Evaluating the multimodal capabilities of ChatGPT for kidney disease diagnosis: integrating text and medical images

Abstract

ChatGPT is advancing nephrology diagnostics as a multimodal AI chatbot capable of processing both textual data and medical images, such as CT, MRI, and ultrasound scans. This study compares its response accuracy with healthcare professionals across 13 kidney disease (KD) -related text and image-based queries to explore its potential in enhancing health literacy, supporting clinical decision-making, and complementing diagnostic tools. The study achieves an average similarity score of 16.0%. The results highlight challenges such as query complexity, phrasing inconsistencies, the need for repeated refinements, response limitations, and network issues, all of which hinder clinical implementation. Integrating patient history, medical records, and retrieval-augmented generation (RAG) could further enhance accuracy and engagement.

Ähnliche Arbeiten