Citation

Tirdad K, Dela Cruz A, Austin C, Sadeghian A, Mousavi Nia S, Cusimano M. Comput. Methods Programs Biomed. Update 2021; 1: e100026.

Copyright

(Copyright © 2021, Elsevier Publishing)

DOI

10.1016/j.cmpbup.2021.100026

PMID

unavailable

Abstract

Many concussions, the mildest form of TBI, go unreported; so the true incidence of TBI makes it the commonest or second most common neurological condition, next to migraines. A concussion can interfere with the transfer of information across the connecting axons in the brain that can be disrupted by TBI, thus resulting in a wide range of symptoms and signs of injury. Although it is known that rapid eye movements, called saccades, can be affected by TBI, the ability to distinguish different phases during the recovery or non-recovery from mild TBI like concussion is not well studied. This research aimed to develop a Machine Learning(ML)-based model that could classify stages after concussions through saccadic eye movement. A dataset of 34 mild traumatic brain injury (mTBI), 27 persisting symptoms called post-trauma syndrome (PTS), and 31 healthy (Control) participants with no prior history of acquired head trauma was collected. Each participant underwent Step, Anti-saccade, and Go/No-Go saccade test. Statistical analysis of each trial's features generated 3450 additional engineered features. An ensemble model, which consisted of various random forest classifiers, was implemented and trained on selected features to classify TBI based on the 116 selected features. The final model classified mTBI vs. PTS vs. Control with an accuracy of 87.8% and TBI (mTBI and PTS) vs. Control with an accuracy of 91.1%. The application of ML allowed the analysis of complex nonlinear patterns in saccadic eye movement to be distinguished and patients' classified as mTBI, PTS, or Control.

Language: en

Keywords

Ensemble model; Machine learning; Random forest; Saccade; Traumatic brain injury