Automatic diagnosis system for xerophthalmia based on image processing

Abstract

The invention relates to the field of medical image processing, in particular to an automatic xerophthalmia diagnosis system based on image processing, which introduces a differential geometry theory, regards the surface of a tear film as a dynamically evolved curved surface structure, and realizes high-precision quantitative evaluation of the stability of the tear film. The system comprises an image acquisition module, a tear film curved surface reconstruction module, a curvature feature extraction module, a multi-scale analysis module, a region connectivity evaluation module, a tear film fracture prediction module and a diagnosis report generation module, and by constructing a parameterized curved surface model of the tear film surface, curvature distribution and time-varying characteristics thereof are calculated; in combination with multi-scale analysis and regional connectivity evaluation, the system can predict the time, position and mode of tear film rupture, provides an objective basis for early diagnosis of xerophthalmia, breaks through the limitation of traditional two-dimensional pixel analysis, not only improves the diagnosis accuracy, but also provides a multi-dimensional evaluation index of tear film rupture.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical image processing, in particular to an automatic dry eye diagnosis system based on image processing, which is applied to the evaluation of dry eye in ophthalmic clinical diagnosis. BACKGROUND

[0002] Dry eye is a common ocular surface disease characterized by tear film instability, ocular surface damage and ocular discomfort symptoms. In clinical practice, tear film break-up time (TBUT) measurement is an important indicator for evaluating tear film stability. The traditional measurement method mainly relies on the subjective judgment of doctors, which has a large error. In recent years, with the development of computer vision technology, automatic diagnosis methods based on image processing have been gradually applied to the field of dry eye diagnosis.

[0003] Existing automatic dry eye diagnosis systems mainly use pixel-based image processing methods to monitor the brightness changes of tear film images to determine tear film breakage. Such methods are often limited to two-dimensional plane analysis and cannot accurately represent the three-dimensional geometric characteristics of the tear film surface, resulting in a large influence of the diagnosis results by factors such as image quality and environmental light. In addition, existing systems cannot realize the prediction analysis of tear film breakage and can only diagnose after the actual tear film breakage, which cannot provide a basis for early intervention.

[0004] Therefore, there is an urgent need for an automatic diagnosis system that can accurately represent the geometric characteristics of the tear film surface and realize the prediction analysis of tear film breakage, providing a scientific basis for the accurate diagnosis and treatment of dry eye. SUMMARY

[0005] The purpose of the present application is to provide an automatic dry eye diagnosis system based on image processing, which introduces differential geometry theory and regards the tear film surface as a dynamically evolving curved surface structure, to realize high-precision quantitative evaluation of tear film stability, thereby improving the accuracy and forward-looking of dry eye diagnosis.

[0006] The present application provides an automatic dry eye diagnosis system based on image processing, which comprises:

[0007] An image acquisition module for acquiring a tear film image sequence of a patient's ocular surface and extracting a tear point marker position;

[0008] A tear film curved surface reconstruction module connected to the image acquisition module for receiving the tear film image sequence and constructing a parameterized curved surface model representing the geometric characteristics of the tear film surface;

[0009] A curvature feature extraction module connected to the tear film curved surface reconstruction module for calculating the curvature distribution of the tear film surface based on the parameterized curved surface model, identifying feature points in the curvature field, and generating curvature feature data representing the geometric stability of the tear film;

[0010] a multi-scale analysis module, connected with the curvature feature extraction module, configured to perform multi-scale decomposition on the curvature feature data, extract tear film stability features at different spatial scales, and form a multi-scale feature vector;

[0011] a regional connectivity evaluation module, connected with the curvature feature extraction module, configured to divide the tear film surface into multiple regions based on the curvature feature data, construct a regional connectivity graph, and monitor changes in regional connectivity over time;

[0012] a tear film rupture prediction module, connected with the multi-scale analysis module and the regional connectivity evaluation module respectively, configured to receive the multi-scale feature vector and the regional connectivity change data, calculate the rupture risk index of each region of the tear film surface, and predict the time, location and mode of tear film rupture;

[0013] a diagnostic report generation module, connected with the tear film rupture prediction module, configured to generate a diagnostic report containing tear film rupture time, rupture mode and dry eye grading according to the tear film rupture prediction result.

[0014] Preferably, the image acquisition module comprises:

[0015] an image acquisition sub-module configured to acquire continuous dynamic images of the patient's ocular surface through a slit lamp microscope system;

[0016] an image preprocessing sub-module, connected with the image acquisition sub-module, configured to perform noise reduction, illumination correction and contrast enhancement processing on the continuous dynamic images;

[0017] a reference point positioning sub-module, connected with the image preprocessing sub-module, configured to identify the position of the lacrimal punctum in the continuous dynamic images and establish an ocular surface coordinate system.

[0018] Preferably, the tear film surface reconstruction module comprises:

[0019] a height map construction unit configured to convert the luminance gradient information of the preprocessed tear film image into relative height values and generate an initial height map of the tear film surface;

[0020] a spatial regularization unit, connected with the height map construction unit, configured to apply spatial filtering to the initial height map to eliminate noise and artifacts;

[0021] a parameterized surface conversion unit, connected with the spatial regularization unit, configured to convert the discrete height map into a continuous parameterized surface representation;

[0022] a time-varying surface tracking unit, connected with the parameterized surface conversion unit, configured to track changes in the parameterized surface over time and construct a spatiotemporal surface model.

