Intellectual Property India Publishes Patent Application for A Lightweight Vision-Based Framework For Automated Retinopathy Of Prematurity Screening From Multi-Device Retinal Fundus Images

MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641112205 A) filed by Priyadarshini R; Kanimozhi T; Sadhana K; and Sam V George on September 18, 2026, for A Lightweight Vision-Based Framework For Automated Retinopathy Of Prematurity Screening From Multi-Device Retinal Fundus Images.

Inventor includes Sam V George.

The application for the patent was published on September 25, 2026, under issue no. 39/2026.

Abstract: A lightweight vision-based framework is disclosed for automated screening assistance of Retinopathy of Prematurity (ROP) from retinal fundus images acquired using different imaging devices. The framework receives retinal fundus images and performs preprocessing including retinal or circular cropping, resizing, normalization, contrast enhancement, illumination correction, and suitable augmentation to obtain standardized image representations. A lightweight artificial intelligence-based feature extraction and enhancement approach is then applied to generate relevant retinal feature representations for ROP/Normal classification with an associated confidence or probability. In one embodiment, GhostNetV2 integrated with an SE-GEM feature enhancement mechanism is employed to provide computationally efficient analysis. The framework further incorporates Grad-CAM-based visual explanation to identify image regions contributing to the generated prediction. A Vision- Language Model (VLM) may receive the retinal image together with the classification output, confidence information, and visual evidence to generate a structured, human-readable screening interpretation. The framework may employ five-fold cross-validation and cross-device evaluation using images acquired from RetCam and Neo to assess reliability and generalization. The disclosed framework combines automated classification, visual explanation, and language-based interpretation into a unified screening-support workflow suitable for resource-constrained retinal image analysis and teleophthalmology applications.

Disclaimer: Curated by HT Syndication.