VestAid: A Tablet-Based Technology for Objective Exercise Monitoring in Vestibular Rehabilitation.

(1) Background: Current vestibular rehabilitation therapy is an exercise-based approach aimed at promoting gaze stability, habituating symptoms, and improving balance and walking in patients with mild traumatic brain injury (mTBI). A major component of these exercises is the adaptation of the vestibulo-ocular reflex (VOR) and habituation training. Due to acute injury, the gain of the VOR is usually reduced, resulting in eye movement velocity that is less than head movement velocity. There is a higher chance for the success of the therapy program if the patient (a) understands the exercise procedure, (b) performs the exercises according to the prescribed regimen, (c) reports pre- and post-exercise symptoms and perceived difficulty, and (d) gets feedback on performance. (2) Methods: The development and laboratory evaluation of VestAid, an innovative, low-cost, tablet-based system that helps patients perform vestibulo-ocular reflex (VORx1) exercises correctly at home without therapist guidance, is presented. VestAid uses the tablet camera to automatically assess patient performance and compliance with exercise parameters. The system provides physical therapists (PTs) with near real-time, objective (head speed and gaze fixation compliance), and subjective (perceived difficulty and pre- and post- exercise symptoms) metrics through a web-based provider portal. The accuracy of the head-angle and eye-gaze compliance metrics was evaluated. The accuracy of estimated head angles calculated via VestAid's low-complexity algorithms was compared to the state-of-the-art deep-learning method on a public dataset. The accuracy of VestAid's metric evaluation during the VORx1 exercises was assessed in comparison to the output of an inertial measurement unit (IMU)-based system. (3) Results: There are low mean interpeak time errors (consistently below 0.1 s) across all speeds of the VORx1 exercise, as well as consistently matching numbers of identified peaks. The spatial comparison (after adjusting for the lag measured with the cross-correlation) between the VestAid and IMU-based systems also shows good matching, as shown by the low mean absolute head angle error, in which for all speeds, the mean is less than 10 degrees. (4) Conclusions: The accuracy of the system is sufficient to provide therapists with a good assessment of patient performance. While the VestAid system's head pose evaluation model may not be perfectly accurate as a result of the occluded facial features when the head moves further towards an extreme in pitch and yaw, the head speed measurements and associated compliance measures are sufficiently accurate for monitoring patients' VORx1 exercise compliance and general performance.