the subject and the DVM of the normal subject;

[0086] And if the amount of at least one CRMB in a subject exceeds the amount of said CRMB in a normal subject, then the subject's retinal disease is treated.

[0087] In another embodiment, the preprocessing step of the method includes applying an early treatment study grid for diabetic retinopathy to the DVM, and quantifying at least one CRMB including quantifying the fractal dimension index (FD); quantifying the capillary perfusion density index; quantifying the macrovascular perfusion density index; quantifying the vascular dispersion index; quantifying the vascular diameter index; quantifying the vascular tortuosity index; quantifying the area index of the foveal avascular region; quantifying the perimeter index of the foveal avascular region; and / or quantifying the non-circularity index of the foveal avascular region.

[0088] In some embodiments, the present invention provides an automated quantification of computational retinal microvascular biomarkers (CRMBs) of the deep retinal capillary plexus from a subject suspected of having retinal disease, and a method for treating the subject if the CRMB quantification indicates that the subject has retinal disease, the method comprising:

[0089] Recording optical coherence tomography (OCTA) image data;

[0090] Preprocess OCTA image data;

[0091] Generate a digital vascular map (DVM) of the deep capillary plexus in the retina;

[0092] Provides retinal DVM of deep capillary plexus in normal subjects;

[0093] Quantify at least one CRMB in the DVM of the subject and the DVM of the normal subject;

[0094] And if the amount of at least one CRMB in a subject exceeds the amount of said CRMB in a normal subject, then the subject's retinal disease is treated.

[0095] In some embodiments, the preprocessing step of the method includes applying an early treatment study grid for diabetic retinopathy to a DVM, and quantifying at least one CRMB including quantifying the fractal dimension index; quantifying the capillary perfusion density index; quantifying the macrovascular perfusion density index; quantifying the vascular dispersion index; quantifying the vascular diameter index; quantifying the vascular tortuosity index; quantifying the area index of the foveal avascular region; quantifying the perimeter index of the foveal avascular region; and / or quantifying the non-circularity index of the foveal avascular region.

[0096] In a specific embodiment, the retinal disease is selected from diabetic retinopathy, retinal vein occlusion, and age-related macular degeneration.

[0097] In some embodiments, treatment includes administration of antibodies against angiogenesis factor, aptamers against angiogenesis factor, anti-angiogenic steroids, antioxidant supplements, laser coagulation therapy, transpupil thermotherapy, and / or eye surgery.

[0098] In a preferred embodiment, the present invention provides an automated quantification of computational retinal microvascular biomarkers (CRMBs) of superficial retinal capillary plexuses from a subject suspected of having an extraretinal disease, and a method for treating the subject if the CRMB quantification indicates that the subject has an extraretinal disease, the method comprising:

[0099] Recording optical coherence tomography (OCTA) image data;

[0100] Preprocess OCTA image data;

[0101] Generate digital vascular maps (DVMs) of the superficial capillary plexus in the retina;

[0102] Provides retinal DVM for normal subjects;

[0103] Quantify at least one CRMB in the DVM of the subject and the DVM of the normal subject;

[0104] And if the amount of at least one CRMB in a subject exceeds the amount of said CRMB in a normal subject, then the subject is treated for extraretinal disease.

[0105] In a specific embodiment, the preprocessing step of the method includes applying an early treatment study grid for diabetic retinopathy to a DVM, and quantifying at least one CRMB including quantifying the fractal dimension index; quantifying the capillary perfusion density index; quantifying the macrovascular perfusion density index; quantifying the vascular dispersion index; quantifying the vascular diameter index; quantifying the vascular tortuosity index; quantifying the area index of the avascular region of the fovea; quantifying the perimeter index of the avascular region of the fovea; and / or quantifying the non-circularity index of the avascular region of the fovea.

[0106] In some embodiments, extraretinal diseases are cardiovascular diseases, systemic arterial hypertension, sickle cell anemia, anemia, preeclampsia, arteritis, aortic aneurysm, diabetes, and / or hypercoagulability.

[0107] In some embodiments, treatment includes administration of antihypertensive drugs, cholesterol-regulating drugs, aspirin, beta-blockers, alpha-blockers, alpha-2 receptor agonists, central agonists, peripheral adrenergic inhibitors, calcium channel blockers, nitroglycerin, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, anticoagulants, blood thinners, antiplatelet agents, digitalis preparations, diuretics, vasodilators, iron supplements, steroids, antidiabetic drugs, endovascular stenting, and / or vascular bypass surgery.

[0108] In some embodiments, treatment includes administration of metformin, sulfonylurea, meglitinide, thiazolidinedione, DPP-4 inhibitor, GLP-1 receptor agonist, SGLT2 inhibitor, insulin, and / or iron supplementation.

[0109] In a preferred embodiment, the present invention provides a system for automated quantitative assessment of retinal microvascular systems and generation of retinal digital vascular maps, comprising:

[0110] A memory configured to store one or more executable instructions;

[0111] A processor configured to execute one or more executable instructions to perform the following steps:

[0112] Recording optical coherence tomography (OCTA) image data;

[0113] Preprocessing OCTA image data; and

[0114] Generate a digital vascular map (DVM) of the retina.

[0115] In some embodiments, preprocessing includes: applying an early treatment study grid for diabetic retinopathy to a DVM; calculating the fractal dimension index; and calibrating retinal vessels.

[0116] In some embodiments, the preprocessing further includes: calculating the vascular dispersion index; separating the primary and secondary vascular segments; calculating the vascular diameter index; calculating the vascular tortuosity index; calibrating the avascular region of the fovea; calculating the area index of the avascular region of the fovea; and calculating the non-circularity index of the avascular region of the fovea.

[0117] In a preferred embodiment, the present invention provides a system for extracting computational retinal microvascular biomarkers (CRMBs) from a DVM, comprising:

[0118] A memory configured to store one or more executable instructions;

[0119] A processor configured to execute one or more executable instructions to extract computational retinal microvascular biomarkers (CRMBs) from the aforementioned DVM.

