MRI RADIOMICS REFLECT MICROSTRUCTURAL PATTERNS ASSOCIATED WITH ACETABULAR COVERAGE IN YOUNG ADULTS

Abnormal hip morphology, bone microstructure, and cartilage composition have independently been linked to OA development at an early age. However, it remains unclear whether hip morphology is associated with underlying bone and cartilage microstructural properties. Understanding this relationship could reveal early mechanistic pathways in OA and support imaging-based risk assessment.

The aim of this study was to determine whether MRI derived radiomic features of femoral bone, acetabular bone, and cartilage were associated with different measures of acetabular coverage and to identify which radiomic feature contributed most to this association.

We analyzed data from 2448 participants from the Generation R Cohort (mean ± standard deviation age and BMI: 18.4±0.6 years and 22.7±3.8 kg/m²) comprising 1196 males and 1252 females. For the analysis, a T1-weighted Dixon (Lava Flex) sequence was acquired on a 3T system, with an in-plane resolution of 0.89 × 0.89 mm² and a slice thickness of 1.2 mm. Femoral and acetabular bone, and cartilage were automatically segmented using a previously trained nn-UNet. 481 radiomic features were extracted per structure, from the in-phase outputs, using the open-source Workflow for Optimal Radiomics Classification (WORC) framework. Six acetabular coverage metrics representing global and regional morphology were computed from the segmentations. Associations between radiomic features and coverage measures were assessed using a linear (ElasticNet) and nonlinear (TabPFN) regression approach, with 10-fold cross validation. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) and performance was quantified using cross validated R² in Python.

Across all analyzed structures, posterolateral and anteromedial coverage measures showed the strongest associations with radiomic features (R² = 0.39-0.76) for both the ElasticNet and TabPFN modelling approaches. Texture based features, particularly those derived from the gray level cooccurrence matrix and Gabor filters, consistently demonstrated the highest predictive value.

Radiomic features derived from T1-weighted MRI partially explain variation in acetabular coverage, suggesting that hip morphology is already linked to distinct microstructural tissue properties in young adults. The findings further highlight that radiomics captures information beyond morphology alone, highlighting its potential as a complementary, non-invasive tool for studying OA and exploring additional tissue-level biomarkers of disease risk and progression.