UMass-BioNLP at MEDIQA-M3G 2024: DermPrompt -- A Systematic Exploration of Prompt Engineering with GPT-4V for Dermatological Diagnosis

TL;DR

This work investigates diagnosing dermatological conditions from images and patient context using a GPT-4V–based retrieval–re-ranking pipeline augmented with a Multi-Agent Conversation framework. By comparing context-independent and context-dependent retrieval, Naive CoT and expert guidelines CoT prompts, and a MAC-based critique loop, the study demonstrates that context-rich prompts and agent-based critique substantially improve diagnostic accuracy. The results show strong retrieval performance around 85% with context, improved re-ranking when including clinical guidelines, and a notable MAC advantage over single-step reasoning, alongside an aligner that benefits from automatic prompt optimization. The findings have practical implications for scalable, peri-clinical dermatology workflows, education, and patient care, while highlighting privacy, cost, and data-quality challenges that warrant further research.

Abstract

This paper presents our team's participation in the MEDIQA-ClinicalNLP2024 shared task B. We present a novel approach to diagnosing clinical dermatology cases by integrating large multimodal models, specifically leveraging the capabilities of GPT-4V under a retriever and a re-ranker framework. Our investigation reveals that GPT-4V, when used as a retrieval agent, can accurately retrieve the correct skin condition 85% of the time using dermatological images and brief patient histories. Additionally, we empirically show that Naive Chain-of-Thought (CoT) works well for retrieval while Medical Guidelines Grounded CoT is required for accurate dermatological diagnosis. Further, we introduce a Multi-Agent Conversation (MAC) framework and show its superior performance and potential over the best CoT strategy. The experiments suggest that using naive CoT for retrieval and multi-agent conversation for critique-based diagnosis, GPT-4V can lead to an early and accurate diagnosis of dermatological conditions. The implications of this work extend to improving diagnostic workflows, supporting dermatological education, and enhancing patient care by providing a scalable, accessible, and accurate diagnostic tool.

Figures (2)

• Figure 1: Overview of the AI-assisted dermatology diagnosis pipeline, from initial patient input through to the GPT-4V generated final diagnosis and treatment plan.
• Figure 2: Multi-Agent Conversation (MAC) Setup