Abstract

Algorithms for human pose estimation (HPE) are advancing rapidly and offer much promise for rehabilitation research and clinical practice. However, there remain numerous barriers including challenges integrating these algorithms into clinically useful pipeline, limited clinical relevance of the outputs, and uncertainty about how well they will generalize to patient populations. In this talk I will describe a computational and data management pipeline that enables applying cutting edge HPE algorithms to large numbers of clinical videos. I will also review the different types of algorithms and highlight situations where they struggle to generalize to patient populations. The outputs can be made more clinically relevant for gait analysis by training novel algorithms on data obtained in a clinical gait analysis laboratory. This gait transformer produces accurate estimates of temporal and spatiotemporal gait parameters. Finally, I will describe a smartphone app for collecting wearable sensor data and video and describe how these two modalities provide complementary data that can be fused together. This is demonstrated both for gait characterization during therapy as well as tracking changes in spasticity following botulinum toxin injections.

Biography

R. James Cotton is a Physician-Scientist at Shirley Ryan AbilityLab and Assistant Professor in the Northwestern University Department of Physical Medicine and Rehabilitation. His background is in electrical engineering (undergrad) and systems neuroscience (PhD). He completed his residency in PM&R at SRAlab, which included an additional year for research, where he developed a wearable sensor platform that records electomyography and kinematics to provide biofeedback and help guide therapies. His lab continues this, combining wearable sensor and video data with machine learning to develop new outcome measures and therapeutic approaches.