MOSAIC: A Multilingual, Taxonomy-Agnostic, and Computationally Efficient Approach for Radiological Report Classification

2025-08-29

AI Summary

Existing radiology report classification methods suffer from three key limitations: rule-based approaches lack generalizability; supervised models require extensive labeled data; and large language model (LLM)-based solutions are predominantly closed-source, computationally expensive, and restricted to English and single-label, unimodal settings. To address these, we propose the first open-source, multilingual, label-agnostic, and lightweight classification framework built on MedGemma-4B (4B parameters). Our method integrates zero- and few-shot prompting with efficient fine-tuning and introduces domain-specific data augmentation. With only dozens of training samples, the model achieves expert-level performance across seven multilingual, multimodal datasets: it attains an average macro-F1 of 88 on five chest X-ray tasks and a weighted F1 of 82 on Danish reports using just 80 samples. The framework is deployable on consumer-grade GPUs (24 GB VRAM), significantly lowering the deployment barrier for LLMs in clinical practice.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Radiology reports contain rich clinical information that can be used to train imaging models without relying on costly manual annotation. However, existing approaches face critical limitations: rule-based methods struggle with linguistic variability, supervised models require large annotated datasets, and recent LLM-based systems depend on closed-source or resource-intensive models that are unsuitable for clinical use. Moreover, current solutions are largely restricted to English and single-modality, single-taxonomy datasets. We introduce MOSAIC, a multilingual, taxonomy-agnostic, and computationally efficient approach for radiological report classification. Built on a compact open-access language model (MedGemma-4B), MOSAIC supports both zero-/few-shot prompting and lightweight fine-tuning, enabling deployment on consumer-grade GPUs. We evaluate MOSAIC across seven datasets in English, Spanish, French, and Danish, spanning multiple imaging modalities and label taxonomies. The model achieves a mean macro F1 score of 88 across five chest X-ray datasets, approaching or exceeding expert-level performance, while requiring only 24 GB of GPU memory. With data augmentation, as few as 80 annotated samples are sufficient to reach a weighted F1 score of 82 on Danish reports, compared to 86 with the full 1600-sample training set. MOSAIC offers a practical alternative to large or proprietary LLMs in clinical settings. Code and models are open-source. We invite the community to evaluate and extend MOSAIC on new languages, taxonomies, and modalities.

Problem

• Classifying radiological reports across languages and taxonomies efficiently
• Overcoming limitations of rule-based and resource-intensive supervised models
• Enabling clinical deployment without costly annotations or proprietary systems

Innovation

• Multilingual open-access language model MedGemma-4B
• Supports zero-shot prompting and lightweight fine-tuning
• Achieves expert-level performance with minimal GPU memory