{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T06:17:20Z","timestamp":1790835440075,"version":"4.1.0"},"reference-count":51,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2025,9,24]],"date-time":"2025-09-24T00:00:00Z","timestamp":1758672000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100010046","name":"University of Essex","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100010046","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Bioeng. Biotechnol."],"abstract":"<jats:sec><jats:title>Objective<\/jats:title><jats:p>Deep learning (DL) has introduced new possibilities for estimating human joint moments - a surrogate measure of joint loads. However, traditional methods typically require extensive synchronised joint angle and moment data for model training, which is challenging to collect in real-world applications. This study aims to improve the accuracy and data efficiency of knee joint moment estimation via leveraging self-supervised learning techniques to automatically extract human motion representations from large-scale unlabeled joint angle datasets.<\/jats:p><\/jats:sec><jats:sec><jats:title>Method<\/jats:title><jats:p>We proposed a joint moment estimation method based on self-supervised learning (SSL), using a Transformer auto-encoder architecture. The model was pre-trained on large-scale unlabeled joint angle data with masked reconstruction to effectively capture spatiotemporal features of human motion. Subsequently, we fine-tuned the model using a small amount of labeled joint moment data, enabling accurate mapping from joint angles to joint moments. We evaluated this method on a dataset of 55 normally developing children and compared the performance of the pre-trained SSL model fine-tuned with different amounts of labeled data to a baseline model.<\/jats:p><\/jats:sec><jats:sec><jats:title>Results<\/jats:title><jats:p>The Fine-tuned model significantly outperformed the baseline model, especially in scenarios with scarce labeled data. MSEs were reduced from 24.00% to 45.16% (with an average reduction of 36.29%), and MAE from 18.18% to 37.80% (with an average reduction of 26.48%). The proposed SSL model exceeded the performance of the baseline model trained with 100% data, using only 20% of the data in the labeled dataset during fine-tuning. When both models were fine-tuned using only 5% of the labeled data, the proposed SSL achieved four-fold better performance than the baseline model.<\/jats:p><\/jats:sec><jats:sec><jats:title>Conclusion<\/jats:title><jats:p>This study demonstrates that self-supervised learning significantly improves the accuracy and data efficiency of joint moment estimation, providing a more efficient solution for biomechanical evaluation. The proposed model can reduce the burden of collecting data and expand clinical applications.<\/jats:p><\/jats:sec>","DOI":"10.3389\/fbioe.2025.1633513","type":"journal-article","created":{"date-parts":[[2025,9,24]],"date-time":"2025-09-24T05:42:07Z","timestamp":1758692527000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":3,"title":["Self-supervised learning enhances accuracy and data efficiency in lower-limb joint moment estimation from gait kinematics"],"prefix":"10.3389","volume":"13","author":[{"given":"Yifan","family":"Li","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiayu","family":"He","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Bernard","family":"Liew","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"David S.","family":"Hollinger","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Qichang","family":"Mei","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Behnam","family":"Gholami","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Maria","family":"Fasli","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Klaus","family":"McDonald-Maier","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiaojun","family":"Zhai","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2025,9,24]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"73","DOI":"10.1016\/s0966-6362(00)00099-0","article-title":"Comparison of inverse dynamics calculated by two-and three-dimensional models during walking","volume":"13","author":"Alkjaer","year":"2001","journal-title":"Gait & Posture"},{"key":"B2","doi-asserted-by":"publisher","first-page":"1215770","DOI":"10.3389\/fbioe.2023.1215770","article-title":"Performance of multiple neural networks in predicting lower limb joint moments using wearable sensors","volume":"11","author":"Altai","year":"2023","journal-title":"Front. Bioeng. Biotechnol."