How mygithth27/af-risk-prediction-by-ecg-dnn works
This repository supports research workflows for three-class atrial-fibrillation risk prediction from 12-lead ECGs, plus survival-model analysis. The main workflow formulates patient/exam labels and splits, trains a residual network from ECG batches, and evaluates a saved model to produce class probabilities. It also includes ECG-subset preparation and survival-analysis notebooks. The training and evaluation links to their core model components are visible in the excerpts; survival-model internals were not sampled, so that branch is documented but not expanded.
Main entry points: evaluate.py Model evaluation · train.py Model training · generate_subhdf.py ECG subset preparation · af_prediction_formulation.py AF labels and splits
Components 13
Label preparation
- AF labels and splits
af_prediction_formulation.py - Age-based splits
archives/formulate_problem.py
Model prediction
- Model training
train.py - Batch loader
dataloader.py - Residual network
resnet.py - Model evaluation
evaluate.py
ECG data
- ECG subset preparation
generate_subhdf.py - ECG HDF5 data
- Exam metadata CSV
- Prediction CSV
Risk analysis
- Survival analysis
notebooks/survival_model.ipynb - Comorbidity analysis
notebooks/survival_model_comorbidities.ipynb
Other components
- Researcher
Connections 16
- Researcher → AF labels and splits: formulates labels
README.md - AF labels and splits → Exam metadata CSV: writes metadata
af_prediction_formulation.py - Researcher → Age-based splits: creates splits
README.md - Age-based splits → Model training: provides splits
- Researcher → Model training: trains model
README.md - Model training → Batch loader: batches data
train.py - Model training → Residual network: trains network
train.py - Researcher → Model evaluation: evaluates model
README.md - Model evaluation → Residual network: runs inference
evaluate.py - Model evaluation → ECG HDF5 data: reads traces
evaluate.py - Model evaluation → Prediction CSV: writes probabilities
evaluate.py - Researcher → ECG subset preparation: prepares subset
generate_subhdf.py - ECG subset preparation → ECG HDF5 data: reads traces
generate_subhdf.py - ECG subset preparation → Exam metadata CSV: reads labels
generate_subhdf.py - Researcher → Survival analysis: explores survival
README.md - Researcher → Comorbidity analysis: explores comorbidities
README.md