[0120] In some embodiments, CRMB is selected from blood fractal dimension index, capillary perfusion density index, large vessel perfusion density index, vascular dispersion index, vessel diameter index, vessel tortuosity index, foveal avascular area index, foveal avascular perimeter index, and foveal avascular non-circularity index.

[0121] In a preferred embodiment, the present invention provides an automated quantification system for computational retinal microvascular biomarkers (CRMBs) of superficial retinal capillary plexuses from a subject suspected of having retinal disease, and a system for treating the subject if the CRMB quantification indicates that the subject has retinal disease, the system comprising:

[0122] A memory configured to store one or more executable instructions;

[0123] A processor configured to execute one or more executable instructions to perform the following steps:

[0124] Recording optical coherence tomography (OCTA) image data;

[0125] Preprocess OCTA image data;

[0126] Generate digital vascular maps (DVMs) of the superficial capillary plexus in the retina;

[0127] Provides retinal DVM for normal subjects;

[0128] Quantify at least one CRMB in the DVM of the subject and the DVM of the normal subject; and

[0129] The subject is treated for retinal disease if the amount of at least one CRMB in the subject exceeds the amount of said CRMB in a normal subject.

[0130] In some embodiments, the preprocessing includes: applying an early treatment study grid for diabetic retinopathy to the DVM, and wherein quantifying at least one CRMB includes quantifying the fractal dimension index; quantifying the capillary perfusion density index; quantifying the macrovascular perfusion density index; quantifying the vascular dispersion index; quantifying the vascular diameter index; quantifying the vascular tortuosity index; quantifying the area index of the foveal avascular region; quantifying the perimeter index of the foveal avascular region; and / or quantifying the non-circularity index of the foveal avascular region.

[0131] In some embodiments, the retinal disease is selected from diabetic retinopathy, retinal vein occlusion, and age-related macular degeneration, and treatment includes administration of antibodies against angiogenesis factors, aptamers against angiogenesis factors, anti-angiogenic steroids, antioxidant supplements, laser coagulation therapy, transpupillary thermotherapy, and / or ocular surgery.

[0132] In a preferred embodiment, the present invention provides an automated quantification system for computational retinal microvascular biomarkers (CRMBs) of the deep retinal capillary plexus from a subject suspected of having retinal disease, and a system for treating the subject if the CRMB quantification indicates that the subject has retinal disease, the system comprising:

[0133] A memory configured to store one or more executable instructions;

[0134] A processor configured to execute one or more executable instructions to perform the following steps:

[0135] Recording optical coherence tomography (OCTA) image data;

[0136] Preprocess OCTA image data;

[0137] Generate a digital vascular map (DVM) of the deep capillary plexus in the retina;

[0138] Provides retinal DVM of deep capillary plexus in normal subjects;

[0139] Quantify at least one CRMB in the DVM of the subject and the DVM of the normal subject; and

[0140] The subject is treated for retinal disease if the amount of at least one CRMB in the subject exceeds the amount of said CRMB in a normal subject.

[0141] In some embodiments, the preprocessing includes applying an early treatment study grid for diabetic retinopathy to the DVM, and quantifying at least one CRMB includes: quantifying the fractal dimension index; quantifying the capillary perfusion density index; quantifying the macrovascular perfusion density index; quantifying the vascular dispersion index; quantifying the vascular diameter index; quantifying the vascular tortuosity index; quantifying the area index of the foveal avascular region; quantifying the perimeter index of the foveal avascular region; and / or quantifying the non-circularity index of the foveal avascular region.

[0142] In some embodiments, the retinal disease is selected from diabetic retinopathy, retinal vein occlusion, and age-related macular degeneration, and treatment includes administration of antibodies against angiogenesis factors, aptamers against angiogenesis factors, anti-angiogenic steroids, antioxidant supplements, laser coagulation therapy, transpupillary thermotherapy, and / or ocular surgery.

[0143] In a preferred embodiment, the present invention provides an automated quantification system for computational retinal microvascular biomarkers (CRMBs) of superficial retinal capillary plexuses from a subject suspected of having an extraretinal disease, and a system for treating the subject if the CRMB quantification indicates that the subject has an extraretinal disease, the system comprising:

[0144] A memory configured to store one or more executable instructions;

[0145] A processor configured to execute one or more executable instructions to perform the following steps:

[0146] Recording optical coherence tomography (OCTA) image data;

[0147] Preprocess OCTA image data;

[0148] Generate digital vascular maps (DVMs) of the superficial capillary plexus in the retina;

[0149] Provides retinal DVM for normal subjects;

[0150] Quantify at least one CRMB in the DVM of the subject and the DVM of the normal subject; and

[0151] If the amount of at least one CRMB in a subject exceeds the amount of said CRMB in a normal subject, then the subject is treated for extraretinal disease.

[0152] In some embodiments, the preprocessing includes applying an early treatment study grid for diabetic retinopathy to the DVM, and quantifying at least one CRMB includes: quantifying the fractal dimension index; quantifying the capillary perfusion density index; quantifying the macrovascular perfusion density index; quantifying the vascular dispersion index; quantifying the vascular diameter index; quantifying the vascular tortuosity index; quantifying the area index of the foveal avascular region; quantifying the perimeter index of the foveal avascular region; and / or quantifying the non-circularity index of the foveal avascular region.

[0153] In some embodiments, extraretinal disease is selected from cardiovascular disease, systemic hypertension, sickle cell anemia, anemia, preeclampsia, arteritis, aortic aneurysm, diabetes, and / or hypercoagulability, and treatment includes administration of antihypertensive drugs, cholesterol-regulating drugs, aspirin, beta-blockers, calcium channel blockers, nitroglycerin, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, blood thinners; iron supplements, steroids, antidiabetic drugs, endovascular stenting, and / or vascular bypass surgery.

[0154] In a preferred embodiment, the present invention provides a computer-readable storage medium comprising a computer program that can be executed by a processor to implement the steps of the method of the present invention.

[0155] Materials and methods

[0156] All patents, patent applications, provisional applications and publications mentioned or cited herein are incorporated herein by reference in their entirety (including all figures and tables) without contradicting the express teachings of this specification.

