Automated epileptic seizures detection using multi-features and multilayer perceptron neural network

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Epileptic seizure detection from long-term electroencephalogram (EEG) signals is crucial for patient treatment.
• Manual analysis of EEG is time-consuming, prone to errors, and requires expert neurologists.
• Automated seizure detection systems are essential for accurate and efficient clinical diagnosis.

Purpose of the Study:

• To develop and evaluate an automated algorithm for detecting epileptic seizures using multi-channel EEG signals.
• To assess the efficacy of a multilayer perceptron neural network (MLPNN) classifier combined with extracted signal features.
• To provide a reliable tool for real-time seizure recognition, aiding clinical diagnosis.

Main Methods:

• Collected multi-channel EEG data from patients after ethical approval.
• Preprocessed EEG signals to remove noise and artifacts.
• Extracted four key features: power spectral density, Shannon entropy, Renyi entropy, and Teager energy.
• Utilized a multilayer perceptron neural network (MLPNN) classifier with single and multi-feature inputs.
• Developed a MATLAB-based graphical user interface for automated analysis.

Main Results:

• The proposed algorithm demonstrated high performance using multi-features.
• Achieved a sensitivity of 97.1% and specificity of 97.8%.
• Reported a false detection rate of 1 event per hour (1 h⁻¹).

Conclusions:

• The developed multi-feature MLPNN algorithm is effective for real-time epileptic seizure recognition from EEG.
• The system provides an automated biomarker for distinguishing normal and epileptic EEG signals.
• This automated tool can significantly assist neurologists in diagnosing epilepsy and improving patient care.