Multimodal Clinical Fusion — Safety Demo

Move the sliders. The system fuses what's present, quantifies disagreement, routes each positive biomarker to therapy, escalates on contraindications, refuses to over-recommend, and hash-signs every output. Illustrative inputs — the point is the safety behavior, not accuracy. Built for clinics, not terminals.

HER2 — fused from imaging + pathology + genomics

Imaging 0.86 Pathology 0.91 Genomics (ERBB2) 0.84

Other biomarkers (per-biomarker confidence)

ER 0.12 PR 0.08 BRCA (germline) 0.05

Patient context (drives contraindications)

LVEF %

Pregnancy

Menopausal

autoimmune history

Recommendation (live)

Same logic as demo.py in this repo — fusion re-normalizes over present modalities (graceful degradation), a negative biomarker triggers no therapy, contraindications (including missing data) escalate, and each recommendation carries a real SHA-256 content hash. Reference demonstration only — not clinically validated, not deployed.