A compact spatial attention model for automated epileptic seizure detection using multichannel EEG

Epilepsy is a non-communicable brain disorder that affects a wide population across the globe. It is diagnosed by the visual analysis of EEG which is tedious and time consuming. Automated seizure detection based on multichannel EEG is crucial for the timely identification and treatment initialisation of the disease. A seizure detection model which incorporates compact spatial attention convolutional neural network is proposed in this paper. Model is composed of 3 blocks. A CNN block that automatically captures EEG spatial features, an attention block which focus on discriminative features and a classifier which successfully classifies the seizure and non-seizure classes. Experimenting on CHBMIT dataset, our model achieved test accuracy of 91.35 %, F1score of 97.09%, sensitivity of 96.15, selectivity of 98.08%, MCC of 94% and precision of 98.04%.