Intelligent monitoring and early warning assessment method for risk of diabetic complications

Abstract

The application provides a diabetes complication risk intelligent monitoring and early warning evaluation method, comprising the following steps: capturing a retinal microvessel image of a diabetes patient, combining blood flow velocity distribution, identifying microvessel tumor signs and bleeding point positions, and obtaining original image data; performing feature extraction according to a hierarchical progressive relationship from a blood vessel trunk to a distal branch, identifying an angle change over-limit area, and re-evaluating a blood flow distribution state in combination with a shear stress level to obtain a blood flow stasis risk; associating early warning signals of diabetic retinopathy and blood flow dynamics abnormalities caused by blood glucose fluctuations of diabetes to determine an evaluation standard; and applying the evaluation standard to a collection process of an imaging module and a velocity detection module, identifying a continuous stasis area, and then outputting an early risk evaluation report of diabetic retinopathy.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for intelligent monitoring and early warning assessment of the risk of diabetic complications. Background Technology

[0002] Diabetic retinopathy, one of the most common microvascular complications of diabetes, directly threatens patients' visual health and can even lead to blindness, making it extremely crucial in the field of diabetic complication risk monitoring. Early detection of abnormal changes in retinal microvessels and accurate early warning can significantly slow disease progression and protect patients' visual function. Current monitoring methods mainly rely on fundus images to observe changes in vessel diameter, generally believing that increased diameter is beneficial for blood flow and improved tissue blood supply. However, this judgment often deviates in the face of complex vascular structures, especially when the bifurcation morphology changes. Simply relying on diameter changes cannot accurately reflect the true blood flow status, leading to inconsistencies between risk assessment results and actual disease progression. While there is a close relationship between changes in vessel diameter and local hemodynamics, they often exhibit contradictory phenomena. Under normal circumstances, increased vessel diameter should improve blood delivery capacity, but when the bifurcation angle also shifts, the distribution of blood flow at the bifurcation point is completely altered. Imagine a main blood vessel branching into tributaries like a river. If the bifurcation angle changes from an acute angle to an obtuse angle, the main blood flow will deviate from the tributary inlet, forming a stagnant zone near the tributary's outer wall, similar to a stagnant pool. The flow velocity drops sharply, easily creating low-shear regions that slow blood flow or even cause stagnation. In this situation, even if the diameter of the main branch vessel increases significantly, according to the principle of blood flow continuity, the reduced local velocity leads to a decrease in the effective blood flow through that area per unit time. The efficiency of oxygen and nutrient delivery decreases, and the retinal tissue around the bifurcation point may still face insufficient perfusion. This contradiction stems from the profound impact of the bifurcation angle shift on hemodynamics, yet it has not been fully considered, making it difficult for monitoring systems to detect hidden low-perfusion areas caused by changes in vascular morphology. For example, in fundus images, a segment of retinal microvessels may be observed to have a significantly wider diameter than in previous examinations. According to traditional standards, this would be considered an improvement in blood flow. However, actual blood flow velocity testing shows a significant decrease in velocity downstream of the bifurcation point, resulting in reduced blood supply to local tissues. This potential risk masked by superficial improvement is easily overlooked. For example, in early fundus screening of diabetic patients, an increased diameter and shift in bifurcation angle at a retinal vein bifurcation point, changing from approximately 45 degrees to 70 degrees, might be traditionally interpreted as adaptive dilation to improve blood supply. However, Doppler flow velocity measurements reveal a 30% decrease in velocity downstream of the bifurcation, indicating reduced local tissue oxygenation and masking potential risks. Another common scenario involves peripheral retinal microvascular networks where densely bifurcerated areas exhibit collective diameter increases accompanied by angular distortion. The main blood flow bypasses certain branches, creating multiple points of stasis, which traditional diameter monitoring completely overlooks, leading to early detection only when the lesions progress to macular edema. Therefore, comprehensively considering the interaction between vessel diameter changes and bifurcation angle shifts in retinal microvascular monitoring, and accurately identifying areas of localized blood flow stasis caused by morphological changes, has become a key issue for achieving early risk warning of diabetic retinopathy. Summary of the Invention

[0003] This invention provides a method for intelligent monitoring and early warning assessment of the risk of diabetic complications, mainly including: Retinal microvascular images of diabetic patients were captured, and microaneurysm signs and hemorrhage locations were identified by combining blood flow velocity distribution, yielding raw image data. Based on the raw image data, the main branch vessel structure was separated, and shear stress levels were identified according to the gradient difference in blood flow velocity on the vessel wall. The preliminary boundaries of potential blood flow stasis areas were determined by combining the distribution location of microaneurysms in diabetic retinopathy. Feature extraction was performed according to the hierarchical progression from the main vessel trunk to the terminal branches, identifying areas with excessive angular changes and reassessing the blood flow distribution status based on shear stress levels to obtain the risk of blood flow stasis. Finally, the risk of blood flow stasis was combined with blood flow velocity distribution, and the distribution of low-shear regions was analyzed. The risk assessment criteria were adjusted to accommodate the microcirculatory disturbances caused by diabetes, resulting in identification labels for occult hypoperfusion areas. Based on these labels, combined with the initial boundaries and velocity distribution in the original image data, areas within the hypoperfusion area where the stasis intensity exceeds the risk assessment criteria were identified and marked as high-risk areas, thus determining early warning signals for diabetic retinopathy. The early warning signals for diabetic retinopathy were correlated with hemodynamic abnormalities caused by blood glucose fluctuations in diabetes to determine the assessment criteria. These criteria were then applied to the acquisition process of the imaging module and velocity detection module, and an early risk assessment report for diabetic retinopathy was output after identifying areas of persistent stasis.

[0004] Furthermore, by capturing retinal microvascular images of diabetic patients and combining them with blood flow velocity distribution, microaneurysm signs and hemorrhage locations were identified, yielding raw image data, including: The acquired retinal image is deconvolutionally processed to compensate for the stretching deformation of the image edge region based on the retinal curvature, resulting in a geometrically corrected retinal microvessel image. The Frangi vascular enhancement filtering algorithm is then used to extract the vessel centerline from the retinal microvessel image. The peak gray-level gradient along the centerline normal vector direction is searched to determine the vessel wall boundary. The bifurcation nodes of the main and branch vessels are located through the intersection of the centerlines. The vessel diameter is obtained by measuring the distance between the vessel walls on both sides of the bifurcation node, and the bifurcation angle is calculated using the vector angle. Based on the bifurcation node location and the vessel diameter, regions of abrupt velocity changes are identified.

