1- Department of Sports Biomechanics, Faculty of Sports Sciences, Shahid Bahonar University of Kerman, Kerman, Iran.
2- Department of Sports Biomechanics, Faculty of Educational Sciences and Psychology, University of Mohaghegh Ardabili, Ardabil, Iran.
3- Department of Sports Biomechanics, Faculty of Educational Sciences and Psychology, University of Mohaghegh Ardabili, Ardabil, Iran. ,amiralijafarnezhad@gmail.com
2- Department of Sports Biomechanics, Faculty of Educational Sciences and Psychology, University of Mohaghegh Ardabili, Ardabil, Iran.
3- Department of Sports Biomechanics, Faculty of Educational Sciences and Psychology, University of Mohaghegh Ardabili, Ardabil, Iran. ,
Abstract: (9 Views)
Objective: Artificial intelligence has made a significant breakthrough with its ability to analyze complex data and identify hidden patterns, especially in the field of gait biomechanics. The aim of this study was to review and classify different AI approaches in gait biomechanics analysis.
Methods: The present study was a systematic review. Searches were conducted in the Web of Science, SID, Magiran, Scopus, ISC, PubMed, and Google Scholar databases between 2020 and 2025 in both Persian and English. Inclusion criteria comprised experimental or applied studies using artificial intelligence for the analysis of human gait, involving data from healthy individuals or patients with neurological, muscular, or musculoskeletal conditions; reporting model performance metrics (e.g., accuracy, sensitivity, specificity) or measurable biomechanical parameters; employing direct motion capture or wearable/imaging sensors; and utilizing cross-sectional designs, algorithm validation, or clinical prediction/diagnostic approaches. Exclusion criteria included theoretical studies, narrative reviews without original data, editorials, case reports, animal studies or simulations without human data, studies relying solely on classical statistical methods without artificial intelligence, articles with incomplete data or limited accessibility, studies focusing on activities other than gait, and studies with low methodological quality. Of the 85 identified articles, 13 studies met the eligibility criteria. Study quality was assessed using the Downs and Black questionnaire.
Findings: Based on the review of 14 studies on artificial intelligence and gait (including 7 studies on healthy individuals and 7 on patients), it was found that more than half of the studies (53.8%) used wearable sensors, approximately 23.1% employed markerless systems such as KinaTrax, 23% utilized machine/deep learning methods, and 10% applied conventional motion analysis systems. Cumulative analysis indicated that wearable sensors, particularly when combined with machine learning models such as Stack and SVR, were highly capable of classifying gait episodes (mean sensitivity 0.961 and MAE% below 2.1% for key parameters), reflecting the relatively high accuracy of these approaches. In studies using machine learning, the combination of ResNet101 and Naïve Bayes performed well in classifying body posture (sensitivity 0.87), and LSTM models also yielded notable results for gait path prediction in VR environments, particularly for short-term predictions (14.5 mm error), although long-term predictions required additional data such as eye-tracking. Furthermore, the use of smart insoles combined with the RF algorithm demonstrated the feasibility of extracting digital biomarkers for managing conditions such as sarcopenia, although the number of studies was limited and sample sizes were small.
Conclusion: The present findings indicate that the combination of wearable sensors particularly self-powered triboelectric sensors with machine and deep learning methods (such as SVR, ResNet101, and LSTM) has considerable potential for analyzing biomechanical gait parameters and predicting movement trajectories in laboratory and simulated environments, such as VR. However, the existing evidence is limited to small sample sizes and controlled conditions, and few studies have been conducted on actual patients or in clinical settings. Therefore, any clinical or rehabilitation applications of these technologies still require further research and validation in real-world environments.
Methods: The present study was a systematic review. Searches were conducted in the Web of Science, SID, Magiran, Scopus, ISC, PubMed, and Google Scholar databases between 2020 and 2025 in both Persian and English. Inclusion criteria comprised experimental or applied studies using artificial intelligence for the analysis of human gait, involving data from healthy individuals or patients with neurological, muscular, or musculoskeletal conditions; reporting model performance metrics (e.g., accuracy, sensitivity, specificity) or measurable biomechanical parameters; employing direct motion capture or wearable/imaging sensors; and utilizing cross-sectional designs, algorithm validation, or clinical prediction/diagnostic approaches. Exclusion criteria included theoretical studies, narrative reviews without original data, editorials, case reports, animal studies or simulations without human data, studies relying solely on classical statistical methods without artificial intelligence, articles with incomplete data or limited accessibility, studies focusing on activities other than gait, and studies with low methodological quality. Of the 85 identified articles, 13 studies met the eligibility criteria. Study quality was assessed using the Downs and Black questionnaire.
Findings: Based on the review of 14 studies on artificial intelligence and gait (including 7 studies on healthy individuals and 7 on patients), it was found that more than half of the studies (53.8%) used wearable sensors, approximately 23.1% employed markerless systems such as KinaTrax, 23% utilized machine/deep learning methods, and 10% applied conventional motion analysis systems. Cumulative analysis indicated that wearable sensors, particularly when combined with machine learning models such as Stack and SVR, were highly capable of classifying gait episodes (mean sensitivity 0.961 and MAE% below 2.1% for key parameters), reflecting the relatively high accuracy of these approaches. In studies using machine learning, the combination of ResNet101 and Naïve Bayes performed well in classifying body posture (sensitivity 0.87), and LSTM models also yielded notable results for gait path prediction in VR environments, particularly for short-term predictions (14.5 mm error), although long-term predictions required additional data such as eye-tracking. Furthermore, the use of smart insoles combined with the RF algorithm demonstrated the feasibility of extracting digital biomarkers for managing conditions such as sarcopenia, although the number of studies was limited and sample sizes were small.
Conclusion: The present findings indicate that the combination of wearable sensors particularly self-powered triboelectric sensors with machine and deep learning methods (such as SVR, ResNet101, and LSTM) has considerable potential for analyzing biomechanical gait parameters and predicting movement trajectories in laboratory and simulated environments, such as VR. However, the existing evidence is limited to small sample sizes and controlled conditions, and few studies have been conducted on actual patients or in clinical settings. Therefore, any clinical or rehabilitation applications of these technologies still require further research and validation in real-world environments.
Type of Study: Systematic Review |
Subject:
Sport Pathology and Corrective Movements
Received: 3/12/2025 | Accepted: 2/02/2026
Received: 3/12/2025 | Accepted: 2/02/2026
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
Copyright © The Author(s);
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC-By-NC), which permits use, distribution, and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
Contact Information
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC-By-NC), which permits use, distribution, and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
Contact Information
