Research utilizes machine learning to improve gait analysis in rehabilitation

Photo credit: Neil Ta

The latest research published in a peer-reviewed article showcases a pioneering approach to gait analysis, a crucial aspect of rehabilitation and clinical diagnosis. Clinicians have long relied on gait indicators like step length, stride velocity, and joint angles to assess and treat gait issues. However, traditional statistical methods have limitations in analyzing the vast sets of data generated by instrumented gait analysis techniques.

This finding was published in the journal PLOS one.

The study, led by Professor Jan Andrysek and his team member Dr. Mohammad Pourmahmood Aghababa at the Holland Bloorview Children’s Hospital, introduces Machine Learning (ML) as a game-changer in the field of gait analysis. ML models offer flexibility, scalability, and independence from inferred assumptions, addressing the shortcomings of traditional statistical methods. These models can efficiently handle large datasets and complex interactions between predictors and dependent variables.

The research explores the application of ML-based classifiers to differentiate between two types of prosthetic interventions using gait parameters as inputs. Additionally, it employs explainable ML techniques to interpret ML models and identify the most influential gait parameters, providing valuable insights for clinicians and researchers.

“By leveraging ML techniques, we aim to enhance our understanding of gait changes related to physical rehabilitation,” said Professor Jan Andrysek, the corresponding author of the study. “Our objective is not only to differentiate between prosthetic interventions but also to develop explainable models that can provide insights about important gait parameters at both population and sample levels. We hope that this will help to improve interpretability of gait analysis data, and streamline clinical prosthetic gait analysis approaches.”

The study underscores the significance of explainable ML in gait analysis, particularly in clinical settings. Techniques such as Permutation Feature Importance (PFI) and Shapley Additive Explanations (SHAP) offer comprehensive insights into the underlying mechanisms of ML models, enabling clinicians to make informed decisions about patient care.

“Our research demonstrates the potential of ML in transforming gait analysis and rehabilitation,” added Mohammad, the first author of the  research. “With explainable ML, we can not only improve diagnosis and treatment but also understand the intricate relationship between gait parameters and prosthetic interventions.”

The findings of this study pave the way for future advancements in gait analysis and rehabilitation, offering hope for more personalized and effective treatments for individuals with gait issues.