MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621074019 A) filed by Mrs. Shilpa Dileep Chindamwar; and Dr. Rajendra Rewatkar on June 15, 2026, for "method For Automated Diabetic Retinopathy Classification Using Retina Topology Encoding, Pathology Evolution Intelligence, Neuro-Symbolic Clinical Reasoning, Bayesian Digital Twin Validation, And Federated Continual Ophthalmic Learning'.
Inventors include Mrs. Shilpa Dileep Chindamwar; and Dr. Rajendra Rewatkar.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The present invention relates to an intelligent computer-implemented system and method for automated diabetic retinopathy classification, disease progression assessment, clinical reasoning validation, and adaptive ophthalmic intelligence generation using retinal fundus images. More particularly, the invention introduces a multi-stage integrated framework that combines topological lesion representation, disease evolution learning, neuro-symbolic clinical reasoning, digital twin validation, and federated continual learning to improve the accuracy, reliability, explainability, and scalability of diabetic retinopathy diagnosis. The proposed system initially acquires retinal fundus images and performs lesion localization, vascular structure extraction, optic disc identification, macular region analysis, and pathological feature detection. The extracted anatomical and pathological entities are transformed into a heterogeneous retinal topology graph through a RetinoGraph Topological Lesion Encoding Network (RTLEN), wherein lesion interactions, vascular connectivity, and anatomical relationships are represented as graph-based disease structures. The generated retinal topology representation is subsequently processed by a Pathology Evolution Transformer Memory Network (PETMN), which learns latent disease progression patterns and severity transition characteristics through a memory-enhanced attention mechanism. The resulting disease evolution embeddings are further analyzed by a Neuro-Symbolic Clinical Reasoning Engine (NSCRE), wherein artificial intelligence predictions are validated against a structured ophthalmic knowledge graph containing clinical diagnostic rules, lesion severity thresholds, and disease progression criteria to generate clinically interpretable diagnostic outcomes. The invention further incorporates a Bayesian Digital Twin Retinal Validation System (BDTRVS) that constructs patient-specific virtual retinal replicas and performs uncertainty estimation, robustness evaluation, pathological perturbation analysis, and disease simulation under varying clinical scenarios. The validated outputs are subsequently processed through a Federated Continual Ophthalmic Intelligence Network (FCOIN), which enables privacy-preserving collaborative learning across distributed healthcare institutions while simultaneously detecting imaging drift, demographic variation, and disease distribution shifts. The framework continuously updates diagnostic intelligence without requiring centralized patient data sharing. The invention provides enhanced diabetic retinopathy classification accuracy, improved lesion sensitivity, clinically explainable predictions, robust uncertainty quantification, adaptive disease progression forecasting, and secure multi-institutional deployment, thereby facilitating next-generation intelligent ophthalmic screening, diagnosis, and decision-support systems.
Disclaimer: Curated by HT Syndication.
