Design and analysis of an adaptive compressive sensing architecture for epileptic seizure detection

Epileptic detection techniques rely heavily on the Electroencephalography (EEG) as a representative signal carrying valuable information pertaining to the current brain state. In this work, we investigate the stability of time domain EEG features while varying the channel conditions. We identify the feature sets that would provide the most robust EEG classification accuracy. Moreover, an embedded Compressive Sensing (CS)-based EEG encoding system whose complexity is adapted to the channel condition is proposed. We also propose a framework called Classification Accuracy-Compression Ratio-Signal to Noise Ratio (CA-CR-SNR) that adapts compression ratio according to the channel condition. Simulation results show that selecting appropriate EEG feature combinations can relatively overcome the impact of bad channel conditions; however, this simple solution is still inadequate. The proposed adaptive algorithm reconfigures the compression ratio based on a channel feedback signal to further improve the classification accuracy.