Towards AI-enabled Insoles to Assess Gait Function in Persons with Neuromuscular Disease in Real-life Environments


Topic:

Other

Poster Number: 121

Author(s):

Ton Duong MEng, David Uher M.S., Sally Dunaway Young PT, DPT, Kayla Cornett PhD, Tina Duong MPT, PhD, Ashley Goodwin M.S., Jacqueline Montes PT, EdD, Damiano Zanotto PhD

Institutions:

1. Department of Mechanical Engineering, Stevens Institute of Technology, 2. Department of Rehabilitation and Regenerative Medicine, Columbia University Irving Medical Center, 3. Department of Neurology, Stanford University School of Medicine, 4. The University of Sydney & The Children's Hospital at Westmead, 5. Department of Neurology, Stanford University School of Medicine, 6. Department of Rehabilitation and Regenerative Medicine, Columbia University Irving Medical Center, 7. Department of Neurology, Columbia University Irving Medical Center, 8. Department of Mechanical Engineering, Stevens Institute of Technology

Traditional capacity measures have been used to capture ambulatory function in spinal muscular atrophy (SMA) and Duchenne muscular dystrophy (DMD) in controlled settings. However, these measures do not necessarily reflect function in patients’ real-life environments. As new disease-modifying treatments have recently become available for SMA and DMD, showing encouraging results in motor and ambulatory function, there is a compelling need for sensitive and accurate measures of performance to assess gait function in these patients in real-life settings.

This project focuses on the development of a new instrumented insole system capable of capturing spatiotemporal gait parameters during in-clinic assessments and in free-living conditions. The flexible, unobtrusive insoles fit inside persons’ own footwear and are controlled using a smartphone. The system can be synchronized with external sensors and can be delivered to patients and clinicians in a ruggedized “lab-on-the-go” case. A short user guide was developed for clinicians and patients.

Spatiotemporal gait parameters are extracted from raw sensor data using a new computational framework that combines traditional integration-based data processing methods with novel machine-learning abstraction models based on Genetic Algorithm and Support Vector Regression. To validate the models, gait data from 10 participants with SMA, DMD, and healthy controls were simultaneously collected with the insoles and with gold-standard gait analysis equipment (Zeno electronic walkway) as participants completed a battery of tests (6MWT, 10MWR, TUG, gait initiation/termination, curved walking) at Columbia University Irving Medical Center. The mean absolute errors were determined as follows: stride length (4.024 +/- 1.819 cm), stride velocity (4.416 +/- 2.363 cm/s), stride time (0.010 +/- 0.009 s), swing time (0.018 +/- 0.016 s), terminal double support (0.019 +/- 0.016 s). A proof-of-concept usability test was conducted at a satellite remote clinical site (Stanford University School of Medicine), demonstrating that the system can be operated without engineering support.

Future work will include system validation in real-life environments during extended-time measurements.