Table 1 Classification results for all the input modalities and network structures.

Input forms	Network structures	Accuracy	Sensitivity	Specificity	F1 score	AUC	FDR
Raw time-series EEG	FCNN	0.985	0.963	0.985	0.031	0.983	0.020
	RNN	0.993	0.966	0.993	0.066	0.989	0.018
	1D CNN	0.996	0.965	0.996	0.118	0.990	0.015
Periodogram	FCNN	0.985	0.963	0.985	0.046	0.984	0.020
	RNN	0.982	0.967	0.982	0.026	0.985	0.024
	1D CNN	0.996	0.962	0.996	0.108	0.989	0.016
Image of STFT	2D CNN	0.998	0.967	0.998	0.194	0.991	0.011
40 × 250 image of EEG	2D CNN	0.999	0.966	0.999	0.407	0.993	0.009
40 × 750 image of EEG	2D CNN	0.999	0.969	0.999	0.492	0.998	0.008
O’shea et al.²⁶	1D CNN	0.997	0.959	0.997	0.136	0.990	0.012
Zhou et al.²⁵	1D CNN	0.995	0.957	0.995	0.089	0.989	0.017
Cao et al.²⁹	2D CNN	0.997	0.962	0.997	0.136	0.990	0.015

AUC: area under the curve, CNN: convolutional neural network, EEG: electroencephalogram, FCNN: fully connected neural network, FDR: false detection rate (n/h), RNN: recurrent neural network, STFT: short-time Fourier transform.

Quick links

Search