Double Attention-Based Deep Convolutional Neural Network for Seizure Detection Using EEG Signals

Epilepsy is a common neurological disorder caused by sudden and transient excessive excitation of neurons in the brain due to abnormal electrical dis-charge. Existing diagnosis methods using brain magnetic resonance imaging (MRI) and computed tomography (CT) scans have limitations that they rely on episodic symptoms and the subjectivity of doctors. Although electroencephalography (EEG) signals only record the electrical activity on the surface of the cerebral cortex and require appropriate timing for examination to capture abnormal brain electrical activity, EEG examinations are non-invasive and safe for patients. They can be used for real-time monitoring and evaluating treatment effects, playing an important role in the diagnosis and treatment of epilepsy. In this study, we proposed a Double Attention-based Convolutional Neural Network (CNN) for Seizure detection. It computed the Pearson Correlation Coefficient (PCC) of each channel, and mapping them to a correlation coefficient matrix for positional encoding, with attention-based CNN model feature extraction to obtain the final classification results. From AUBMC dataset and the CHB-MIT dataset, our proposed model achieved classification accuracies of 90.88% and 93.69% respectively.