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

2026-06-06 · Source: cs.AI updates on arXiv.org · Field: Health & Wellbeing — Health & Medical Research, Medical Devices & Health Technology, Artificial Intelligence & Machine Learning · Depth: Expert, extended

Summary

EasyLens is a training-free, plug-and-play subtle-lesion representation amplifier designed for medical vision-language models (VLMs). It addresses the limited sensitivity of frozen medical VLMs to subtle lesions, which are often sparse, low-contrast, and underrepresented in global image embeddings. EasyLens comprises three components: EasyBank, a pathology-anatomy prototype space; EasyTag, a counterfactual prototype reasoning module for selecting lesion-relevant patches; and EasyAmplifier, which strengthens these selected patch representations through morphology-guided residual enhancement. Experiments on ReXGroundingCT, LIDC-IDRI, and AbdomenAtlas 3.0 Mini, using frozen MedGemma1.5, LLaVA-Med, RadFM, Lingshu, and MedGemma backbones, demonstrate consistent improvements in subtle-lesion detection, region selection, and report generation without model fine-tuning.

Key takeaway

For AI Scientists and Machine Learning Engineers working with medical VLMs, EasyLens offers a critical solution to improve subtle lesion detection without costly retraining. If your current frozen models struggle with sparse, low-contrast abnormalities, you should consider integrating this training-free, plug-and-play amplifier. EasyLens enhances lesion-relevant visual evidence at inference time, enabling more reliable diagnostic support and report generation for micro-lesions, thereby increasing the clinical utility of your existing VLM deployments.

Key insights

Training-free EasyLens amplifies subtle lesion cues in frozen VLM patch representations, improving detection and report generation.

Principles

• Subtle lesion cues are often preserved at patch-level but diluted in global VLM embeddings.
• Prototype spaces can differentiate pathological patterns from normal anatomical variations.
• Morphology-guided residual enhancement strengthens lesion signals without disrupting context.

Method

EasyLens constructs EasyBank (prototype space), uses EasyTag for counterfactual patch selection, then EasyAmplifier applies morphology-guided residual enhancement to selected patches.

In practice

• Integrate EasyLens as an inference-time adapter for existing medical VLMs.
• Build pathology-anatomy prototype banks for fine-grained lesion discrimination.
• Apply morphology-guided residual updates to enhance subtle lesion representations.

Topics

• Medical Vision-Language Models
• Subtle Lesion Detection
• Representation Amplification
• Prototype Learning
• Training-Free Inference
• Radiology AI

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• EasyLens: A Training-Free Plug-and-Play Subtle-Lesion Representation Amplifier for Medical Vision-Language Models — EasyLens enhances medical VLM sensitivity to subtle lesions by amplifying relevant patch representations without retraining.
• Intelligent Healthcare Imaging Platform: A VLM-Based Framework for Automated Medical Image Analysis and Clinical Report Generation — A VLM-based framework automates medical image analysis and clinical report generation across multiple imaging modalities.
• Looking Where the Eye Cannot See: Improving MS Diagnosis with Deep Learning — DeepMS improves MS diagnosis by analyzing subtle microstructural changes in normal-appearing white matter using deep learning.
• Evidential learning driven Breast Tumor Segmentation with Stage-divided Vision-Language Interaction — TextBCS enhances breast tumor segmentation in MRI by combining text prompts with evidential learning to address low contrast and blurred boundaries.
• Parameter-Efficient Subspace Decoupling ViT for Mitigating Multi-Task Negative Transfer in Histological Scoring — Subspace-decoupled ViT with orthogonal adapters mitigates negative transfer in multi-task histological scoring, improving efficiency.
• Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images — Contrastive learning, especially with expert-guided hard negatives, improves fine-grained pediatric brain tumor classification from WSIs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.