MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641091959 A) filed by Sr University on July 29, 2026, for Corrective Multimodal Retrieval-Augmented Generation System With Dynamic Pubmed Evidence Integration For Medical Visual Question Answering.
Inventors include Dr. B. Omkar Lakshmi Jagan; Dr. Durgesh Nandan; and Dr S Satyanarayana Ph. D. , Pdfai.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: ABSTRACT [0010] The invention aims to provide an adaptive corrective retrieval mechanism that dynamically evaluates retrieval quality and supplements insufficient knowledge using real-time PubMed literature, thereby reducing hallucinations and improving clinical reliability. The invention employs a Self-RAG relevance assessment module that evaluates the quality of retrieved medical context. If the relevance score falls below a predefined threshold, a corrective retrieval loop automatically queries the PubMed database to obtain the latest peer-reviewed medical literature. The newly retrieved evidence is integrated with the original retrieval results before answer generation by the Vision-Language Model. This adaptive retrieval strategy improves factual accuracy, incorporates current medical knowledge, and substantially reduces clinical hallucinations in Medical Visual Question Answering systems.
Disclaimer: Curated by HT Syndication.
