MedHEval: Benchmarking Hallucinations and Mitigation Strategies in Medical Large Vision-Language Models

2025-03-04

AI Summary

Medical large vision-language models (Med-LVLMs) suffer from pervasive hallucination, primarily stemming from visual misinterpretation, knowledge gaps, and context mismatch—yet existing benchmarks lack etiology-oriented fine-grained evaluation and systematic validation of mitigation strategies. To address this, we introduce the first hallucination-focused benchmark for Med-LVLMs, proposing a causality-driven taxonomy categorizing hallucinations into three types: visual, knowledge, and context-related. We conduct the first comprehensive evaluation across 11 state-of-the-art Med-LVLMs and 7 representative mitigation techniques. Additionally, we release a multimodal medical visual question answering dataset and a hybrid automatic–human evaluation framework. Empirical results reveal severe deficiencies in current models’ handling of knowledge- and context-related hallucinations, while mainstream mitigation methods exhibit poor generalizability across model architectures and hallucination types. This work establishes a foundational, causality-guided evaluation paradigm and provides open-source resources to advance domain-specific hallucination mitigation.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Vision Language Models (LVLMs) are becoming increasingly important in the medical domain, yet Medical LVLMs (Med-LVLMs) frequently generate hallucinations due to limited expertise and the complexity of medical applications. Existing benchmarks fail to effectively evaluate hallucinations based on their underlying causes and lack assessments of mitigation strategies. To address this gap, we introduce MedHEval, a novel benchmark that systematically evaluates hallucinations and mitigation strategies in Med-LVLMs by categorizing them into three underlying causes: visual misinterpretation, knowledge deficiency, and context misalignment. We construct a diverse set of close- and open-ended medical VQA datasets with comprehensive evaluation metrics to assess these hallucination types. We conduct extensive experiments across 11 popular (Med)-LVLMs and evaluate 7 state-of-the-art hallucination mitigation techniques. Results reveal that Med-LVLMs struggle with hallucinations arising from different causes while existing mitigation methods show limited effectiveness, especially for knowledge- and context-based errors. These findings underscore the need for improved alignment training and specialized mitigation strategies to enhance Med-LVLMs' reliability. MedHEval establishes a standardized framework for evaluating and mitigating medical hallucinations, guiding the development of more trustworthy Med-LVLMs.

Problem

• Evaluates hallucinations in Medical Large Vision-Language Models (Med-LVLMs)
• Assesses mitigation strategies for visual, knowledge, and context errors
• Introduces MedHEval benchmark for systematic evaluation and improvement

Innovation

• Introduces MedHEval benchmark for medical hallucinations
• Categorizes hallucinations into three underlying causes
• Evaluates 7 state-of-the-art hallucination mitigation techniques