Artificial neural network for detecting drowsiness from EEG recordings

We describe a novel method for classifying alert vs. drowsy states from one-second long sequences of full spectrum EEG recordings. This method uses time series of inter-hemispeheric and intra-hemispheric cross spectral densities of full spectrum EEG as input to an artificial neural network (ANN) with two discrete outputs: drowsy and alert. The experimental data were collected from 17 subjects. Two experts in EEG interpretation visually inspected the data and provided the necessary expertise for the training of an ANN. After several experiments we selected the learning vector quantization (LVQ) as the most suitable neural network and used the data from 5 subjects for the training. Classification properties of LVQ were validated using the data recorded from the remaining 12 subjects, whose EEG recordings have not been used for the training of the ANN. The statistics were used as a measure of potential applicability of the LVQ: the t-distribution showed that in 95% (confidence interval) of the target group the matching between the human assessment and the network output was 94, 37±1.95 percent.