MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641100697 A) filed by Jeppiaar Institute Of Technology on August 20, 2026, for A Multimodal Explainable Artificial Intelligence System For Clinical Decision Support, Diagnostic Re.
Inventors include Akshaya R; Sangannithra K; Saranya S; Gokul Priyan V; Rahul Krishna R; and Sathya Chezhiyan S.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: 6. ABSTRACT OF THE INVENTION (to be given along with complete specification on separate page) Existing artificial intelligence tools in clinical diagnostics are largely confined to single-modality report generation and treat inconsistency detection, confidence estimation, and evidence retrieval as separate, independently engineered components, none of which reason jointly over the full body of a patient's multimodal record. The present invention discloses a multimodal, explainable artificial intelligence system whose core inventive mechanism is a Cross-Modal Clinical Evidence Graph (CCEG), a unified graph-theoretic data structure that represents magnetic resonance imaging (MRI) findings, corn puted tomography (CT) findings, blood investigation values, prescribed medications, and diagnostic statements as interconnected nodes joined by weighted, typed relational edges denoting support, contradiction, or dependency. An Adaptive Evidence Weighting mechanism dynamically updates these edge weights based on source reliability, cross-modal corroboration, recency, and concordance with retrieved clinical guideline content. Clinically significant contradictions are identified through structured graph traversal of the CCEG, while differential diagnoses are ranked and assigned calibrated confidence scores through weighted path aggregation and propagation across the graph, rather than through uncorroborated model-internal token probability. Because provenance is retained natively at every graph node and edge, every generated diagnosis and flagged inconsistency remains traceable to its originating source passage, image finding, or laboratory value. These outputs are synthesized into a structured, clinician-friendly report suitable for direct clinical review. By unifying inconsistency detection, confidence estimation, and evidence linking as properties of a single evolving evidence graph rather than as disconnected modules, the disclosed invention
Disclaimer: Curated by HT Syndication.
