Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Supervised and Unsupervised Deep Learning Approaches for EEG Seizure Prediction

Zakary Georgis-Yap, Milos R. Popovic, Shehroz S. Khan

TL;DR

Predicting epileptic seizures from EEG is challenging due to variable preictal patterns and data scarcity. The authors compare supervised (CNN, CNN-LSTM, TCN) and unsupervised (CNN autoencoder, CNN-LSTM autoencoder, TCN autoencoder) deep learning approaches trained on interictal or preictal data from two large EEG datasets using leave-one-seizure-out, patient-specific evaluation. They show that both paradigms are feasible but exhibit substantial per-patient variability; unsupervised methods can match or exceed supervised performance in some cases, particularly on the SWEC-ETHZ dataset, while CHB-MIT results are more variable. The work underscores the potential for personalized seizure prediction and identifies practical challenges, including hyperparameter tuning and threshold selection for deployment. The findings motivate further exploration of larger hyperparameter searches, thresholds, and alternative architectures.

Abstract

Epilepsy affects more than 50 million people worldwide, making it one of the world's most prevalent neurological diseases. The main symptom of epilepsy is seizures, which occur abruptly and can cause serious injury or death. The ability to predict the occurrence of an epileptic seizure could alleviate many risks and stresses people with epilepsy face. We formulate the problem of detecting preictal (or pre-seizure) with reference to normal EEG as a precursor to incoming seizure. To this end, we developed several supervised deep learning approaches to identify preictal EEG from normal EEG. We further develop novel unsupervised deep learning approaches to train the models on only normal EEG, and detecting pre-seizure EEG as an anomalous event. These deep learning models were trained and evaluated on two large EEG seizure datasets in a person-specific manner. We found that both supervised and unsupervised approaches are feasible; however, their performance varies depending on the patient, approach and architecture. This new line of research has the potential to develop therapeutic interventions and save human lives.

Supervised and Unsupervised Deep Learning Approaches for EEG Seizure Prediction

TL;DR

Predicting epileptic seizures from EEG is challenging due to variable preictal patterns and data scarcity. The authors compare supervised (CNN, CNN-LSTM, TCN) and unsupervised (CNN autoencoder, CNN-LSTM autoencoder, TCN autoencoder) deep learning approaches trained on interictal or preictal data from two large EEG datasets using leave-one-seizure-out, patient-specific evaluation. They show that both paradigms are feasible but exhibit substantial per-patient variability; unsupervised methods can match or exceed supervised performance in some cases, particularly on the SWEC-ETHZ dataset, while CHB-MIT results are more variable. The work underscores the potential for personalized seizure prediction and identifies practical challenges, including hyperparameter tuning and threshold selection for deployment. The findings motivate further exploration of larger hyperparameter searches, thresholds, and alternative architectures.

Abstract

Epilepsy affects more than 50 million people worldwide, making it one of the world's most prevalent neurological diseases. The main symptom of epilepsy is seizures, which occur abruptly and can cause serious injury or death. The ability to predict the occurrence of an epileptic seizure could alleviate many risks and stresses people with epilepsy face. We formulate the problem of detecting preictal (or pre-seizure) with reference to normal EEG as a precursor to incoming seizure. To this end, we developed several supervised deep learning approaches to identify preictal EEG from normal EEG. We further develop novel unsupervised deep learning approaches to train the models on only normal EEG, and detecting pre-seizure EEG as an anomalous event. These deep learning models were trained and evaluated on two large EEG seizure datasets in a person-specific manner. We found that both supervised and unsupervised approaches are feasible; however, their performance varies depending on the patient, approach and architecture. This new line of research has the potential to develop therapeutic interventions and save human lives.
Paper Structure (20 sections, 16 figures, 9 tables)

This paper contains 20 sections, 16 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Supervised Seizure Prediction
  4. CNN
  5. CNN-LSTM
  6. TCN
  7. Unsupervised Seizure Prediction
  8. Data Processing
  9. Datasets
  10. Data Preprocessing
  11. Time-Frequency Transform
  12. Experimental Setting and Results
  13. Supervised Prediction
  14. Comparison with Fixed Parameters
  15. Comparison with Other Architectures
  16. ...and 5 more sections

Figures (16)

  • Figure 1: (a) Convolutional block, (b) CNN architecture with convolution blocks and fully connected (FC) layers
  • Figure 2: CNN-LSTM architecture showing time-frequency (TF) transform, CNN layers, LSTM and fully connected (FC) layer.
  • Figure 3: (a) TCN block description. TCN: temporal convolutional network. ReLU: rectified linear unit. (b) Supervised TCN architecture overview. TCN: temporal convolutional network. FC layer: fully connected layer.
  • Figure 4: Autoencoders comprising of (a) Convolution layers only, (b)CNN-LSTM, and (c) TCN
  • Figure 5: (a) Labelling of the preictal and interictal periods with parameters. (b) Simplified visualization of LOSO test partitioning by withholding the last seizure.
  • ...and 11 more figures