A 1-D CNN-FSVM Model with a multi-scale sub-band feature learning for automated seizure detection

Seizure is a sudden surge in the electrical movement of the brain. The recurrent seizures will leads to the most common brain disorder called epilepsy. Hence, an automatic seizure detection technique is essential to the doctors for accurate seizure diagnosis and improves the life span of the patient. This work presents a deep learning-based multi-scale frequency band feature fusion method to classify seizure EEG signals. All the input signals are transformed into frequency sub-band domain using wavelet transform. Then three different sub-band combinations are formed and are separately given to the convolutional neural network (CNN) to extract the convoluted features. These obtained individual feature maps are combined to form a multi-scale frequency band feature vector. Finally, the Fuzzy based SVM is trained with the resultant feature vector for seizure classification. The experimental analysis shows the efficiency of the proposed method with 99.02% of accuracy in the case of healthy and ictal class.