Epileptic Seizures Classification Based on Deep Neural Networks

Abstract

Epileptic seizure is a chronic and non-communicable disease which occurs in people of all ages. In the detection of epileptic seizures, electroencephalography (EEG) plays a vital role. Due to advantages like low cost, portability, etc., EEG is preferred over other brain acquisition techniques. To analyze EEG data with the increased number of segments, it will take additional time for the neurophysician to classify the seizures. To reduce the time in analyzing EEG data, two deep neural network-based models are proposed. In this paper, one model using the “long short-term memory (LSTM) model” and another using “bi-directional long short-term memory (Bi-LSTM)” are anticipated to classify whether the given data is normal or partial seizures or generalized seizure data. The EEG data of 40 patients is collected, and each patient’s data is split into 10-second segments. The algorithm is written in MATLAB® by giving the 10-second segments of all patients. The training is done on the personal computer with an improved system configuration which reduces training time from days to a few hours. The training time to train the network with the LSTM model is 82 min, and the network with Bi-LSTM model is 122 min. The accuracy achieved with LSTM and Bi-LSTM model is 90.89% and 90.92%, respectively.