EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Transfer Learning with Customized XceptionNet Architecture for Osteoarthritis Detection in Knee Radiographic Images

Shashikala H K, Suresh M B

Journal of Robotics and Control (JRC) · 07.11.2025 · 6 (6) · 2653-2665

DOI: 10.18196/jrc.v6i6.25987 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Robotics and Control (JRC), online erschienen am 07.11.2025. Verfasst von Shashikala H K und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Knee osteoarthritis stands as a formidable global health challenge, impacting millions worldwide and lacking a definitive cure. Timely detection plays a pivotal role in managing its progression, with X-ray imaging serving as a cornerstone diagnostic modality. However, manual interpretation of X-rays introduces variability and potential errors due to differences in radiologists' expertise. Recent developments in machine and deep learning have spurred automated systems for osteoarthritis prediction from knee radiographic images. Nevertheless, achieving higher predictive accuracy remains critical for early-stage detection. In this paper, we propose OsteoXceptionNet, a deep learning approach which integrates transfer learning with the XceptionNet architecture for automated grading of osteoarthritis severity level. An efficient deep learning model, OsteoXceptionNet was developed and validated to improve automated grading accuracy for knee osteoarthritis. Proposed model has been trained and validated using Osteoarthritis Initiative (OAI) data set, a publicly available and diverse database containing over 9786 annotated knee radiographs graded according to Kellgren-Lawrence (KL) grading. Presented model is benchmarked against ResNet50, DenseNet169, and EfficientNetB5 for performance comparison. Additionally, comparative evaluation was conducted with recent advanced models like OSTEOHRNet and VGG16, where OsteoXceptionNet demonstrated improved diagnostic precision and robustness, particularly in severe OA (KL-4) cases with limited sample representation. OsteoXceptionNet achieved 97% accuracy, 97.8% precision, 97.6% recall, 97.6% F1-measure and 95.94%, Cohen’s kappa, outperforming all baseline models. While the model demonstrates excellent performance on the OAI dataset, its effectiveness in heterogeneous, real-world clinical environments remains to be validated. Differences in imaging quality, population demographics, and clinical workflows could affect generalizability and will be the focus of future work. This study contributes to advancing AI-assisted diagnosis by introducing a high-performing, scalable solution for knee OA grading and supports incorporation of deep learning frameworks to orthopedic clinical setting.

Zitieren

APA

K, S. H., & B, S. M. (2025). Transfer Learning with Customized XceptionNet Architecture for Osteoarthritis Detection in Knee Radiographic Images. Journal of Robotics and Control (JRC), 6(6), 2653-2665. https://doi.org/10.18196/jrc.v6i6.25987

Vancouver

K SH, B SM. Transfer Learning with Customized XceptionNet Architecture for Osteoarthritis Detection in Knee Radiographic Images. Journal of Robotics and Control (JRC). 2025;6(6):2653-2665. doi:10.18196/jrc.v6i6.25987

BibTeX
@article{K2025,
  author = {K, Shashikala H and B, Suresh M},
  title = {Transfer Learning with Customized XceptionNet Architecture for Osteoarthritis Detection in Knee Radiographic Images},
  journal = {Journal of Robotics and Control (JRC)},
  year = {2025},
  volume = {6},
  number = {6},
  pages = {2653--2665},
  doi = {10.18196/jrc.v6i6.25987}
}
Kostenlos starten →