Feature fusion ensemble classification approach for epileptic seizure prediction using electroencephalographic...

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a neurological disorder characterized by recurrent seizures, significantly impacting patients' quality of life.
• Accurate prediction of epileptic seizures using electroencephalogram (EEG) signals remains challenging due to noise, dimensionality, and feature extraction complexities.

Purpose of the Study:

• To develop a robust and computationally efficient method for accurate epileptic seizure prediction.
• To address preprocessing, channel selection, and feature extraction challenges in EEG-based seizure prediction.

Main Methods:

• A novel preprocessing pipeline involving Butterworth, wavelet, and Fourier transforms for EEG signal denoising.
• Dimensionality reduction using an optimal spatial filter and feature extraction combining handcrafted and 1D CNN-based features.
• Ensemble classification using a model-agnostic meta-learner with LSTM as the base classifier for interictal and preictal state prediction.

Main Results:

• The proposed methodology achieved 99.34% sensitivity and 98.67% specificity on the CHB-MIT dataset.
• A low false positive alarm rate of 0.039 was recorded.
• The method demonstrated superior performance compared to existing techniques.

Conclusions:

• The developed method offers a significant advancement in epileptic seizure prediction accuracy and efficiency.
• The computational efficiency makes the proposed approach suitable for real-time seizure prediction applications.