Performance of an Artificial Intelligence–Based Software for Automated Kellgren-Lawrence Grading of Knee Osteoarthritis: A Multicenter Cohort Study
The Journal of Arthroplasty · 2026
DOI: 10.1016/j.arth.2026.01.078 ↗Kurz gefasst
Multizentrische Studie in der Zeitschrift The Journal of Arthroplasty, 2026. Verfasst von Byung Sun Choi und 3 weiteren Autor:innen. Beteiligte Einrichtung: Department of Orthopedic Surgery, Seoul National University College of Medicine, Seoul, Republic of Korea. Hauptthemen laut MEDLINE: Osteoarthritis, Knee, Artificial Intelligence, Software, Knee Joint. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 35 Literaturangaben.
Abstract (Originalsprache)
BACKGROUND: The Kellgren-Lawrence (KL) grading system is the standard for assessing knee osteoarthritis (OA) severity. However, it is limited by major observer variability. Artificial intelligence (AI) may standardize grading, yet external validation is limited. This study evaluated the diagnostic efficacy of AI-based software on a large, independent, multicenter clinical dataset. METHODS: This multicenter, retrospective, pivotal study included 2,546 knee radiographs from 1,273 patients across two tertiary hospitals in Korea. A reference standard was established by an expert consensus panel, with KL grades 0 and 1 consolidated into a single KL ≤ 1 category. The AI software was trained exclusively on public United States datasets (Osteoarthritis Initiative and the Multicenter Osteoarthritis Study) and validated on this separate Korean dataset. The primary outcomes were grade-specific sensitivity and specificity for four categories (≤ 1, 2, 3, and 4). The secondary outcomes included accuracy, the area under the receiver operating characteristics curve, and binary diagnostic performance for radiographic OA (KL ≥ 2). RESULTS: The AI met all prespecified noninferiority endpoints. For KL ≤ 1, sensitivity was 90.5% (95% confidence interval (CI), 87.9 to 92.8) and specificity was 96.6% (95% CI, 95.0 to 97.9). For KL grade 4, sensitivity was 97.7% (95% CI, 97.5 to 99.1), and specificity was 98.4% (95% CI, 97.5 to 99.1). For KL grade 2, sensitivity was 77.2% and specificity was 95.3%. In the binary classification of radiographic OA, the AI achieved an area under the curve of 0.94 (95% CI, 0.92 to 0.96), sensitivity of 96.6%, specificity of 90.5%, and accuracy of 94.2%. CONCLUSIONS: In a large-scale, multicenter external validation using a dataset entirely independent of its training data, the AI-based software demonstrated high and robust diagnostic performance for KL grading. These findings support the software's potential for clinical integration to improve the consistency, objectivity, and efficiency of knee OA severity assessment. LEVEL OF EVIDENCE: Level III.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Orthopedic Surgery, Seoul National University College of Medicine, Seoul, Republic of Korea.
- Department of Radiology, Seoul National University College of Medicine, Seoul, Republic of Korea.
- Department of Orthopedic Surgery, Chung-Ang University Hospital, Chung-Ang University College of Medicine, Seoul, Republic of Korea.
Zitieren
APA
Choi, B. S., Hong, S. H., Lee, H. J., & Kim, S. H. (2026). Performance of an Artificial Intelligence–Based Software for Automated Kellgren-Lawrence Grading of Knee Osteoarthritis: A Multicenter Cohort Study. The Journal of Arthroplasty. https://doi.org/10.1016/j.arth.2026.01.078
Vancouver
Choi BS, Hong SH, Lee HJ, Kim SH. Performance of an Artificial Intelligence–Based Software for Automated Kellgren-Lawrence Grading of Knee Osteoarthritis: A Multicenter Cohort Study. The Journal of Arthroplasty. 2026. doi:10.1016/j.arth.2026.01.078
BibTeX
@article{Choi2026,
author = {Choi, Byung Sun and Hong, Sung Hwan and Lee, Han-Jun and Kim, Seong Hwan},
title = {Performance of an Artificial Intelligence–Based Software for Automated Kellgren-Lawrence Grading of Knee Osteoarthritis: A Multicenter Cohort Study},
journal = {The Journal of Arthroplasty},
year = {2026},
doi = {10.1016/j.arth.2026.01.078}
}
EUVIMED