In-Context Learning for Wound Classification with Small Multimodal Language Models

2026-07-21

AI Summary

This work addresses the limitations of traditional wound image classification, which relies heavily on large annotated datasets and task-specific fine-tuning, rendering it inflexible to shifts in label space or deployment conditions. For the first time, the study introduces retrieval-augmented in-context learning to this domain, proposing a training-free classification approach that leverages small multimodal language models (e.g., Qwen, Ministral, Gemma). The method constructs informative support-set contexts through k-nearest neighbor (kNN) retrieval, followed by Maximal Marginal Relevance (MMR)-based reranking and a weighted kNN strategy. Experiments demonstrate that the approach achieves accuracies of 0.872 and 0.678 on the Kaggle and Medetec datasets, respectively, with performance scaling favorably with model size. Notably, it approaches performance saturation with only 8–10 support examples and substantially outperforms naive nearest-neighbor voting, enabling flexible and efficient zero-shot deployment.

Abstract

Wound image classification is often treated as a task-specific supervised learning problem, requiring substantial amounts of manually labelled data and retraining when the label space or deployment setting changes. This study evaluated whether small multimodal language models (SMLMs) can provide a training-free alternative for wound classification through retrieval-based in-context learning (ICL). Experiments used two public wound-image datasets: the Kaggle wound dataset (1469 images, 10 classes) and the Medetec dataset (560 images, 9 classes). Eleven SMLMs from the Qwen 3.5, Ministral 3, and Gemma 4 families were evaluated under zero-shot prompting and few-shot prompting with random support examples, embedding-based k-nearest-neighbour (kNN) retrieval, and kNN retrieval followed by maximal marginal relevance reranking (MMR). Retrieval-only weighted-kNN controls, support-set reduction experiments, and support-context size sweeps were used to assess the effects of retrieval, model scale, and prompt length. Query-conditioned ICL consistently outperformed zero-shot and random few-shot prompting. On the Kaggle dataset, the best result was achieved by Qwen 3.5 27B with kNN+MMR, reaching 0.872 accuracy and 0.871 F1 score. On Medetec, Qwen 3.5 27B with kNN+MMR reached 0.678 accuracy and 0.670 F1. Larger models exceeded matched weighted-kNN controls, indicating use of retrieved examples beyond nearest-neighbour voting. Retrieval-based ICL degraded modestly under support-set reduction, and most gains saturated with 8-10 support images. Retrieval-based ICL allows SMLMs to perform adaptable wound image classification without task-specific retraining. Compact retrieved contexts may support practical and privacy-conscious deployment, although performance remains dependent on model scale, retrieval strategy, and dataset difficulty.

Problem

• wound classification
• in-context learning
• multimodal language models
• few-shot learning
• retrieval-based learning

Innovation

• in-context learning
• multimodal language models
• retrieval-based prompting
• wound classification
• training-free adaptation