timedcy/aied

Automated detector for the detection of intracranial IEDs.

Dartmouth ECoG Lab Automated Spike Detector

This repository contains an automated intracranial spike detector called "AiED: Artificial intelligence for the detection of intracranial interictal epileptiform discharges"

Our automated method consists of a template-matching algorithm and convolutional neural network (CNN) for the detection of intracranial IEDs. The final network had an F1-score of 0.95 (95% CI: 0.91-0.98) and an average Area Under the Receiver Operating Characteristic of 0.98 (95% CI: 0.96-1.00).

More information can be found in the following publications: Horak et al., 2015 and Quon et al., 2021.

Brief description of notebooks

These python notebooks can be accessed using JupyterLab. Please refer to the JupyterLab documentation for install and starting instructions.

"preprocessing.ipynb"

This notebook was created to preprocess intracranial EEG data. Input is a .EDF iEEG file. Output is a .csv file containing the preprocessed iEEG.

"aied.ipynb"

This notebook contains our automated IED detector. Before using this detector, please download the "model_aied.pt" file, containing our pretrained network, and indicate where this file is stored in the detector notebook.

Sample data

"sample_eegdata.csv"

This file contains preprocessed iEEG where rows correspond to unique channels and columns correspond to time points (fs = 200).

"sample_finalspikes.csv"

This file contains the IEDs detected from the sample EEG file with our automated detector.