Unsupervised analysis of gait recovery after spinal cord injury
DOI:
https://doi.org/10.17979/ja-cea.2026.47.13773Keywords:
Spinal Cord Injury, Gait, Neurorehabilitation, Multimodal Data, Dimensionality Reduction, Machine LearningAbstract
Gait recovery in patients with incomplete spinal cord injury exhibits high inter-individual variability, which is difficult to characterize using conventional clinical metrics. The integration of clinical, biomechanical, and neurophysiological variables enables an objective and multidimensional characterization of motor recovery, capturing both functional performance and the underlying mechanisms of motor control. However, this global view generates high-dimensional feature spaces that require specific computational tools for exploration. This study proposes an unsupervised dimensionality reduction using Uniform Manifold Approximation and Projection (UMAP), applied to longitudinal multimodal data from four patients with similar levels of functional independence. Results show subject-specific groupings in the two-dimensional projection with no common longitudinal pattern, reflecting the heterogeneity of the recovery process. This work represents a first exploratory step towards the development of machine learning models for individualized classification and prediction of gait recovery.
References
Becht, E. et al. (2019) «Dimensionality reduction for visualizing single-cell data using UMAP», Nature Biotechnology, 37(1), pp. 38-44. Disponible en: https://doi.org/10.1038/nbt.4314.
Cai, T.T. y Ma, R. (2021) «Theoretical Foundations of t-SNE for Visualizing High-Dimensional Clustered Data». arXiv. Disponible en: https://doi.org/10.48550/ARXIV.2105.07536.
Cherubini, A. et al. (2024) «Multi-Level Characterization of the Recovery Process of a Stroke Survivor After 2 Months of Robotic Therapy with the Walkbot Robot», 2024 10th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob). 2024 10th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob), Heidelberg, Germany: IEEE, pp. 913-918. Disponible en: https://doi.org/10.1109/BioRob60516.2024.10719925.
Cunningham, J.P. y Ghahramani, Z. (sin fecha) «Linear Dimensionality Reduction: Survey, Insights, and Generalizations».
Ditunno, J.F. et al. (2007) «Validity of the walking scale for spinal cord injury and other domains of function in a multicenter clinical trial», Neurorehabilitation and Neural Repair, 21(6), pp. 539-550. Disponible en: https://doi.org/10.1177/1545968307301880.
Ditunno, J.F. et al. (2013) «The Walking Index for Spinal Cord Injury (WISCI/WISCI II): nature, metric properties, use and misuse», Spinal Cord, 51(5), pp. 346-355. Disponible en: https://doi.org/10.1038/sc.2013.9.
Groppa, S. et al. (2012) «A practical guide to diagnostic transcranial magnetic stimulation: Report of an IFCN committee», Clinical Neurophysiology, 123(5), pp. 858-882. Disponible en: https://doi.org/10.1016/j.clinph.2012.01.010.
Hermens, H.J. et al. (2000) «Development of recommendations for SEMG sensors and sensor placement procedures», Journal of Electromyography and Kinesiology: Official Journal of the International Society of Electrophysiological Kinesiology, 10(5), pp. 361-374. Disponible en: https://doi.org/10.1016/s1050-6411(00)00027-4.
Jamieson, A. et al. (2023) «Unsupervised Cluster Analysis of Walking Activity Data for Healthy Individuals and Individuals with Lower Limb Amputation», Sensors, 23(19), p. 8164. Disponible en: https://doi.org/10.3390/s23198164.
Kesar, T. (2023) «The Effects of Stroke and Stroke Gait Rehabilitation on Behavioral and Neurophysiological Outcomes:: Challenges and Opportunities for Future Research», Delaware Journal of Public Health, 9(3), pp. 76-81. Disponible en: https://doi.org/10.32481/djph.2023.08.013.
Lee, K.E. y Jeoung, B. (2025) «An analysis of the effectiveness of rehabilitation protocols for patients with spinal cord injury: A systematic review», Journal of Public Health, 33(7), pp. 1461-1470. Disponible en: https://doi.org/10.1007/s10389-023-02115-9.
López Gómez, S. et al. (2025) Longitudinal characterization of the electromyographic evolution of gait in stroke survivors during robotic therapy. Escola d’Enginyeria de Barcelona Est. Disponible en: https://hdl.handle.net/2117/449487.
Lu, Y. et al. (2024) «Global incidence and characteristics of spinal cord injury since 2000-2021: a systematic review and meta-analysis», BMC medicine, 22(1), p. 285. Disponible en: https://doi.org/10.1186/s12916-024-03514-9.
McInnes, L., Healy, J. y Melville, J. (2020) «UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction». arXiv. Disponible en: https://doi.org/10.48550/arXiv.1802.03426.
van Middendorp, J.J. et al. (2011) «A clinical prediction rule for ambulation outcomes after traumatic spinal cord injury: a longitudinal cohort study», Lancet, 377(9770), pp. 1004-1010. Disponible en: https://doi.org/10.1016/S0140-6736(10)62276-3.
Panier, M. et al. (2026) «Classification of Walking Function After Spinal Cord Injury Using Longitudinal Multimodal Clinical, Strength, and Neurophysiological Data». 2026 IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics (BioRob).
Spinal cord injury (sin fecha). Disponible en: https://www.who.int/news-room/fact-sheets/detail/spinal-cord-injury (Accedido: 28 de abril de 2026).
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Mathilde Panier, Clara B. Sanz-Morère, Jorge A. Gómez-García, Sara González-Expósito, Juan C. Moreno
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
