EUVIMEDEUVIMED entdecken
Crossref

Explainable radiomics and clinical models for prediction of total knee arthroplasty in knee osteoarthritis: data from Osteoarthritis Initiative

Atefe Pooyan, Ehsan Alipour, Nasim Eshraghi, Peyman Mirghaderi, Farhad Pishgar, Matthew J Nyflot, Chankue Park, Hyun Su Kim, Peter C Thurlow, Thomas M Link, Majid Chalian

Skeletal Radiology · 11.11.2026

DOI: 10.1007/s00256-026-05419-5 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Skeletal Radiology, online erschienen am 11.11.2026. Verfasst von Atefe Pooyan und 10 weiteren Autor:innen. Beteiligte Einrichtung: Department of Radiology, Mercy Catholic Medical Center, Darby, PA, USA. Die Arbeit nennt 29 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

OBJECTIVE: To develop and evaluate MRI radiomics models, with and without clinical variables, for predicting total knee arthroplasty (TKA) within 9 years in patients with knee osteoarthritis. MATERIALS AND METHODS: This retrospective exploratory study included 507 participants from the Osteoarthritis Initiative-Zuse Institute Berlin dataset with manually segmented knee MRIs; 66 underwent TKA. Four hundred radiomics features from bone and cartilage were combined with 13 clinical variables. Logistic regression, least absolute shrinkage and selection operator (LASSO) regression, and Extreme Gradient Boosting (XGBoost) were evaluated using radiomics-only, clinical-only, and combined feature sets in an 80/20 development-holdout analysis and stratified nested cross-validation. Preprocessing, feature selection, and tuning were restricted to training folds. RESULTS: On the held-out test set, radiomics-only LASSO had the highest observed AUC-ROC of 0.76 (95% CI, 0.61-0.88), followed closely by combined LASSO at 0.75 (0.61-0.87). Combined logistic regression and XGBoost achieved AUC-ROC values of 0.73 (0.60-0.86) and 0.72 (0.57-0.84), respectively. Differences among the leading models were modest. In nested cross-validation, combined LASSO showed the highest discrimination (AUC-ROC, 0.786 [0.722-0.841]; AUC-PR, 0.343 [0.239-0.451]; F1 score, 0.405 [0.321-0.481]), with modest improvement over radiomics-only LASSO (AUC-ROC, 0.781 [0.717-0.834]). Consistently retained predictors included femoral bone sphericity, femoral cartilage first-order skewness, and WOMAC disability. CONCLUSION: MRI radiomics features, alone or combined with clinical variables, showed potential for predicting TKA risk. Performance differences among leading models were modest, and these exploratory findings require external validation before clinical application.

Schlagwörter

Stichwörter der Autor:innen

Knee osteoarthritisMagnetic resonance imagingRadiomicsTotal knee arthroplasty prediction

Autor:innen und Institutionen

  • Department of Radiology, Mercy Catholic Medical Center, Darby, PA, USA.
  • Division of Musculoskeletal Imaging and Intervention, Department of Radiology, University of Washington, UW Radiology-Roosevelt Clinic, 4245 Roosevelt Way NE, Box 354755, Seattle, WA, 98105, USA.
  • Department of Radiology and Biomedical Imaging, University of California-San Francisco, San Francisco, CA, USA.
  • Division of Musculoskeletal Imaging and Intervention, Department of Radiology, University of Washington, UW Radiology-Roosevelt Clinic, 4245 Roosevelt Way NE, Box 354755, Seattle, WA, 98105, USA. mchalian@uw.edu.

Zitieren

APA

Pooyan, A., Alipour, E., Eshraghi, N., Mirghaderi, P., Pishgar, F., Nyflot, M. J., Park, C., Kim, H. S., Thurlow, P. C., Link, T. M., & Chalian, M. (2026). Explainable radiomics and clinical models for prediction of total knee arthroplasty in knee osteoarthritis: data from Osteoarthritis Initiative. Skeletal Radiology. https://doi.org/10.1007/s00256-026-05419-5

Vancouver

Pooyan A, Alipour E, Eshraghi N, Mirghaderi P, Pishgar F, Nyflot MJ, et al. Explainable radiomics and clinical models for prediction of total knee arthroplasty in knee osteoarthritis: data from Osteoarthritis Initiative. Skeletal Radiology. 2026. doi:10.1007/s00256-026-05419-5

BibTeX
@article{Pooyan2026,
  author = {Pooyan, Atefe and Alipour, Ehsan and Eshraghi, Nasim and Mirghaderi, Peyman and Pishgar, Farhad and Nyflot, Matthew J and Park, Chankue and Kim, Hyun Su and Thurlow, Peter C and Link, Thomas M and Chalian, Majid},
  title = {Explainable radiomics and clinical models for prediction of total knee arthroplasty in knee osteoarthritis: data from Osteoarthritis Initiative},
  journal = {Skeletal Radiology},
  year = {2026},
  doi = {10.1007/s00256-026-05419-5}
}
Kostenlos starten →