},{"key":"B3","first-page":"1597","article-title":"A simple framework for contrastive learning of visual representations","volume-title":"International conference on machine learning","author":"Chen","year":""},{"key":"B4","first-page":"699","article-title":"Adversarial robustness: from self-supervised pre-training to fine-tuning","volume-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition","author":"Chen","year":""},{"key":"B5","doi-asserted-by":"publisher","first-page":"144180","DOI":"10.1109\/access.2023.3344531","article-title":"Applications of self-supervised learning to biomedical signals: a survey","volume":"11","author":"Del Pup","year":"2023","journal-title":"IEEE Access"},{"key":"B6","doi-asserted-by":"publisher","first-page":"1940","DOI":"10.1109\/tbme.2007.901024","article-title":"Opensim: open-source software to create and analyze dynamic simulations of movement","volume":"54","author":"Delp","year":"2007","journal-title":"IEEE Trans. Biomed. Eng."},{"key":"B7","doi-asserted-by":"publisher","first-page":"074505","DOI":"10.1115\/1.4046713","article-title":"Modifying stride length in isolation and in combination with foot progression angle and step width can improve knee kinetics related to osteoarthritis; a preliminary study in healthy subjects","volume":"142","author":"Edd","year":"2020","journal-title":"J. Biomechanical Eng."},{"key":"B8","volume-title":"Rethinking supervised pre-training for better downstream transferring","author":"Feng","year":"2021"},{"key":"B9","volume-title":"A review on discriminative self-supervised learning methods","author":"Giakoumoglou","year":"2024"},{"key":"B10","doi-asserted-by":"publisher","first-page":"1431","DOI":"10.1002\/jbmr.2483","article-title":"Loading of hip measured by hip contact forces at different speeds of walking and running","volume":"30","author":"Giarmatzis","year":"2015","journal-title":"J. Bone Mineral Res."},{"key":"B11","doi-asserted-by":"publisher","first-page":"615","DOI":"10.1002\/jor.24226","article-title":"Three-dimensional motion of the knee-joint complex during normal walking revealed by mobile biplane x-ray imaging","volume":"37","author":"Gray","year":"2019","journal-title":"J. Orthop. Research\u00ae"},{"key":"B12","first-page":"21271","article-title":"Bootstrap your own latent-a new approach to self-supervised learning","volume":"33","author":"Grill","year":"2020","journal-title":"Adv. neural Inf. Process. Syst."},{"key":"B13","doi-asserted-by":"publisher","first-page":"1034","DOI":"10.1109\/jbhi.2021.3107532","article-title":"Cross-domain self-supervised complete geometric representation learning for real-scanned point cloud based pathological gait analysis","volume":"26","author":"Gu","year":"2021","journal-title":"IEEE J. Biomed. Health Inf."},{"key":"B14","doi-asserted-by":"publisher","first-page":"928","DOI":"10.1038\/s42003-023-05310-2","article-title":"A self-supervised deep learning method for data-efficient training in genomics","volume":"6","author":"G\u00fcnd\u00fcz","year":"2023","journal-title":"Commun. Biol."},{"key":"B15","article-title":"Deep learning scaling is predictable, empirically","author":"Hestness","year":"2017"},{"key":"B16","doi-asserted-by":"publisher","first-page":"222","DOI":"10.1016\/0966-6362(95)01057-2","article-title":"Scaling gait data to body size","volume":"3","author":"Hof","year":"1996","journal-title":"Gait & posture"},{"key":"B17","doi-asserted-by":"publisher","first-page":"2829","DOI":"10.1109\/jbhi.2023.3262164","article-title":"Estimation of lower extremity joint moments and 3d ground reaction forces using imu sensors in multiple walking conditions: a deep learning approach","volume":"27","author":"Hossain","year":"2023","journal-title":"IEEE J. Biomed. health Inf."