Integrating Unsupervised and Supervised Learning for Robust Identification of Hip Osteoarthritis Morphological Features
Journal of Orthopaedic Research · 2026-11 · 44 (11)
DOI: 10.1002/jor.70315 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Orthopaedic Research, online erschienen am 18.11.2026. Verfasst von Signe Folkmane und 6 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
This case‐control observational study aimed to determine whether an integrated multivariate analytical framework combining complementary unsupervised and supervised learning methods can identify reproducible radiographic and clinical features associated with hip osteoarthritis (OA) status. A total of 216 participants (155 OA patients and 61 controls) underwent clinical assessment and standardized radiographic evaluation, with morphological measurements obtained from anteroposterior, Dunn and frog‐leg lateral (LL) views. A two‐stage analytical approach combined Principal Component Analysis and Hierarchical Cluster Analysis with Orthogonal Partial Least Squares Discriminant Analysis and Random Forest. Model robustness was assessed using cross‐validation, permutation testing and class‐balancing strategies. Unsupervised analyses revealed modest but significant morphological differences between OA patients and controls, with substantial overlap reflecting disease heterogeneity. Supervised models demonstrated moderate predictive performance. Clinical variables, particularly functional status and pain measures, were the strongest discriminators between groups. Among morphological features, the alpha angle consistently showed the highest variable importance across all projections and analytical approaches, alongside the Singh index, head‐neck offset ratio and pistol grip deformity. Inclusion of LL measurements did not improve model performance and in some cases reduced discriminatory power. Hip OA was associated with subtle but consistent morphological differences rather than distinct structural separation from controls. Across complementary multivariate analyses, proximal femoral morphology, particularly the alpha angle, showed the most consistent association with OA status. These findings suggest that integrated multivariate analytical approaches can help identify reproducible OA‐associated radiographic features and provide a framework for future studies investigating morphological phenotypes of hip OA.
Zitieren
APA
Folkmane, S., Dzerins, A., Zolovs, M., Moritis, E. T., Paegle, K., Studers, P., & Zolova, A. (2026). Integrating Unsupervised and Supervised Learning for Robust Identification of Hip Osteoarthritis Morphological Features. Journal of Orthopaedic Research, 44(11). https://doi.org/10.1002/jor.70315
Vancouver
Folkmane S, Dzerins A, Zolovs M, Moritis ET, Paegle K, Studers P, et al. Integrating Unsupervised and Supervised Learning for Robust Identification of Hip Osteoarthritis Morphological Features. Journal of Orthopaedic Research. 2026;44(11). doi:10.1002/jor.70315
BibTeX
@article{Folkmane2026,
author = {Folkmane, Signe and Dzerins, Andris and Zolovs, Maksims and Moritis, Eduards Toms and Paegle, Kristine and Studers, Peteris and Zolova, Alīna},
title = {Integrating Unsupervised and Supervised Learning for Robust Identification of Hip Osteoarthritis Morphological Features},
journal = {Journal of Orthopaedic Research},
year = {2026},
volume = {44},
number = {11},
doi = {10.1002/jor.70315}
}
EUVIMED