Reproduce “Retrieval-Augmented Multimodal Language Model with Heterogeneous Evidence Integration for Clinical Prognosis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hamza Shafiq, Jialu Tang, Ting Dang, Jun Hu, Aaqib Saeed
PRISM combines numerical EHR features, current and prior ECGs, and retrieved encounters as soft tokens for a shared MedGemma decoder. It pools the three nearest encounters separately in EHR and ECG spaces and recency-weights prior ECG embeddings. With matched five-stream inputs it reaches 0.9092 macro-of-macro AUROC versus 0.7652 for a compact model; benchmark-window performance is 0.8895, below UniPACT's 0.8937.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.