[0005] Furthermore, the process of separating the main branch vessel structure based on the original image data, identifying the shear stress level based on the gradient difference in blood flow velocity on the vessel wall, and determining the preliminary boundary of the potential blood flow stasis area in conjunction with the distribution location of microaneurysms in diabetic retinopathy includes: Morphological opening operations are applied to the original image data to eliminate interference from small blood vessel branches. The main blood vessel path is identified by a connected component labeling algorithm based on the blood vessel centerline. The main path is traced until a diameter change point or bifurcation point is encountered. The hierarchical relationship between the main blood vessel and the branch blood vessels is determined based on the diameter being greater than a preset threshold, and the separated main and branch blood vessel structure map is obtained. Multiple sampling points are selected along the centerline of each blood vessel segment. The direction vector of each blood vessel segment is obtained by calculating the principal components of the sampling point sequence. The bifurcation angle is obtained by calculating the angle between the direction vectors of the main blood vessel and the branch blood vessel. Furthermore, the process of separating the main branch vessel structure based on the original image data, identifying the shear stress level based on the gradient difference of blood flow velocity on the vessel wall, and determining the preliminary boundary of the potential blood flow stasis area in combination with the distribution location of microaneurysms in diabetic retinopathy includes: using the ratio of the upstream main vessel diameter to the downstream branch vessel diameter as the diameter ratio at the bifurcation node; determining the low shear region based on the bifurcation angle and the diameter ratio; performing spatial overlay analysis on the low shear region and the distribution location of the microaneurysms; when a low shear region exists around the microaneurysm, starting from the center point of the microaneurysm, expanding pixel by pixel in the direction of increasing shear stress value, stopping when the shear stress returns to normal or reaches the vessel wall boundary, connecting all stopping points to form a closed contour, and determining the preliminary boundary of the potential blood flow stasis area.

[0006] Furthermore, the feature extraction is performed according to the hierarchical progression from the main vascular trunk to the terminal branches, identifying regions where angular changes exceed limits and reassessing the blood flow distribution status in conjunction with shear stress levels to obtain the risk of blood stasis, including: A vascular tree topology was constructed, starting from the main vessel at the optic disc and labeling the vascular levels according to the bifurcation order, with the main vessel marked as the first level. The level increased with each bifurcation. A breadth-first search algorithm was used to traverse the topology, extracting the mean bifurcation angle and diameter ratio distribution for each level to obtain a level feature vector. Based on the level feature vector, the levels of vascular morphology associated with co-occurring lesions were determined. Based on the vascular levels of these co-occurring lesions, vascular segments with bifurcation angles deviating beyond the upper limit of the lesion staging standard were identified as angle variation excess regions. Shear stress distribution data for these regions was extracted to determine the redistribution ratio of blood flow velocity at the bifurcation, thus obtaining the blood flow distribution state. Based on the blood flow distribution state, the proportion of blood flow velocity below normal values ​​within the angle variation excess region was calculated as the stasis volume ratio to determine the risk of blood flow stasis.

[0007] Furthermore, the step of determining the synergistic lesions of the vascular morphology at each level based on the hierarchical feature vector includes: for the hierarchical feature vector, comparing with the vascular morphology reference values ​​in the staging criteria for diabetic retinopathy, calculating the deviation of angle change and the deviation of diameter increase at each level, calculating the correlation between the two deviations using the Pearson correlation coefficient, and determining that the vascular morphology at that level has synergistic lesions if the correlation coefficient exceeds a preset threshold.

[0008] Furthermore, by combining the risk of blood stasis with blood flow velocity distribution, and adjusting the risk assessment criteria according to the distribution range of low-shear regions to adapt to the characteristics of microcirculatory disturbances caused by diabetes, identification labels for occult low-perfusion regions are obtained, including: Based on the blood flow stasis risk value and blood flow velocity distribution map, areas with velocities below a preset threshold are extracted as low-velocity zones. The shear stress value distribution within the low-velocity zones is statistically analyzed to determine the areas to be evaluated. Spatial coordinate matching is performed between the areas to be evaluated and the low-shear zones, and the ratio of the intersection volume to the union volume of the two areas is calculated as an overlap index. If the overlap exceeds a preset ratio, it is determined to be a suspected low-perfusion area. Based on the suspected low-perfusion areas and their severity, identification labels for hidden low-perfusion areas are obtained.

[0009] Furthermore, the step of obtaining identification labels for occult low-perfusion areas based on the suspected low-perfusion areas and their severity includes: performing morphological closure operations using circular structural elements based on the suspected low-perfusion areas and their severity, with the radius of the structural elements set according to the vessel diameter, filling the internal cavities of the area and smoothing the boundaries, and assigning different categories of identifiers to the processed connected regions according to their severity to obtain identification labels for occult low-perfusion areas.

[0010] Furthermore, after obtaining the identification tag for the concealed low-perfusion region, the process includes: Based on the spatial position of the identification tag and the preliminary boundary, the blood flow velocity time series data of the corresponding area in the original image data is extracted. The average velocity of the same position in healthy people is obtained as the baseline velocity value. The attenuation amplitude is obtained by calculating the difference between the current velocity value and the baseline velocity value and dividing it by the baseline velocity value. The dwell time is obtained by multiplying the number of consecutive frames with velocities below the threshold in consecutive frame images by the frame interval time. Based on the attenuation amplitude and the dwell time, the blood flow stasis intensity in the low perfusion area is evaluated by multiplying the attenuation amplitude by the first weighting coefficient and adding the dwell time by the second weighting coefficient.

[0011] Furthermore, by associating early warning signs of diabetic retinopathy with hemodynamic abnormalities caused by blood glucose fluctuations in diabetes, assessment criteria were determined, including: Based on the high-risk area location information in the warning signal, the bifurcation angle and diameter ratio of the corresponding vascular segment in each monitoring cycle are retrieved from the historical monitoring database and arranged in chronological order to form an angle time series and a diameter time series. The change in adjacent time points is calculated by difference operation to obtain the morphological change rate. Based on the time correspondence between abnormal hemodynamic periods and morphological change rates, the sliding window method is used to count the number of times the morphological change rate exceeds the historical average during abnormal periods. The proportion of this number to the total number of abnormal periods is calculated as the synchronous occurrence frequency. The correlation strength between the degree of morphological deviation and the relative change rate of viscosity is evaluated by calculating the mutual information value between the two. Based on the correlation strength and the synchronous occurrence frequency, the bifurcation angle, diameter ratio, morphological change rate, and relative change rate of viscosity are used as monitoring indicators. The Pearson correlation coefficient between each indicator is calculated to construct a correlation matrix. The matrix is ​​decomposed by eigenvalue to extract the eigenvector corresponding to the largest eigenvalue. The weight coefficients of the corresponding indicators in the evaluation criteria are adjusted according to the proportion of each component of the eigenvector.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an intelligent monitoring and early warning assessment method for the risk of diabetic complications, focusing on solving the problem of identifying the correlation between microvascular morphological changes and hemodynamic abnormalities through high-resolution imaging and blood flow velocity detection technology. This invention captures retinal microvascular images and blood flow velocity distribution, extracts features such as vessel bifurcation angles and diameter ratios, and combines shear stress levels and microaneurysm distribution to accurately identify areas of blood stasis risk. It also assesses the intensity of stasis based on the spatial overlap and velocity attenuation of low-perfusion areas, marking high-risk areas. Simultaneously, this invention retrospectively examines the correlation between morphological changes and blood glucose fluctuations based on historical monitoring data to generate optimized assessment criteria, improving the sensitivity of microvascular abnormality identification, and ultimately outputting a comprehensive risk assessment report. The technical effect of this invention is that it achieves efficient early warning of diabetic retinopathy, provides a scientific basis for clinical intervention, and significantly improves the monitoring and management capabilities of microcirculatory disorders. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for intelligent monitoring and early warning assessment of the risk of diabetic complications according to the present invention.

