| Artificial intelligence in gait analysis of adult musculoskeletal disorders: Applications and limitations: A narrative review |
| Paper ID : 1106-ISCSR3 (R1) |
| Authors |
|
Afaf Mohamed Tahoon1, Youssef Samy Alsaeed2, Mohamed Saeed Mobarez2, Abdelrahman Mohamed Badran *2 1Afaf Mohamed Tahoon - lecturer at physical therapy for orthopedics department, Cairo university 2Faculty of physical therapy for orthopedics department, Cairo university |
| Abstract |
| Background: The role of Machine Learning (ML) in gait analysis has significantly emerged in decision-making and treatment plan adjustment. However, there is a lack of conclusive literature integrating ML in the gait analysis of musculoskeletal disorders (MKD), highlighting a need for a comprehensive review. Purpose: Provide an overview of ML applications in gait analysis of adult MKD patients Methods: Cochrane, Web of Science, Google Scholar, and PubMed were the databases searched for ML in MSD among adults, covering the period from inception to March 2025, with restrictions to English language. Results: A total of 10 articles in gait analysis based on multiple ML models in MKD address lower limb joint impairments, spinal disorders, joint replacement, and fractures. For gait classification, ML with inertial measurement units achieved over 91% accuracy in classifying knee and ankle impairments. K-nearest Neighbor (KNN) demonstrated 96.6% accuracy in classifying gait through ground reaction forces (GRF). Support Vector Machine achieved an accuracy of 80.4% in lumbar radiculopathies gait classification and about 97-100% in total hip arthroplasty. EXtreme gradient boost (XGBoost) and random forest models demonstrated 88.69% and 93.7% accuracy of osteopenia and sarcopenia respectively Regarding, predicting recovery post-fracture, XGBoost achieved 86% accuracy highlighting ML’s ability to analyze gait patterns as biomarkers for recovery risks. A Long Short-Term Memory Autoencoder model shows promising results in monitoring the recovery status of different lower-limb impairments patients through reconstructing GRF. Despite the high accuracy of ML in gait analysis, some limitations prevent its integration, such as standardized datasets of MSK pathologies, explainable ML interfaces for clinicians, and cost-benefit burden on the healthcare system. Conclusion: ML has the potential ability to classify gait-related MKD, monitor recovery status, predicting post-fracture recovery. However, the applicability of ML has been limited by a lack of real-world clinical translation to bridge research and clinical practice. |
| Keywords |
| Gait analysis, machine learning (ML), musculoskeletal disorders (MSD), orthopedic gait deviations, adult |
| Status: Abstract Accepted |