A FUZZY-DEEP LEARNING APPROACH FOR MEASURING EPILEPSY SEVERITY USING EEG SIGNAL

Abstract

Epilepsy is a neurological disorder that affects more than 75 million people in the world. It has resulted in an increase in the mortality rate, especially in Sub-Saharan Africa, due to the lack of experienced medical experts to diagnose the disease, leading to misdiagnosis and time-consuming diagnoses. Several automatic epileptic seizure detection methods have been used to extract features from EEG signals but lack the capacity to calibrate the characterizing features of epileptic and non-epileptic EEG signals overlap. Hence, in this paper a deep learning Long Short-Term Memory (LSTM) algorithm and fuzzy system are proposed using electroencephalograph signals (EEG) with 24 epileptic subjects containing 18 EEG channels each. The EEG signals were pre-processed to remove artifacts generated during EEG recordings using a Notch filter for a band stop of 64 Hz and a band pass of 32 Hz. The deep-learning model based on LSTM is used for training of 100 segments per channel epileptic signals and 33 segments used for recognizing epileptic signals, with performance metrics of accuracy, time, precision, recall, and F1 score used to evaluate performance. The extracted parameters from the epileptic signals, Signal Energy (SE) and Logarithmic Band Power (LBP), serve as input to the fuzzy inference system. A triangle membership function that fuzzifies the extracted features to establish intensity scales using nine (9) fuzzy rules in a fuzzy inference system (FIS) was used to characterize each of the disease severities as low, medium, and high in the FIS, and the result showed that the proposed model has potential in classifying epileptic EEG signals.