Hallucination Filtering in Radiology Vision-Language Models Using Discrete Semantic Entropy

2025-10-10

AI Summary

Black-box vision-language models (VLMs) frequently generate hallucinations in radiology visual question answering (VQA), compromising clinical reliability. Method: We propose a non-intrusive, question-level filtering method based on Discrete Semantic Entropy (DSE), which quantifies semantic inconsistency without accessing model internals. DSE is computed via high-temperature multi-sample generation followed by bidirectional entailment checking for semantic clustering, augmented by bootstrap resampling and confidence interval estimation. Contribution/Results: The approach is model-agnostic and clinically deployable. Evaluated on 706 radiology image–question pairs, filtering questions with DSE > 0.3 significantly improved accuracy: GPT-4o increased from 51.7% to 76.3%, and GPT-4.1 from 54.8% to 63.8%—both improvements statistically significant (p < 0.001). This demonstrates DSE’s effectiveness in mitigating hallucination-prone queries while preserving interpretability and practical utility in medical AI applications.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Natural Language Processing: Question Answering

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

To determine whether using discrete semantic entropy (DSE) to reject questions likely to generate hallucinations can improve the accuracy of black-box vision-language models (VLMs) in radiologic image based visual question answering (VQA). This retrospective study evaluated DSE using two publicly available, de-identified datasets: (i) the VQA-Med 2019 benchmark (500 images with clinical questions and short-text answers) and (ii) a diagnostic radiology dataset (206 cases: 60 computed tomography scans, 60 magnetic resonance images, 60 radiographs, 26 angiograms) with corresponding ground-truth diagnoses. GPT-4o and GPT-4.1 answered each question 15 times using a temperature of 1.0. Baseline accuracy was determined using low-temperature answers (temperature 0.1). Meaning-equivalent responses were grouped using bidirectional entailment checks, and DSE was computed from the relative frequencies of the resulting semantic clusters. Accuracy was recalculated after excluding questions with DSE > 0.6 or > 0.3. p-values and 95% confidence intervals were obtained using bootstrap resampling and a Bonferroni-corrected threshold of p < .004 for statistical significance. Across 706 image-question pairs, baseline accuracy was 51.7% for GPT-4o and 54.8% for GPT-4.1. After filtering out high-entropy questions (DSE > 0.3), accuracy on the remaining questions was 76.3% (retained questions: 334/706) for GPT-4o and 63.8% (retained questions: 499/706) for GPT-4.1 (both p < .001). Accuracy gains were observed across both datasets and largely remained statistically significant after Bonferroni correction. DSE enables reliable hallucination detection in black-box VLMs by quantifying semantic inconsistency. This method significantly improves diagnostic answer accuracy and offers a filtering strategy for clinical VLM applications.

Problem

• Detecting hallucinations in radiology vision-language models using semantic entropy
• Improving diagnostic accuracy of black-box VLMs in medical visual question answering
• Filtering unreliable questions to enhance clinical VLM application reliability

Innovation

• Uses discrete semantic entropy for hallucination detection
• Filters high-entropy questions to improve VLM accuracy
• Quantifies semantic inconsistency in black-box vision-language models