Presented at the 2015 Military Health System Research symposium, Ft. Lauderdale, FL (August 2015).
Background: US military medical personnel may be deployed to a variety of operational environments where their success saving lives depends on their ability to act quickly and effectively, both as individuals and as teams. Therefore, effective training must go beyond individual skills to include interactions among team members, and how those interactions transfer to operational environments. Currently, trainers must infer competence by observation alone—a challenging task. Automatically sensing indicators of cognitive workload can augment performance observations, offering insight into factors underlying that performance.
Method: We designed and demonstrated a system to augment training by Monitoring, Extracting, and Decoding Indicators of Cognitive workload (MEDIC). Our effort combines: (1) a multimodal suite of unobtrusive, field-ready neurophysiological, physiological, and behavioral sensors; (2) complex event processing to extract and fuse the best indicators of cognitive workload and team dynamics from the multiple, high-volume data streams originating from the sensor suite; and (3) probabilistic modeling techniques to interpret those indicators for easy understanding.
Results and Conclusions: We designed a device to be mounted on the inside of the trainee’s helmet to measure indicators of cognitive load, a body band device that includes additional sensors to measure indicators of physical activity, and a user interface (UI) through which the trainer can enter details of the simulation into the system prior to the actual exercise. Through this UI the trainer can take notes and record time-tagged photos and videos to record performance details as the scenario unfolds. During our pilot study, we successfully extracted and fused data from a team of two individuals across both physically (simulated CPR) and cognitively (arithmetic and n-back cognitive task) demanding conditions. We concisely identified complex sequences of events in raw data using causal event histories, identification of complex event patterns, and mining of causal and temporal relationships among events. MEDIC is capable of processing both continuous (heart rate, oxygenated hemoglobin) and discrete (communication events, trainer observations on performance) variables.We explored five methods of data modeling, and determined that for our pilot data weighted averaging was most accurate for estimating individual state (e.g., stress and cognitive workload), but Bayesian Networks and Causal Influence Models were most accurate for estimating team state (e.g., team dynamics). Finally, we designed a separate UI that displays these interpreted data after the simulation is complete, including information about each state for each individual and for the team as a whole. Future work will include validation of each component of MEDIC using both controlled laboratory studies and testing in realistic training environments (simulation centers).
1 Charles River Analytics
2 McLean Hospital/Harvard Medical School
3 Arizona State University
This work was supported by the U.S. Army Medical Research and Material Command under Contract No. W81XWH-14-C-0018. The views, opinions and/or findings contained in this report are those of the author(s) and should not be construed as an official Department of the Army position, policy or decision unless so designated by other documentation.
For More Information
(Please include your name, address, organization, and the paper reference. Requests without this information will not be honored.)