Efficient Seizure Prediction and EEG Channel Selection Based on Multi-Objective Optimization

Epileptic seizures are unpredictable events due to sudden abnormal electrical activities in the brain of epilepsy patients. A seizure can be predicted by analyzing the EEG signals to prevent unwanted life risks. The goal of this paper is to implement a method that will apply to design a lightweight, wearable, and efficient seizure prediction device. The proposed method will satisfy two objectives. The first objective is relevant feature extraction for the classification of EEG signals with excellent accuracy. The second objective is the use of fewer EEG channels. In this paper, one 1D-CNN is applied for feature extraction and classification of raw EEG signals for early prediction of seizure events. The 1D-CNN is faster compared to 2D-CNN, which uses fewer trainable parameters. Hence, it is suitable to implement a low-power energy-efficient seizure prediction device. In this paper, the NSGA-II algorithm is applied to get the optimum set of EEG channels for seizure prediction. The NSGA-II algorithm identifies a set of three EEG channels from twenty-two channels as the optimum channel set. The proposed method optimizes the EEG channels from 22 to 3, i.e., 86.36% channel reduction. It provides the classification accuracy, sensitivity, and specificity of 0.9651, 0.9655, and 0.9647, respectively. The proposed method is better than the state-of-the-art works under the condition of using three channels. The proposed method provides excellent performance using only three EEG channels, which will be applicable to design a lightweight, low-power, and wearable seizure prediction device.