[0023] As preferred, the curvature feature extraction module comprises:

[0024] a discrete curvature computation unit configured to apply a discrete differential operator to the parametric surface to compute a mean curvature and a Gaussian curvature distribution of the tear film surface;

[0025] a feature point identification unit connected to the discrete curvature computation unit and configured to extract local extreme points and saddle points from the curvature field;

[0026] a curvature variation rate computation unit connected to the discrete curvature computation unit and configured to compute a variation rate of the curvature over time to form a variation rate field representing the stability of the tear film;

[0027] a feature trajectory analysis unit connected to the feature point identification unit and configured to track the evolution trajectory of the feature points over time to identify unstable regions.

[0028] As preferred, the multi-scale analysis module comprises:

[0029] a scale space construction unit configured to construct a multi-level scale representation based on the curvature field of the tear film surface;

[0030] a scale feature extraction unit connected to the scale space construction unit and configured to extract curvature features and region segmentation at each scale level;

[0031] a cross-scale association unit connected to the scale feature extraction unit and configured to establish a correspondence between features of different scale levels to form a feature propagation graph;

[0032] a scale stability analysis unit connected to the cross-scale association unit and configured to analyze the stability of the features in the scale space to identify abnormal regions consistent across multiple scales.

[0033] As preferred, the regional connectivity evaluation module comprises:

[0034] a region division unit configured to divide the tear film surface into multiple regions based on the curvature features;

[0035] a connectivity graph construction unit connected to the region division unit and configured to compute similarity and connectivity strength between regions to construct a region connectivity graph;

[0036] a connectivity variation monitoring unit connected to the connectivity graph construction unit and configured to track the evolution of the connectivity graph over time to detect significant changes in edge weights;

[0037] a rupture location prediction unit connected to the connectivity variation monitoring unit and configured to identify region pairs with rapidly decreasing connectivity to determine potential rupture locations.

[0038] As preferred, the tear film rupture prediction module comprises:

[0039] a multi-source feature fusion unit configured to integrate the multi-scale feature vectors and the regional connectivity change data;

[0040] a risk index calculation unit connected to the multi-source feature fusion unit and configured to calculate a tear film surface point rupture risk index;

[0041] a rupture prediction unit connected to the risk index calculation unit and configured to predict a tear film rupture time and position based on the risk distribution;

[0042] a rupture mode classification unit connected to the rupture prediction unit and configured to classify tear film rupture modes associated with different types of dry eye syndrome;

[0043] Preferably, the rupture mode classification unit classifies tear film rupture modes based on the area, shape and diffusion rate of the rupture region, including three basic modes of punctate rupture, regional rupture and merged rupture.

[0044] Preferably, the diagnostic report generation module includes:

[0045] a tear film rupture time calculation unit configured to calculate a standardized tear film rupture time index based on the predicted tear film rupture event;

[0046] a dry eye syndrome grading unit connected to the tear film rupture time calculation unit and configured to grade dry eye syndrome according to tear film rupture time and rupture mode;

[0047] a visualization result generation unit connected to the dry eye syndrome grading unit and configured to generate visualization data including 2D / 3D charts, heat maps and time series graphs;

[0048] a treatment recommendation generation unit connected to the dry eye syndrome grading unit and configured to provide personalized treatment recommendations based on dry eye syndrome grading and rupture characteristics.

[0049] Preferably, the application further includes:

[0050] a result correction module connected to the tear film rupture prediction module and the diagnostic report generation module, configured to receive tear film rupture prediction results from multiple examinations, analyze result consistency, correct abnormal prediction results and improve diagnostic reliability;

[0051] a historical data management module connected to the diagnostic report generation module and configured to store patient historical examination results, support pre- and post-treatment comparative analysis and evaluate treatment effectiveness.

[0052] The present application combines differential geometry with ocular surface image processing technology to construct a dynamic geometric model of the tear film surface, breaking through the limitations of traditional two-dimensional pixel analysis and achieving the following beneficial effects:

[0053] 1. Diagnosis accuracy is significantly improved: the system diagnosis accuracy rate reaches 92.5%, which is about 20% higher than the traditional manual evaluation, effectively reducing misdiagnosis and missed diagnosis;

[0054] 2. Tear film breakage prediction is realized: the system can predict the breakage position and time 3-5 seconds before the actual tear film breakage, advance the diagnosis time, and provide the possibility for preventive treatment;

[0055] 3. Multi-dimensional evaluation is provided: compared with the traditional method which only outputs a single breakage time value, the system provides multi-dimensional indexes such as tear film breakage time, position, mode and degree, and provides a scientific basis for individualized treatment plan;

[0056] 4. Early screening ability is enhanced: the detection sensitivity of the system to slight tear film instability is improved by 42%, which can effectively identify early dry eye symptoms and help early intervention of the disease. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 The overall architecture diagram of the dry eye automatic diagnosis system based on image processing provided by the embodiment of the present application is provided;

[0058] Figure 2 The workflow diagram of the tear film curved surface reconstruction module provided by the embodiment of the present application is provided;

[0059] Figure 3 The workflow diagram of the curvature feature extraction module provided by the embodiment of the present application is provided;

[0060] Figure 4 The workflow diagram of the multi-scale analysis module provided by the embodiment of the present application is provided;

[0061] Figure 5 The workflow diagram of the regional connectivity evaluation module provided by the embodiment of the present application is provided;

[0062] Figure 6 The workflow diagram of the tear film breakage prediction module provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0063] Please refer to Figures 1-6 , the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the following description is only exemplary and does not limit the present application.

