Deep learning classification models demonstrate high accuracy and clinical potential in radiograph interpretation in the arthroplasty clinical pathway: a systematic review and meta-analysis
Author(s)
Type
Journal Article
Abstract
Purpose
Imaging is a cornerstone of the osteoarthritis (OA)-arthroplasty clinical pathway: the continuum of care that patients with OA undergo, from initial diagnosis through to arthroplasty and postoperative follow-up. With growing interest from the orthopaedic community, this meta-analysis broadly evaluates the performance of deep learning algorithms in interpreting radiographs and cross-sectional imaging in this pathway. The authors hypothesise that deep learning algorithms will have comparable performance to clinicians when interpreting radiographs, but not cross-sectional imaging, with negligible difference in diagnostic and prognostic tasks.
Methods
Ovid Medline, Ovid Embase, Scopus and Web of Science were searched for studies published between January 1, 2012, and April 1, 2024, evaluating deep learning algorithms for diagnostic and prognostic tasks along the pathway. Eligible studies included those that used established diagnostic or surgical candidacy assessments as ground truth. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 tool, and pooled sensitivity and specificity were determined. Hierarchical summary receiver operating characteristic curves assessed
diagnostic performance.
Results
The meta-analysis of artificial intelligence (AI) interpretation included 66 studies, with a pooled sensitivity of 0.88 (95% confidence interval [CI]: 0.81–0.92) and a pooled specificity of 0.91 (95% CI: 0.87–0.94). Sensitivity and specificity values were higher for AI interpretation of radiographs (55 studies) compared to cross-sectional imaging, with no significant difference in performance between diagnostic and prognostic tasks. For clinician interpretation, 11 studies showed a pooled sensitivity of 0.76 (95% CI: 0.64–0.85) and a pooled specificity of 0.79 (95% CI: 0.59–0.90).
Conclusions
This meta-analysis highlights the potential of deep learning algorithms to improve efficiency in OA classification and prognosis in the arthroplasty pathway based on low-to-moderate quality evidence. Although the results are not generalisable, the findings suggest deep learning models have the potential to be adopted in OA treatment pathways, warranting further exploration of its role in patient care.
Imaging is a cornerstone of the osteoarthritis (OA)-arthroplasty clinical pathway: the continuum of care that patients with OA undergo, from initial diagnosis through to arthroplasty and postoperative follow-up. With growing interest from the orthopaedic community, this meta-analysis broadly evaluates the performance of deep learning algorithms in interpreting radiographs and cross-sectional imaging in this pathway. The authors hypothesise that deep learning algorithms will have comparable performance to clinicians when interpreting radiographs, but not cross-sectional imaging, with negligible difference in diagnostic and prognostic tasks.
Methods
Ovid Medline, Ovid Embase, Scopus and Web of Science were searched for studies published between January 1, 2012, and April 1, 2024, evaluating deep learning algorithms for diagnostic and prognostic tasks along the pathway. Eligible studies included those that used established diagnostic or surgical candidacy assessments as ground truth. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 tool, and pooled sensitivity and specificity were determined. Hierarchical summary receiver operating characteristic curves assessed
diagnostic performance.
Results
The meta-analysis of artificial intelligence (AI) interpretation included 66 studies, with a pooled sensitivity of 0.88 (95% confidence interval [CI]: 0.81–0.92) and a pooled specificity of 0.91 (95% CI: 0.87–0.94). Sensitivity and specificity values were higher for AI interpretation of radiographs (55 studies) compared to cross-sectional imaging, with no significant difference in performance between diagnostic and prognostic tasks. For clinician interpretation, 11 studies showed a pooled sensitivity of 0.76 (95% CI: 0.64–0.85) and a pooled specificity of 0.79 (95% CI: 0.59–0.90).
Conclusions
This meta-analysis highlights the potential of deep learning algorithms to improve efficiency in OA classification and prognosis in the arthroplasty pathway based on low-to-moderate quality evidence. Although the results are not generalisable, the findings suggest deep learning models have the potential to be adopted in OA treatment pathways, warranting further exploration of its role in patient care.
Date Issued
2025-07-01
Date Acceptance
2025-04-30
Citation
Journal of Experimental Orthopaedics, 2025, 12 (3)
ISSN
2197-1153
Publisher
SpringerOpen
Journal / Book Title
Journal of Experimental Orthopaedics
Volume
12
Issue
3
Copyright Statement
© 2025 The Author(s). Journal of Experimental Orthopaedics published by John Wiley & Sons Ltd on behalf of European Society of Sports Traumatology, Knee Surgery and Arthroscopy. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
License URL
Publication Status
Published
Article Number
ARTN e70342
Date Publish Online
2025-07-13