AI-powered system for predicting and managing eye health risks in diabetes

Abstract

A system for predicting and managing eye health risks in diabetes, consisting of a data acquisition interface for receiving ophthalmic images and longitudinal health data of a diabetes patient, a feature extraction module for deriving imaging and clinical features, a trained risk assessment engine for classifying the patient into risk or stable categories and generating alerts according to predefined thresholds, and a communication module for providing personalized recommendations and notifications to patient and physician interfaces.

Description

Application area of ​​the invention

[0001] The invention relates to medical informatics systems that integrate data from diabetes patients and ophthalmic imaging with artificial intelligence to predict the risk of eye diseases and support proactive management via patient and physician interfaces. Background of the invention

[0002] Diabetic retinopathy and related eye complications are among the leading causes of preventable vision loss. Guidelines therefore recommend annual retinal examinations for people with diabetes to detect conditions requiring treatment early. AI systems that analyze retinal images enable autonomous screenings and can improve the performance of eye examinations. Teleophthalmology and multimodal risk models that consider blood glucose control, blood pressure, blood lipids, and lifestyle factors improve prognostic prediction. However, many of these systems are fragmented or require specialized infrastructure. There is therefore a need for an integrated system that captures longitudinal clinical and behavioral data, generates transparent risk assessments, provides personalized recommendations and alerts, and ensures data privacy and security for routine care. Summary of the invention

[0003] The invention provides an AI-supported system that includes interfaces for data acquisition for ophthalmic imaging and health data, a feature extraction and risk assessment engine using trained models, and communication modules for providing personalized recommendations and alerts to patients and healthcare professionals. The system classifies individuals into risk groups (e.g., high-risk patients vs. stable patients) and continuously updates the assessments as new data arrives. This creates a feedback loop that adapts recommendations for eye examinations, blood glucose control, and lifestyle interventions.

[0004] In its various implementations, the system employs privacy-preserving mechanisms such as on-device inference for mobile retinal imaging or federated learning; it supports physician oversight and audit protocols in accordance with healthcare AI guidelines; and it integrates into teleophthalmological workflows to expand access while maintaining quality. The architecture enables proactive eye protection through early warnings and personalized guidance. Detailed description

[0005] The system includes data interfaces for capturing multimodal input data: ophthalmological images (e.g., fundus photographs), structured health data including HbA1c value, blood pressure, lipid levels, medications and comorbidities, as well as patient-reported or wearable-captured activity and nutritional data. The data acquisition layer performs validation, anonymization where necessary, unit harmonization, and timestamp matching to generate longitudinal patient timelines for model input.

[0006] A feature extraction module preprocesses retinal images (e.g., illumination correction, vascular segmentation) and derives imaging biomarkers from them. Tabular pipelines calculate features from laboratory and vital parameter trends (e.g., HbA1c variability, blood pressure control, dyslipidemia) as well as behavioral metrics. These features feed trained models that estimate the risk of developing or progressing diabetic retinopathy or macular edema over a defined period and output calibrated probabilities and confidence intervals. The literature demonstrates improved prognosis through the combination of imaging and clinical risk factors.

[0007] The risk assessment engine enhances traceability by ranking the contribution of features and image areas to the outcome and providing thresholds for categorizing patients as high-risk or stable. To increase participation in preventive screenings, the system can trigger prompts for AI-assisted eye exams directly at the point of care or initiate referrals. Autonomous AI-assisted preventive screenings have been shown to increase completion rates among adolescents, suggesting benefits for interventions aimed at promoting participation.

[0008] A recommendation module assigns personalized measures to risk groups, such as examination intervals, blood glucose target values ​​in consultation with physicians, and informational materials on nutrition and exercise. Alerts are delivered via secure patient apps and physician dashboards. Escalation guidelines for high-risk findings and images of insufficient quality trigger repeat imaging or referral. The integration of teleophthalmology enables remote image acquisition and interpretation.

[0009] The system implements data protection and security measures: encryption during transmission and at rest, role-based access control, logging of processes, and consent management in accordance with healthcare data protection regulations. It supports human oversight and discloses the use of AI, data processing, and risks such as algorithmic bias. It thus complies with ethical checklists and regulations that restrict fully automated medical decisions without physician involvement in the EU.

[0010] Deployment options include clinic-based interfaces to electronic health records, mobile data collection for population screening, and cloud-based analytics with regional data storage. For model maintenance, the platform supports performance monitoring across different demographic groups, deviation detection, and approval processes for model updates.

[0011] For protection as a German utility model, the claims relate to the system device and its modules, and not to a pure method, since German utility model law provides exceptions for methods; however, product-related claims with functional limitations are permissible. The application requires a description and claims, as well as drawings where applicable; the formal examination leads to relatively quick registration.

Claims

[1] A system for predicting and managing eye health risks in diabetes, consisting of a data acquisition interface for receiving ophthalmic images and longitudinal health data of a diabetes patient, a feature extraction module for deriving imaging and clinical features, a trained risk assessment engine for classifying the patient into risk or stable categories and generating alerts according to predefined thresholds, and a communication module for providing personalized recommendations and notifications to patient and physician interfaces. [2] System according to claim 1, wherein the data acquisition interface is further configured to capture signals relating to physical activity and nutrition and the risk assessment engine integrates these signals with imaging and clinical features to update the risk assessments in a continuous feedback loop. [3] System according to claim 1, wherein the system implements data protection and security measures, including encryption during transmission and at rest, role-based access, logging of audit processes, consent management and human oversight of AI-supported decisions communicated to physicians and patients. [4] System according to claim 1, wherein the system is integrated into teleophthalmological workflows to trigger the scheduling of eye examinations, the repetition of imaging in case of insufficient quality or referrals, thereby increasing compliance with the recommended screening intervals.