[0064] Referring to Figure 1The application provides an image processing-based dry eye automatic diagnosis system, which comprises an image acquisition module 1, a tear film curved surface reconstruction module 2, a curvature feature extraction module 3, a multi-scale analysis module 4, a regional connectivity evaluation module 5, a tear film rupture prediction module 6, a diagnosis report generation module 7, a result correction module 8 and a historical data management module 9.

[0065] The image acquisition module 1 is used for acquiring a tear film image sequence of the ocular surface of a patient and extracting a tear punctum mark position. The tear film curved surface reconstruction module 2 is connected with the image acquisition module 1 and is used for receiving the tear film image sequence and constructing a parameterized curved surface model representing the geometric characteristics of the tear film surface. The curvature feature extraction module 3 is connected with the tear film curved surface reconstruction module 2 and is used for calculating the curvature distribution of the tear film surface based on the parameterized curved surface model, identifying feature points in the curvature field and generating curvature feature data representing the geometric stability of the tear film. The multi-scale analysis module 4 is connected with the curvature feature extraction module 3 and is used for performing multi-scale decomposition on the curvature feature data, extracting tear film stability features at different spatial scales and forming a multi-scale feature vector. The regional connectivity evaluation module 5 is connected with the curvature feature extraction module 3 and is used for dividing the tear film surface into multiple regions based on the curvature feature data, constructing a regional connectivity graph and monitoring the change of the regional connectivity with time. The tear film rupture prediction module 6 is connected with the multi-scale analysis module 4 and the regional connectivity evaluation module 5 respectively and is used for receiving the multi-scale feature vector and the regional connectivity change data, calculating the rupture risk index of each region of the tear film surface and predicting the time, position and mode of the tear film rupture. The diagnosis report generation module 7 is connected with the tear film rupture prediction module 6 and is used for generating a diagnosis report containing the tear film rupture time, rupture mode and dry eye classification according to the tear film rupture prediction result.

[0066] The specific implementation modes of the modules of the system will be described in detail below.

[0067] The image acquisition module 1 comprises an image acquisition sub-module 11, an image preprocessing sub-module 12 and a reference point positioning sub-module 13.

[0068] The image acquisition sub-module 11 is used for acquiring continuous dynamic images of the ocular surface of a patient through a slit lamp microscope system. Preferably, the frame rate is set to 30 frames / second and the resolution is 1280x720 pixels, so as to ensure the image quality and time accuracy. During the acquisition process, the patient needs to keep the eyes naturally open, and the tear film surface is continuously observed and recorded through the slit lamp microscope system.

[0069] The image preprocessing submodule 12 is connected to the image acquisition submodule 11 and is used to perform noise reduction, illumination correction, and contrast enhancement processing on continuous dynamic images. In one embodiment of the present invention, the noise reduction processing uses Gaussian filtering with a filter kernel size of 5×5 and a standard deviation σ=1.5; the illumination correction uses an adaptive histogram equalization method, dividing the image into 8×8 sub-blocks for histogram equalization; and the contrast enhancement uses logarithmic transformation to enhance subtle changes on the tear film surface.

[0070] The reference point localization submodule 13 is connected to the image preprocessing submodule 12 and is used to identify the lacrimal punctum position in continuous dynamic images and establish an ocular surface coordinate system. The lacrimal punctum is a relatively fixed anatomical structure on the ocular surface and can be used as a reference point for the coordinate system. This invention uses a template matching-based method to identify the lacrimal punctum, achieving an accuracy of over 96%. The established ocular surface coordinate system has the lacrimal punctum as the origin, the horizontal direction as the x-axis, and the vertical direction as the y-axis, providing a unified reference system for subsequent tear film surface analysis.

[0071] The tear film surface reconstruction module 2 includes: a height map construction unit 21, a spatial regularization unit 22, a parametric surface transformation unit 23, and a time-varying surface tracking unit 24.

[0072] The height map construction unit 21 is used to convert the preprocessed tear film image brightness gradient information into relative height values ​​to generate an initial height map of the tear film surface. In an embodiment of the present invention, the conversion relationship between brightness gradient and relative height can be expressed as:

[0073] ,

[0074] in, For point The relative height value at a location, in micrometers (µm). m); The value represents the image brightness at the corresponding location, ranging from [0, 255]. This is the proportionality coefficient, in units of... m, determined based on experimental calibration, has a typical value of 0.05. m. This logarithmic transformation can effectively enhance details in low-brightness areas and more accurately reflect subtle changes on the tear film surface.

