MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202641118837 A) filed by Lavanya Kanaparthi; and G. Madhukar Rao on October 05, 2026, for Lava-Vit: A Lesion-Aware Vision Transformer Based System And Method For Explainable And Uncertainty-Aware Diabetic Retinopathy Analysis.
Inventors include G. Madhukar Rao; and Lavanya Kanaparthi.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: A lesion-aware vision transformer based system and method for the explainable and uncertainty-aware analysis of diabetic retinopathy is disclosed. The system comprises an image preprocessing module that denoises and contrast-enhances an acquired retinal fundus or optical coherence tomography image; a lesion proposal network that generates a soft lesion probability map identifying candidate pathological regions; a vision transformer backbone that encodes the image as patch tokens; a lesion-aware attention module that re-weights the transformer's self-attention using the lesion probability map so that clinically relevant regions receive disproportionately greater attention while global retinal context is preserved; classification, regression and segmentation prediction heads that jointly output a disease- severity grade, a continuous severity score and a pixel-level lesion mask; a calibration and evidential uncertainty estimation module that produces calibrated confidence values and flags low-confidence predictions for expert review; a domain adaptation module that improves robustness across differing imaging sources; and an explainability and visualization module that generates a lesion-attention overlay derived intrinsically from the model's own attention weights. By jointly integrating lesion-aware attention, multi-task prediction, calibrated uncertainty estimation, cross-dataset domain adaptation and intrinsic explainability within a single architecture, the invention provides a clinically transparent, reliability-aware and broadly deployable diagnostic tool for diabetic-retinopathy screening.
Disclaimer: Curated by HT Syndication.
