EEG Correlates of Task Engagement and Mental Workload in Vigilance, Learning, and Memory Tasks

Review

In 2007, Berka et al published their article, EEG Correlates of Task Engagement and Mental Workload in Vigilance, Learning, and Memory Tasks. With the aim to improve our ‘capability to continuously monitor an individual’s level of fatigue, attention, task engagement, and mental workload in operational environments using physiological parameters’, they present the following:

  • A new EEG metric for task engagement
  • A new EEG metric for mental workload
  • A hardware and software solution for real-time acquisition and analysis of EEG using these metrics
  • The results of study of the use of these systems and metrics

The article focuses primarily on two related concepts: task engagement and mental workload. As they put it:

Both measures increase as a function of increasing task demands but the engagement measure tracks demands for sensory processing and attention resources while the mental workload index was developed as a measure of the level of cognitive processes generally considered more the domain of executive function.

Using features derived from the signals acquired using a wireless, twelve channel EEG headset, Berka et al trained a model using linear and quadratic discriminant function analysis to identify and quantify cognitive state changes. For engagement, the model gives probabilities for each of high engagement, low engagement, relaxed wakefulness, and sleep onset. For workload, the model gives probabilities for both low and high mental workload. (They appear to consider cognitive states as unlabeled combinations of probabilities of each of these classes.) The aim of their simplified model was generalizability across various subjects and scenarios, as well as the ability to implement the model in wireless, real-time systems.

They trained the model using 13 subjects performing a battery of tasks, and cross-validated it with 67 additional subjects performing a similar battery of tasks. Task order was not randomized in either training or cross validation. The batteries of tasks encompass a range of task types and difficulties. Unfortunately, the authors struggle to present these batteries of tasks as a cohesive whole and to argue for relationship between the tasks.

In general, Berka et al found that for the indexes they developed:

[T]he EEG engagement index is related to processes involving information-gathering, visual scanning, and sustained attention. The EEG-workload index increases with working memory load and with increasing difficulty level mental arithmetic and other problem-solving tasks.

My primary issue with this article revolves around the authors’ statement:

During [some] multi-level tasks, EEG-engagement showed a pattern of change that was variable across tasks, levels, and participants.

Indeed, these tasks represented a large portion of the task battery. The authors argue for the effectiveness of their engagement index, but never thoroughly address why this index is inconsistent across tasks, levels, and participants. At the very least, this might have been included in the authors’ suggestions for future work.

Open Questions

  • The authors gave very few details on the specifics of their wireless EEG system. Many recent products in this area have been of questionable usefulness, at best…
  • Why did the authors not control for ordering effects?
  • Why the different protocols for training and cross-validation? More than this, why modify tasks that were common across both protocols. Finally, if the authors were going to modify common tasks, why not modify those that seemed particularly problematic–at least as they presented them in the paper (e.g., “Trails”)?