@inproceedings{mahapatra-etal-2026-lens,
title = "{LENS}: Exploiting the Cross-View Stability Gap for Hallucination Mitigation in Medical Vision{--}Language Models",
author = "Mahapatra, Dwarikanath and
Das, Abhijit and
Roy, Sudipta and
Bozorgtabar, Behzad and
Razzak, Imran",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.88/",
pages = "2071--2081",
ISBN = "979-8-89176-459-0",
abstract = "Medical VLMs frequently hallucinate clinically unsupported findings. We identify a \textit{stability gap}: visually grounded tokens maintain relatively stable logits under semantics-preserving image perturbations, whereas visually misgrounded tokens exhibit substantially larger shifts. We introduce \textbf{LENS} (\textbf{L}ogit-\textbf{EN}hanced \textbf{S}tability), which exploits this signal at both training and inference. During training, \textit{Semantics-Preserving Preference Optimization} (SPPO) scores preference pairs using clinical faithfulness and cross-view stability; at inference, stability-aware contrastive decoding penalizes tokens whose logits diverge across original and counterfactual views. Constrained counterfactuals use attribution-guided masking and feature-space proximity guards to reduce disruption of lesion- and device-level evidence. Across radiology VQA, chest X-ray reporting, and gastrointestinal endoscopy, LENS consistently outperforms strong decoding-time and preference-optimization baselines. On MIMIC-CXR, entity-level RadGraph recall improves from 21.0 to 22.8 over PCCD, while expert evaluation shows a hallucination-rate reduction from 17.0{\%} to 8.5{\%}."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mahapatra-etal-2026-lens">
<titleInfo>
<title>LENS: Exploiting the Cross-View Stability Gap for Hallucination Mitigation in Medical Vision–Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dwarikanath</namePart>
<namePart type="family">Mahapatra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abhijit</namePart>
<namePart type="family">Das</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sudipta</namePart>
<namePart type="family">Roy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Behzad</namePart>
<namePart type="family">Bozorgtabar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Imran</namePart>
<namePart type="family">Razzak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Medical VLMs frequently hallucinate clinically unsupported findings. We identify a stability gap: visually grounded tokens maintain relatively stable logits under semantics-preserving image perturbations, whereas visually misgrounded tokens exhibit substantially larger shifts. We introduce LENS (Logit-ENhanced Stability), which exploits this signal at both training and inference. During training, Semantics-Preserving Preference Optimization (SPPO) scores preference pairs using clinical faithfulness and cross-view stability; at inference, stability-aware contrastive decoding penalizes tokens whose logits diverge across original and counterfactual views. Constrained counterfactuals use attribution-guided masking and feature-space proximity guards to reduce disruption of lesion- and device-level evidence. Across radiology VQA, chest X-ray reporting, and gastrointestinal endoscopy, LENS consistently outperforms strong decoding-time and preference-optimization baselines. On MIMIC-CXR, entity-level RadGraph recall improves from 21.0 to 22.8 over PCCD, while expert evaluation shows a hallucination-rate reduction from 17.0% to 8.5%.</abstract>
<identifier type="citekey">mahapatra-etal-2026-lens</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.88/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2071</start>
<end>2081</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T LENS: Exploiting the Cross-View Stability Gap for Hallucination Mitigation in Medical Vision–Language Models
%A Mahapatra, Dwarikanath
%A Das, Abhijit
%A Roy, Sudipta
%A Bozorgtabar, Behzad
%A Razzak, Imran
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F mahapatra-etal-2026-lens
%X Medical VLMs frequently hallucinate clinically unsupported findings. We identify a stability gap: visually grounded tokens maintain relatively stable logits under semantics-preserving image perturbations, whereas visually misgrounded tokens exhibit substantially larger shifts. We introduce LENS (Logit-ENhanced Stability), which exploits this signal at both training and inference. During training, Semantics-Preserving Preference Optimization (SPPO) scores preference pairs using clinical faithfulness and cross-view stability; at inference, stability-aware contrastive decoding penalizes tokens whose logits diverge across original and counterfactual views. Constrained counterfactuals use attribution-guided masking and feature-space proximity guards to reduce disruption of lesion- and device-level evidence. Across radiology VQA, chest X-ray reporting, and gastrointestinal endoscopy, LENS consistently outperforms strong decoding-time and preference-optimization baselines. On MIMIC-CXR, entity-level RadGraph recall improves from 21.0 to 22.8 over PCCD, while expert evaluation shows a hallucination-rate reduction from 17.0% to 8.5%.
%U https://aclanthology.org/2026.emnlp-main.88/
%P 2071-2081
Markdown (Informal)
[LENS: Exploiting the Cross-View Stability Gap for Hallucination Mitigation in Medical Vision–Language Models](https://aclanthology.org/2026.emnlp-main.88/) (Mahapatra et al., EMNLP 2026)
ACL