EEG-based lapse detection with high temporal resolution

A warning system capable of reliably detecting lapses in responsiveness (lapses) has the potential to prevent many fatal accidents. We have developed a system capable of detecting lapses in real-time with second-scale temporal resolution. Data was from 15 subjects performing a visuomotor tracking task for two 1-hour sessions with concurrent electroencephalogram (EEG) and facial video recordings. The detector uses a neural network with normalized EEG log-power spectrum inputs from two bipolar EEG derivations, though we also considered a multichannel detector. Lapses, identified using a combination of video rating and tracking behavior, were used to train our detector. We compared detectors employing tapped delay-line linear perceptron, tapped delay-line multilayer perceptron (TDL-MLP), and long short-term memory (LSTM) recurrent neural networks operating continuously at 1 Hz. Using estimates of EEG log-power spectra from up to 4 s prior to a lapse improved detection compared with only using the most recent estimate. We report the first application of a LSTM to an EEG analysis problem. LSTM performance was equivalent to the best TDL-MLP network but did not require an input buffer. Overall performance was satisfactory with area under the curve from receiver operating characteristic analysis of 0.84 +/- 0.02 (mean +/- SE) and area under the precision-recall curve of 0.41 +/- 0.08.