Detection of lapses in responsiveness from the EEG.

A system capable of reliably detecting lapses in responsiveness ('lapses') has the potential to increase safety in many occupations. We have developed an approach for detecting the state of lapsing with second-scale temporal resolution using data from 15 subjects performing a one-dimensional (1D) visuomotor tracking task for two 1 h sessions while their electroencephalogram (EEG), facial video, and tracking performances were recorded. Lapses identified using a combination of facial video and tracking behaviour were used to train the classification models. Linear discriminant analysis was used to form detection models based on individual subject data and stacked generalization was utilized to combine the outputs of multiple classifiers to obtain the final prediction. The performance of detectors estimating the lapse/not-lapse state at 1 Hz based on power spectral features, approximate entropy, fractal dimension, and Lempel-Ziv complexity of the EEG was compared. Best lapse state estimation performance was achieved using the detector model created using power spectral features with an area under the curve from receiver operating characteristic analysis of 0.86 ± 0.03 (mean±SE) and an area under the precision-recall curve of 0.43 ± 0.09. A novel technique was developed to provide improved estimation of accuracy of detection of variable-duration events. Via this approach, we were able to show that the detection of lapse events from spectral power features was of moderate accuracy (sensitivity = 73.5%, selectivity = 25.5%).