[0014] Figure 2 This is a schematic diagram of an intelligent monitoring and early warning assessment method for the risk of diabetic complications according to the present invention.

[0015] Figure 3 This is another schematic diagram of the intelligent monitoring and early warning assessment method for the risk of diabetic complications according to the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0017] like Figures 1-3 This embodiment of a method for intelligent monitoring and early warning assessment of the risk of diabetic complications may specifically include: Step S101: Capture retinal microvascular images of diabetic patients, combine blood flow velocity distribution, identify microaneurysm signs and hemorrhage locations, and obtain raw image data.

[0018] The retinal image acquired by the high-resolution imaging module is deconvolved using the point spread function of the optical system preset by the fundus monitoring device. Stretching deformation of the image edge region is compensated according to the retinal curvature to obtain a geometrically corrected retinal microvascular image. The Frangi vascular enhancement filtering algorithm is used to extract the vessel centerline from the retinal microvascular image. The peak gray-level gradient along the normal vector direction of the centerline is searched to determine the vessel wall boundary. The bifurcation nodes of the main and branch vessels are located through the intersection of the centerlines. The vessel diameter is obtained by measuring the distance between the vessel walls on both sides of the bifurcation node, and the bifurcation angle is calculated using the vector angle. Based on the bifurcation node location and vessel diameter, a blood flow perfusion map is obtained using laser speckle imaging technology from the velocity detection module. Regions with abrupt velocity changes are identified based on the blood flow velocity distribution gradient in the perfusion map. If the abrupt region exhibits a cystic expansion morphology, it is marked as a microaneurysm location. High absorbance regions are detected by calculating the difference between the red and green channels of the image. Regions with a difference exceeding a preset threshold are identified as hemorrhage points, resulting in raw image data containing the degree of vessel diameter increase and changes in the bifurcation angle.

[0019] In one embodiment, the point spread function of the optical system of the fundus monitoring device is pre-determined using a calibration plate. This function describes the spatial distribution characteristics of the imaging system's response to a point light source. By performing Wiener filtering deconvolution on the acquired retinal image, the blurring and degradation caused by the optical system are recovered based on the inverse operation of the point spread function. Simultaneously, a distortion correction mapping table is constructed based on the retinal curvature parameters, and bilinear interpolation compensation is performed on the edge regions.

[0020] Specifically, the Frangi blood vessel enhancement filtering algorithm enhances blood vessel contrast based on the tubular structure features of blood vessels by calculating the eigenvalues ​​of the Hessian matrix in the image. The algorithm calculates the Hessian matrix for each pixel at multiple scales, where the eigenvalues ​​λ1 and λ2 reflect the principal curvature directions of the local structure. and When the point is located within a bright blood vessel structure on a dark background, it indicates that the point is situated within a bright blood vessel structure on a dark background. This is achieved by constructing a blood vessel similarity function. ,in Indicates the eccentricity of the tubular structure. The structural strength of the image is represented by α and β, which are control parameters. A similarity function is calculated at different scales σ, and the maximum response value is taken to obtain the enhanced vessel centerline. Along the detected centerline, the extreme points of the first derivative of the grayscale value are searched in the normal vector direction at each point; these extreme points correspond to the boundary positions of the vessel wall.

[0021] Preferably, the intersection point of multiple centerlines is found by tracing the connectivity of the centerlines; this intersection point is the vessel bifurcation node. Positions 5 pixels away from the node are selected on both sides, and the Euclidean distance between the vessel wall boundaries at these points is measured as the vessel diameter. The precise value of the bifurcation angle is obtained by calculating the angle between the tangent vectors of the two vessel centerlines at the bifurcation point.

[0022] In one embodiment, laser speckle imaging technology uses changes in the speckle pattern produced when coherent light illuminates the retina to detect blood flow velocity. When a laser illuminates flowing red blood cells, the temporal correlation of the speckle pattern changes with blood flow velocity. A blood perfusion map is constructed by calculating the contrast K = σ / I of consecutive frames of speckle images, where σ is the intensity standard deviation and I is the average intensity; the contrast is inversely proportional to blood flow velocity.

[0023] For example, the identification of microaneurysms is based on localized abnormalities in blood flow velocity distribution. The velocity gradient is calculated in the perfusion map. When the gradient value exceeds a preset threshold and the corresponding area exhibits cystic dilatation with a diameter greater than 1.5 times that of a normal blood vessel, the area is marked as a potential microaneurysm. Hemorrhage detection is achieved by separating the red and green channels in the RGB color space and calculating the difference image D=RG. Blood in fundus images exhibits characteristics of high red components and low green components. When the D value exceeds the dynamic threshold T=μ+2σ, it is determined to be a hemorrhage area.

[0024] Step S102: Separate the main branch vessel structure based on the original image data, identify the shear stress level based on the gradient difference of blood flow velocity on the vessel wall, and determine the preliminary boundary of the potential blood flow stasis area in combination with the distribution location of microaneurysms in diabetic retinopathy.

[0025] Morphological opening operations are used to eliminate interference from small blood vessel branches in the original image data. A connected component labeling algorithm along the vessel centerline is used to identify the main vessel path. The main path is traced until a diameter abrupt change or bifurcation point is encountered. The hierarchical relationship between the main and branch vessels is determined based on the diameter exceeding a preset threshold, resulting in a separated main-branch vessel structure map. From this main-branch vessel structure map, two branch vessel segments and one main vessel segment are extracted from the bifurcation node. Multiple sampling points are selected along the centerline of each vessel segment. The principal components of the sampling point sequence are calculated to obtain the direction vector of each vessel segment. The angle between the direction vectors of the main and branch vessels is calculated to obtain the bifurcation angle. The diameters of the upstream main vessel and downstream branch vessels at the bifurcation node are measured, their ratio is calculated, and the bifurcation node coordinates are recorded. Based on the bifurcation angle and diameter ratio, the radial distribution law of blood flow velocity is derived according to Poiseuille's law. The blood flow velocity v and radial position r satisfy a parabolic distribution relationship. The velocity gradient value is calculated at the vessel wall, and the wall shear stress is obtained by multiplying the blood viscosity and the velocity gradient. If the shear stress is lower than a preset threshold, the vessel segment is marked as a low-shear region. Spatial overlay analysis is performed on the low-shear region and the distribution location of the microaneurysm. When a low-shear region exists around the microaneurysm, the expansion is performed pixel by pixel starting from the center point of the microaneurysm and expanding in the direction of increasing shear stress value. The expansion stops when the shear stress returns to normal or when the blood vessel wall boundary is reached. All stopping points are connected to form a closed contour, thus determining the preliminary boundary of the potential blood stasis region.

[0026] In one implementation, the morphological opening operation uses a circular structuring element to perform erosion followed by dilation on the original image data. The radius of the structuring element is set to be less than half the diameter of the smallest blood vessel, thereby eliminating interference from capillaries and noise points while preserving the main vascular structure. An eight-connected-domain labeling algorithm is used to traverse the binarized vascular image, assigning a unique identifier to each continuous vascular region. The main vascular path is identified based on the number of pixels and the extension length of the connected domains.

