Meta-analysis: AI detects femoral head avascular necrosis with high accuracy but limited external validation

Pooled sensitivity 0.888 (95% CI 0.789-0.944) and specificity 0.937 (0.906-0.958) for AI detection of avascular necrosis of the femoral head across nine studies; MRI-based models outperformed radiography (p=0.027). External validation remains limited.

• Only five of the nine pooled estimates involved external test sets, limiting generalizability of the results.
• MRI-based AI models showed significantly higher sensitivity than radiography-based models (p=0.027); no differences by disease-stage focus or validation strategy.
• Risk of bias and between-study heterogeneity were present, and the authors stress the need for prospective, multicenter validation before clinical use.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.