Med-StepBench: A Hierarchical Reasoning Framework for Evaluating Hallucinations in Medical Vision-Language Models

TL;DR

Med-StepBench is a comprehensive benchmark for evaluating step-wise hallucination detection in 3D medical vision-language models, revealing critical limitations in current models' grounding and reasoning capabilities.

Contribution

It introduces the first large-scale, step-wise hallucination benchmark for 3D medical imaging, with clinician-verified annotations and detailed evaluation of model failure modes.

Findings

Current VLMs are highly susceptible to adversarial explanations.

Systematic failure modes are revealed through step-level evaluation.

Grounding multi-step reasoning remains a fundamental challenge.

Abstract

Large vision-language models (VLMs) demonstrate strong performance in medical image understanding, but frequently generate clinically plausible yet incorrect statements, raising significant safety concerns. Existing medical hallucination benchmarks primarily focus on 2D imaging with one-shot diagnostic questions, offering limited insight into whether predictions are grounded in correct localization and abnormality identification, allowing critical reasoning errors to remain hidden behind seemingly correct diagnoses. We introduce Med-StepBench, the first large-scale benchmark for step-wise hallucination detection in 3D oncological PET/CT, comprising over 12,000 images and more than 1,000,000 image-statement pairs across volumetric and multi-view 2D data, which decomposes clinical reasoning into four expert-designed diagnostic stages. Using clinician-verified annotations, we perform the first step-level evaluation of general-purpose and medical VLMs, revealing systematic failure modes obscured by aggregate accuracy metrics. Furthermore, we show that current VLMs are highly susceptible to adversarial yet clinically plausible intermediate explanations, which significantly amplify hallucinations despite contradictory visual evidence. Together, our findings highlight fundamental limitations in grounding multi-step clinical reasoning and establish Med-StepBench as a rigorous benchmark for developing safer and more reliable medical VLMs.