[0027] Preferably, pixel-level tracking is performed along the main blood vessel path, and the local blood vessel diameter is calculated at each pixel. When the diameter change rate exceeds a preset threshold or multiple directions of blood vessel extension are detected, it is determined as a bifurcation point or a diameter abrupt change point. By comparing the average diameter values ​​of the blood vessels on both sides of the bifurcation point, the larger diameter is marked as the main blood vessel, and the smaller diameter is marked as a branch blood vessel.

[0028] Specifically, the direction vector extraction process is based on the principle of principal component analysis. Starting from the bifurcation node, a predetermined length is extended outward along the centerline of each vessel segment, and multiple coordinate points are uniformly collected along that segment to form a point set. Calculate the covariance matrix C of the point set, whose elements... Let v1 represent the covariance of the i-th and j-th dimensions of the coordinates. Eigenvalue decomposition is performed on the covariance matrix; the eigenvector corresponding to the largest eigenvalue is the principal direction vector of the vessel segment. The advantage of this method is that it can accurately extract the overall direction even if the vessel has slight bends or local deformations. When the principal vessel direction vector is v1 and the branch vessel direction vector is v2, the direction is calculated... The precise value of the bifurcation angle is obtained. Compared with the simple endpoint connection method, the method is more robust to local perturbations of blood vessel morphology, and is particularly suitable for retinal vascular distortion and deformation in diabetic patients.

[0029] It should be noted that the diameter measurements upstream and downstream of the bifurcation node employ the equivalent circle diameter method based on cross-sectional area. At a position a certain number of pixels upstream of the bifurcation node, the number of pixels in the vessel's cross-section is counted perpendicular to the vessel's centerline. The equivalent circle diameter, D1, is calculated based on the area. Similarly, the diameter D2 of the downstream branch vessel is obtained. The diameter ratio R = D2 / D1 reflects the morphological characteristics at the bifurcation point.

[0030] In one possible implementation, Poiseuille's law describes the laminar flow of a viscous fluid in a circular pipe. For a blood vessel of radius r, the blood flow velocity v follows a parabolic distribution along the radial direction: ,in The maximum speed is at the center line. ρ is the radial distance from the centerline. At the vessel wall, ρ=r, the velocity is zero, and the velocity gradient is... v / ρ reaches its maximum value at the wall. According to Newtonian fluid constitutive relations, wall shear stress It equals the product of blood dynamic viscosity μ and velocity gradient. When the bifurcation angle increases, blood flow undergoes streamline deflection at the bifurcation point, causing the velocity distribution at the branch vessel inlet to become asymmetrical, forming a low-velocity backflow zone near the outer wall, and significantly reducing shear stress. Combining the bifurcation angle and diameter ratio, a correction factor is applied. Adjust the theoretical shear stress value, where θ is the bifurcation angle, to obtain an estimate of the actual wall shear stress.

[0031] For example, the calculated shear stress values ​​are compared with the pathological threshold of diabetic retinopathy. Studies have shown that when the wall shear stress is below a certain critical value, vascular endothelial cell function is impaired, promoting the expression of inflammatory factors and platelet aggregation. Based on a preset shear stress threshold, consecutive low-shear stress vascular segments are grouped and marked as low-shear regions, which are high-risk locations for potential blood flow stasis.

[0032] Understandably, spatial overlay analysis is achieved by constructing a two-dimensional mask matrix. Low-shear regions are mapped to a binary mask M1 in the image coordinate system, and the locations of microaneurysms are mapped to mask M2. A logical AND operation is then performed on the two masks. This yields an initial seed region that overlaps with low-shear hemangiomas. This overlay not only considers hemodynamic abnormalities but also incorporates pathological morphological features, improving the accuracy of identifying stasis areas.

[0033] For example, the region growth process begins at the center of each microaneurysm in the seed region and expands outward along the gradient direction of the shear stress value. In each iteration, the eight neighbors of the current boundary pixel are examined. If the shear stress value of a neighboring pixel is still below a threshold and has not reached the vessel wall, it is included in the stasis region. When all boundary pixels can no longer expand, the boundary points are connected to form a closed contour, determining the initial boundary of the potential blood flow stasis region.

[0034] Step S103: Feature extraction is performed according to the hierarchical progression from the main vascular trunk to the terminal branches. Regions with excessive angle changes are identified, and the blood flow distribution is reassessed in conjunction with the shear stress level to obtain the risk of blood stasis.

[0035] A vascular tree topology is constructed based on bifurcation angles and diameter ratios. Starting from the main vessels at the optic disc, vascular levels are labeled according to bifurcation order, with the main vessels designated as the first level. The level increases with each bifurcation. A breadth-first search algorithm is used to traverse the topology, extracting the mean bifurcation angle and diameter ratio distribution for each level to obtain a level feature vector. For these level feature vectors, the deviations in angle change and diameter increase are calculated at each level, using reference values ​​for vascular morphology in the diabetic retinopathy staging criteria. The formula for calculating the angle change deviation is as follows: Where Da is the angle change deviation, Ai is the average bifurcation angle of the current level, and Ar is the reference value; the formula for calculating the diameter increase deviation is... Where Dd represents the diameter increase deviation, Di represents the average diameter ratio of the current level, and Dr is the reference value. The correlation between the two deviations is calculated using the Pearson correlation coefficient. If the correlation coefficient exceeds 0.8, the vascular morphology at that level is considered to have a synergistic lesion. Based on the vascular level of the synergistic lesion, vascular segments with bifurcation angles deviating beyond the upper limit of the lesion staging standard are identified as angle change excess regions. Shear stress distribution data of these regions are extracted, and the redistribution ratio of blood flow velocity at the bifurcation is determined according to the fluid continuity equation. Combined with the shear stress distribution, the blood flow distribution state is obtained. Based on the blood flow distribution state, the volume ratio of blood flow velocity below normal values ​​within the angle excess region is calculated as the stasis volume ratio. Combined with the shear stress deviation, a risk score is determined by a weighted sum of the stasis volume ratio and the shear stress deviation. When the risk score exceeds a preset risk threshold, it is considered a high stasis risk, thus obtaining the blood flow stasis risk.

[0036] In one implementation, the vascular tree topology is constructed based on the natural bifurcation characteristics of retinal vessels. The central artery and central vein at the optic disc serve as root nodes, each bifurcation point as an internal node of the tree structure, and terminal vessels as leaf nodes. This tree-like representation clearly presents the hierarchical relationship of the vascular network, enabling hierarchical analysis of vascular morphological parameters.

[0037] Specifically, the breadth-first search algorithm visits vascular nodes layer by layer, starting from the root node. The algorithm maintains a queue data structure. Initially, the main vascular node at the optic disc is added to the queue and marked as the first level. When the queue is not empty, the head node is removed, and all its branching child nodes are checked. Unvisited child nodes are marked with a level value equal to the parent node's level plus one and added to the end of the queue. This traversal method ensures that vascular nodes at the same level are processed consecutively. During the traversal, the branching angle and diameter ratio of each node are recorded. For all nodes at the k-th level, the arithmetic mean of the branching angles at that level is calculated as the angle feature, and the frequency distribution of the diameter ratio is statistically analyzed. The mode, median, and interquartile range of the distribution are taken as the diameter feature. These feature values ​​are combined to form the feature vector for that level. In diabetic patients, retinal vessels typically exhibit significant morphological abnormalities at the third to fifth levels, manifested as increased branching angles and imbalanced diameter ratios. By extracting features layer by layer, the vascular level at which the lesion occurs can be accurately located. The hierarchical feature vector contains multi-dimensional information. The mean angle reflects the average deviation of the bifurcation of the blood vessels at that level, and the dispersion of the diameter ratio distribution reflects the consistency of the changes in blood vessel diameter. These parameters together constitute a complete description of the morphology of blood vessels.