[0075] Spatial regularization unit 22 is connected to height map construction unit 21 and is used to apply spatial filtering to the initial height map to eliminate noise and artifacts. This invention uses bilateral filtering for spatial regularization, which can both preserve the edge features of the tear film surface and smooth noise. The filtering parameters are set as follows: spatial domain standard deviation. Pixel, range, standard deviation m, the window size is 9×9 pixels.

[0076] The parameterized surface conversion unit 23 is connected with the spatial regularization unit 22, and is used for converting the discrete height map into a continuous parameterized surface representation. The present application adopts a B-spline surface representation method, and fits the discrete height map into a continuous parameterized surface :

[0077] ,

[0078] wherein, is the three-dimensional coordinate of the point on the parameterized surface, and are the parameter space coordinates, and the value range is [0, 1]; and are the B-spline basis functions of the order and order; is the control point grid, and each control point is a three-dimensional coordinate ; is the control point number. In actual application, the cubic B-spline is selected (three B-spline), and the control point number is determined according to the image resolution, and is usually 1 / 10 of the image size, that is, for the image of 1280x720, the control point number is set to 128x72.

[0079] The time-varying surface tracking unit 24 is connected with the parameterized surface conversion unit 23, and is used for tracking the change of the parameterized surface with time, and constructing a time-space surface model. In order to capture the dynamic change of the tear film surface, the present application expands the surface model into , wherein represents the time parameter, and the unit is second (s). By tracking the change trajectory of the control point with time, a complete time-varying surface model is constructed:

[0080] ,

[0081] wherein, is the three-dimensional coordinate of the point on the parameterized surface at time t; is the control point coordinate at time t. The time-varying surface model can effectively capture the slight change of the tear film surface with time, and provides a basis for subsequent stability analysis.

[0082] ​With reference to claim 4, the curvature feature extraction module 3 comprises a discrete curvature computation unit 31, a feature point identification unit 32, a curvature variation rate computation unit 33 and a feature trajectory analysis unit 34.

[0083] The discrete curvature computation unit 31 is configured to apply a discrete differential operator to the parameterized surface, and to compute the mean curvature and Gaussian curvature distribution of the tear film surface. The present application employs the theory of discrete differential geometry to compute the curvature based on the local geometric properties of the parameterized surface. For a parameterized surface , the computation of its mean curvature and Gaussian curvature involves the fundamental forms of the surface:

[0084] ,

[0085] ,

[0086] where is the mean curvature with unit ; is the Gaussian curvature with unit ; , , are the coefficients of the first fundamental form, which describe the metric properties of the surface; , , are the coefficients of the second fundamental form, which describe the bending properties of the surface. These coefficients are computed through the partial derivatives of the surface:

[0087] ,

[0088] ,

[0089] where and are the partial derivatives of with respect to and , respectively, representing the tangent vectors of the surface; , and are the second-order partial derivatives of ; is the unit normal vector of the surface, computed as , where denotes the vector cross product operation. On the discrete grid, the finite difference method is employed to approximate the computation of these partial derivatives, realizing the fast computation of the curvature field. The computation window size is set to 5x5 grid cells, and the mirror extension method is employed for the boundary treatment.

[0090] The feature point recognition unit 32 is connected with the discrete curvature calculation unit 31, and is used for extracting local extreme points and saddle points from the curvature field. The application defines three types of key feature points: curvature maximum points (high curvature area, potential tear film weak point), curvature minimum points (low curvature area, relatively stable area) and curvature saddle points (curvature changes sharply, potential stress concentration point). The feature point extraction adopts a non-maximum suppression algorithm, the window size is 9*9 pixels, and the extraction threshold is set to 2.5 times the local average curvature. These feature points constitute the skeleton structure of the tear film surface, reflecting the geometric characteristics of the tear film.

[0091] The curvature change rate calculation unit 33 is connected with the discrete curvature calculation unit 31, and is used for calculating the change rate of the curvature with time, forming a change rate field representing the stability of the tear film. The curvature change rate is defined as the derivative of the curvature with respect to time:

[0092] ,

[0093] Wherein, is the average curvature with respect to time , the unit is μm⁻¹s⁻¹; Δt is the time increment, the unit is second (s). On the discrete time sequence, the change rate is calculated by using forward difference approximation:

[0094] ,

[0095] Wherein, Δt is the time interval between adjacent frames, usually 1 / 30 second (corresponding to the acquisition frame rate of 30fps). The curvature change rate field directly reflects the dynamic stability of the tear film surface, and the area with high change rate usually corresponds to the unstable area of the tear film, which is a potential rupture initiation point.

[0096] The feature trajectory analysis unit 34 is connected with the feature point recognition unit 32, and is used for tracking the evolution trajectory of the feature points with time and identifying unstable areas. The application adopts a feature point tracking algorithm based on optical flow to calculate the displacement vector of the feature points between consecutive frames. For each feature point, the change of its position (x(t, y(t) and curvature value H(t) with time is recorded to form a feature trajectory. The trajectory analysis focuses on the following indexes:

[0097] Position stability: the standard deviation of the change of the position of the feature point;

[0098] Curvature stability: the standard deviation of the change of the curvature value of the feature point;

[0099] Trajectory length: the number of frames that the feature point can be continuously tracked.

[0100] By setting thresholds (standard deviation of position change > 5 pixels, standard deviation of curvature change > 0.3 μm-1, trajectory length < 10 frames), unstable feature points are identified, which are often precursors to tear film rupture.

