One Model Fits All? Rethinking Medical AI Chatbots with Specialized Retrieval

Abstract

Medical diagnostics and treatment planning are increasingly complex, demanding sophisticated tools for healthcare professionals. AI chatbots in healthcare often use generalized models, leading to suboptimal predictions, especially with multimodal data. This research introduces a multimodal AI chatbot with specialized model retrieval for customized medical insights. The system selects relevant models for disease prediction, diagnostics, or drug interactions based on input type, integrating EHR data via FHIR APIs and using UMLS and SNOMED CT. Compared to generalized systems, preliminary tests show improved accuracy in diagnosis, treatment planning, and drug safety. This approach enhances diagnostic precision, reduces medication errors, and personalizes patient care.

Ähnliche Arbeiten