Deterministic Hallucination Detection in Medical VQA via Confidence-Evidence Bayesian Gain

Mohammad Asadi; Tahoura Nedaee; Jack W. O'Sullivan; Euan Ashley; Ehsan Adeli

cs.AIMar 23, 2026

Deterministic Hallucination Detection in Medical VQA via Confidence-Evidence Bayesian Gain

Authors: Mohammad Asadi, Tahoura Nedaee, Jack W. O'Sullivan, Euan Ashley, Ehsan Adeli

Organizations: Department of Electrical Engineering, Stanford University, CA, USA · Department of Biology, Stanford University, CA, USA · Division of Cardiology, Department of Medicine, Stanford University, CA, USA · Department of Biomedical Data Science, Stanford University, CA, USA ·

Abstract

Multimodal large language models (MLLMs) have shown strong potential for medical Visual Question Answering (VQA), yet they remain prone to hallucinations, defined as generating responses that contradict the input image, posing serious risks in clinical settings. Current hallucination detection methods, such as Semantic Entropy (SE) and Vision-Amplified Semantic Entropy (VASE), require 10 to 20 stochastic generations per sample together with an external natural language inference model for semantic clustering, making them computationally expensive and difficult to deploy in practice. We observe that hallucinated responses exhibit a distinctive signature directly in the model's own log-probabilities: inconsistent token-level confidence and weak sensitivity to visual evidence. Based on this observation, we propose Confidence-Evidence Bayesian Gain (CEBaG), a deterministic hallucination detection method that requires no stochastic sampling, no external models, and no task-specific hyperparameters. CEBaG combines two complementary signals: token-level predictive variance, which captures inconsistent confidence across response tokens, and evidence magnitude, which measures how much the image shifts per-token predictions relative to text-only inference. Evaluated across four medical MLLMs and three VQA benchmarks (16 experimental settings), CEBaG achieves the highest AUC in 13 of 16 settings and improves over VASE by 8 AUC points on average, while being fully deterministic and self-contained. https://github.com/masadi-99/CEBaG

Figures & tables

Figure 1: Overview of CEBaG for hallucination detection in medical VQA. (a) The CEBaG pipeline: the model generates answer

r

, scored with and without the image to yield two signals: token-level predictive variance

\sigma

and evidence magnitude

E{=}|G|/L

. CEBaG combines these as

\sigma\cdot(1{+}E)

. (b) Per-token probabilities (

100\,P(\text{token})

) with vs. without the image, and CEBaG scores for a grounded and a hallucinated example.

MedGemma-4b-it [ 21 ]
	AvgProb [ 10 ]		SE [ 4 ]		VASE [ 16 ]		RadFlag [ 22 ]		CEBaG (Ours)
Dataset	AUC	AUG	AUC	AUG	AUC	AUG	AUC	AUG	AUC	AUG
VQA-RAD (Open)	39.9	48.0	61.9	65.3	60.8	63.3	58.4	63.6	61.0	64.8
VQA-RAD (All)	42.9	54.6	52.2	58.0	57.6	59.9	54.7	58.9	60.8	64.6
SLAKE	35.3	52.9	54.9	64.2	62.6	70.1	57.0	66.2	71.9	75.9
PathVQA	46.4	45.1	58.0	48.9	61.5	49.6	56.8	47.4	71.9	59.0

Table 1: Hallucination detection performance: AUC (%)

\uparrow

and AUG (%)

\uparrow

across four medical MLLMs and three VQA datasets. Best results per row are in bold ; second best are underlined . CEBaG achieves the best AUC in 13 of 16 settings.

Method	Forward passes	External model	Deterministic	Hyperparams	$\overline{AUC}$	$\overline{AUG}$
AvgProb [ 10 ]	1 gen.	None	Yes	None	51.3	47.6
SE [ 4 ]	$M$ gen. + NLI	DeBERTa-MNLI	No	$M$ , $T$	57.4	51.2
VASE [ 16 ]	$2M$ gen. + NLI	DeBERTa-MNLI	No	$M$ , $T$ , $\alpha$	59.6	52.4
RadFlag [ 22 ]	$M$ gen. + NLI	DeBERTa-MNLI	No	$M$ , $T$	57.8	50.6
CEBaG (Ours)	1 gen. + 2 evals	None	Yes	None	67.9	57.2

Table 2: Computational cost and performance. Despite an order of magnitude fewer forward passes, CEBaG achieves the highest average AUC and AUG.

MedGemma-4b-it
	AUC (%)				AUG (%)
Setting	$\sigma$	$E$	CEBaG	CEBaG λ	$\sigma$	$E$	CEBaG	CEBaG λ
SLAKE	71.1	51.6	71.9	73.5	74.8	64.9	75.9	74.6
PathVQA	67.8	63.4	71.9	77.5	55.6	57.0	59.0	61.7
MedGemma-1.5-4b-it
VQA-RAD (Open)	68.2	63.9	72.9	75.7	44.7	40.8	47.2	47.0

Table 3: Ablation study: AUC (%)

\uparrow

of CEBaG components.

