Predicting Future Errors During Skill Performance

Study Description:

Human motor skills are composed of sequences of individual actions performed with utmost precision. However, even highly skilled human behavior is susceptible to errors. When these errors occur, they may have serious consequences, for example, when pilots are manually landing a plane or when surgeons control robotic devices during surgery. In such cases, the ability to predict and prevent these upcoming errors from occurring would clearly be advantageous. We recently utilized a withinindividual machine learning strategy to characterize brain activity predictive of future motor skill performance errors, in a manner consistent with accepted practices in the field. Implementation of this novel approach combining brain oscillatory activity (particularly in the low frequency delta range) and behavior (keypress transition times, KTT) in our previous work showed that we can predict up to 70% of single future erroneous keypress actions within 0.1s before they occur. One limitation of this work is that 0.1s preceding errors does not give enough time for subjects to stop an upcoming erroneous action. This protocol aims to characterize using this within-individual machine learning approach, already demonstrated to be effective in our own lab, to predict future erroneous actions up to approximately 1s before they occur. We intend to eventually develop a warning signal that allows subjects to stop upcoming skill errors. This development would allow communication to rapidly inform subjects to stop potential future errors. To this effect, we will record neural magnetoencephalography (MEG) activity while human participants perform sequences of motor actions.

Objectives:

The primary objective is to detect future skill errors up to approximately 1s before they occur. To predict these future errors, we will evaluate how brain activity and behavioral features preceding an error differ from those preceding correct sequence keypresses. We will also explore the feasibility of providing a feedback signal to participants when brain activity encodes future errors in real-time. Secondarily, we will evaluate spatial and temporal features of neural representations and replay and their relationship with erroneous and correct sequences.

Endpoints:

The primary endpoint measure will be predictive classification accuracy of upcoming erroneous keypresses from combined brain oscillatory and behavioral features. The exploratory endpoint measures are the characterization of neural representations and replay associated with correct and erroneous keypresses.