To cope with the lack of highly skilled professionals, machine leaning with proper signal techniques is a key to establishing automated diagnostic-aid technologies to conduct epileptic electroencephalogram (EEG) testing. In particular, frequency filtering with appropriate passbands is essential to enhance biomarkers[-]such as epileptic spike waves[-]that are noted in the EEG. This paper introduces a novel class of convolutional neural networks (CNNs) having a bank of linear-phase finite impulse response filters at the first layer. These may behave as bandpass filters that extract biomarkers without destroying waveforms because of linear-phase condition. The proposed CNNs were trained with a large amount of clinical EEG data, including 15,899 epileptic spike waveforms recorded from 50 patients. These have been labeled by specialists. Experimental results show that the trained data-driven filter bank with supervised learning is dyadic like discrete wavelet transform. Moreover, the area under the curve achieved above 0.9 in most cases.
bioRxiv Subject Collection: Neuroscience