Creating an autoencoder single summary metric to assess gait quality to compare surgical outcomes in children with cerebral palsy: The Shriners Gait Index (SGI)

• Published in: Journal of biomechanics Vol. 168; p. 112092
• Main Authors: Wang, Shou-Jen, Tabashum, Thasina, Kruger, Karen M., Krzak, Joseph J., Graf, Adam, Chafetz, Ross S., Linton, Judi, Davids, Jon, Bagley, Anita, Bengani, Kanav, Albert, Mark V.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.05.2024; Elsevier Limited
• Subjects: Cerebral palsy; Deep learning; Developmental disabilities; Gait; Gait analysis; Kinematics; Machine learning; Movement disorders; Parameters; Therapeutic applications; Cerebral palsy; Gait analysis; Machine learning
• ISSN: 0021-9290; 1873-2380; 1873-2380
• DOI: 10.1016/j.jbiomech.2024.112092

Gait for individuals with movement disorders varies widely and the variability makes it difficult to assess outcomes of surgical and therapeutic interventions. Although specific joints can be assessed by fewer individual measures, gait depends on multiple parameters making an overall assessment metric difficult to determine. A holistic, summary measure can permit a standard comparison of progress throughout treatments and interventions, and permit more straightforward comparison across varied subjects. We propose a single summary metric (the Shriners Gait Index (SGI)) to represent the quality of gait using a deep learning autoencoder model, which helps to capture the nonlinear statistical relationships among a number of disparate gait metrics. We utilized gait data of 412 individuals under the age of 18 collected from the Motion Analysis Center (MAC) at the Shriners Children’s - Chicago. The gait data includes a total of 114 features: temporo-spatial parameters (7), lower extremity kinematics (64), and lower extremity kinetics (43) which were min–max normalized. The developed SGI score captured more than 89% variance of all 144 features using subject-wise cross-validation. Such summary metrics holistically quantify an individual’s gait which can then be used to assess the impact of therapeutic interventions. The machine learning approach utilized can be leveraged to create such metrics in a variety of contexts depending on the data available. We also utilized the SGI to compare overall changes to gait after surgery with the goal of improving mobility for individuals with gait disabilities such as Cerebral Palsy.