[0157] The following are examples illustrating procedures for practicing this invention. These examples should not be construed as limiting. Unless otherwise stated, all percentages are by weight, and all solvent mixture proportions are by volume.

[0158] Example 1 – A System for Computerized Automated OCTA Image Analysis

[0159] An automated analysis system for OCTA image analysis using a cloud platform was designed. Figure 1 The system was designed for use by ophthalmologists and researchers at medical centers.

[0160] To this end, OCTA images of patients are obtained from various OCTA machines available at medical centers, including OPTOUVUE, Cirrus 5000, and SPECTRALIS OCT2, comprising two consecutive 3×3mm frontal OCTA images of superficial and deep capillary plexuses. In addition to ophthalmologists, patients can also access the system's cloud platform and scan their OCTA reports from hospital visits. OCTA images are uploaded to the cloud via an Internet Application Programming Interface (API), anonymized by removing certain types of metadata, and stored in the cloud with server-side encryption. The cloud infrastructure can accommodate, manage, and analyze large-scale imaging data. Advantageously, the computerized automated OCTA image analysis system of this invention determines CRMB based on the image and returns a report to the user within minutes. Therefore, clinicians can discuss OCTA results with patients during outpatient visits. Furthermore, the system enables seamless re-imaging when conclusive results cannot be obtained using the initial OCTA report. Moreover, users in development can perform effective downstream statistical analysis for quantitative assessment of biomarkers. Advantageously, cloud databases can store images from thousands of patients, including images from data uploaded by patients themselves, images collected as part of clinical studies for disease or drug discovery, or images collected during routine clinical workflows where a group of patients is analyzed in batch mode. Data obtained from these real-time systems is used to develop predictive features through machine learning techniques, including deep learning and gradient-driven decision trees.

[0161] Example 2 – A Method for Computerized Automated OCTA Image Analysis in Fractal Dimensions

[0162] A process for extracting different CRMBs according to embodiments of this technology has been generated. Figure 2 [1]). This process acquires OCTA image data ( Figure 2 Box [2] and Figures 3A to 3B This process calibrates each raw OCTA image so that it is in the correct orientation. Figure 2 [3] This is often necessary when the original image is a scanned version of the printout from an OCTA machine. For example, when the image is in portrait view, it is rotated 90 degrees. Then, the scanned image is converted from RGB (R,G,B) to Lab color space (L,a,b) and divided into left and right halves. If the left half has more If the pixels are not large enough, rotate the image clockwise. Otherwise, rotate the image counterclockwise. After rotation, all scanned images are in landscape view and correctly oriented.

[0163] The image, thus oriented, is further rotated by a small angle to ensure the reference lines are strictly horizontal / vertical. Subsequently, the scanned image is converted to grayscale, Gaussian blurred to reduce noise, thresholded, and edge-processed using a Canny edge detector. Then, the rotation angle is determined by detecting the orientation of the reference lines in the edge-processed image using Hough line transform.

[0164] This process extracts and generates a digital vascular system map (DVM) from calibrated OCTA images. Figure 2 , box [4]). For this purpose, the OCTA image is first converted to grayscale and thresholded using Otsu's binarization. All vertical / horizontal line segments in the obtained image are identified by Hough line transform and the detected line segments are combined into a new image. The contours in the new image correspond to the rectangles in the original OCTA image. The top left rectangle with a width / height greater than 0.27 times the width of the original OCTA image is extracted. If the rectangle does not have equal length and width, it is cropped into a square, where the crop coordinates are determined by performing Harris angle detection on a padded, Gaussian blurred and thresholded version of the image. Thus, this square is the DVM extracted from the calibrated OCTA image. The generation process of the DVM is as follows: Figure 4 As shown in the image.

[0165] This process also applies the Early Treatment for Diabetic Retinopathy (ETDRS) mesh to the DVM generated as described above. Figure 2 [5] To discard most of the vascular signals, the image was converted to the Lab color space (L,a,b) and replaced with white pixels. All pixels are extracted to retain only the color components. The resulting image is then converted to grayscale and skeletonized, and a reference circle in the ETDRS mesh is identified using the Hough circle transform, where the inner reference circle C... 内 Contains center (x) 内 ,y 内 ) and r 内 The radius, and the outer reference circle C 外 Contains center (x) 外 ,y 外 )=(x 内 ,y 内 ) and r 外 The radius of C. 内 The area within roughly corresponds to the central concave region, while the secondary central concave region refers to C. 内 and C 外 The annular region between them. Figure 5A The process of applying the ETDRS mesh to the DVM is shown in the figure.

[0166] Next, the subfovea is divided into superior, inferior, nasal, and temporal segments by calculating the segment boundaries using the parameters recorded on the reference circle in the ETDRS mesh, based on whether the eye under study is the right eye (OD) or the left eye (OS). For example, noting that the coordinates of the upper left corner of the DVM are (0,0), the endpoints of the line segments dividing the superior and nasal segments of the right eye are: and Figure 5B The section is shown as defined by the ETDRS mesh in the left and right eyes.

[0167] Figure 2 The process shown in the diagram determines the fractal dimension index (FD) at box [6] based on the DVM from box [4] to quantify the complexity of the retinal microvascular structures in both the superficial capillary plexus (SCP) and the deep capillary plexus (DCP). First, the generated DVM is binarized by adaptive thresholding, and the FD is calculated by applying a box counting algorithm to the resulting graph.

[0168] This process automatically maps retinal vessels in the superficial capillary plexus. Figure 2 , box [7]). Figure 6 Image A to Figure 6 Image G illustrates the process of depicting retinal vessels. (Original vascular system diagram) Figure 6 Image A) is first binarized through adaptive global thresholding, thus producing Figure 6 Image B of B. Based on the applied ETDRS mesh, only the central concave region and the subcentral concave region are preserved (see image B). Figure 2 Next, Sato’s filter is used to detect continuous ridges or blood vessels in the resulting vascular system map (see [5]). Figure 6 Image C). The ridge-filtered image is then subjected to adaptive thresholding based on the median of the sub-central fovea pixel intensity (see image C). Figure 6 The image D is binarized, and for better robustness, the binarized image (see [image D]) is further binarized. Figure 6 Image E) removes small objects that may be irrelevant, isolated vascular branches or noise. Finally, the resulting image is labeled (see [image description missing]). Figure 6 All contours in image F). Each contour corresponds to a depicted retinal vessel, such as... Figure 6 The image G is shown.

