Abstract
Abstract
Background
Developmental dysplasia of the hip (DDH) requires accurate and standardized ultrasound assessment in early infancy. The Graf method is widely used but remains susceptible to measurement variability, particularly among clinicians with differing levels of experience. Artificial intelligence (AI)-assisted image analysis has the potential to improve measurement consistency and efficiency.
Objectives
To evaluate the accuracy, reliability, and efficiency of AI-assisted hip ultrasound measurements compared with manual assessment using the Graf method, with exploratory analysis of femoral head coverage (FHC).
Methods
This retrospective diagnostic agreement and reliability study included 215 infants (309 hips) younger than 6 months who underwent hip ultrasound screening. Manual measurements of the α angle and FHC were performed in a blinded manner by senior and junior pediatric orthopedic physicians and postgraduate trainees. AI-assisted measurements were obtained using the ZKyK software. Measurement agreement was assessed using intraclass correlation coefficients (ICCs), Spearman correlation analysis, paired
t
tests, and Bland–Altman analysis. Measurement time was recorded to compare efficiency.
Results
AI-assisted α-angle measurements showed good agreement with manual measurements, with no statistically significant difference between methods. Inter-observer reliability for α-angle and FHC measurements was good to excellent. Bland–Altman analysis demonstrated a small mean difference between manual and AI-assisted α-angle measurements, with most data points within the 95% limits of agreement. A moderate correlation was observed between the α angle and FHC for both manual and AI-assisted measurements. AI-assisted analysis required substantially less time than manual measurement.
Conclusions
AI-assisted hip ultrasound measurement achieves accuracy and reliability comparable to manual assessment while markedly improving measurement efficiency. AI tools such as the ZKyK software may serve as reliable adjuncts in DDH screening, particularly by reducing inter-observer variability and supporting clinicians with varying levels of experience.