Systematic review and meta-analysis of AI for infant hip ultrasound accuracy

Photo by Rebeca Alvidrez / Unsplash

Pediatric radiology Published May 2, 2026 Medically reviewed May 2, 2026 Study authors: Azmi Aryan, McArthur Adam, McDonald Steel, Hareendranathan Abhilash, Jaremko Jacob PubMed ↗ DOI ↗ By Dr. Sofia Müller, MD · Lifespan & Whole-Person Care

Key Takeaway

Consider AI-assisted ultrasound for infant hip screening but note high bias and need for external validation.

This is a systematic review and diagnostic test-accuracy meta-analysis of AI applied to two-dimensional or three-dimensional hip ultrasound for developmental dysplasia of the hip in infants 12 months or younger. The analysis synthesized data from 9 studies comprising 6,351 hips.

The authors report a pooled sensitivity of 0.92 (95% CI 0.86-0.95) and a pooled specificity of 0.96 (95% CI 0.91-0.98) for AI-assisted ultrasound compared with expert Graf-based interpretation or follow-up consensus. Feasibility signals included operator training times and scan acquisition time reductions, though specific values were not reported.

Key limitations noted by the authors include frequently high or unclear risk of bias for patient selection and the index test, and limited economic reporting. The authors state that larger multicenter studies with external validation and robust economic evaluation are needed.

Practice relevance is restrained; the authors suggest AI-assisted ultrasound may help standardize hip imaging and facilitate safe use by nonexpert operators. However, they caution against inferring superiority beyond the reported diagnostic accuracy and against clinical implementation without external validation and economic evaluation.

Study Details

Study typeMeta analysis

EvidenceLevel 1

PublishedMay 2026

PMID41961108

View Original Abstract ↓

BACKGROUND: Ultrasound is the standard imaging test for infant developmental dysplasia of the hip (DDH) but is highly operator-dependent, leading to variable image quality and classification. Artificial intelligence (AI)-assisted ultrasound may standardize acquisition and interpretation and support DDH screening beyond specialist centers. OBJECTIVE: To evaluate the diagnostic accuracy and feasibility of AI-assisted ultrasound for infant DDH. MATERIALS AND METHODS: We performed a systematic review and diagnostic test-accuracy meta-analysis of studies enrolling infants (≤12 months) undergoing hip ultrasound, in which the index test was AI applied to two-dimensional (2D) or three-dimensional (3D) ultrasound and the reference standard was expert Graf-based interpretation or follow-up consensus. Risk of bias was assessed with QUADAS-2 (diagnostic accuracy bias tool). Sensitivity and specificity were pooled with a bivariate random-effects model. RESULTS: Twenty-nine studies were eligible; nine provided 2×2 data (6,351 hips) for pooling. Pooled sensitivity was 0.92 (95% CI 0.86-0.95) and specificity 0.96 (95% CI 0.91-0.98). Risk of bias was frequently high or unclear for patient selection and the index test. Feasibility signals included short operator training times (approx. 1-2 h) and scan acquisition time reductions (approx. 20-50%), while economic reporting was limited. CONCLUSION: AI-assisted ultrasound demonstrates high diagnostic accuracy for infant DDH and may help standardize hip imaging and facilitate safe use by nonexpert operators, but larger multicenter studies with external validation and robust economic evaluation are needed.