744 - Longitudinal Monitoring of Gait Parameters for New Lower Limb Amputees with Physical Therapy using Video-Based Gait Analysis

Objectives: Gait training in physical therapy is a common standard of practice for new lower limb prosthetic users. Progress is typically assessed through functional outcome measures such as the 10-meter or 6-minute walk tests. While these tests can measure walking speed and endurance, they fall short of capturing gait quality or quantifying gait parameters, which literature shows to be of therapeutic value. Routine access to quantitative gait assessment could provide clinicians with benchmarks to optimize treatment interventions. Quantified analysis could also provide additional evidence to insurance companies that an individual is progressing in therapy even when other measures have plateaued. Traditional gait analysis requires specialized equipment in a motion analysis lab making it very resource-intensive and inconvenient for the patient.

Design: Using human pose estimation techniques, we have developed and trained a custom gait analysis system that allows us to measure spatiotemporal gait parameters from video. Lower limb prosthetic users who were undergoing physical therapy at the Shirley Ryan AbilityLab were recorded during their regularly scheduled therapy sessions via a custom app on an Android phone (n= 25) with a target frequency of 1x/week. Processing of the videos was performed using our custom gait analysis pipeline to assess spatiotemporal parameters including cadence, velocity and temporal symmetry. Due to the diverse activities being performed in therapy, manual annotation of activity, assistive device, and algorithm performance were required.

Results: Longitudinal tracking across numerous subjects showed improvements in cadence and velocity parameters that accurately reflected coinciding functional outcome measures.

Conclusions: This pilot study shows the potential for video-based gait analysis to enable quantitative monitoring of gait parameters in clinical settings. Future work will explore ways to use the manual annotation to further automate our system, investigate how spatiotemporal information can be used to optimize individualized treatment plans and explore additional parameters that could aid in clinical decision-making.