[0169] Example 3 – A method for computerized automated OCTA image analysis of pupillary perfusion density

[0170] Figure 2The process shown in the figure determines the capillary perfusion density index (PDC) based on the generated DVM and its accompanying ETDRS grid at box [8]. First, the image is converted to grayscale, where pixel intensity values ​​range from 0 (black) to 255 (white). Next, the PDC is calculated as the average intensity of all undepicted vascular pixels in the considered segment of the image. Since brighter pixels (with larger intensity values) generally correspond to vascular structures in OCTA images, an image with a larger calculated PDC has denser capillaries.

[0171] Example 4 – Computerized Automated OCTA Image Analysis Method for Perfusion Density of Large Vessels

[0172] Figure 2 The process shown in box [9] determines the large vessel perfusion density index (PDL) based on the DVM generated in the SCP and its ETDRS grid. For the segment of the graph under consideration, the PDL is calculated as the proportion of pixels belonging to the vessels or larger vessels depicted therein. When the graph has a larger calculated PDL, it has a denser network of large vessels.

[0173] Example 5 – A Computerized Automated OCTA Image Analysis Method for Vascular Dispersion

[0174] The process described in Example 2 is also used to determine the vascular dispersion index (VDisp). Figure 2 , box

[10] and Figure 6 Image G). In a healthy eye, the parafoveal vessels are more centripetal towards the center of the fovea. Vascular dispersion refers to the degree of centripetal tendency of the parafoveal vessels. The greater the vascular dispersion, the less centripetal the vessels have on average. To determine VDisp, firstly... Figure 6 The image shown in image F is filled and cropped so that the subfoveal region is centered. Next, for each depicted retinal vessel V, i (i = 1, ..., N, where N is the total number of blood vessels depicted), determining the regions with the same characteristics as V. i Ellipse E with the same second moment covering i If V i If it is completely centripetal, then it is connected to E. i The line segments at the centroid and the center of the central concave are oriented in the same direction. Therefore, VDisp is defined as E i spindle and connection E i The center of mass and C 内 The average of all angles between the line segments at the center (which corresponds to the central concave area), as indicated when applying the ETDRS mesh to the DVM (as shown in the image). Figure 2 In the middle, as shown in box [5], the process of VDisp calculation is in Figure 7 As shown in the image. Figure 9A The image shows an example of the defined VDisp for each blood vessel. For example, as shown... Figure 9A As shown, the total VDisp of the instance DVM is the average of all individual VDisp calculated and displayed in the label.

[0175] This process also separated the primary and secondary retinal vascular segments. Figure 2 , box

[11] ). In Figure 6 Image F to Figure 6 Image I illustrates the process of separating large and small vessel segments. First, a diagram of the vascular system depicting the retinal vessels is shown. Figure 6 Image F and Figure 2 The frame [7] is skeletonized. Figure 8 The algorithm shown in the image detects skeletonized images (see [link]). Figure 6 The endpoints and branch points in the image (H) are then identified. The identified branch points are then masked from the image, and all contours in the masked image are found. Each identified contour corresponds to a vessel segment. The total number of branch points is represented as N. bP Then the total number of blood vessel segments N seg Equals 1 + 2N bP Furthermore, the geodesic length of the line segment is expressed as F. seg And the geodesic length of the main blood vessel to which it belongs is denoted as L. ves .if If the segment is marked as a minor vessel segment, then it is marked as a major vessel segment.

[0176] Example 6 – A Computerized Automated OCTA Image Analysis Method for Blood Vessel Diameter Index

[0177] The described process is also used to determine the vessel diameter index (VDiam). Figure 2 , frame

[12] , Figure 6 Image G and Figure 6 The image H). VDiam is defined as the average ratio between the pixel area and geodesic length of all depicted retinal vessels. Let S represent the side length of the digital vascular map (DVM), N represent the number of depicted vessels, and V... i Let VDiam be the i-th depicted blood vessel in the image. The formula for VDiam is:

[0178]

[0179] Figure 9B The image shows instances of the identified VDiam for each vessel. The total VDiam of the DVM is the average of all labeled individual VDiams in the image divided by S.

[0180] Example 7 – A Computerized Automated OCTA Image Analysis Method for Vascular Torque

[0181] The above process is further used to determine the vascular tortuosity index (VT). Figure 2 , box

[13] and Figure 6 Image I). Vessel tortuosity (VT) is quantified as the average ratio between the geodesic length and Euclidean length of the separated vessel segments. M1 represents the number of primary vessel segments, M2 represents the number of secondary vessel segments, and S... i 主要 Represented as the i-th major vascular segment and S i 次要 Let VT be the i-th minor vascular segment. The formula for VT is:

[0182]

[0183] Where a = 0.8 is the weighting factor. Because secondary vessel segments typically correspond to secondary vessels, they have lower importance when calculating VT. If the vessels in the graph are more tortuous and convoluted on average, the graph will have a larger VT. Figure 9C The figure shows an example of the calculated VT for each vessel segment. The total VT of the example DVM is the weighted average of the individual VTs marked in the figure.

[0184] Example 8 – A method for computerized automated OCTA image analysis of avascular areas in the fovea.

[0185] The above process is further used to identify the foveal avascular zone (FAZ) in the DVM. Figure 2

[14] ). The FAZ is defined as the non-capillary region within the innermost ring of the subcentral foveal capillary plexus. Reference circles and lines representing the ETDRS grid in the DVM are masked using median-filtered versions of their corresponding regions in the figure. Median filtering is then applied to suppress noise. The masked and denoised image (see, for example,

[14] ). Figure 10 Image B) is adaptively thresholded and closed (dilated, then eroded), followed by removal of small objects. Then, bridging (setting 0-value pixels to 1 if they have two non-contiguous non-zero neighboring pixels) and thinning are applied to the resulting image (see, for example, [link to image B]). Figure 10 Image C). Again, small isolated objects are removed from the image to improve the robustness of the FAZ identification method.

