EasyLens: A Training-Free Plug-and-Play Subtle-Lesion Representation Amplifier for Medical Vision-Language Models

NeurIPS, 2026

Hao Wang^1,2*, Qiwei Zeng^1*, Jinghao Lin^3*, Shuchang Ye², Yuezhe Yang^2,4, Yige Peng⁴, Haoyuan Che^1†, Jinman Kim^2†, Lei Bi^4†

^*Equal Contribution ^†Corresponding author

¹Jilin University, ²The University of Sydney, ³Northeastern University, ⁴Shanghai Jiao Tong University

Abstract

Medical vision-language models can miss subtle lesions because sparse, low-contrast cues are underrepresented when local visual tokens are aggregated. EasyLens is a training-free, plug-and-play representation amplifier for frozen medical VLMs. It constructs a pathology-anatomy prototype space, selects lesion-relevant patches using counterfactual prototype reasoning, and strengthens their representations through morphology-guided residual enhancement. Experiments on multiple medical image datasets and frozen VLM backbones show improved subtle-lesion detection without model fine-tuning.

Highlights

Build pathology-anatomy prototypes for comparing suspicious image patches.
Select lesion-relevant patches and amplify subtle morphological cues.
Improve subtle-lesion recognition without fine-tuning frozen medical VLMs.

Results

Quantitative Results

Model	ReX			LIDC			Abdomen
Model	Stat.	Sel.	Gen.	Stat.	Sel.	Gen.	Stat.	Sel.	Gen.
MedGemma1.5	42.86	23.33	4.41	20.45	27.82	41.93	15.12	49.04	38.18
EasyLens	66.67	31.11	5.15	30.30	36.09	45.86	52.33	55.77	40.67

Comparison on subtle-lesion status recognition, region selection, and report generation.

Qualitative Results

EasyLens case study on a small pulmonary nodule

Case study of subtle-lesion perception and lesion-aware report generation.

Citation

@article{wang2026easylens, title={EasyLens: A Training-Free Plug-and-Play Subtle-Lesion Representation Amplifier for Medical Vision-Language Models}, author={Wang, Hao and Zeng, Qiwei and Lin, Jinghao and Ye, Shuchang and Yang, Yuezhe and Peng, Yige and Che, Haoyuan and Kim, Jinman and Bi, Lei}, journal={Advances in Neural Information Processing Systems}, year={2026}, url={https://arxiv.org/abs/2606.06379} }