Date of Award
Doctor of Philosophy
Kristian M. O'Connor
Stephen C. Cobb, Kevin G. Keenan, Allison Hyngstrom, Jake Luo
Accelerometers, Falls, Obstacle Avoidance, Trips, Walking, Wearable Devices
Falls remain a significant problem for stroke patients. Tripping, the main cause of falls, occurs when there is insufficient clearance between the foot and ground. Based on an individual’s gait deficits, different joint angles and coordination patterns are necessary to achieve adequate foot clearance during walking. However, gait deficits are typically only quantified in a research or clinical setting, and it would be helpful to use wearable devices – such as accelerometers – to quantify gait disorders in real-world situations. Therefore, the objective of this project was to understand gait characteristics that influence the risk of tripping, and to detect these characteristics using accelerometers.
Thirty-five participants with a range of walking abilities performed normal walking and attempted to avoid tripping on an unexpected object while gait characteristics were quantified using motion capture techniques and accelerometers. Multiple regression was used to identify the relationship between joint coordination and foot clearance, and multiple analysis of variance was used to determine characteristics of gait that differ between demographic groups, as well as those that enable obstacle avoidance. Machine learning techniques were employed to detect joint angles and the risk of tripping from patterns in accelerometer signals.
Measures of foot clearance that represent toe height throughout swing instead of at a single time point are more sensitive to changes in joint coordination, with hip-knee coordination during midswing having the greatest effect. Participants with a history of falls or stroke perform worse than older non-fallers and young adults on many factors related to falls risk, however, there are no differences in the ability to avoid an unexpected obstacle between these groups. Individuals with an inability to avoid an obstacle have lower scores on functional evaluations, exhibit limited sagittal plane joint range of motion during swing, and adopt a conservative walking strategy.
Machine learning processes can be used to predict knee range of motion and classify individuals at risk for tripping based on an ankle-worn accelerometer. This work is significant because a portable device that detects gait characteristics relevant to the risk of tripping without expensive motion capture technology may reduce the risk of falls for stroke patients.
Benson, Lauren, "Identifying Gait Deficits in Stroke Patients Using Inertial Sensors" (2016). Theses and Dissertations. 1336.