nseeuws / EventBasedModeling

Avoiding post-processing with event-based detection in biomedical signals

Code repository accompanying our publication on event-based modeling.

The repo is split among our synthetic data experiments and the real-world data experiments, since both rely on separate dependencies (synthetic uses PyTorch, real-world Tensorflow). The supplementary material can be found in supp.pdf.

Synthetic data

The files event_train_and_evaluation.py and epoch_train_and_evaluate.py run the event-based and epoch-based training runs respectively. The epoch-based post-processing can be set using command line arguments.

Real-world data

EEG artefact training runs can be started with run_event_artefact.py and run_epoch_artefact.py. One can choose how many signal recordings each run trains on.

EEG seizures training runs can be started with run_event_seizure.py and run_epoch_seizure.py. One can choose how many signal recordings each run trains on.