A Comparative Analysis of Seizure Detection via Scalogram using GoogLeNet, AlexNet and SqueezeNet

Nowadays, neural disorders have become a familiar and complex global health issue. Automated recognition of epileptic seizures using electroencephalogram (EEG) data may allow for more accurate diagnosis and better clinical care. An epileptology specialist could detect this abnormal brain activity directly by visually observing EEG data. If so, seizure identification may benefit from Convolution Neural Networks (CNN) because of their visual perception characteristics and the inherent ability to learn the discriminative features. This work has emphasized comparing the results obtained by applying the EEG data on three pre-trained networks i.e., GoogLeNet, AlexNet, and SqueezeNet. Among the three networks, AlexNet has shown the best performance having an accuracy of 95.61% followed by GoogLeNet with an accuracy of 94.99%, and SqueezeNet having an accuracy of 94.09%. The variations in performance metrics were observed by varying the hyperparameters.