Research Area:  Machine Learning
Monitoring and recording brain activities using Electroencephalograms (EEGs) has become the foremost wide applied tool by physicians for epilepsy diagnosis due to viable reasons like its availability, simplicity, and low cost. In this paper, we propose an automatic epileptic seizure detection framework based on deep learning techniques that are applied to raw EEG signals recordings without the overhead of features extraction. The proposed framework uses one-dimensional deep convolutional autoencoder for features extraction and dimensionality reduction. Three different neural networks systems classifiers are evaluated. Classification between normal and ictal cases has achieved 100% accuracy on all systems. The best classification results between the normal, interictal and ictal cases accomplished a 99.33% average overall accuracy using Bidirectional Long Short-Term Memory.
Author(s) Name:  Ahmed M. Abdelhameed; Hisham G. Daoud; Magdy Bayoumi
Conferrence name:  IEEE International Workshop on Signal Processing Systems (SiPS)
Publisher name:  IEEE