RAG4Outcome: AI framework predicts osteomyelitis prognosis using RAG

Chronic osteomyelitis poses serious prognostic challenges due to high recurrence risk and complex recovery. Traditional manual scoring systems are inconsistent and not scalable, while existing multimodal AI approaches often require aligned inputs and large annotated datasets that struggle with heterogeneous clinical data. To address this, Daqian Shi and colleagues introduce RAG4Outcome, a retrieval-augmented generation (RAG) framework that synthesizes diverse data types — including PET-CT imaging reports, structured surgical records, and unstructured follow-up notes — without requiring perfect alignment. By combining a domain-specific retrieval corpus with expert-guided prompting, RAG4Outcome produces interpretable, evidence-grounded prognostic assessments.

Early experiments on real-world cases demonstrate that RAG4Outcome achieves clinically relevant predictions, outperforming manual scoring in consistency and detail. The framework’s modular design allows it to incorporate new patient data as it becomes available, making it adaptable for continuous learning. This approach not only improves prediction accuracy but also provides clinicians with traceable reasoning for each prognosis, addressing the critical need for transparency in AI-assisted medical decision-making. The work represents a significant step toward practical AI deployment in infection management and postoperative care for chronic osteomyelitis.