An Efficient Transfer Learning Based Deep Learning Approach for Automatic Classification of Epileptic Seizure
In real-life situations, electroencephalograms (EEG)-based seizure detection systems face many challenges. The second most common brain disorder following Migraine is epileptic seizure. The automatic detection of seizures can significantly improve the quality of life of the patients. Effective use of these signals is very important in terms of both time and cost, particularly in the detection of disease. EEGs are non-stationary signals and the seizure patterns are different between the various patients and sessions. In addition, EEG data is susceptible to several types of noise that negatively impact the precise detection of epileptic diseases. In order to tackle these challenges, we use a thorough learning approach which learns the discriminatory EEG characteristics of seizures automatically. An EEG signal classification method based on a ResNet, Bi-LSTM and SupportVector Machine (SVM) is proposed to significantly lower the specimen rate and improve detection efficiency. Bi-LSTM can receive all the important information in practise at a lower sample rate than the sampling principle. ResNetis commonly used to automatically extract higher layer features. SVM is used to construct the generalised, Optimal Classification Hyper plane, to ensure good classification for the EEG database. In particular, the time-series EEG data are segmented first into a sequence of overlapping eras to reveal the relationship between successive data samples, and then using ResNet the data is trained. Secondly, the Bi-directional Long Short-Term Memory network (Bi-LSTM) uses high-level representation of normal patterns and EEG seizure patterns to understand. Thirdly, these representations are fed into the Support Vector Machine Training/Classification function.