Patient-specific method of sleep electroencephalography using wavelet packet transform and Bi-LSTM for epileptic seizure prediction

Epileptic seizures during sleep increase the probability of complications and sudden death in patients. Effective epileptic seizure prediction in sleep can assist doctors (patients) in administering (receiving) effective treatments to reduce the abovementioned probability. Most existing prediction methods do not consider electroencephalogram (EEG) features during sleep, leading to low accuracies in predicting seizures during sleep. Furthermore, traditional machine-learning based approaches might be susceptible to a high false alarm rate. In this paper, a novel patient-specific method based on deep learning and sleep scalp EEG is proposed to predict epileptic seizures. Raw EEG data are preprocessed to remove noise and reduce the complexity. Wavelet energy is used to reveal the features of the EEG signals in the time-frequency domain as the input of the classification phase. Bidirectional long short-term memory (Bi-LSTM) networks are applied to excavate the most discriminative features to obtain classification results combined with leave-one-out cross-validation method. Subsequently, a two-step post-processing process optimises the prediction results. Based on four patients reasonably selected from the CHB-MIT scalp database, our experiments with various frequency band constraints demonstrate that the delta and gamma-band signals are important factors affecting seizure prediction performance. The classification performance of Bi-LSTM in our method with an accuracy of 99.47%, a sensitivity of 99.34%, and a specificity of 99.60%, is higher than that achieved with other neural networks. Additionally, compared with the existing methods for seizure prediction during sleep, the proposed method has a better prediction performance.