Diagnostic performance of deep-learning algorithms in radiographic grading of knee osteoarthritis: a systematic review and meta-analysis.
Xiaolu Ren, Xinyu Jin, ZhiJun Wang et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated the diagnostic performance of deep learning (DL) algorithms in grading knee osteoarthritis (KOA) using radiographs. Analyzing 32 studies, the pooled sensitivity for Kellgren-Lawrence (K-L) grades 0-4 was 0.90, 0.66, 0.80, 0.87, and 0.88, respectively. The study found that DL models performed poorly for early-stage KOA detection, particularly for K-L grade 1.
Key findings
- Pooled sensitivity for K-L grade 1 was 0.66, indicating limited reliability for early-stage KOA detection.
- Pooled precision for K-L grade 1 was 0.71, highlighting challenges in accurately grading early KOA.
- High heterogeneity was observed, especially in sensitivity for K-L grades 1 and 2.
- Transfer learning and data splitting were identified as potential sources of heterogeneity.
Why it matters
Understanding the diagnostic performance of DL algorithms in KOA grading is crucial for clinicians, as accurate early detection can influence treatment decisions and patient outcomes. However, current limitations in sensitivity for early-stage KOA necessitate caution in clinical application.
What to keep in mind
The study noted substantial heterogeneity across outcomes and grades, particularly for sensitivity in K-L grades 1 and 2.