[0038] Preferably, the Pearson correlation coefficient is used to quantify the linear correlation between angular deviation and diameter deviation. In the calculation, the angular deviation and diameter deviation values ​​at each level are treated as two separate variable sequences, and the correlation coefficient is obtained by dividing the covariance by the product of the standard deviations of the two variables. The correlation coefficient ranges from -1 to 1. When the coefficient is close to 1, it indicates a strong positive correlation between the two deviations; that is, as the angle increases, the diameter also increases accordingly. This synergistic change pattern is a typical characteristic of diabetic retinopathy.

[0039] For example, the determination of synergistic lesions is based on a comparison of the correlation coefficient with a preset threshold. When the correlation coefficient of a certain level exceeds 0.7, a synergistic lesion is determined to occur at that level. This determination method is more accurate than assessing changes in angle or diameter alone because it captures the inherent correlation between the two morphological parameters.

[0040] In one possible implementation, the fluid continuity equation describes the conservation of blood flow at bifurcation points. According to the law of conservation of mass, the blood flow into the bifurcation point equals the total blood flow out. As the bifurcation angle changes, the momentum distribution of blood flow at the bifurcation point also changes. A larger angle requires a greater deflection of the blood flow, resulting in additional pressure loss. Using a simplified form of Bernoulli's equation, the pressure loss is proportional to the square of the velocity and inversely proportional to the cosine of the bifurcation angle. In the region exceeding the angle limit, the redistribution of blood flow velocity follows the principle of minimum energy loss. Blood flow in the main branch tends to continue in its original direction, while the blood flow entering the branch vessels decreases. Combining shear stress distribution data, low-shear regions correspond to locations where blood flow velocity decreases. By integrating the velocity field, the actual blood flow rate of each vessel segment is calculated and compared with the theoretical blood flow rate under normal conditions to obtain the degree of deviation in blood flow distribution. This assessment method comprehensively considers both geometric morphology and hydrodynamic factors.

[0041] For example, the risk score is obtained through a weighted summation of multiple dimensions. The stasis volume ratio reflects the spatial proportion of areas with low-velocity blood flow; this value is calculated by dividing the number of voxels with velocities below a threshold by the total number of voxels in the area. The shear stress deviation is represented by the ratio of the actual shear stress to the normal reference value. The two indicators are assigned different weighting coefficients, determined based on clinical case statistics; typically, the weight for the stasis volume ratio is 0.6, and the weight for the shear stress deviation is 0.4.

[0042] Step S104: Combining the risk of blood stasis with the distribution of blood flow velocity, the risk assessment criteria are adjusted according to the distribution range of low shear areas to adapt to the characteristics of microcirculatory disorders caused by diabetes, thus obtaining identification labels for occult low perfusion areas.

[0043] Based on the blood flow stasis risk value and blood flow velocity distribution map, areas with velocities below a preset threshold are extracted as low-velocity regions. The shear stress distribution within these low-velocity regions is statistically analyzed. The original risk threshold is adjusted according to the degree of vascular wall thickening caused by diabetic microcirculatory disturbances. The adjusted threshold is obtained by multiplying the original threshold by a correction coefficient. Regions with shear stress below the adjusted threshold are marked as areas to be evaluated. Spatial coordinate matching is performed between the areas to be evaluated and the low-shear regions. The ratio of the intersection volume to the union volume of the two regions is calculated as the overlap. If the overlap exceeds a preset ratio, it is determined as a suspected low-perfusion region. The severity of the suspected low-perfusion region is determined by classifying it into high, medium, and low levels based on the blood flow stasis risk value. Based on the suspected low-perfusion regions and their severity, morphological closure operations are performed using circular structuring elements. The radius of the structuring element is set according to the vessel diameter. The internal voids of the region are filled and the boundaries are smoothed. Different categories of identifiers are assigned to the processed connected regions according to their severity, resulting in identification labels for occult low-perfusion regions.

[0044] In one implementation, the blood flow velocity distribution map displays velocity values ​​in different regions using color coding, with a preset threshold determined based on the statistical lower limit of retinal blood flow velocity in healthy individuals. Low-velocity regions are extracted using a threshold segmentation method, marking pixels with velocity values ​​below the threshold and forming a binary mask.

[0045] Specifically, the correction factor is determined based on the ratio of the thickened blood vessel wall in diabetic patients to the normal blood vessel wall thickness. Thickened blood vessels reduce the effective diameter of the lumen and increase blood flow resistance, necessitating adjustments to the risk assessment criteria. Blood vessel wall thickness is measured using optical coherence tomography (OCT), and the ratio of the diseased vessel wall thickness to the normal reference value is calculated; this ratio serves as the base correction factor. Considering the cumulative impact of diabetes duration on microcirculation, a time-weighted factor is introduced, increasing the correction factor by 0.1 for every 5 years of disease duration. The original risk threshold is multiplied by the comprehensive correction factor to obtain the adjusted threshold, which reflects the individualized pathological state. When the shear stress value is lower than the adjusted threshold, it indicates that the hemodynamic state of that area has deviated from the normal range and is marked as an area to be assessed. This dynamic adjustment mechanism allows the risk assessment criteria to adapt to patients with different disease durations and lesion degrees. Spatial coordinate matching is achieved by establishing a unified image coordinate system. The area to be assessed and the low-shear area are represented as binary matrices in the coordinate system, with a value of 1 indicating belonging to the area and a value of 0 indicating not belonging. The intersection volume is obtained by performing an element-wise AND operation on the two matrices, and the union volume is obtained by performing an element-wise OR operation. The overlap index is the number of intersection elements divided by the number of union elements.

[0046] Preferably, the severity grading is determined based on the distribution range of blood stasis risk values. The risk value range is divided into three intervals: below the 33rd percentile is low risk, between the 33rd and 67th percentiles is medium risk, and above the 67th percentile is high risk. This grading method is based on statistical distribution and can objectively reflect the relative level of risk.

[0047] For example, the choice of structuring element in the morphological closing operation affects the processing effect. A circular structuring element with a radius set to one-quarter of the average diameter of blood vessels can fill small voids without excessively altering the region boundaries. The closing operation first performs an expansion operation to broaden the region boundaries, then performs an erosion operation to restore the original size. During this process, internal voids are filled, and the boundaries become smooth. The processed connected regions are assigned unique identifiers using a connected component labeling algorithm: high-risk regions are labeled in red, medium-risk regions in yellow, and low-risk regions in green.