[0101] The multi-scale analysis module 4 includes: a scale space construction unit 41, a scale feature extraction unit 42, a cross-scale correlation unit 43, and a scale stability analysis unit 44.

[0102] Scale space construction unit 41 is used to construct a multi-level scale representation of the curvature field based on the tear film surface. This invention uses the solution of the thermonuclear equation as the scale space to perform multi-scale decomposition of the curvature field:

[0103] ,

[0104] in, Let H be the partial derivative of the curvature field H with respect to the scale parameter t; The Laplace operator is defined as follows: , representing the sum of the second-order partial derivatives with respect to the x and y directions; t is the scale parameter (different from time), in pixels². In the discrete implementation, Gaussian convolution is used to approximate the solution of the heat kernel equation:

[0105] ,

[0106] in, The scale parameter is The curvature field at that time; The standard deviation is The two-dimensional Gaussian kernel is defined as ; Indicates the convolution operation; The standard deviation of the Gaussian kernel, in pixels, is related to the scale parameter. The relationship is This invention constructs five scale layers, with scale parameters... The resolutions are 1.0, 2.0, 4.0, 8.0, and 16.0 pixels, covering a range of spatial scales from micro to macro.

[0107] The scale feature extraction unit 42 is connected to the scale space construction unit 41 and is used to extract curvature features and segment regions at each scale level. At each scale level, feature points are extracted using the same method as the curvature feature extraction module, but the threshold parameters are appropriately adjusted to adapt to the characteristics of different scales. Simultaneously, region segmentation is performed based on curvature values, dividing the tear film surface into high-curvature regions, low-curvature regions, and transition regions. The segmentation threshold is determined using the OTSU adaptive thresholding method to ensure reasonable segmentation results at different scales.

[0108] The cross-scale association unit 43 is connected to the scale feature extraction unit 42 to establish the correspondence between features at different scale levels, forming a feature propagation map. This invention defines the scale-space trajectory of feature points, tracking the evolution of feature points from fine to coarse scales. Feature points at two different scale levels... and A correspondence is considered to exist if and only if their spatial distance is less than 1 / 4 of the current scale resolution:

[0109] ,

[0110] in, For feature points and The Euclidean distance between them, in pixels; represents the scale parameter of the coarse-scale layer, in pixels. By establishing this cross-scale correspondence, a feature propagation map is formed, describing the evolution pattern of features in scale space.

[0111] The scale stability analysis unit 44 is connected to the cross-scale correlation unit 43 to analyze the stability of features in the scale space and identify anomalous regions consistent across multiple scales. This invention defines a scale stability index for features to measure their persistence in the scale space:

[0112] ,

[0113] in, For feature points The scale stability index has a value range of [0,1], and the larger the value, the higher the stability. For feature points The number of layers appearing in the scale space; Total number of scale layers (in this system) High scale stability (). Features that appear at least across three scales are considered intrinsic properties of the tear film surface, rather than noise or artifacts. Furthermore, analyzing the spatial distribution and temporal evolution of these stable features helps identify consistent anomalous regions across multiple scales, which are often reliable indicators of tear film instability.

[0114] Referring to claim 6, the regional connectivity assessment module 5 includes: a regional division unit 51, a connectivity graph construction unit 52, a connectivity change monitoring unit 53, and a break location prediction unit 54.

[0115] Region partitioning unit 51 is used to divide the tear film surface into multiple regions based on curvature characteristics. This invention employs a region growing algorithm based on curvature fields to divide the tear film surface into connected regions with similar curvature characteristics. The seed points for region growing are selected from curvature extrema, and the growth conditions are based on curvature similarity: when a point... curvature difference of the region is less than a threshold , the region is merged into the region :

[0116] ,

[0117] wherein is the average curvature of the point , in units of m ; is the average curvature of the region , in units of m , calculated as the arithmetic mean of the curvatures of all points in the region; the threshold is set to 0.3 m (the optimal value verified by experiments). The region division result is stored as a label map, with each pixel associated to a specific region identifier. Preferably, the number of regions is controlled to be between 10 and 20, to balance the detail expression and computational efficiency.

[0118] The connected graph construction unit 52 is connected to the region division unit 51, for calculating the similarity and connected strength between regions, and constructing a region connected graph. The region connected graph is represented as an undirected weighted graph , wherein the vertex set corresponds to the region centers, the edge set represents the connection relationship between regions, and the weight set represents the connection strength. The connected strength between region i and region j is defined as:

[0119] ,

[0120] wherein is the connected strength between region and region , with a value range of [0, 1], and the greater the value, the stronger the connectivity; and are the average curvatures of region and region , in units of m ; is the Euclidean distance between the centers of region and region , in units of pixels; and are standardization parameters, each set to 0.5 m and 100 pixels, used to adjust the weight of curvature difference and spatial distance on connectivity strength.

