Bracken, B.1, Elkin-Frankston, S.1, Palmon, N.1, deB Frederick, B.2
Presented at the Military Health System Research Symposium, Kissimmee, FL (August 2017).
Background
Army medic training often includes high-fidelity simulations. Trainers currently infer competence by observation alone—a challenging task. We designed a system to continuously monitor physiological proxies for cognitive workload to augment performance observations. Assessing cognitive workload using functional near-infrared spectroscopy (fNIRS) when individuals are seated is well established (Ayaz, et al., 2012), but methods to assess cognitive workload during normal activities are only recently emerging. Even in these studies, typical motion includes nothing more strenuous than walking (McKendrick, et al., 2017).
Method
We designed an fNIRS sensor to be mounted on the inside of a baseball cap or standard issue helmet to measure indicators of cognitive workload. We collected data from 19 teams of three undergraduate participants completing an obstacle course of physical and cognitive challenges. Tasks were previously validated for effects on cognitive and physical workload to allow us to validate MEDIC estimates of these measures in conditions similar to medic training simulations. These included: (1) baseline (sitting quietly); (2) word list memorization; (3) balance board; (4) 20 questions; (5) puzzle; (6) hot potato; (7) logic problems; (8) moving boxes; (9) word list recall; and (10) synchronized jump rope. Results: Because MEDIC is designed to work under much more strenuous conditions than those faced by most researchers assessing cognitive workload, the motion artifacts in our data were larger than normal. To address this, we preprocessed our raw data to remove the sometimes extreme motion artifacts. Fortunately other than the large transients in our data, the band-pass filtered raw data was of reasonable quality (i.e., the shape of the raw data was preserved). We needed a technique that would allow us to salvage data during high motion events. Much work has been done comparing techniques for rejecting NIRS motion artifacts (Cooper, et al., 2012). We evaluated several standard techniques, including MARA spline motion correction, (Scholkmann and Wolf, 2009), wavelet-based filtering (Molavi and Dumont, 2012), and combined MARA and wavelet filtering. Wavelet-based filtering improved data quality to the point that we were able to salvage data from even the highest-motion conditions. MARA spline motion correction did not further improve data quality.
Conclusions
This work has resulted in a fully-functional Data Fusion Engine that corrects for motion, de-noises, processes, and time-aligns data across multiple teammates. We are now analyzing our time-series data to determine whether information on oxygenated blood volume alone is enough to assess cognitive workload during physical activity as with computer-based tasks, or whether the effect of physical activity requires us to use additional derived variables to assess cognitive workload.
Acknowledgements
This work was supported by USA MRMC Contract W81XWH-14-C-0018. Views, opinions and/or findings herein are those of the author(s) and should not be construed as an official Department of the Army position, policy or decision. In conduct of research where humans are subjects, investigators adhered to policies regarding protection of human subjects prescribed by CFR Title 45, Volume 1, Part 46; Title 32, Chapter 1, Part 219; and Title 21, Chapter 1, Part 50 (Protection of Human Subjects).
Learning objectives
Describe the data handling techniques required of high-motion data; describe the techniques used to validate MEDIC; describe the results of the validation experiment
1 Charles River Analytics
2 McLean Hospital/Harvard Medical School
For More Information
To learn more or request a copy of a paper (if available), contact Bethany Bracken.
(Please include your name, address, organization, and the paper reference. Requests without this information will not be honored.)