[0186] The largest polygon in the avascular region of the fovea was obtained, i.e., the polygon that does not cover any of the positive vascular signals (see, for example, [link]). Figure 10 Image D). The position (0,0) corresponds to the top left corner of the DVM. The center of the concave region is represented as (x...内 ,Y 内 )(exist Figure 2 (marked at box [5]), and the side length of the DVM is represented as s. For Θ = 0, 3, 6, 9, ..., 357 (i.e., a total of 120 Θ), consider starting from (x 内 ,y 内 ) to (x 内 +s·sinθ,y 内 All pixels within the image on the directed line segment (+s·cosθ). The first pixel with a positive signal (i.e., part of the blood vessel) is recorded as one of the endpoints of the FAZ. Half of these endpoints are located on... Figure 10 The point is shown as a red dot in image E. If all pixels on a line segment within the image have a value of zero, a specific Θ is skipped. If fewer than three searches return a positive signal, an error is reported. The polygon defined by connecting all n = 120 (or fewer, if some of the searches do not find any positive pixels) endpoints in the order they were found is an approximate FAZ (see, for example, see...). Figure 10 Image F). The entire process of calibrating FAZ is in Figure 10 As shown in the image.

[0187] Example 9 – A Computerized Automated OCTA Image Analysis Method for the Area Index of Avascular Zones in the Fovea

[0188] The above process is used to determine the FAZ area index (FAI). Figure 2

[15] , The vertices of the FAZ polygon are arranged in clockwise order as follows Figure 10 The symbols shown are represented and labeled as (a1,b1), (a2,b2), ..., (a... n ,b n The area of ​​a polygon is calculated using the shoelace formula:

[0189]

[0190] FAI is then defined as the area of ​​the FAZ polygon divided by the pixel area of ​​the DVM. Advantageously, if the DVM is always focused on a 3×3 region, then FAI is comparable between images of different sizes.

[0191] Example 10 – A Computerized Automated OCTA Image Analysis Method for Perimeter Index of Avascular Zones in the Fovea

[0192] The above process is used to determine the FAZ perimeter index (FPI). Figure 2 , box

[16] ). According to Figure 2 The perimeter of the FAZ polygon, indicated by the box symbol

[15] , is calculated using the following formula:

[0193]

[0194] FPI is then defined as the perimeter of the FAZ polygon divided by the side length of the DVM (in pixels). Advantageously, if the DVM is always focused on a 3×3 region, FPI is comparable between images of different sizes.

[0195] Example 11 – A Computerized Automated OCTA Image Analysis Method for Non-Circularity Index of Avascular Regions in the Fovea

[0196] The described process is used to determine the FAZ non-circularity index (FACI). Figure 2 , box

[17] ). FACI quantifies the irregularity of the calibrated FAZ shape and is calculated using the following formula:

[0197]

[0198] When FACI equals one, the FAZ is a perfect circle. A larger FACI value indicates that the FAZ has a more irregular shape.

[0199] Example 12 – Measurement of Computational Retinal Microvascular Biomarkers

[0200] Using the techniques discussed above, different computational retinal microvascular biomarkers (CRMBs) were automatically identified based on OCTA images. These included FD on the superficial capillary plexus (SCP) and deep capillary plexus (DCP), PDC on the SCP and DCP, PDL, VDisp, VDiam, and VT on the SCP, and FAI, FPI, and FACI on both the SCP and DCP. To illustrate the usefulness of the proposed computerized automated analysis system, CRMB analysis was performed on 43 OCTA images from RVO patients and 30 OCTA images from healthy subjects. All OCTA images used in the study were 3mm × 3mm scans, acquired from both the SCP and DCP, and analyzed using the automated system of this invention. Different CRMBs were identified, representing the degree of retinal abnormality from both the SCP and DCP layers in the macula. Summary statistics were generated for the different CRMBs, and Student's t-tests were performed to compare the amplitude of CRMBs between the normal and RVO groups.

[0201] The fractal dimension index (FD) of both the SCP and DCP layers in eyes with RVO was significantly lower than that in control eyes (Table 1 and 2). Figure 11 Additionally, the FAI, FPI, and FACI of both layers in eyes with RVO were significantly greater than those in control eyes.

[0202] The vascular dispersion in eyes with RVO was significantly greater than that in control eyes. These results demonstrate that the proposed analytical system effectively captures the clinically important characteristics of RVO. CRMB has also been used to investigate other types of eye diseases to gain deeper insights into the potential clinical interpretations of these conditions.

[0203] Table 1. Comparison of CRMB between the normal group and the RVO group

[0204]

[0205] Example 13 – Computational Retinal Microvascular Biomarkers

[0206] To evaluate the validity of the defined CRMB, Pearson correlation tests were performed using the system of this invention between ten parafoveal LD ​​phenotypes (SCP and DCP overall, temporal, superior, nasal, and inferior) and the FAZ region provided by the machine's built-in OCTA software. Figure 11 As shown, all correlations are strongly positive (ranging from 0.70 to 0.88) and significantly different from zero at the adjusted p-value cutoff of 0.01. This demonstrates the reliability of the computerized automated analysis system of the present invention and the CRMB calculated therefrom.

[0207] It should be understood that the examples and embodiments described herein are for illustrative purposes only, and various modifications or variations based on these examples and embodiments are to be suggested to those skilled in the art and will be included within the spirit and scope of this application. Furthermore, any element or limitation of any invention or embodiment disclosed herein may be combined with any and / or all other elements or limitations disclosed herein (alone or in any combination) or any other invention or embodiment thereof, and all such combinations are considered to be within the scope of the invention, but not limited thereto.

