eess.IVNov 10, 2026

Counterfactual Analysis of Extent Leakage from Annotation-Derived Prompts in Vision-Language Medical Image Segmentation

Authors: Niran Nataraj, Ankita Das, C. Krishna Mohan, Kenji Kawashima

Organizations: Department of Information Physics and Computing, Graduate School of Information Science and Technology, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan · Department of Computer Science and Engineering, Indian Institute of Technology Hyderabad, Kandi, Sangareddy, Telangana, 502284, India

Abstract

Text-guided segmentation models receive an image and a short caption about the lesion, and on public benchmarks the caption improves accuracy considerably. On colonoscopy and breast ultrasound, however, the size, count and location words of each caption were computed from the mask that later scores the prediction, and the chest X-ray captions were written by benchmark annotators. We tested what two published models take from them. On colonoscopy and chest X-ray, it is mostly the lesion's extent. Keeping only the size word or the lung-zone phrase recovers 84 to 98 percent of the improvement over an empty caption, and removing it loses 52 to 77 percent. The rest matters little, and a polyp called a rock is segmented just as well. We call this extent leakage. On breast ultrasound, nearly half the captions also name the diagnosis, which carries part of the benefit. Retraining on three datasets shows that models pick up this habit from their training captions, though only partly on breast ultrasound, and that it weakens their segmentation from the image alone. The leak can be measured without training a model, since caption words alone predict lesion area (coefficient of determination 0.54 to 0.77 on six of seven test sets). Without the test mask, the best method we tested, which estimates extent from the image, reaches 74.59 Dice on colonoscopy against the published 77.23. We propose an evaluation protocol that keeps what a caption says about the image apart from what it copies from the label.

Figures & tables

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Nov 8, 2026

Seeing Is Not Believing: Counterfactual Lesion Erasure Teaches a Small Medical Vision-Language Agent to Report Only What the Image Shows

Medical vision-language models (VLMs) often report a finding because the question names it; standard benchmarks miss this. From public segmentation datasets we build counterfactual views without manual annotation: the lesion erased by inpainting, a same-shape control edit elsewhere, and unedited lesion-free images. The evidence sensitivity gap (ESG) is the rate of absent-or-refer answers after erasure minus the rate after the control edit; it is zero for a model that ignores the image or reacts to both edits alike. MedScope-4B (Qwen3.5-4B, LoRA, image tools) was trained on label-derived trajectories, selected by a pre-registered rule and tested once against nine frontier and six open VLMs; four pre-registered checks then probed the mechanism. On 507 internal cases across five imaging types, MedScope-4B reached ESG 0.758 and claimed an erased lesion in 9.6% of cases, against 0.27-0.47 and 41-62% for the frontier models (all p<0.001), without significant loss of accuracy on real lesions. Without counterfactual views ESG fell to 0.279+-0.107 (three seeds); the effect replicated on MedGemma 1.5 4B and two unseen inpainting tools. At four external centres the advantage held but frontier models detected more real lesions; on a fresh external test MedScope-4B detected 0 of 48 pneumothorax films, with net utility below gemini-4-argon's (-0.36 [-0.50, -0.23]). Matching the control edit was not shown to matter; over three seeds no fresh-external ESG advantage over removal-based DPO was established; a paired-view consistency objective did not beat a matched-compute control, while that continuation alone cut erased-lesion claims on unseen tools at every seed, at a cost in sensitivity at one. Training on erased and edited images gives a small model an evidence-gated disposition that prompting does not give frontier models; it does not improve perception, and rare critical findings need coverage.
May 18, 2026

Rad-VLSM: A Cross-Modal Framework with Semantics-Assisted Prompting for Medical Segmentation and Diagnosis

Medical image segmentation is more clinically valuable when it supports diagnosis rather than merely producing lesion masks. However, diagnostically relevant lesion cues are often subtle and localized, while existing models may be distracted by background tissues, acoustic artifacts, and irrelevant visual correlations. To address this problem, we propose Rad-VLSM, a two-stage cross-modal framework for semantics-assisted lesion focusing, robust segmentation, and visually grounded diagnosis. In the first stage, a BLIP-2-based vision-language alignment module identifies lesion-related candidate regions under semantic guidance and converts them into box prompts. In the second stage, these prompts are fed into a SAM-based multitask network, where a multi-candidate region aggregation strategy improves prompt stability and guides lesion segmentation. The predicted masks are then used as spatial priors for diagnosis, and a visual-radiomics fusion head integrates lesion-aware visual features with selected radiomics descriptors. By using semantic information for localization rather than direct prediction, Rad-VLSM reduces text-to-diagnosis dependence and grounds diagnosis in lesion-level evidence. Experiments on a private clinical breast ultrasound dataset and public benchmarks show that Rad-VLSM achieves strong segmentation and diagnostic performance with favorable generalization.
Oct 11, 2026

Lesion-Aware Attention for Mammography Segmentation with Heterogeneous Supervision

Accurate lesion segmentation in mammography is important for computer-aided breast cancer diagnosis, but dense pixel-level annotations remain costly and limited. Existing public segmentation datasets usually cover only a small set of lesion types with manually delineated masks, whereas larger mammography datasets often provide only image-level diagnostic labels. This annotation mismatch makes it difficult to train segmentation models that can generalize to rare or weakly annotated lesion categories. To address this problem, we study open-vocabulary mammographic lesion segmentation under a mixed-supervision setting, where pixel-level masks are available for base lesion categories and image-level labels are available for a broader set of categories. We propose a two-branch framework that jointly learns from segmentation data and classification data. The segmentation branch learns lesion boundaries from annotated masks, while the classification branch introduces additional image-level semantic supervision. To better transfer image-level supervision to lesion-relevant regions, we further introduce a Lesion-wise Attentive Classifier that maintains category-specific feature prototypes and generates lesion-aware attention maps for foreground feature pooling. This design reduces the influence of background tissue and encourages more compact lesion representations. Experiments on multiple mammography datasets show that the proposed method consistently improves open-vocabulary segmentation baselines on both base and novel lesion categories, while reducing the gap to fully supervised models that require dense target-domain masks. Qualitative results further indicate that our method produces more complete lesion localization for small, low-contrast, and visually subtle mammographic findings.