Vibrotactile Sensory Augmentation and Machine Learning Based Approaches for Balance Rehabilitation

Vestibular disorders and aging can negatively impact balance performance. Currently, the most effective approach for improving balance is exercise-based balance rehabilitation. Despite its effectiveness, balance rehabilitation does not always result in a full recovery of balance function. In this dissertation, vibrotactile sensory augmentation (SA) and machine learning (ML) were studied as approaches for further improving balance rehabilitation outcomes. Vibrotactile SA provides a form of haptic cues to complement and/or replace sensory information from the somatosensory, visual and vestibular sensory systems. Previous studies have shown that people can reduce their body sway when vibrotactile SA is provided; however, limited controlled studies have investigated the retention of balance improvements after training with SA has ceased. The primary aim of this research was to examine the effects of supervised balance rehabilitation with vibrotactile SA. Two studies were conducted among people with unilateral vestibular disorders and healthy older adults to explore the use of vibrotactile SA for therapeutic and preventative purposes, respectively. The study among people with unilateral vestibular disorders provided six weeks of supervised in-clinic balance training. The findings indicated that training with vibrotactile SA led to additional body sway reduction for balance exercises with head movements, and the improvements were retained for up to six months. Training with vibrotactile SA did not lead to significant additional improvements in the majority of the clinical outcomes except for the Activities-specific Balance Confidence scale. The study among older adults provided semi-supervised in-home balance rehabilitation training using a novel smartphone balance trainer. After completing eight weeks of balance training, participants who trained with vibrotactile SA showed significantly greater improvements in standing-related clinical outcomes, but not in gait-related clinical outcomes, compared with those who trained without SA. In addition to investigating the effects of long-term balance training with SA, we sought to study the effects of vibrotactile display design on people’s reaction times to vibrational cues. Among the various factors tested, the vibration frequency and tactor type had relatively small effects on reaction times, while stimulus location and secondary cognitive task had relatively large effects. Factors affected young and older adults’ reaction times in a similar manner, but with different magnitudes. Lastly, we explored the potential for ML to inform balance exercise progression for future applications of unsupervised balance training. We mapped body motion data measured by wearable inertial measurement units to balance assessment ratings provided by physical therapists. By training a multi-class classifier using the leave-one-participant-out cross-validation method, we found approximately 82% agreement among trained classifier and physical therapist assessments. The findings of this dissertation suggest that vibrotactile SA can be used as a rehabilitation tool to further improve a subset of clinical outcomes resulting from supervised balance rehabilitation training. Specifically, individuals who train with a SA device may have additional confidence in performing balance activities and greater postural stability, which could decrease their fear of falling and fall risk, and subsequently increase their quality of life. This research provides preliminary support for the hypothesized mechanism that SA promotes the central nervous system to reweight sensory inputs. The preliminary outcomes of this research also provide novel insights for unsupervised balance training that leverage wearable technology and ML techniques. By providing both SA and ML-based balance assessment ratings, the smart wearable device has the potential to improve individuals’ compliance and motivation for in-home balance training.PHDMechanical EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttps://deepblue.lib.umich.edu/bitstream/2027.42/143901/1/baotian_1.pd