Revolutionizing Clinical Diagnosis: LLMs Outperform Neurologists in Generalizable Multimodal Reasoning

This research introduces an incredibly exciting advancement in clinical AI by seamlessly translating complex, fragmented electronic health records into natural language for Large Language Models (LLMs). By utilizing a Multimodal framework that combines tabular data with MRI scans, the system achieves Zero-shot transfer capabilities without the need for manual feature engineering. Most impressively, this innovative approach significantly outperformed board-certified neurologists in retrospective dementia diagnosis, showcasing the immense Scalability of AI in real-world healthcare.