[0208] References

[0209] Adhi, M., Filho, MA, Louzada, RN, Kuehlewein, L., de Carlo, TE, Baumal, CR, … Waheed, NK, “Retinal Capillary Network and Foveal Avascular Zone in Eyes with Vein Occlusion and Fellow Eyes Analyzed With Optical Coherence Tomography Angiography,” *Invest Ophthalmol Vis Sci*, October 2016, 57(9), 486-494. doi:10.1167 / iovs.15-18907

[0210] Aghamohamadian-Sharbaf, M., Pourreza, HR, & Banae, T., “A Novel Curvature-Based Algorithm for Automatic Grading of Retinal Blood Vessel Tortuosity,” IEEE Journal of Biomedical and Health Informatics, 2016, 20(2), 586-595. doi:10.1109 / jbhi.2015.2396198

[0211] Arrigo, A., Aragona, E., Capone, L., Pierro, L., Romano, F., Bandello, F., & Parodi, MB, “Advanced Optical Coherence Tomography Angiography Analysis of Age-related Macular Degeneration Complicated by Onset of Unilateral Choroidal Neovascularization,” *American Journal of Ophthalmology*, 2018, 195, 233-242. doi:10.1016 / j.ajo.2018.08.001

[0212] Arrigo, A., Romano, F., Albertini, G., Aragona, E., Bandello, F., & Battaglia, Parodi, M. "Vascular Patterns in Retinitis Pigmentosa on Swept-Source Optical Coherence Tomography Angiography," *Journal of Clinical Medicine*, 2019, 5(9). doi:10.3390 / jcm8091425

[0213] Brar, M., Sharma, M., Grewal, SPS, & Grewal, DS. “Quantification of retinal microvasculature and neurodegeneration changes in branch retinal vein occlusion after resolution of cystoid macular edema on optical coherence tomography angiography,” Indian Journal of Ophthalmology, 2019, 67(11), 1864-1869. doi:10.4103 / ijo.IJO_1554_18

[0214] Conrath, J., Valat, O., Giorgi, R., Adel, M., Raccah, D., Meyer, F., & Ridings, B., “Semi-automated detection of the foveal avascular zone in fluorescein angiograms in diabetes mellitus,” *Clinical and Experimental Ophthalmology*. "Optical Coherence Tomography Angiography in Retinal Vein Occlusion: Evaluation of Superficial and Deep Capillary Plexa", American Journal of Ophthalmology, 2016, 161, 160-171. doi:10.1111 / j.1442-9071.2006.01138.x Coscas, F., Glacet-Bernard, A., Miere, A., Caillaux, V., Uzzan, J., Fupidi, M., …Souied, EH. "Optical Coherence Tomography Angiography in Retinal Vein Occlusion: Evaluation of Superficial and Deep Capillary Plexa", American Journal of Ophthalmology, 2016, 161, 160-171.e161-162. doi:10.1016 / j.ajo.2015.10.008

[0215] Costanzo, E., Parravano, M., Gilardi, M., Cavalleri, M., Sacconi, R., Aragona, E., …Querques, G., “Microvascular Retinal and Choroidal Changes in Retinal Vein Occlusion Analyzed by Two Different Optical Coherence Tomography Angiography Devices”, *Ophthalmologica*, 2019, 242(1), 8-15. doi:10.1159 / 000496195

[0216] Deng, Y., Cai, X., Zhang, S., Su, L., Chen, H., Lin, Y., … Chi, W. “Quantitative Analysis of Retinal Microvascular Changes after Conbercept Therapy in Branch Retinal Vein Occlusion Using Optical Coherence Tomography Angiography”, *Acta Ophthalmologica Sinica*, 2019, 242(2), 69-80. doi:l0.1159 / 000499608

[0217] Diaz, M., Novo, I, Cutrin, P., Gomez-Ulla, F., Penedo, MG, & Ortega, M., “Automatic segmentation of the fovealavascular zone in ophthalmological OCT-A images,” PLoSOne, 2019, 14(2), e0212364. doi:10.1371 / journal.pone.0212364

[0218] Dumitrascu, OM, Demaerschalk, BM, Valencia Sanchez, C., Almader-Douglas, D., O'Carroll, CB, Aguilar, MI, ... Kumar, G. "Retinal Microvascular Abnormalities as Surrogate Markers of Cerebrovascular Ischemic Disease: A Meta-Analysis", *Journal of Stroke and Cerebrovascular Disease*, 2018, 27(1), 1960-1968. doi:10.1016 / j.jstrokecerebrovasdis.2018.02.041

[0219] Grisan, E., Foracchia, M., & Ruggeri, A., “A novel method for the automatic grading of retinal vessel tortuosity,” IEEE TransMedImaging, 2008, 27(3), 310-319. doi:10.1109 / tmi.2007.904657

[0220] Hanada, N., Iijima, H., Sakurada, Y., & Imasawa, M., “Recurrence of macular edema associated with branch retinal vein occlusion after intravitreal bevacizumab injection”, *Japanese Journal of Ophthalmology*, 2012, 56(2), 165-174. doi:10.1007 / s10384-011-0113-4

[0221] Hao, H., Sasongko, MB, Wong, TY, CheAzemin, MZ, Aliahmad, B., Hodgson, L., …Kumar, DK. “Does retinal vascular geometry vary with cardiac cycle?”, *Ophthalmic Research and Vision*, 2012, 53(9), 5799-5805. doi:10.1167 / iovs.l1-9326

[0222] Hartnett, ME, Martiniuk, D., Byfield, G., Geisen, P., Zeng, G., & Bautch, VL., “Neutralizing VEGF decreases tortuosity and alters endothelial cell division orientation in arterioles and veins in a rat model of ROP: relevance to plus disease,” *Ophthalmic Research and Vision*, 2008, 49(1), 3107-3114. doi:10.1167 / iovs.08-1780

[0223] Huang, D., Jia, Y., Gao, SS, Lumbroso, B., & Rispoli, M., “Optical Coherence Tomography Angiography Using the Optovue Device,” *Dev Ophthalmol*, 2016, 56, 6-12. doi:10.1159 / 000442770