[0048] Step S105: Based on the identification tags combined with the velocity distribution in the preliminary boundary and the original image data, areas in the low perfusion area where the stasis intensity exceeds the risk assessment criteria are identified and marked as high-risk areas, thus determining early warning signals for diabetic retinopathy.

[0049] Based on the spatial location of the identification tag and the preliminary boundary, the time-series blood flow velocity data of the corresponding region in the original image data is extracted. The average velocity of the same location in healthy individuals is obtained as the baseline velocity value. The attenuation amplitude is obtained by calculating the difference between the current velocity value and the baseline velocity value and dividing it by the baseline velocity value. The residence time is obtained by multiplying the number of consecutive frames in a continuous frame image where the velocity is below the threshold by the frame interval. For the attenuation amplitude and residence time, a stasis intensity value is constructed by multiplying the attenuation amplitude by a first weighting coefficient and adding the residence time by a second weighting coefficient. The first and second weighting coefficients are determined based on clinical case statistics. If the stasis intensity value exceeds the risk assessment standard, the region is marked as a high-risk area. The risk assessment standard is determined based on clinical case statistics and is 0.7, which is independent of the risk threshold used for shear stress mentioned above, and is used to assess the risk of stasis intensity. Based on the high-risk area and stasis intensity value, the proportion of the area with stasis intensity value exceeding the severe threshold to the total retinal area is statistically analyzed, and the nearest distance of the area to the macula center is calculated. When the proportion exceeds the preset area proportion threshold or the distance is less than the preset safe distance threshold, it is determined as an early warning signal requiring medical intervention, thus identifying an early warning signal of diabetic retinopathy.

[0050] In one implementation, identification tags and preliminary boundaries jointly define the retinal regions requiring focused monitoring. Blood flow velocity values ​​at corresponding spatial locations are precisely extracted from the time series of raw image data using coordinate mapping, forming a velocity time-series curve for each pixel. Baseline velocity values ​​are determined using a large-sample statistical method. Samples matching the patient's age group are selected from a database of fundus images of healthy individuals, and blood flow velocity data at the same vascular level and distance from the optic disc are extracted. After performing a normality test on these data, the mean and standard deviation are calculated, and the mean is used as the baseline velocity reference value for that location. This method considers the influence of vascular location on blood flow velocity, as the normal blood flow velocity gradually decreases with distance from the heart. The attenuation magnitude is quantified by the ratio of the difference between the current velocity and the baseline velocity, reflecting the relative decrease in blood flow velocity. Dwell time is calculated based on continuous frame analysis; when the blood flow velocity at a certain location remains below a threshold, the accumulated number of frames multiplied by the frame interval of the imaging system equals the dwell time. This parameter reflects the persistence of the hypoperfusion state and is an important indicator for assessing the risk of tissue hypoxia. The first and second weighting coefficients were determined through regression analysis. Historical data of patients diagnosed with diabetic retinopathy were collected, with the attenuation magnitude and duration as independent variables and the severity of the disease as the dependent variable. The weighting coefficients were obtained through multiple linear regression.

[0051] Preferably, the calculation of the stasis intensity value employs normalization to ensure a reasonable combination of parameters with different dimensions. The attenuation amplitude is expressed as a percentage, and the retention time is converted into a proportion relative to the normal blood flow cycle. The two normalized parameters are multiplied by their corresponding weighting coefficients and then summed to obtain a stasis intensity value between 0 and 1. When this value exceeds 0.7, it is identified as a high-risk area.

[0052] For example, the determination of early warning signals takes into account both spatial distribution and location factors. The area ratio threshold is set at 5% of the total retinal area, and the safe distance threshold is set at twice the radius of the macula's center.

[0053] Step S106: Associate early warning signals of diabetic retinopathy with hemodynamic abnormalities caused by blood glucose fluctuations in diabetes to determine the assessment criteria.

[0054] Based on the location information of high-risk areas in the early warning signal, the bifurcation angle and diameter ratio of the corresponding vascular segment in each monitoring cycle are retrieved from the historical monitoring database and arranged in chronological order to form angle time series and diameter time series. The change in adjacent time points is calculated through differential operation to obtain the morphological change rate. For the morphological change rate, the blood glucose monitoring records of patients during the same period are obtained, and the difference between the postprandial blood glucose peak and the fasting blood glucose is extracted as the blood glucose fluctuation amplitude. The osmotic pressure change value is calculated based on the linear relationship between blood glucose concentration and plasma osmotic pressure. The osmotic pressure change value is divided by the baseline osmotic pressure to obtain the relative change rate of blood viscosity. When the relative change rate of viscosity exceeds a preset threshold, it is marked as a period of abnormal hemodynamics. Based on the time correspondence between the abnormal hemodynamic periods and the morphological change rate, the sliding window method is used to count the number of times the morphological change rate exceeds the historical average within the abnormal period. The proportion of this number to the total number of abnormal periods is calculated as the synchronous occurrence frequency. The correlation strength between the degree of morphological deviation and the relative change rate of viscosity is evaluated by calculating the mutual information value of the two, where the degree of morphological deviation D is defined as the absolute value of the difference between the morphological change rate R and the historical average change rate A, i.e. Based on the correlation strength and frequency of synchronous occurrence, the bifurcation angle, diameter ratio, morphological change rate, and relative viscosity change rate are used as monitoring indicators. The Pearson correlation coefficient between each indicator is calculated to construct a correlation matrix. The matrix is ​​then decomposed to extract the eigenvector corresponding to the largest eigenvalue. The weight coefficients of the corresponding indicators in the evaluation criteria are adjusted according to the proportion of each component of the eigenvector to obtain the evaluation criteria.

[0055] In one implementation, the historical monitoring database is organized and stored using a three-dimensional index based on patient number, examination date, and vessel location. Each record contains multiple parameters such as bifurcation angle, diameter ratio, and blood flow velocity. The temporal resolution is based on monthly examinations, while the spatial resolution is accurate to individual vessel bifurcation points, providing a complete data foundation for long-term trend analysis. The rate of morphological change is calculated using a first-order difference method. For the time series of bifurcation angles... angular change at adjacent time points The rate of change of angle is obtained by dividing by the time interval. The rate of change of diameter ratio is calculated in the same way. This differential operation can capture the dynamic evolution of vascular morphology and identify different patterns of rapid change or slow progression.

[0056] Specifically, the relationship between blood glucose fluctuations and blood viscosity is based on an osmotic pressure regulation mechanism. When blood glucose concentration increases, plasma osmotic pressure increases accordingly; for every 1 mmol / L increase in blood glucose, osmotic pressure increases by approximately 1 mOsm / kg. High osmotic pressure leads to dehydration and shrinkage of red blood cells, reducing their deformability, while plasma protein concentration relatively increases. These factors work together to increase blood viscosity and flow resistance. In diabetic patients, postprandial blood glucose peaks often exceed 11.1 mmol / L, with a difference of 5-8 mmol / L compared to fasting blood glucose. The corresponding osmotic pressure change is sufficient to cause a relative change in blood viscosity of 15-20%. Increased viscosity leads to decreased blood flow velocity, altered shear stress, and abnormal mechanical stimulation of vascular endothelial cells, triggering vascular remodeling responses. Long-term, repeated blood glucose fluctuations create periodic hemodynamic disturbances, gradually altering the geometry of blood vessels. This pathophysiological process explains why patients with poorly controlled blood glucose are more prone to increased vascular bifurcation angles and abnormal diameters.

