Long Short-Term Memory Framework for Classification of Seizure Types Using A Different Format of EEG Signal

Accurate classification of seizure types is crucial for epileptic seizures diagnosis, medication selection, and medical care. However, most recent works are mainly surrounded by analysis of seizures, in comparison to seizure types. This work proposes, an advanced deep learning (DL) pipeline, long short-term memory (LSTM) based framework to classify different types of seizures using multichannel electroencephalogram (EEG). This framework concurrently uses time and spectral formats of EEG as input for extraction of distinct and pertinent features to classify complex partial, focal non-specific, generalized non-specific, myoclonic, tonic-clonic seizures, and seizure-free. For validation, we utilized the Temple University Hospital EEG dataset (TUH, v1.5.2). The framework achieved remarkable results with a classification accuracy of 97.7%, recall of 98.0%, and weighted F1-score of 98.0%.