[0224] Kadomoto, S., Muraoka, Y., Ooto, S., Miwa, Y., Iida, Y., Suzuma, K., … Yoshimura, N. "Evaluation of macular ischemia in eyes with branch retinal vein occlusion: An Optical Coherence Tomography Angiography Study", *Retina*, 2018, 38(2), 272-282. doi:10.1097 / iae.0000000000001541

[0225] Khodabandeh, A., Shahraki, K., Roohipoor, R., Riazi-Esfahani, H., Yaseri, M., Faghhihi, H., & Bazvand, F., “Quantitative measurement of vascular density and flow using optical coherence tomography angiography (OCTA) in patients with central retinal vein occlusion: Can OCTA help in distinguishing ischemic from non-ischemic type?”, *International Journal of Retina and Vitreous*, 2018, 4, 47. doi:10.1186 / s40942-018-0152-9

[0226] Kulikov, AN, Maltsev, DS, & Burnasheva, MA, “Improved analysis of foveal avascular zone area with optical coherence tomography angiography,” GraefesArch Clin Exp Ophthalmol, 2018, 256(12), 2293-2299. doi:10.1007 / s00417-018-4139-x

[0227] Lee, H, Fee, M., Chung, H, & Kim, HC, “Quantification of Retinal Vessel Torque in Diabetic Retinopathy Using Optical Coherence Tomography Angiography”, *Retina*, 2018, 38(5), 976-985. doi:10.1097 / iae.0000000000001618

[0228] Lee, HE, Wang, Y., Fayed, AE, & Fawzi, AA, “Exploring the relationship between collaterals and vessel density in retinal vein occlusions using optical coherence tomography angiography,” *PLOS ONE*, 2019, 14(1), e0215790. doi:10.1371 / journal.pone.0215790

[0229] McIntosh, RL, Rogers, SL, Lim, L., Cheung, N., Wang, JJ, Mitchell, P., … Wong, TY, “Natural history of central retinal vein occlusion: an evidence-based systematic review”, *Ophthalmology*, 2010, 117(6), 1113-1123. el 115. doi:10.1016 / j.ophtha.2010.01.060

[0230] Murray, CD, “The Physiological Principle of Minimum Work: I. The Vascular System and the Cost of Blood Volume,” Proceedings of the National Academy of Sciences of the United States of America (ProcNatlAcad Sci USA), 1926, 12(3), 207-214. doi:10.1073 / pnas.12.3.207

[0231] Murray, CD, “The Physiological Principle of Minimum Work: II. Oxygen Exchange in Capillaries,” Proceedings of the National Academy of Sciences, 1926, 12(5), 299-304. doi:10.1073 / pnas.12.5.299

[0232] Newman, AR, Andrew, NH, & Casson, RJ, “Review of paediatric retinal microvascular changes as a predictor of cardiovascular disease,” *Clinical and Experimental Ophthalmology*, 2017, 45(1), 33-44. doi:l0.1111 / ceo.l2773

[0233] Papadia, M., Mistelie, M., Jeannin, B., & Herbort, CP. “The influence of anti-VEGF therapy on present day management of macular edema due to BRVO and CRVO: a longitudinal analysis on visual function, injection time interval and complications”, *International Journal of Ophthalmology*, 2014, 34(6), 1193-1201. doi:10.1007 / sl0792-014-0002-l

[0234] Qian, T., Zhao, M., & Xu, X. “Comparison between anti-VEGF therapy and corticosteroid or laser therapy for macular oedema secondary to retinal vein occlusion: A metaanalysis,” *Journal of Clinical Pharmacy and Therapy*, 2017, 42(5), 519-529. doi:10.1111 / jcpt.12551

[0235] Rispoli, M., Savastano, MC, & Lumbroso, B. “Capillary network anomalies in branch retinal vein occlusion on optical coherence tomographhyangiographhy”, Retina, 2015, 35(11), 2332-2338. doi:10.1097 / iae.0000000000000845

[0236] Rogers, S., McIntosh, RL, Cheung, N., Lim, L., Wang, JJ, Mitchell, P., …Wong, TY. “The prevalence of retinal vein occlusion: pooled data from populationstudies from the United States, Europe, Asia, and Australia”, *Ophthalmology*, 2010, 117(2), 313-319. e311. doi:10.1016 / j.ophtha.2009.07.017

[0237] Rogers, SL, McIntosh, RL, Lim, L., Mitchell, P., Cheung, N., Kowalski, JW, … Wong, TY, “Natural history of branch retinal vein occlusion: an evidence-based systematic review”, *Ophthalmology*, 2010, 117(6), 1094-1101. el095.doi:10.1016 / j.ophtha.2010.01.058

[0238] Sasongko, MB, Hodgson, LA, Wong, TY, Kawasaki, R., Cheung, CY, Hsu, W., …Wang, JJ. “Correlation and reproducibility of retinal vascular geometric measurements for stereoscopic retinal images of the same eyes,” *Ophthalmic Epidemiol*, 2012, 19(5), 322-327. doi:10.3109 / 09286586.2012.702258

[0239] Sasongko, MB, Wang, JJ, Donaghue, KC, Cheung, N., Benitez-Aguirre, P., Jenkins, A., …Wong, TY. "Alterations in retinal microvascular geometry in young type 1 diabetes," *Diabetes Care*, 2010, 33(6), 1331-1336. doi:10.2337 / dcl0-0055

[0240] Seknazi, D., Coscas, F., Sellam, A., Rouimi, F., Coscas, G., Souied, EH, & Glacet-Bernard, A., “Optical coherence tomography angiography in retinal vein occlusion: Correlations between macular vascular density, visual acuity, and peripheral nonperfusion area on fluorescein angiography,” *Retina*, 2018, 35(8), 1562-1570. doi:10.1097 / iae.0000000000001737

[0241] Spaide, RF, “Microvascular flow abnormalities associated with retinal vasculitis: a potential mechanism of retinal injury,” *Retina*, 2017, 37(6), 1034-1042. doi:10.1097 / iae.0000000000001359

[0242] Spaide, RF, Klancnik, JM, Jr., & Cooney, MJ, “Retinal vascular layers imaged by fluorescein angiography and optical coherence tomography angiography”, JAMA Ophthalmol, 2015, 133(1), 45-50. doi:10.1001 / jamaophthalmol.2014.3616