},{"key":"B18","volume-title":"Predicting kinetics using musculoskeletal modeling and inertial motion capture","author":"Karatsidis","year":"2018"},{"key":"B19","first-page":"2","article-title":"Bert: pre-training of deep bidirectional transformers for language understanding","volume":"1","author":"Kenton","year":"2019","journal-title":"Proc. naacL-HLT"},{"key":"B20","volume-title":"Selffed: self-supervised federated learning for data heterogeneity and label scarcity in iomt","author":"Khowaja","year":"2023"},{"key":"B21","volume-title":"Adam: a method for stochastic optimization","author":"Kingma","year":"2014"},{"key":"B22","doi-asserted-by":"publisher","first-page":"1135","DOI":"10.1088\/0305-4470\/25\/5\/020","article-title":"Generalization in a linear perceptron in the presence of noise","volume":"25","author":"Krogh","year":"1992","journal-title":"J. Phys. A Math. General"},{"key":"B23","volume-title":"Sgdr: stochastic gradient descent with warm restarts","author":"Loshchilov","year":"2016"},{"key":"B24","doi-asserted-by":"publisher","first-page":"3253","DOI":"10.1007\/s11517-023-02890-3","article-title":"Estimation of lower limb joint moments based on the inverse dynamics approach: a comparison of machine learning algorithms for rapid estimation","volume":"61","author":"Mansour","year":"2023","journal-title":"Med. & Biol. Eng. & Comput."},{"key":"B25","doi-asserted-by":"publisher","first-page":"617","DOI":"10.1136\/ard.61.7.617","article-title":"Dynamic load at baseline can predict radiographic disease progression in medial compartment knee osteoarthritis","volume":"61","author":"Miyazaki","year":"2002","journal-title":"Ann. rheumatic Dis."},{"key":"B26","volume-title":"HBM2 - reference manual","author":"Motek","year":"2018"},{"key":"B27","doi-asserted-by":"publisher","first-page":"21","DOI":"10.1016\/s0966-6362(96)01078-8","article-title":"Gait study of patients with patellofemoral pain syndrome","volume":"5","author":"Nadeau","year":"1997","journal-title":"Gait & posture"},{"key":"B28","doi-asserted-by":"publisher","first-page":"77","DOI":"10.1007\/s10462-018-09679-z","article-title":"Machine learning and deep learning frameworks and libraries for large-scale data mining: a survey","volume":"52","author":"Nguyen","year":"2019","journal-title":"Artif. Intell. Rev."},{"key":"B29","first-page":"75","article-title":"Deep learning can recognize complex relationships","volume-title":"Artificial intelligence: what is behind the technology of the future?","author":"Paa\u00df","year":"2023"},{"key":"B30","doi-asserted-by":"publisher","first-page":"7912","DOI":"10.3390\/s24247912","article-title":"Measurement properties of wearable kinematic-based data collection systems to evaluate ball kicking in soccer: a systematic review with evidence gap map","volume":"24","author":"Palucci Vieira","year":"2024","journal-title":"Sensors"},{"key":"B31","volume-title":"Self-supervised learning for image super-resolution and deblurring","author":"Scanvic","year":"2023"},{"key":"B32","doi-asserted-by":"publisher","first-page":"109142","DOI":"10.1016\/j.dib.2023.109142","article-title":"Dataset of 3d gait analysis in typically developing children walking at three different speeds on an instrumented treadmill in virtual reality","volume":"48","author":"Senden","year":"2023","journal-title":"Data brief"},{"key":"B33","doi-asserted-by":"publisher","first-page":"12513","DOI":"10.1007\/s00521-023-08379-2","article-title":"Estimation of joint torques using an artificial neural network model based on kinematic and anthropometric data","volume":"35","author":"Serbest","year":"2023","journal-title":"Neural Comput. Appl."},{"key":"B34","doi-asserted-by":"publisher","first-page":"5876","DOI":"10.3390\/s21175876","article-title":"The use of synthetic imu signals in the training of deep learning models significantly improves the accuracy of joint kinematic predictions","volume":"21","author":"Sharifi Renani","year":"2021","journal-title":"Sensors"},{"key":"B35","volume-title":"On efficient training of large-scale deep learning models: a literature review","author":"Shen","year":"2023"},{"key":"B36","first-page":"1929","article-title":"Dropout: a simple way to prevent neural networks from overfitting","volume":"15","author":"Srivastava","year":"2014","journal-title":"J. Mach. Learn. Res."