Confidence Calibration for Multimodal LLMs: An Empirical Study through Medical VQA

2026-06-18

AI Summary

This study addresses the critical issue of miscalibration—where confidence diverges from accuracy—in medical multimodal large language models (MLLMs) performing visual question answering (VQA), a problem that poses significant risks of misdiagnosis. The authors present the first systematic analysis of this phenomenon and propose a Multi-Strategy Fusion-Based Interrogation (MS-FBI) framework integrated with auxiliary expert large language models to recalibrate model confidence. By synthesizing diverse questioning strategies and leveraging external expert evaluation, the method effectively aligns predicted confidence with true accuracy. Evaluated on three medical VQA benchmarks, the approach reduces Expected Calibration Error (ECE) by 40% on average, substantially enhancing model reliability and offering a novel pathway toward trustworthy medical AI systems.

Abstract

Multimodal Large Language Models (MLLMs) show great potential in medical tasks, but their elicited confidence often misaligns with actual accuracy, potentially leading to misdiagnosis or overlooking correct advice. This study presents the first comprehensive analysis of the relationship between accuracy and confidence in medical MLLMs. It proposes a novel method that combines Multi-Strategy Fusion-Based Interrogation (MS-FBI) with auxiliary expert LLM assessment, aiming to improve confidence calibration in Medical Visual Question Answering (VQA). Experiments demonstrate that our method reduces the Expected Calibration Error (ECE) by an average of 40\% across three Medical VQA datasets, significantly enhancing MLLMs' reliability. The findings highlight the importance of domain-specific calibration for MLLMs in healthcare, offering a more trustworthy solution for AI-assisted diagnosis.

Problem

• Confidence Calibration
• Multimodal LLMs
• Medical VQA
• Expected Calibration Error
• AI-assisted Diagnosis

Innovation

• Confidence Calibration
• Multimodal LLMs
• Medical VQA
• MS-FBI
• Expected Calibration Error