How ayush-dev-001/cardio works
CardioScope is an interactive CAD decision-support and education application. A user enters routine clinical measurements, receives overall CAD and LAD/LCX/RCA stenosis estimates, sees risk mapped onto a 3D heart, and can inspect SHAP explanations and what-if changes. The repository also includes the data preparation and model-training workflow behind the serving models. The principal execution path is documented, while several frontend and ML internals are represented at their source-directory boundary because their implementation was not sampled.
Main entry points: App.tsx Dashboard · service.py Prediction service · train.py Training and evaluation · main.py HTTP API
Components 22
User experience
- Landing page
frontend/src/landing/Landing.tsx - Dashboard
frontend/src/App.tsx - Patient inputs
frontend/src/components/PatientForm.tsx - API client
frontend/src/api/client.ts - 3D heart viewer
frontend/src/scene/HeartViewer.tsx - Risk and SHAP results
frontend/src/components/Results.tsx - What-if analysis
frontend/src/components/WhatIfPanel.tsx
Prediction service
- HTTP API
backend/app/main.py - Prediction service
backend/app/service.py - Request and response schemas
backend/app/schemas.py
ML pipeline
- Model predictor
ml/cardioscope_ml/predictor.py - SHAP explanations
ml/cardioscope_ml/explain.py - Probability calibration
ml/cardioscope_ml/calibration.py - Training and evaluation
ml/cardioscope_ml/train.py - Nested cross-validation
ml/cardioscope_ml/nested_cv.py - Model fitting
ml/cardioscope_ml/modeling.py - Trained models
Data and configuration
- Data preparation
ml/cardioscope_ml/data.py - Feature and target definitions
ml/cardioscope_ml/config.py - Derived clinical features
ml/cardioscope_ml/derived.py - Clinical dataset
Other components
- Clinician or learner
Connections 26
- Clinician or learner → Landing page: opens
README.md - Landing page → Dashboard: opens app
frontend/src/landing/Landing.tsx - Clinician or learner → Patient inputs: enters data
README.md - Patient inputs → Dashboard: updates inputs
frontend/src/App.tsx - Dashboard → API client: requests predictions
frontend/src/App.tsx - API client → HTTP API: sends HTTP
README.md - HTTP API → Request and response schemas: validates and shapes
backend/app/main.py - HTTP API → Prediction service: dispatches
backend/app/main.py - Prediction service → Model predictor: predicts
backend/app/service.py - Prediction service → Derived clinical features: derives inputs
backend/app/service.py - Prediction service → SHAP explanations: builds explanations
backend/app/service.py - Prediction service → Feature and target definitions: reads definitions
backend/app/service.py - Model predictor → Probability calibration: calibrates probabilities
ml/cardioscope_ml/predictor.py - Model predictor → Trained models: loads models
- Dashboard → 3D heart viewer: renders probabilities
frontend/src/App.tsx - Dashboard → Risk and SHAP results: renders results
frontend/src/App.tsx - Dashboard → What-if analysis: offers what-if
frontend/src/App.tsx - Risk and SHAP results → 3D heart viewer: maps risk to anatomy
- Training and evaluation → Data preparation: prepares data
ml/cardioscope_ml/train.py - Data preparation → Clinical dataset: reads clinical records
- Data preparation → Feature and target definitions: uses definitions
ml/cardioscope_ml/data.py - Training and evaluation → Nested cross-validation: evaluates models
ml/cardioscope_ml/train.py - Nested cross-validation → Model fitting: fits fold models
ml/cardioscope_ml/nested_cv.py - Model fitting → Probability calibration: fits calibrator
ml/cardioscope_ml/modeling.py - Training and evaluation → SHAP explanations: creates explainers
ml/cardioscope_ml/train.py - Training and evaluation → Trained models: saves trained models