Coronary Probability Map

Coronary artery disease risk is usually a single score or a wall of numbers. Clinicians and learners still have to map those numbers onto anatomy in their heads. Coronary Probability Map turns one clinical record into four probabilities — overall CAD plus stenosis risk for the LAD, LCX, and RCA — and paints only those three arteries on an interactive 3D heart diagram.

How it works. We fit four logistic models on the public Extension of Z-Alizadeh Sani dataset (UCI, 303 records, CC BY 4.0). Inputs never include LAD, LCX, RCA, or the cath label. Coefficients are exported to model.json, so the browser can score any record with no server and no paid API. Drag to rotate, scroll to zoom, click an artery for a per-vessel explanation, or edit a field (e.g. typical chest pain) and watch probabilities and colors update. Previous / Next browses all 303 rows; unmodified rows show that row’s own leave-one-fold held-out probability.

Results (held-out, threshold 0.5). CAD accuracy 0.825, precision 0.910, recall 0.838, F1 0.872, AUC 0.900. LAD AUC 0.806; LCX 0.659; RCA 0.706. A Node cross-check of all 303 rows against the page formulas matched within 5.0×10⁻⁷.

Honesty. This is an educational decision-support display, not a diagnosis, not medical advice, and not a treatment recommendation. The heart is a standard diagram, not the patient’s scan; color is model probability, not plaque imaging. Run locally with python3 -m http.server from app/web, or open the GitHub Pages demo.

Built for Multimodal AI Hackathon 2026 Track A. Code assisted by an AI coding assistant (disclosed in Built With and README).

Built With

• ai-coding-assistant
• css
• html
• javascript
• numpy
• openpyxl
• pandas
• scikit-learn
• three.js
• uci-machine-learning-repository