Automated 3D Analysis of Hip Morphology in a Young Adult Population.

Abstract

Purpose: Although many factors contribute to the development of hip OA, structural findings, such as acetabular dysplasia, are recognized as strong contributing factors to early degenerative changes. However, conventional two-dimensional imaging cannot fully capture the joint's three-dimensional anatomy. This study aimed to develop and validate an automated MRI-based method for 3D hip morphology analysis.

Methods: MRI data from 2,454 participants in the Generation R Cohort (1,199 males; 1,255 females; mean age 18.4 years; mean BMI 22.7 kg/m²) were analyzed. An nnU-Net deep learning model, trained on 40 manually segmented hips, automatically segmented femoral bone, acetabular bone, and femoral and acetabular cartilage. Model performance was assessed using Dice scores. From the segmentations, five categories of morphological features were computed: basic geometric metrics, cartilage volumes, angular measurements, coverage metrics, and joint space width. Sex differences were evaluated using independent t-tests, and BMI-related differences using one-way ANOVA.

Results: The model achieved Dice scores of 0.97 for femoral bone, 0.90 for acetabular bone, 0.76 for femoral cartilage, and 0.77 for acetabular cartilage. Reference ranges were established for all biomarkers. Significant differences were observed between males and females and across BMI groups.

Conclusion: Automated 3D hip morphology analysis using MRI is feasible and reliable. The generated reference values enhance understanding of hip development and anatomical variation in young adults and support future large-scale, reproducible assessments. All developed code is made openly available at https://gitlab.com/radiology/msk/genr/hip/hip-segmentation-and-biomarkers.