[0121] Connectivity change monitoring unit 53 is connected with connectivity graph construction unit 52, used to track the evolution of connectivity graph over time, detect significant changes in edge weight. By comparing the connectivity graphs of adjacent time points and , the change rate of edge weight is calculated:

[0122] ,

[0123] where, is the change rate of connectivity strength between region at time and region , usually in the range [-1, +∞); and are the connectivity strengths of region at time and region respectively; is the time increment, usually 1 / 30 second. When the change rate is lower than the threshold -0.5 (i.e. the connectivity strength drops more than 50%) and lasts for at least 3 frames, it is determined that the connectivity between region and region has a significant drop, which may indicate that the tear film is about to break at this region.

[0124] Break location prediction unit 54 is connected with connectivity change monitoring unit 53, used to identify the region pairs with rapidly decreasing connectivity and determine the potential break location. The present invention defines a region vulnerability index to characterize the likelihood of a region as a potential break point:

[0125] ,

[0126] where, is the vulnerability index of region , the larger the value, the higher the risk of break; is the set of regions adjacent to region ; represents the larger value between the negative value of connectivity strength change rate and 0, only considering the case of decreasing connectivity strength. Regions with higher vulnerability index ( ) are considered as potential break regions. Further, by analyzing the geometric properties (such as area, perimeter, shape factor, etc.) and temporal evolution patterns of these regions, the most likely break initiation location and propagation direction are predicted.

[0127] The tear film breakage prediction module 6 comprises a multi-source feature fusion unit 61, a risk index calculation unit 62, a breakage prediction unit 63 and a breakage mode classification unit 64.

[0128] The multi-source feature fusion unit 61 is used to integrate multi-scale feature vectors and regional connectivity change data. The present application adopts a feature-level fusion strategy to splice and normalize feature vectors from different modules, ensuring that features from different sources have the same dimension and weight. The normalization adopts a minimum-maximum normalization method:

[0129] ,

[0130] wherein, is the normalized feature value, with a value range of [0, 1]; is the original feature value; and are the minimum and maximum values of the feature, respectively. The fused feature vector has a high dimension (usually 100-200 dimensions) and contains comprehensive feature descriptions of the tear film surface.

[0131] The risk index calculation unit 62 is connected with the multi-source feature fusion unit 61 and is used to calculate the breakage risk index of each point on the tear film surface. The present application constructs a breakage risk model based on the fused features to calculate the breakage risk index of each point on the tear film surface :

[0132] ,

[0133] wherein, is the breakage risk index of point , with a value range of [0, 1], and the larger the value, the higher the breakage risk; is the curvature change rate at point , which has been normalized to the interval [0, 1]; is the scale stability of point , with a value range of [0, 1]; is the vulnerability index of the region where point is located, which has been normalized to the interval [0, 1]; , and are weight coefficients, satisfying , and are set to 0.4, 0.3 and 0.3 (optimal configuration verified by experiments), respectively.

[0134] The rupture prediction unit 63 is connected with the risk index calculation unit 62, and is used for predicting the time and position of tear film rupture based on the risk distribution. When the risk index of a certain area exceeds the threshold value 0.75, the area is determined as a high-risk area, and the rupture may occur in a short time. The present application adopts a time sequence analysis method, establishes an evolution model of the risk index with time, and predicts the rupture time:

[0135] ,

[0136] wherein, is the predicted rupture time at the point , in seconds (s); is the current time, in seconds (s); is the rupture threshold value, which is set as 0.9, indicating that the rupture is considered when the risk index reaches this value; is the time derivative of the risk index , in s , indicating the rate of change of the risk index with time. The predicted rupture position corresponds to the local maximum point of the risk index, which is usually the weakest position of the tear film.

[0137] The rupture mode classification unit 64 is connected with the rupture prediction unit 63, and is used for classifying the tear film rupture mode and associating it to different types of dry eye. According to claim 8, the present application classifies the tear film rupture mode into three basic types based on the area, shape and diffusion rate of the rupture area:

[0138] Point rupture: the rupture area is small (<1 mm²), the shape is approximately circular, and the diffusion rate is low (<0.5 mm² / s), which is usually related to the over-evaporation type of dry eye;

[0139] Regional rupture: the rupture area is medium (1-5 mm²), the shape is irregular, and the diffusion rate is medium (0.5-2 mm² / s), which is usually related to the mixed type of dry eye;

[0140] Fusion rupture: the rupture area is large (>5 mm²), formed by the rapid fusion of multiple small areas, and the diffusion rate is high (>2 mm² / s), which is usually related to the aqueous deficiency type of dry eye.

[0141] The rupture mode classification adopts a support vector machine algorithm, and classifies based on the geometric features (area, perimeter, circularity, rectangularity, etc.) and dynamic features (diffusion rate, diffusion direction, etc.) of the rupture area, with an accuracy of 88%. The classification result serves as an important basis for judging the type of dry eye, and provides guidance for personalized treatment.

[0142] With reference to claim 9, the diagnostic report generation module 7 comprises: a tear film break-up time calculation unit 71, a dry eye grading unit 72, a visualization result generation unit 73, and a treatment recommendation generation unit 74.

[0143] The tear film break-up time calculation unit 71 is configured to calculate a standardized tear film break-up time indicator based on the predicted tear film break-up event. The present application defines the tear film break-up time (TBUT) as the time elapsed from the start of eye opening to the first appearance of a broken area with an area exceeding 1 mm²:

[0144]

[0145] wherein, TBUT is the tear film break-up time, in seconds (s); t is the break-up occurrence time, in seconds (s); topen is the eye opening time, in seconds (s). By automatically detecting the blinking action, the system can accurately determine the eye opening time, realizing the automatic calculation of TBUT. Compared with traditional manual measurement, the automatically calculated TBUT has higher accuracy and repeatability.

