Virtual reality, machine learning, and statistical modeling reveal an association between multiple concussion history and impaired perceptual decision-making
- 1Department of Health & Human Performance, University of Tennessee at Chattanooga, Chattanooga, Tennessee, United States of America
- 2Department of Business Analytics & Information Systems, Harbert College of Business, Auburn University, Auburn, Alabama, United States of America
Abstract
Current clinical assessment methods are insufficiently sensitive for detection of subtle post-concussion impairments in perceptual-cognitive function. The purpose of this study was to search for any perceptual response metrics associated with concussion history that might increase understanding of altered brain-behavior relationships. Immersive virtual reality test data were aggregated for 202 healthy adolescents and young adults (116 males, 86 females) who participated in different studies over a 2-year period. Left- versus right-directed neck rotation, arm reach, and step-lunge responses to sequential presentations of 2 types of horizontally moving visual stimuli were measured in terms of time to initiation of body segment movement (perceptual latency [PL]), as well as response completion (response time [RT]). Speed-accuracy tradeoff was represented by rate correct per second for PL (RCS-PL) and RT (RCS-RT) of neck and arm movements, and across-trial inconsistency was represented by PL variability (PLV) and RT variability (RTV). Both supervised machine learning and theory-based statistical regression methods were used to identify metrics that best discriminated between participants who reported a history of no concussion (NC), single concussion (SC), NC + SC, or multiple concussions (MC). Additionally, statistical regression was used to assess a theoretical relationship between metrics believed to align with components of the drift-diffusion computational model of decision-making. The best metric for discrimination between NC + SC and MC was Neck RCS-PL. Neck PLV values demonstrated a strong inverse logarithmic correlation with Neck RCS-PL (r = –0.796, P < 0.001). The Neck RCS-PL and Neck PLV behavioral metrics may have relevance to the two components of the drift-diffusion computational model of perceptual decision-making, and their combination may be associated with a cumulative and persisting deficiency after having sustained more than one lifetime concussion.
Related
- Sensorimotor learning compensates for irreducible perceptual bias
- Perceptual lengthening of visual time throughout learning of hand-reaching actions
- No practice and still perfect: Different routes for promoting offline gains in perceptual learning
- Distinct basal ganglia contributions to learning from implicit and explicit value signals in perceptual decision-making
- Temporal confounds emulate multivariate fMRI measures of perceptual learning
BibTeX
@article{wilkerson2026virtual,
title = {{Virtual reality, machine learning, and statistical modeling reveal an association between multiple concussion history and impaired perceptual decision-making}},
author = {Gary B. Wilkerson and Mohammad Joghataee and Ashish Gupta},
journal = {PLOS One},
year = {2026},
month = sep,
doi = {10.1371/journal.pone.0358042},
url = {https://doi.org/10.1371/journal.pone.0358042},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Virtual reality, machine learning, and statistical modeling reveal an association between multiple concussion history and impaired perceptual decision-making
AU - Gary B. Wilkerson
AU - Mohammad Joghataee
AU - Ashish Gupta
PY - 2026
DA - 2026/09/21
JO - PLOS One
DO - 10.1371/journal.pone.0358042
UR - https://doi.org/10.1371/journal.pone.0358042
AB - Current clinical assessment methods are insufficiently sensitive for detection of subtle post-concussion impairments in perceptual-cognitive function. The purpose of this study was to search for any perceptual response metrics associated with concussion history that might increase understanding of altered brain-behavior relationships. Immersive virtual reality test data were aggregated for 202 healthy adolescents and young adults (116 males, 86 females) who participated in different studies over a 2-year period. Left- versus right-directed neck rotation, arm reach, and step-lunge responses to sequential presentations of 2 types of horizontally moving visual stimuli were measured in terms of time to initiation of body segment movement (perceptual latency [PL]), as well as response completion (response time [RT]). Speed-accuracy tradeoff was represented by rate correct per second for PL (RCS-PL) and RT (RCS-RT) of neck and arm movements, and across-trial inconsistency was represented by PL variability (PLV) and RT variability (RTV). Both supervised machine learning and theory-based statistical regression methods were used to identify metrics that best discriminated between participants who reported a history of no concussion (NC), single concussion (SC), NC + SC, or multiple concussions (MC). Additionally, statistical regression was used to assess a theoretical relationship between metrics believed to align with components of the drift-diffusion computational model of decision-making. The best metric for discrimination between NC + SC and MC was Neck RCS-PL. Neck PLV values demonstrated a strong inverse logarithmic correlation with Neck RCS-PL (r = –0.796, P < 0.001). The Neck RCS-PL and Neck PLV behavioral metrics may have relevance to the two components of the drift-diffusion computational model of perceptual decision-making, and their combination may be associated with a cumulative and persisting deficiency after having sustained more than one lifetime concussion.
SN - 1932-6203
ER -