\sigma

only uses token-level variance alone;

E

only uses absolute per-token gain alone; CEBaG combines both (parameter-free); CEBaG λ adds a tunable weight (upper bound).

Figure 2: Sensitivity to ground-truth threshold. CEBaG maintains high AUC across all GREEN thresholds (0.3–1.0), whereas VASE and SE degrade significantly at lower thresholds. This indicates CEBaG is robust to the strictness of the hallucination definition.

Model	Corr(Real, Noisy)	Corr(Real, Text-only)
HuatuoGPT-Vision-7B	0.995	0.974
LLaVA-Med-v1.5-Mistral-7B	0.999	0.957
MedGemma-1.5-4b-it	0.998	0.975
MedGemma-4b-it	0.938	0.777

Table 4: Correlation of model log-probabilities: Real image vs. Noisy image (VASE protocol) and Real image vs. Text-only (CEBaG protocol). High correlation between real and noisy images suggests that noise injection (as used in VASE) may not sufficiently disrupt the model’s visual prior, limiting the contrastive signal. Text-only priors provide a more distinct reference point.

Hallucination Detection in VLMs

Hallucination detection in vision-language models identifies claims that are unsupported by images, video, or other visual evidence.

See all 72 papers

VIHD: Visual Intervention-based Hallucination Detection for Medical Visual Question Answering
May 20, 2026Jiayi Chen, Benteng Ma, Zehui Liao +3Hallucination Detection in VLMs Healthcare
Department of Data Science & AI, Faculty of Information Technology, Monash University, Melbourne, VIC 3800, Australia · Hong Kong Polytechnic University, Hong Kong SAR, China · Alfred Health Radiology, Alfred Health, Melbourne, VIC 3004, Australia +1
PROVE: Proof-guided Regime-aware Operator Verification for Hallucination Detection in Medical Visual Question Answering
Aug 28, 2026Keyang Zhou, Siyi Li, Zhongnan Shi +3Hallucination Detection in VLMs Medical VQA
Fudan University Shanghai, Shanghai, China · Fudan University Shanghai Cancer Center Shanghai, Shanghai, China
Hallucination Detection and Correction in Medical VLMs via Counter-Evidence Verification
Jun 17, 2026Nan Zhou, Ke Zou, Meng Liu +5Hallucination Detection in VLMs Vision-Language Models
the College of Computer Science, Sichuan University. · Yong Loo Lin School of Medicine, National University of Singapore. · National Key Laboratory of Autonomous Intelligent Unmanned Systems, Beijing2026 Institute of Technology. +2

May 20, 2026J/K move · Enter open · S save

VIHD: Visual Intervention-based Hallucination Detection for Medical Visual Question Answering

Jiayi Chen, Benteng Ma, Zehui Liao, Winston Chong+2

Department of Data Science & AI, Faculty of Information Technology, Monash University, Melbourne, VIC 3800, Australia · Hong Kong Polytechnic University, Hong Kong SAR, China · Alfred Health Radiology, Alfred Health, Melbourne, VIC 3004, Australia+1

Hallucination Detection in VLMs Healthcare Hallucination Detection Hallucination in Language Models VLM Hallucination Medical VQA+2

Open paper AskPDF arXiv

While medical Multimodal Large Language Models (MLLMs) have shown promise in assisting diagnosis, they still frequently generate hallucinated responses that appear linguistically plausible but lack visual evidence. Such hallucinations pose risks to clinical decision-making and necessitate effective detection. Existing introspective detection methods primarily perform uncertainty estimation or logical verification by analyzing model responses conditioned on original or perturbed inputs. However, such external perturbations are often heuristic and context-agnostic, which overlooks the internal cross-modal dependency between generated tokens and related visual tokens during decoding. To address this issue, we propose VIHD, a Visual Intervention-based Hallucination Detection method that leverages targeted visual token masking to calibrate semantic entropy for more effective hallucination detection. VIHD locates visually dominant decoder layers via Visual Dependency Probing (VDP), executes Visual Intervention Decoding (VID) via token masking to calibrate the semantic distribution, and quantifies the resulting Calibrated Semantic Entropy (CSE) as a reliable hallucination signal. Extensive experiments on three medical VQA benchmarks with two medical MLLMs demonstrate that VIHD consistently outperforms state-of-the-art methods, underscoring the importance of fine-grained visual dependency for hallucination detection. The code will be available at https://github.com/Jiayi-Chen-AU/VIHD

Abstract

Hallucination Detection in VLMs

Explore similar work

VIHD: Visual Intervention-based Hallucination Detection for Medical Visual Question Answering

PROVE: Proof-guided Regime-aware Operator Verification for Hallucination Detection in Medical Visual Question Answering

Hallucination Detection and Correction in Medical VLMs via Counter-Evidence Verification