How harshithk9515-dev/cardiotwin works
CardioTwin is a synthetic-data cardiac decompensation prediction prototype. Its workflow generates static EHR baselines and wearable telemetry, builds normalized patient windows, trains a late-fusion model, then serves patient playback and risk scoring in a Streamlit dashboard. The dashboard also presents integrated-gradient explanations and intervention what-if scoring. Training and dashboard wiring are supported by the excerpts and README; details of unsampled dashboard calls are not inferred.
Main entry points: app.py Twin dashboard · train.py Training and evaluation · dataset.py Windowing and scaling · model.py Fusion risk model
Components 11
Synthetic cohort
- Cohort generator
data/generate_synthetic_data.py - Static EHR data
data/static_ehr.csv - Wearable data
data/dynamic_wearables.csv
Model pipeline
- Windowing and scaling
src/dataset.py - Training and evaluation
src/train.py - Fusion risk model
src/model.py - Model checkpoint
models/cardiotwin_weights.pth
Clinician dashboard
- Twin dashboard
src/app.py - Feature explanations
src/explainability.py - What-if scoring
src/app.py
Other components
- Clinician
Connections 13
- Cohort generator → Static EHR data: writes
data/generate_synthetic_data.py - Cohort generator → Wearable data: writes
data/generate_synthetic_data.py - Static EHR data → Windowing and scaling: loads
src/dataset.py - Wearable data → Windowing and scaling: loads
src/dataset.py - Training and evaluation → Windowing and scaling: uses
src/train.py - Training and evaluation → Fusion risk model: trains
src/train.py - Training and evaluation → Model checkpoint: saves
src/train.py - Clinician → Twin dashboard: uses
README.md - Twin dashboard → Model checkpoint: loads
src/app.py - Twin dashboard → Fusion risk model: scores windows
src/app.py - Twin dashboard → Feature explanations: shows drivers
README.md - Twin dashboard → What-if scoring: offers simulation
README.md - What-if scoring → Fusion risk model: re-scores
README.md