[0057] Preferably, a dynamic threshold is used to determine the hemodynamically abnormal period. Considering individual differences, the preset threshold is adjusted based on the patient's baseline blood viscosity level. For patients with high baseline viscosity, the threshold is increased accordingly to avoid oversensitivity. Abnormal periods are marked using a binarization method to facilitate subsequent time-correspondence analysis.

[0058] In one possible implementation, the sliding window is set to a width of 24 hours and a step size of 1 hour, covering the complete blood glucose fluctuation cycle. Within each window, the number of events in which the rate of morphological change exceeds the historical average is counted. The historical average is calculated using data from the previous three months and represents the patient's baseline level of change. The calculation of the frequency of occurrence takes into account the time lag effect, as there may be a lag of several hours to several days between blood glucose fluctuations and morphological changes.

[0059] For example, the calculation of mutual information values ​​quantifies the statistical dependence between two variables. The degree of morphological deviation and the relative rate of change of viscosity are discretized into multiple intervals, respectively, to construct a joint probability distribution matrix. The matrix element P(i,j) represents the probability that the degree of morphological deviation falls within the i-th interval and the rate of change of viscosity falls within the j-th interval. Mutual Information Where P(i) and P(j) are the marginal probability distributions, respectively. A higher mutual information value indicates a stronger association between the two variables. In the context of diabetic retinopathy, a mutual information value exceeding 0.5 indicates a strong correlation between blood glucose fluctuations and vascular morphological changes. This quantitative assessment method captures nonlinear relationships, particularly threshold and saturation effects, better than simple correlation coefficients.

[0060] Understandably, the correlation matrix integrates the interrelationships of multiple monitoring indicators. The matrix is ​​a 4×4 symmetric matrix, with diagonal elements equal to 1 and off-diagonal elements representing the Pearson correlation coefficients of the corresponding indicator pairs. The correlation coefficients are calculated based on standardized data, eliminating the influence of dimensions.

[0061] For example, the eigenvector corresponding to the largest eigenvalue obtained from eigenvalue decomposition reflects the relative importance of each indicator. The larger the absolute value of the eigenvector component, the greater the contribution of the corresponding indicator to the overall change pattern. After normalizing the eigenvector, each component is directly used as a new weighting coefficient, replacing the empirical weights in the original evaluation criteria.

[0062] Step S107: The evaluation criteria are applied to the acquisition process of the imaging module and the velocity detection module. After identifying the persistent stagnation area, an early risk assessment report of diabetic retinopathy is output.

[0063] Based on the weighting coefficients in the evaluation criteria, the spatial sampling density of the imaging module and the temporal sampling interval of the velocity detection module are adjusted proportionally. For regions with high weighting for bifurcation angles, the angle measurement accuracy is increased by increasing the sampling point density. For locations with high weighting for diameter ratios, the diameter measurement frequency is increased by shortening the measurement interval, resulting in enhanced image data with optimized parameters. For this enhanced image data, regions with a correlation strength exceeding 0.8 in the correlation coefficient calculation results between morphological parameters and blood flow velocity are identified as key monitoring targets. Persistent stasis regions with blood flow velocities consistently below the baseline value and fixed locations for five consecutive monitoring cycles are identified. Blood flow velocity values ​​and shear stress distributions are summarized according to risk level, calculated from blood flow velocity and diameter, and morphological parameter change trends are obtained from continuous monitoring. A comprehensive risk assessment report is output, enabling early risk warning and processing for diabetic retinopathy.

[0064] In one implementation, the weighting coefficients in the evaluation criteria are stored as normalized numerical values, ranging from 0 to 1. A linear mapping relationship is established between the weighting coefficients and the imaging parameters; the higher the weight, the greater the increase in the corresponding sampling density or frequency.

[0065] Specifically, the spatial sampling density is adjusted by modifying the pixel spacing of the imaging module. When the bifurcation angle weight is 0.8, the pixel spacing in that region is reduced to 0.6 times its original value, equivalent to an increase of 1.67 times in the number of sampling points. The adjustment of the temporal sampling interval is reflected in the acquisition frequency of consecutive frames; the position with a high diameter ratio weight is increased from 10 frames per second to 15 frames per second. The determination of persistent stasis regions is based on a dual spatiotemporal standard. Spatially, the positional offset of the stasis region is required to be no more than one pixel; temporally, the blood flow velocity must be continuously lower than the baseline value for more than 5 consecutive monitoring cycles. This strict determination standard excludes temporary blood flow fluctuations.

[0066] For example, the comprehensive risk assessment report is divided into three parts according to risk level: a list of high-risk areas, a distribution map of medium-risk areas, and statistical data of low-risk areas. Each part includes location coordinates, area percentage, and recommended intervention measures.

[0067] Based on the embodiments of the present invention described above, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention. The technical scope of the present invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for intelligent monitoring and early warning assessment of the risk of diabetic complications, characterized in that, The method includes: Capture retinal microvessel images of diabetic patients, combine blood flow velocity distribution, identify microaneurysm signs and hemorrhage locations, and obtain raw image data; The main branch vessel structure is separated based on the original image data, the shear stress level is identified based on the gradient difference of blood flow velocity on the vessel wall, and the preliminary boundary of the potential blood stasis area is determined by combining the distribution location of microaneurysms in diabetic retinopathy. Feature extraction is performed based on the hierarchical progression from the main vascular trunk to the terminal branches. Regions with excessive angle changes are identified, and the blood flow distribution is reassessed in conjunction with the shear stress level to obtain the risk of blood stasis. By combining the risk of blood stasis with the distribution of blood flow velocity, the risk assessment criteria were adjusted according to the distribution range of low shear regions to adapt to the characteristics of microcirculatory disturbances caused by diabetes, thus obtaining identification labels for occult low perfusion regions. Based on the identification tags combined with the velocity distribution in the preliminary boundary and the original image data, areas in the low perfusion region where the stasis intensity exceeds the risk assessment criteria are identified and marked as high-risk areas, thus determining early warning signals for diabetic retinopathy. To identify early warning signs of diabetic retinopathy and hemodynamic abnormalities caused by blood glucose fluctuations in diabetes, and to determine the assessment criteria; The evaluation criteria are applied to the acquisition process of the imaging module and velocity detection module, and an early risk assessment report of diabetic retinopathy is output after identifying areas of persistent stagnation.

2. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 1, characterized in that, The method involves capturing retinal microvessel images of diabetic patients, combining them with blood flow velocity distribution, identifying microaneurysm signs and hemorrhage locations, and obtaining raw image data, including: The acquired retinal image is deconvolutionally processed to compensate for the stretching deformation of the image edge region based on the retinal curvature, resulting in a geometrically corrected retinal microvessel image. The Frangi vascular enhancement filtering algorithm is then used to extract the vessel centerline from the retinal microvessel image. The peak gray-level gradient along the centerline normal vector direction is searched to determine the vessel wall boundary. The bifurcation nodes of the main and branch vessels are located through the intersection of the centerlines. The vessel diameter is obtained by measuring the distance between the vessel walls on both sides of the bifurcation node, and the bifurcation angle is calculated using the vector angle. Based on the bifurcation node location and the vessel diameter, regions of abrupt velocity changes are identified.

3. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 1, characterized in that, The process of separating the main branch vessel structure based on the original image data, identifying the shear stress level based on the gradient difference of blood flow velocity on the vessel wall, and determining the preliminary boundary of the potential blood flow stasis area in combination with the distribution location of microaneurysms in diabetic retinopathy includes: Morphological opening operations are applied to the original image data to eliminate interference from small blood vessel branches. The main blood vessel path is identified by a connected component labeling algorithm based on the blood vessel centerline. The main path is traced until a diameter change point or bifurcation point is encountered. The hierarchical relationship between the main blood vessel and the branch blood vessels is determined based on the diameter being greater than a preset threshold, and the separated main and branch blood vessel structure map is obtained. Multiple sampling points are selected along the centerline of each blood vessel segment. The direction vector of each blood vessel segment is obtained by calculating the principal components of the sampling point sequence. The bifurcation angle is obtained by calculating the angle between the direction vectors of the main blood vessel and the branch blood vessel.

4. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 3, characterized in that, The process involves separating the main branch vessel structure from the original image data, identifying the shear stress level based on the gradient difference in blood flow velocity on the vessel wall, and determining the preliminary boundary of the potential blood flow stasis area in conjunction with the distribution location of microaneurysms in diabetic retinopathy. This includes: using the ratio of the upstream main vessel diameter to the downstream branch vessel diameter at the bifurcation node as the diameter ratio; determining the low-shear region based on the bifurcation angle and the diameter ratio; performing spatial overlay analysis on the low-shear region and the distribution location of the microaneurysms; when a low-shear region exists around the microaneurysm, starting from the center point of the microaneurysm, expanding pixel by pixel in the direction of increasing shear stress value, stopping when the shear stress returns to normal or reaches the vessel wall boundary, connecting all stopping points to form a closed contour, and determining the preliminary boundary of the potential blood flow stasis area.

5. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 1, characterized in that, The feature extraction is performed according to the hierarchical progression from the main vascular trunk to the terminal branches. Regions exceeding the angular change limit are identified, and the blood flow distribution is reassessed in conjunction with shear stress levels to obtain the risk of blood stasis, including: A vascular tree topology was constructed, starting from the main vessel at the optic disc and labeling the vascular levels according to the bifurcation order, with the main vessel marked as the first level. The level increased with each bifurcation. A breadth-first search algorithm was used to traverse the topology, extracting the mean bifurcation angle and diameter ratio distribution for each level to obtain a level feature vector. Based on the level feature vector, the levels of vascular morphology associated with co-occurring lesions were determined. Based on the vascular levels of these co-occurring lesions, vascular segments with bifurcation angles deviating beyond the upper limit of the lesion staging standard were identified as angle variation excess regions. Shear stress distribution data for these regions was extracted to determine the redistribution ratio of blood flow velocity at the bifurcation, thus obtaining the blood flow distribution state. Based on the blood flow distribution state, the proportion of blood flow velocity below normal values ​​within the angle variation excess region was calculated as the stasis volume ratio to determine the risk of blood flow stasis.

6. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 5, characterized in that, The step of determining the synergistic lesions of the vascular morphology at each level based on the hierarchical feature vector includes: for the hierarchical feature vector, comparing with the vascular morphology reference values ​​in the staging criteria for diabetic retinopathy, calculating the deviation of angle change and the deviation of diameter increase at each level, calculating the correlation between the two deviations using the Pearson correlation coefficient, and determining that the vascular morphology at that level has synergistic lesions if the correlation coefficient exceeds a preset threshold.

7. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 1, characterized in that, The risk assessment criteria, which combine the risk of blood stasis with blood flow velocity distribution and adjust the distribution range of low-shear regions to adapt to the characteristics of microcirculatory disturbances caused by diabetes, are used to obtain identification labels for occult low-perfusion regions, including: Based on the blood flow stasis risk value and blood flow velocity distribution map, areas with velocities below a preset threshold are extracted as low-velocity zones. The shear stress value distribution within the low-velocity zones is statistically analyzed to determine the areas to be evaluated. Spatial coordinate matching is performed between the areas to be evaluated and the low-shear zones, and the ratio of the intersection volume to the union volume of the two areas is calculated as an overlap index. If the overlap exceeds a preset ratio, it is determined to be a suspected low-perfusion area. Based on the suspected low-perfusion areas and their severity, identification labels for hidden low-perfusion areas are obtained.

8. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 7, characterized in that, The method for obtaining identification labels for occult low-perfusion areas based on the suspected low-perfusion areas and their severity includes: performing morphological closure operations using circular structural elements based on the suspected low-perfusion areas and their severity, with the radius of the structural elements set according to the vessel diameter, filling the internal cavities of the area and smoothing the boundaries, and assigning different categories of identifiers to the processed connected areas according to their severity to obtain identification labels for occult low-perfusion areas.

9. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 1, characterized in that, After obtaining the identification tag for the concealed low-perfusion region, the process includes: Based on the spatial position of the identification tag and the preliminary boundary, the blood flow velocity time series data of the corresponding area in the original image data is extracted. The average velocity of the same position in healthy people is obtained as the baseline velocity value. The attenuation amplitude is obtained by calculating the difference between the current velocity value and the baseline velocity value and dividing it by the baseline velocity value. The dwell time is obtained by multiplying the number of consecutive frames with velocities below the threshold in consecutive frame images by the frame interval time. Based on the attenuation amplitude and the dwell time, the blood flow stasis intensity in the low perfusion area is evaluated by multiplying the attenuation amplitude by the first weighting coefficient and adding the dwell time by the second weighting coefficient.

10. The intelligent monitoring and early warning assessment method for the risk of diabetic complications according to claim 1, characterized in that, The criteria for assessing early warning signs of diabetic retinopathy and hemodynamic abnormalities caused by blood glucose fluctuations in diabetes were determined, including: Based on the high-risk area location information in the warning signal, the bifurcation angle and diameter ratio of the corresponding vascular segment in each monitoring cycle are retrieved from the historical monitoring database and arranged in chronological order to form an angle time series and a diameter time series. The change in adjacent time points is calculated by difference operation to obtain the morphological change rate. Based on the time correspondence between abnormal hemodynamic periods and morphological change rates, the sliding window method is used to count the number of times the morphological change rate exceeds the historical average during abnormal periods. The proportion of this number to the total number of abnormal periods is calculated as the synchronous occurrence frequency. The correlation strength between the degree of morphological deviation and the relative change rate of viscosity is evaluated by calculating the mutual information value between the two. Based on the correlation strength and the synchronous occurrence frequency, the bifurcation angle, diameter ratio, morphological change rate, and relative change rate of viscosity are used as monitoring indicators. The Pearson correlation coefficient between each indicator is calculated to construct a correlation matrix. The matrix is ​​decomposed by eigenvalue to extract the eigenvector corresponding to the largest eigenvalue. The weight coefficients of the corresponding indicators in the evaluation criteria are adjusted according to the proportion of each component of the eigenvector.

Images