PRIMA: Pre-training with Risk-integrated Image-Metadata Alignment for Medical Diagnosis via LLM

2026-02-26

AI Summary

This work addresses the limitation of existing multimodal medical diagnosis approaches that oversimplify clinical metadata into isolated labels, thereby neglecting their rich semantic content. To overcome this, the study introduces a novel framework that embeds clinical prior knowledge—specifically risk–disease associations—into multimodal pretraining. An expert-curated corpus enhances the text encoder, while a dual-encoder architecture aligns visual and textual features. The authors further propose an innovative multi-granularity soft-label alignment mechanism to mitigate ambiguity in clinical relevance. The model integrates RAG, Clinical ModernBERT, DINOv3, and Qwen-3, optimized with a quadruple complementary loss function. Remarkably, it achieves state-of-the-art performance in robust and accurate image–metadata joint diagnosis without requiring large-scale data or high computational resources.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Medical diagnosis requires the effective synthesis of visual manifestations and clinical metadata. However, existing methods often treat metadata as isolated tags, failing to exploit the rich semantic knowledge embedded in clinical descriptions. We propose PRIMA (Pre-training with Risk-integrated Image-Metadata Alignment), a framework that integrates domain-specific knowledge into multi-modal representation learning. We first curate an expert corpus of risk-disease correlations via Retrieval-Augmented Generation (RAG) to refine Clinical ModernBERT, embedding diagnostic priors into the text encoder. To bridge the modality gap, we introduce a dual-encoder pre-training strategy utilizing DINOv3 and our refined BERT, optimized by a suite of four complementary loss functions. These losses are designed to capture multi-granular semantic alignment and handle the ambiguity of clinical correlations through soft labels. Finally, we leverage Qwen-3 to fuse these aligned features for precise disease classification. Extensive experiments demonstrate that PRIMA effectively harmonizes pixel-level features with abstract clinical expertise, significantly outperforming other state-of-the-art methods. Notably, our framework achieves superior robustness without the need for massive data collection or exhaustive computational resources. Our code will be made public upon acceptance.

Problem

• medical diagnosis
• image-metadata alignment
• clinical metadata
• multi-modal representation
• semantic knowledge

Innovation

• Risk-integrated Alignment
• Retrieval-Augmented Generation (RAG)
• Dual-encoder Pre-training
• Multi-granular Semantic Alignment
• Clinical Metadata Integration