CardioVision 3D

Inspiration

What it doesInspiration

We wanted to make complex cardiac data easier to understand by combining AI, explainable predictions, and 3D heart visualization.

What it does CardioVision 3D predicts coronary artery disease (CAD) risk using 55 clinical features. It explains the factors behind each prediction, visualizes potentially affected coronary arteries in 3D, and extracts candidate values from medical reports for user review.

How we built it We used Python, LightGBM, and TreeSHAP for prediction and explainability; FastAPI for the backend; and React, TypeScript, and Three.js for the interactive interface and 3D anatomy.

Challenges we ran into Handling medical units, extracting accurate values from reports, preventing data leakage, and creating anatomically meaningful visualizations were our main challenges.

Accomplishments we're proud of We integrated AI-based risk prediction, explainable insights, medical document parsing, and interactive 3D anatomy into one prototype. Our reported cross-validation ROC-AUC for overall CAD prediction is 0.907.

What we learned We learned that AI predictions must be explainable, carefully evaluated, and presented with their limitations—not just impressive visualizations and percentages.

What's next for CardioVision 3D We aim to improve vessel-level predictions, validate the models on independent datasets, enhance document extraction, and explore patient-specific cardiac visualization.

CardioVision 3D is a research prototype for education and decision support, not a diagnostic device.