Automatic Seizure Recognition Based on Data Enhancement and 1DCNN-BiLSTM Network Using EEG Signal

Epilepsy is one of the most common life-threatening neurological disorders in the world. The uncertain nature of seizures poses a challenge in the detection and recognition of epilepsy. Aiming at the problem of imbalance between seizure and non-seizure data, an automatic seizure recognition based on data enhancement and 1DCNN-BiLSTM network using scalp EEG is proposed in this paper. The raw EEG signals are denoised by discrete wavelet transform(DWT), and three frequency band sub-signals are selected for restructuring. Then, three efficient data enhancement techniques, MixUp, SuperMix, and Co-MixUp are used to address data imbalance. The model uses a combination of Convolutional Neural Network (CNN) with attention mechanism and Bidirectional Long Short-Term Memory (BiLSTM) to capture key EEG feature capabilities. Finally, the classification results of the three EEG sub-signals are fused to recognize epileptic seizures. Evaluations and experiments are performed on two public EEG datasets, Bonn dataset and CHB-MIT dataset, using five-fold cross-validation to assess the generalization ability of the model. Experimental results show that data enhancement based on Co-MixUp is more suitable for 1DCNN-BiLSTM model.