Adopting Convolutional Long Short-Term Memory Network to Detect Seizures

Recent advances in deep learning have enabled computer-aided medical data processing, thus adding to automated disease diagnosis and detection. Neurologists typically diagnose seizures, also known as epilepsy, through visual evaluation of Electroencephalogram (EEG) waveforms, which include data regarding the brain’s electrical impulses. However, because this is a manual, tardy, and difficult process, automated and precise detection of epileptic seizures can increase efficacy and patient quality of life. In this study, a one-dimensional convolutional long short-term memory (1D CNN-LSTM) architecture is proposed for the automated detection of seizures based on EEG waveforms. The proposed classification model combines the abilities of the convolution neural network (CNN) and the long short-term memory (LSTM) architecture forming a convolutional LSTM network, where the CNN extracts high level robust features from the EEG signal data while the LSTM layers aid with the extraction of temporal features. The effectiveness of the model being suggested is verified on the Bonn dataset and the maximum accuracy achieved by the model on the binary class seizure detection is 100%. To verify the sturdiness of the proposed architecture, it has also been tested on noisy EEG signals.