Unsupervised Seizure Detection in Eeg Using Long Short Term Memory Network and Clustering

Seizures, specifically electroencephalographic seizures, are used as a biomarker for the development of epilepsy, which is one of the most common neurological disorders. Electroencephalogram (EEG) is the most commonly used technique used to detect seizures due to its easy and low-cost collection. Manual analysis of EEG signals is very tedious due to its stochastic nature. The methods proposed for the automatic detection of seizures mostly use supervised learning and hence require large amounts of labeled data to perform well. Unsupervised learning techniques, proposed thus far, have used complex neural networks, such as transformers. In this paper, we aim to accomplish the task of unsupervised classification of seizures using a simple Long Short Term Memory (LSTM) Network and a clustering algorithm. The proposed method outperforms both its supervised and unsupervised counterparts on seizure detection in intracranial EEG from the benchmark dataset from the University of Pennsylvania and Mayo Clinic, publicly available at [1].