CARDIA: Cardiovascular Risk Intelligence

CARDIA is an interactive machine learning prototype for exploring coronary artery disease (CAD) risk and vessel-specific prediction through an explainable, visual interface.

The Problem

Cardiovascular risk assessment involves interpreting multiple patient characteristics, clinical measurements, and diagnostic indicators. Machine learning can help explore patterns across these variables, but predictions need to be presented transparently and with clear limitations.

Our Solution

CARDIA combines four machine learning pipelines with an interactive dashboard and anatomical 3D heart visualization.

• Overall CAD prediction: Estimates the model's prediction for coronary artery disease.
• Vessel-specific analysis: Provides separate predictions for the left anterior descending (LAD), left circumflex (LCX), and right coronary artery (RCA).
• Interactive visualization: Explores predictions through a 3D anatomical heart and coronary artery branches.
• Explainable insights: Presents feature-level explanations and model evaluation metrics.
• Patient input interface: Organizes clinical features into categories for easier exploration.

Technical Approach

The project uses the Z-Alizadeh Sani coronary artery disease dataset and evaluates Logistic Regression, Random Forest, and Gradient Boosting models using stratified five-fold cross-validation. Separate pipelines are used for the overall CAD target and each vessel-specific target.

The frontend is built with React, TypeScript, Vite, and Three.js-based 3D visualization. A Flask API serves the prediction pipelines and exposes endpoints for model metrics, input schema, and sample profiles.

Evaluation

The selected models achieved the following out-of-fold ROC-AUC scores:

These results come from a small dataset of 303 patients. They are internal cross-validation estimates, not evidence of clinical readiness or performance on independent populations.

Limitations and Responsible Use

CARDIA is an educational and research prototype, not a diagnostic device. Its predictions must not replace medical evaluation, clinician judgment, or diagnostic imaging. The vessel visualizations communicate model outputs and do not establish the actual anatomical location or extent of plaque.

Data and References

Dataset: Z-Alizadeh Sani coronary artery disease dataset, UCI Machine Learning Repository, dataset ID 411.

DOI: https://doi.org/10.24432/C5461K

Built With

• css3
• flask
• html
• javascript
• learning
• machine
• python
• react
• scikit-learn
• three.js
• typescript
• vite