Real-Time Alertness Monitoring
List of Relevant Publications: Alertness/Vigilance Monitoring
Basic Paper: Makeig S, and Inlow M, 1993
Latest Paper:Scott Makeig, Tzyy-Ping Jung, and Terrence J. Sejnowski, "Awareness during Drowsiness: Dynamics and Electrophysiological Correlates" (.html | .pdf), Can J Exp Psychol 54(4):266-73, 2000.
A visuomotor task for vigilance research: S. Makeig and M. Jolley, CTT: A compensatory tracking task for monitoring alertness (.pdf 50k), Technical Document 96-3C Naval Health Research Center, San Diego, 1996 Download the executable, technical document, and (Borland C++) source code. NOTE: Satisfactory timing and operation of the code have NOT been verified for post-1996 versions of Windows -- Check its performance carefully before using.
- The Neural Human-System Interface (NHSI) technology concept
- A NIPS95 Post-conference Workshop on NHSI technology.
Estimating Changes in the Level of Alertness from the EEG Power Spectrum
Many studies of vigilance research during the past few decades have shown that, for most or all operators engaged in attention-intensive and monotonous tasks, retaining a constant level of alertness is rare, if not impossible. Alertness deficits may lead to severe consequences in occupations ranging from air traffic control to monitoring of nuclear power plants. Changes in the electroencephalographic (EEG) power spectrum accompany these fluctuations in the level of alertness, as assessed by measuring simultaneous changes in EEG and performance on an auditory monitoring task. By combining power spectrum estimation, principal component analysis and artificial neural networks, we accurately estimate, continuously and near real-time, an operator's global level of alertness using EEG measures from as few as two central scalp sites.
Description of Sample Session
The figure above shows results of alertness estimation in a typical experimental session. During a half-hour session, the operator struggles to remain alert while performing a continuous monitoring task -- pressing a response button each time he hears a noise burst target. The task simulates a submarine sonar operator listening for a sound signature of the presence of another vessel.
The upper panel shows the operator's responses to the targets -- RED pulses represent failures to respond to presented targets, while GREEN downward pulses represent successful detections by the operator. When the operator is alert, he easily detects and responds to each of the targets. As the operator loses alertness and becomes drowsy, he begins to miss targets...
The lower panel shows, in BLUE, the time course of error rate in the session, smoothed from the data for individual targets shown above it. The RED line shows the real-time error rate estimate derived wholly from the subject's EEG recorded at two scalp channels. An individualized model of alertness changes in the EEG for this subject was derived from data recorded in a previous session. This model, incorporating spectral analysis and artificial neural network processing, can accurately estimate the probability of the subject failing to detect targets throughout the session.
During the first ten minutes of this session, the subject experienced a wave of drowsiness. Despite his best efforts to remain alert, he was unable to respond to up to one third of the targets presented. Throughout, the EEG-based error-rate estimate correctly estimated major changes in the operator's alertness level.
In this experiment, no feedback was given to the operator. Experiments showed that EEG-based error rate estimates can be used to deliver feedback to operators which can help them to manage their own alertness and can improve overall monitoring performance.