How bdsp-core/icare-dl works
ICARE is a research implementation for predicting neurological outcomes after cardiac arrest from continuous EEG and admission characteristics using multiscale CNN-LSTM models. The README establishes that high-level workflow; filenames indicate separate clinical, short-IIC, and long-IIC scripts. Source excerpts were unavailable, so exact data loading, feature-extraction, and script-to-script wiring are not asserted.
Main entry points: deep_learning_multiscale_lstm.m Multiscale CNN-LSTM
Components 13
Clinical inputs
- ICARE cohort data
- EDF reader
edfread.m - Incremental EDF reader
edfreadUntilDone.m
Signal analysis
- Feature sequences
features/mgh_features_sequences.mat - EEG analysis functions
Supporting Functions
Outcome modeling
- Multiscale CNN-LSTM
deep_learning_multiscale_lstm.m - Clinical model
deep_learning_multiscale_lstm_iic_clinical.m - Short-IIC model
deep_learning_multiscale_lstm_short_iic.m - Long-IIC model
deep_learning_multiscale_lstm_long_iic.m - Model figure analysis
deep_learning_multiscale_lstm_fig1.m - Model figure analysis
deep_learning_multiscale_lstm_fig2.m - Neurological outcome predictions
- Clinical summary results
results/multi-lstm-summary-clinical_avg.mat
Connections 2
- ICARE cohort data → Multiscale CNN-LSTM: provides EEG and admission data
- Multiscale CNN-LSTM → Neurological outcome predictions: predicts outcome