Hidden flaws behind expert-level accuracy of multimodal GPT-4 vision in medicine

TL;DR

This study critically examines GPT-4V's medical image reasoning, revealing high accuracy but frequent flawed rationales, highlighting the need for thorough evaluation before clinical use.

Contribution

It provides a comprehensive analysis of GPT-4V's reasoning, knowledge recall, and diagnostic steps in medical imaging, uncovering flaws behind its expert-level accuracy.

Findings

GPT-4V achieves 81.6% accuracy on NEJM Image Challenges.

Over 78% accuracy in cases where physicians answered incorrectly.

35.5% of GPT-4V's correct answers are based on flawed rationales.

Abstract

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges - an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.