MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096776 A) filed by J. J College Of Engineering And Technology on August 11, 2026, for Transformer-Based Artificial Intelligence Framework For Automated Rare Disease Identification Using Electronic Health Records And Medical Imaging.
Inventors include Dr I Shahanaz Begum; Jaison Vimalraj T; Dr S Murugesan; S Lakshmi Narasimhan; and Dr. K. Saravana Kumar.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The present invention relates to a transformer-based artificial intelligence framework for automated identification of rare-disease-associated patterns from heterogeneous electronic health record (EHR) information and medical imaging information. The framework comprises an EHR processing module, an EHR transformer module, a medical imaging processing module, an imaging transformer module, a multimodal fusion module, and an inference module. The EHR processing module receives patient-specific clinical information comprising structured and/or unstructured clinical events and generates temporally encoded representations. The EHR transformer module processes the representations using attention mechanisms to generate a contextualized EHR representation capturing relationships among longitudinal clinical events. In parallel, the medical imaging processing module receives one or more patient-specific medical images, performs preprocessing and spatial partitioning, and generates image representations. The imaging transformer module processes the image representations using attention mechanisms to generate a contextualized imaging representation capturing spatial relationships among image regions. The multimodal fusion module combines the EHR representation and imaging representation using attention-based fusion, including cross-attention, to generate a joint patient representation. An inference module processes the joint representation to generate one or more ranked disease-associated scores corresponding to rare-disease-associated categories. The framework may further provide confidence, uncertainty, and explanation-associated outputs identifying relevant clinical events and/or image regions contributing to the computational result. The architecture enables heterogeneous longitudinal clinical information and medical imaging information to be independently encoded and subsequently integrated, thereby providing a computer-implemented mechanism for prioritizing patients exhibiting patterns associated with rare diseases for further professional evaluation and confirmatory investigation.
Disclaimer: Curated by HT Syndication.
