MedGate-Fusion Integrates Narratives and Biomarkers for Stroke Risk
September 23, 2026
MedGate-Fusion uses a multi-modal gated architecture to combine transformer-based embeddings of clinical narratives with ten physiological biomarkers. Evaluated on 102,736 patient records, the model targets five-year stroke risk stratification while using dictionary-based redaction to prevent target leakage from diagnostic terms.
HOW THIS AFFECTS YOU
●
researcherThe use of dictionary-based redaction provides a methodology for handling target leakage in medical NLP.
●
healthThis architecture shows how unstructured clinical notes can be combined with biomarkers for better risk prediction.