[0146] The dry eye grading unit 72 is connected to the tear film break-up time calculation unit 71 and is configured to grade the dry eye according to the tear film break-up time and the break-up pattern. The present application adopts the following grading standards:

[0147] Normal: TBUT > 10 seconds, no obvious break-up pattern;

[0148] Mild dry eye: TBUT 5-10 seconds, mainly with punctate break-up;

[0149] Moderate dry eye: TBUT 3-5 seconds, mixed break-up pattern;

[0150] Severe dry eye: TBUT < 3 seconds, mainly with fusion break-up.

[0151] In addition, considering the spatial distribution of the broken area and the consistency of the break-up pattern, the grading result is fine-tuned to improve the grading accuracy.

[0152] The visualization result generation unit 73 is connected to the dry eye grading unit 72 and is configured to generate visualization data including 2D / 3D charts, heat maps, and time series graphs. The present application provides multiple visualization methods to intuitively display the tear film stability analysis results:

[0153] Curvature field heat map: the curvature distribution of the tear film surface is displayed through color mapping, with red indicating high curvature areas and blue indicating low curvature areas;

[0154] Risk index distribution graph: shows the tear film surface break-up risk distribution, with high-risk areas marked in bright colors;​

[0155] Breakup prediction animation: dynamically display the predicted breakup process, including the breakup starting point and the expansion direction;

[0156] Time series curve: show the trend of key parameters (such as average curvature, risk index, etc.) over time.

[0157] These visualization results not only help doctors understand the basis for diagnosis, but also facilitate patient communication and improve doctor-patient communication.

[0158] The treatment recommendation generation unit 74 is connected with the dry eye grading unit 72, and is used to provide personalized treatment recommendations based on dry eye grading and breakup characteristics. Based on dry eye grading and breakup patterns, the present application provides targeted treatment recommendations:

[0159] Mild dry eye (point breakup): recommend using artificial tears, improving environmental factors, and regular follow-up;

[0160] Moderate dry eye (regional breakup): recommend using moisturizing artificial tears, considering punctal plug, and increasing blink rate training;

[0161] Severe dry eye (fusion breakup): recommend using high-viscosity artificial tears, punctal plug, and consider local anti-inflammatory treatment or secretagogue.

[0162] Treatment recommendations are personalized according to patient-specific conditions (age, gender, occupation, complications, etc.), improving treatment targeting and effectiveness.

[0163] The system of the present application also includes a result correction module 8 and a historical data management module 9.

[0164] The result correction module 8 is connected with the tear film breakup prediction module 6 and the diagnosis report generation module 7, and is used to receive tear film breakup prediction results from multiple examinations, analyze result consistency, correct abnormal prediction results, and improve diagnosis reliability. When the single examination result is significantly different from the patient's historical data (such as a change in TBUT of more than 50%), the system will mark the result as abnormal and adjust it based on historical data. In addition, for multiple repeated examinations of the same patient, the system uses the median filtering method to remove outliers and improve the stability of the results.

[0165] The historical data management module 9 is connected to the diagnostic report generation module 7 to store the patient's past examination results, supporting pre- and post-treatment comparative analysis and evaluating treatment effectiveness. This module implements time-series management of examination results, supporting multi-dimensional queries and statistical analysis by time, dry eye syndrome grade, treatment plan, and other dimensions. By comparing changes in tear film breakup characteristics before and after treatment, the system can objectively assess treatment effectiveness, providing a basis for adjusting treatment plans. Furthermore, historical data can also be used for long-term follow-up management to monitor the progression or improvement of dry eye syndrome.

[0166] The foregoing has described in detail the various modules and working principles of the image processing-based automatic dry eye diagnosis system of this invention. By introducing differential geometry theory, this invention treats the tear film surface as a dynamically evolving curved structure, achieving high-precision quantitative assessment and predictive analysis of tear film stability, thus providing a powerful tool for the accurate diagnosis and personalized treatment of dry eye.

[0167] It should be understood that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. An automatic dry eye diagnosis system based on image processing, characterized by, The application relates to a tear film stability evaluation system, comprising: an image acquisition module for acquiring a tear film image sequence of a patient's ocular surface and extracting a tear punctum marker position; a tear film curved surface reconstruction module connected with the image acquisition module, for receiving the tear film image sequence, and constructing a parameterized curved surface model representing the geometric characteristics of the tear film surface; a curvature feature extraction module connected with the tear film curved surface reconstruction module, for calculating the curvature distribution of the tear film surface based on the parameterized curved surface model, identifying feature points in the curvature field, and generating curvature feature data representing the geometric stability of the tear film; a multi-scale analysis module connected with the curvature feature extraction module, for multi-scale decomposition of the curvature feature data, extracting tear film stability features at different spatial scales, and forming a multi-scale feature vector; a regional connectivity evaluation module connected with the curvature feature extraction module, for dividing the tear film surface into multiple regions based on the curvature feature data, constructing a regional connectivity graph, and monitoring the change of regional connectivity over time; a tear film rupture prediction module connected with the multi-scale analysis module and the regional connectivity evaluation module, for receiving the multi-scale feature vector and the regional connectivity change data, calculating the rupture risk index of each region of the tear film surface, and predicting the time, position and mode of tear film rupture; a diagnosis report generation module connected with the tear film rupture prediction module, for generating a diagnosis report containing the tear film rupture time, rupture mode and dry eye grading according to the tear film rupture prediction result.

