Published In

Authors

Abstract

Purpose:
To assess the diagnostic performances of artificial intelligence (AI)-based software to perform anatomic measurements on anteroposterior and lateral hip radiographs.

Methods:
We retrospectively collected consecutive 117 anteroposterior hip radiographs and 110 lateral hip radiographs from 3 imaging institutions. Two senior musculoskeletal radiologists independently annotated key points to calculate the femoral-neck-shaft (CC’D), the lateral-center-edge (LCE), and acetabular roof angles, and the pelvic obliquity on anteroposterior hip radiographs and the vertical-center-anterior (VCA) angle on lateral hip radiographs. The gold standard was defined as the mean of their two measurements.
Statistical analysis consisted of mean absolute difference (MAE), bias assessed with Bland-Altman analysis between the gold standard and the AI prediction and intraclass coefficient (ICC) between the two manual ratings.

Results:
90 anteroposterior hip radiographs were included and 27 radiographs were excluded. 83 lateral hip radiographs were included and 27 were excluded. MAE for the CC’D angle, LCE angle, acetabular roof angle, and pelvic obliquity on anteroposterior hip radiographs were respectively 2.6° (95% CI: [2.3 ; 2.9], bias=-1.27°, ICC=0.8),
2.5° (95% CI: [2.2 ; 2.8], bias=1.21°, ICC=0.7), 1.9° (95% CI: [1.6 ; 2.1], bias=1.13°, ICC=0.8), 0.5mm (95% CI: [0.4 ; 0.6], bias=0.13mm, ICC=0.99). MAE for the VCA angle on the lateral hip radiograph was 3.7° (95% CI:[2.5 ; 4.8], bias=1.61°, ICC=0.89). Bias and MAE between the gold standard and the AI prediction were low across all measurements. ICC was good across all measurements and excellent for pelvic obliquity.

Conclusion:
AI allows accurate and automatic anatomic measurements on anteroposterior and lateral hip radiographs.

Limitations:
The study is retrospective with a small number of radiographs and no comparison to an independent manual rating.