Classification of EEG signals using neural network and logistic regression.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Epileptic seizures are challenging to diagnose using conventional electroencephalograph (EEG) analysis due to signal non-stationarity.
• Accurate detection of epileptiform discharges in EEG is crucial for epilepsy diagnosis and understanding underlying mechanisms.

Purpose of the Study:

• To develop and compare novel methods for analyzing non-stationary EEG signals for epileptic seizure detection.
• To evaluate the efficacy of wavelet transform combined with artificial neural networks (ANN) and logistic regression (LR) for classifying epileptic seizures.

Main Methods:

• EEG signals were preprocessed using lifting-based discrete wavelet transform (LBDWT) for efficient feature extraction.
• Classification models were developed using logistic regression (LR) and multilayer perceptron neural networks (MLPNN) with LBDWT coefficients as input.
• Classifier performance was evaluated using receiver operating characteristic (ROC) curves and scalar performance measures.

Main Results:

• The LBDWT effectively captured transient features in EEG signals, localizing them in time and frequency.
• The MLPNN-based classifier demonstrated superior accuracy and performance compared to the LR-based classifier.
• The proposed LBDWT and MLPNN architecture provided a novel and reliable method for automatic epileptic seizure classification.

Conclusions:

• Wavelet transform, particularly LBDWT, is effective for analyzing complex, non-stationary EEG signals.
• MLPNNs offer a powerful and accurate approach for classifying epileptic seizures from EEG data.
• The developed system can serve as a valuable tool to support physicians in the epilepsy diagnosis process.