2. The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, characterized by, The image acquisition module comprises: an image acquisition sub-module for acquiring continuous dynamic images of a patient's ocular surface through a slit lamp microscope system; an image preprocessing sub-module connected with the image acquisition sub-module, for noise reduction, illumination correction and contrast enhancement processing of the continuous dynamic images; a reference point positioning sub-module connected with the image preprocessing sub-module, for identifying the position of the tear punctum in the continuous dynamic images and establishing an ocular surface coordinate system. 3.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, characterized in that, The tear film curved surface reconstruction module comprises: a height map construction unit for converting the luminance gradient information of the preprocessed tear film image into relative height values and generating an initial height map of the tear film surface; a spatial regularization unit connected with the height map construction unit, for applying spatial filtering to the initial height map to eliminate noise and artifacts; a parameterized curved surface conversion unit connected with the spatial regularization unit, for converting the discrete height map into a continuous parameterized curved surface representation; a time-varying curved surface tracking unit connected with the parameterized curved surface conversion unit, for tracking the change of the parameterized curved surface over time and constructing a time-space curved surface model. 4.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, characterized in that, The curvature feature extraction module comprises: a discrete curvature calculation unit for applying a discrete differential operator to the parameterized curved surface to calculate the average curvature and Gaussian curvature distribution of the tear film surface; a feature point identification unit connected with the discrete curvature calculation unit, for extracting local extreme points and saddle points from the curvature field; a curvature change rate calculation unit connected with the discrete curvature calculation unit, for calculating the change rate of the curvature over time to form a change rate field representing the stability of the tear film. The feature trajectory analysis unit, connected with the feature point recognition unit, is configured to track the evolution trajectory of the feature points over time and identify unstable regions. 5.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, wherein, The multi-scale analysis module comprises: A scale space construction unit configured to construct a multi-level scale representation based on the curvature field of the tear film surface; A scale feature extraction unit, connected with the scale space construction unit, is configured to extract curvature features and region segmentation at each scale layer; A cross-scale association unit, connected with the scale feature extraction unit, is configured to establish a corresponding relationship between features at different scale layers to form a feature propagation graph; A scale stability analysis unit, connected with the cross-scale association unit, is configured to analyze the stability of the features in the scale space and identify abnormal regions consistent across multiple scales. 6.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, wherein The region connectivity evaluation module comprises: A region division unit configured to divide the tear film surface into multiple regions based on the curvature features; A connected graph construction unit, connected with the region division unit, is configured to calculate the similarity and connectivity strength between regions and construct a region connectivity graph; A connectivity change monitoring unit, connected with the connected graph construction unit, is configured to track the evolution of the connectivity graph over time and detect significant changes in edge weights; A fracture location prediction unit, connected with the connectivity change monitoring unit, is configured to identify region pairs with rapid connectivity decline and determine potential fracture locations. 7.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, wherein, The tear film rupture prediction module comprises: A multi-source feature fusion unit configured to integrate multi-scale feature vectors and region connectivity change data; A risk index calculation unit, connected with the multi-source feature fusion unit, is configured to calculate the rupture risk index of each point on the tear film surface; A rupture prediction unit, connected with the risk index calculation unit, is configured to predict the time and location of tear film rupture based on the risk distribution; A rupture mode classification unit, connected with the rupture prediction unit, is configured to classify the tear film rupture mode and associate it to different types of dry eye. 8.The image processing-based dry eye syndrome automatic diagnosis system according to claim 7, characterized in that, The rupture mode classification unit classifies the tear film rupture mode based on the area, shape and diffusion rate of the rupture region, including three basic modes: punctate rupture, regional rupture and fusion rupture. 9.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, wherein, The diagnostic report generation module comprises: A tear film rupture time calculation unit configured to calculate a standardized tear film rupture time indicator based on the predicted tear film rupture event; A dry eye classification unit, connected with the tear film rupture time calculation unit, is configured to classify dry eye according to the tear film rupture time and rupture mode; A visualization result generation unit, connected with the dry eye classification unit, is configured to generate visualization data including 2D / 3D charts, heat maps and time series graphs; A treatment recommendation generation unit, connected with the dry eye classification unit, is configured to provide personalized treatment recommendations based on the dry eye classification and rupture characteristics. 10.The image processing-based dry eye syndrome automatic diagnosis system according to claim 1, wherein, Further comprising: A result correction module, connected with the tear film rupture prediction module and the diagnostic report generation module, is configured to receive tear film rupture prediction results from multiple examinations, analyze the consistency of the results, correct abnormal prediction results and improve the reliability of the diagnosis; A historical data management module, connected with the diagnostic report generation module, is configured to store the patient's previous examination results, support pre-and post-treatment comparative analysis and evaluate the treatment effect.

Images