MP-SeizNet: A multi-path CNN Bi-LSTM Network for seizure-type classification using EEG

Seizure type identification is essential for the treatment and management of epileptic patients. However, it a difficult process known to be time consuming labor intensive. Automated diagnosis systems, with advancement machine learning algorithms, have potential accelerate classification process, alert patients, support physicians in making quick accurate decisions. In this paper, we present novel multi-path seizure-type deep network (MP-SeizNet), consisting convolutional neural (CNN) bidirectional long short-term memory (Bi-LSTM) an attention mechanism. The objective study was classify specific types seizures, including complex partial, simple absence, tonic, tonic-clonic using only electroencephalogram (EEG) data. EEG data fed our proposed model two different representations. CNN wavelet-based features extracted from signals, while Bi-LSTM raw signals let MP-SeizNet jointly learns representations seizure more information learning. evaluated largest available epilepsy database, Temple University Hospital Corpus, TUSZ v1.5.2. We across patient three-fold cross-validation five-fold cross-validation, achieving F1 scores 87.6% 98.1%, respectively.