[0243] Suzuki, N., Hirano, Y., Yoshida, M., Tomiyasu, T., Uemura, A., Yasukawa, T., & Ogura, Y. "Microvascular Abnormalities on Optical Coherence Tomography Angiography in Macular Edema Associated With Branch Retinal Vein Occlusion," *American Journal of Ophthalmology*, 2016, 161, 126-132. el21. doi:10.1016 / j.ajo.2015.09.038

[0244] Winegarner, A., Wakabayashi, T., Fukushima, Y., Sato, T., Hara-Ueno, C., Busch, C., ... Nishida, K. "Changes in Retinal Microvasculature and Visual Acuity After Antivascular Endothelial Growth Factor Therapy in Retinal Vein Occlusion", *Ophthalmological Research and Vision*, 2018, 59(7), 2708-2716. doi:10.1167 / iovs.17-23437

[0245] Witt, N., Wong, TY, Hughes, AD, Chaturvedi, N., Klein, BE, Evans, R., …Klein, R. “Abnormalities of retinal microvascular structure and risk of mortality from ischemic heart disease and stroke”, Hypertension, 2006, 47(5), 975-981. doi:10.1161 / 01.HYP.0000216717.72048.6c

[0246] Yamada, R., Nishida, A., Shimozono, M., Kameda, T., Miyamoto, N., Mandai, M., & Kurimoto, Y. "Predictive factors for recurrence of macular edema after successful intravitreal bevacizumab therapy in branch retinal vein occlusion," *Japanese Journal of Ophthalmology*, 2015, 59(6), 389-393. doi:10.1007 / s10384-015-0412-2

[0247] Yoo, SJ, Kim, JH, Lee, TG, Kim, JW, Cho, SW, & Han, JI, “Natural Short-term Course of Recurrent Macular Edema Following Intravitreal Bevacizumab Therapy in Branch Retinal Vein Occlusion”, *Korean J Ophthalmol*, 2017, 31(2), 95-101. doi:10.3341 / kjo.2017.31.2.95

[0248] Youm, DJ, Ha, MM, Chang, Y., & Song, SJ, “Retinal vessel caliber and risk factors for branch retinalvein occlusion,” *CurrEyeRes*, 2012, 37(4), 334-338. doi:10.3109 / 02713683.2011.629070

[0249] Lemmens, Sophie et al., “Systematic review on fractal dimension of the retinal vasculature in neurodegeneration and stroke: assessment of a potential biomarker,” *Frontiers in Neuroscience*, 14 (2020): 16.

Claims

1. A method for automated quantitative assessment of the retinal microvascular system and generation of a retinal digital vasculogram, comprising: Record optical coherence tomography (OCTA) angiography image data; Preprocess OCTA image data; as well as A digital vascular image (DVM) of the retina is generated, wherein the DVM is a standardized vascular image in a uniform format for subsequent quantitative assessment. An ETDRS grid for early treatment of diabetic retinopathy is applied to the DVM to define a unified analysis region on the DVM for quantitative calculation of the computational retinal microvascular biomarker CRMB. The ETDRS grid has an inner reference circle and an outer reference circle. The region within the inner reference circle corresponds to the foveal region. The annular region between the outer reference circle and the inner reference circle is the subfoveal region. The subfoveal region is further divided into upper, lower, nasal, and temporal segments. The unified analysis region includes: the region within the inner reference circle corresponding to the foveal region, and the subfoveal region divided into upper, lower, nasal, and temporal segments. Calculate the fractal dimension index, which is used to quantify the complexity of the retinal microvascular structure; and Marking retinal vessels; The preprocessing includes calibrating the OCTA image to a predetermined orientation, converting it to grayscale, applying Gaussian blur to reduce noise, and thresholding using Otsu's binarization. The calibration of the OCTA image to a predetermined orientation includes determining the rotation angle by detecting the orientation of reference lines in the edge-processed image using Hough line transform, ensuring the reference lines are strictly horizontal and vertical. The DVM extracts and crops a square-normalized vascular image from the above-processed OCTA image by: identifying all vertical and horizontal line segments in the image using Hough line transform and combining them into a new image; identifying the contours in the new image; extracting a top-left rectangle with a width / height greater than 0.27 times the width of the original OCTA image; and cropping the rectangle into a square if it does not have equal length and width. After the ETDRS grid is applied to the DVM, it provides a unified analysis area for the quantification of computational retinal microvascular biomarkers (CRMBs) such as fractal dimension index, vascular dispersion index, and foveal avascular region-related parameters, so as to realize the quantitative comparison of CRMBs between different patients and different devices. The CRMBs include fractal dimension index, vascular dispersion index, and foveal avascular region-related parameters, and the fractal dimension index, vascular dispersion index, and foveal avascular region-related parameters are all quantified based on the unified analysis area.

2. The method according to claim 1, further comprising: Calculate the vascular dispersion index; Separate the main vascular segment and the secondary vascular segment; Calculate the vessel diameter index; Calculate the vascular tortuosity index; Mark the avascular area of ​​the central fovea; Calculate the area index of the avascular region of the fovea; and Calculate the non-circularity index of the avascular region of the fovea.

3. A method for extracting computational retinal microvascular biomarkers (CRMBs) from a DVM generated using the method described in claim 1.

4. The method according to claim 3, wherein the CRMB is selected from fractal dimension index, capillary perfusion density index, large vessel perfusion density index, vascular dispersion index, vessel diameter index, vessel tortuosity index, central foveal avascular area index, central foveal avascular perimeter index, and central foveal avascular non-circularity index.

5. An automated analysis system for optical coherence tomography (OCTA) angiography image analysis, comprising: A memory configured to store one or more executable instructions; A processor configured to perform the steps of the method according to any one of claims 1 to 4 by executing the one or more executable instructions.

6. A computer-readable storage medium containing a computer program, wherein the computer program is executable by a processor to perform the steps of the method according to any one of claims 1 to 4.

Images