},{"key":"B37","doi-asserted-by":"publisher","first-page":"2095","DOI":"10.1109\/tbme.2024.3361888","article-title":"Self-supervised learning improves accuracy and data efficiency for imu-based ground reaction force estimation","volume":"71","author":"Tan","year":"2024","journal-title":"IEEE Trans. Biomed. Eng."},{"key":"B38","doi-asserted-by":"publisher","first-page":"333","DOI":"10.1016\/j.gaitpost.2017.08.025","article-title":"Factors influencing knee adduction moment measurement: a systematic review and meta-regression analysis","volume":"58","author":"Telfer","year":"2017","journal-title":"Gait & posture"},{"key":"B39","first-page":"1","volume-title":"Opencap: 3d human movement dynamics from smartphone videos","author":"Uhlrich","year":"2022"},{"key":"B40","doi-asserted-by":"publisher","first-page":"111449","DOI":"10.1016\/j.jbiomech.2023.111449","article-title":"Walking with shorter stride length could improve knee kinetics of patients with medial knee osteoarthritis","volume":"147","author":"Ulrich","year":"2023","journal-title":"J. Biomechanics"},{"key":"B41","doi-asserted-by":"publisher","first-page":"888","DOI":"10.1109\/tnsre.2020.2978537","article-title":"Real-time estimation of knee adduction moment for gait retraining in patients with knee osteoarthritis","volume":"28","author":"Wang","year":"","journal-title":"IEEE Trans. Neural Syst. Rehabilitation Eng."},{"key":"B42","first-page":"2574","article-title":"A survey on large-scale machine learning","volume":"34","author":"Wang","year":"","journal-title":"IEEE Trans. Knowl. Data Eng."},{"key":"B43","doi-asserted-by":"publisher","first-page":"9576","DOI":"10.3390\/s23239576","article-title":"Lower limb joint torque prediction using long short-term memory network and gaussian process regression","volume":"23","author":"Wang","year":"2023","journal-title":"Sensors"},{"key":"B44","volume-title":"Self-supervised learning for electroencephalogram: a systematic survey","author":"Weng","year":"2024"},{"key":"B45","doi-asserted-by":"publisher","first-page":"108016","DOI":"10.1016\/j.compbiomed.2024.108016","article-title":"Integrating an lstm framework for predicting ankle joint biomechanics during gait using inertial sensors","volume":"170","author":"Xiang","year":"2024","journal-title":"Comput. Biol. Med."},{"key":"B46","volume-title":"Evaluating fairness in self-supervised and supervised models for sequential data","author":"Yfantidou","year":"2024"},{"key":"B47","article-title":"How transferable are features in deep neural networks?","volume":"27","author":"Yosinski","year":"2014","journal-title":"Adv. neural Inf. Process. Syst."},{"key":"B48","volume-title":"Bissl: bilevel optimization for self-supervised pre-training and fine-tuning","author":"Zakarias","year":"2024"},{"key":"B49","doi-asserted-by":"publisher","first-page":"600","DOI":"10.1109\/tnsre.2022.3156786","article-title":"Lower-limb joint torque prediction using lstm neural networks and transfer learning","volume":"30","author":"Zhang","year":"","journal-title":"IEEE Trans. Neural Syst. Rehabilitation Eng."},{"key":"B50","doi-asserted-by":"publisher","first-page":"5895","DOI":"10.1109\/jbhi.2022.3207313","article-title":"Ankle joint torque prediction using an nms solver informed-ann model and transfer learning","volume":"26","author":"Zhang","year":"","journal-title":"IEEE J. Biomed. health Inf."},{"key":"B51","doi-asserted-by":"publisher","first-page":"6775","DOI":"10.1109\/tpami.2024.3387317","article-title":"Self-supervised learning for time series analysis: taxonomy, progress, and prospects","volume":"46","author":"Zhang","year":"2024","journal-title":"IEEE Trans. Pattern Analysis Mach. Intell."}],"container-title":["Frontiers in Bioengineering and Biotechnology"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fbioe.2025.1633513\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,9,24]],"date-time":"2025-09-24T05:42:10Z","timestamp":1758692530000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fbioe.2025.1633513\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,9,24]]},"references-count":51,"alternative-id":["10.3389\/fbioe.2025.1633513"],"URL":"https:\/\/doi.org\/10.3389\/fbioe.2025.1633513","relation":{},"ISSN":["2296-4185"],"issn-type":[{"value":"2296-4185","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,9,24]]},"article-number":"1633513"}}