Routine pelvic MRI-derived marrow markers for predicting lumbar QCT-defined bone status using chemical shift-encoded fat quantification and radiomics
文献摘要
Objectives To evaluate whether pelvic MRI-derived IDEAL-IQ parameters and T1-weighted radiomics can predict lumbar QCT-defined bone status in cancer survivors undergoing routine pelvic oncologic MRI. Materials & methods This secondary analysis within a prospective surveillance cohort included 123 patients with pelvic malignancies who underwent MRI (with IDEAL-IQ) and lumbar QCT within three months. QCT-derived volumetric BMD at T12-L2 served as the reference standard to classify subjects as normal, osteopenia, or osteoporotic. Pelvic MRI-derived PDFF and R2* were measured at the femoral head, femoral neck, and ilium; a radiomics pipeline was built on T1-weighted images. Inter-reader reliability used absolute-agreement intraclass correlation coefficients (ICCs). Discrimination was evaluated with one-vs-rest ROC analysis, and clinical utility was assessed by decision-curve analysis using out-of-fold probabilities from five-fold internal validation for two decisions: osteoporosis versus non-osteoporosis and low bone mass versus normal bone status. Results PDFF increased and R2* decreased with worsening BMD at all sites (P < 0.05), with good inter-reader agreement (single-measurement ICCs, 0.756-0.988; mean-of-three-reader ICCs, 0.903-0.996). In internal validation, T1-weighted radiomics and IDEAL-IQ demonstrated comparable discrimination for three-class bone status (AUC 0.73-0.78). Class-wise analysis showed near-chance discrimination for osteopenia, indicating that the overall macro-AUC was mainly driven by normal-versus-osteoporosis separation. Models incorporating MRI-derived markers showed higher validation macro-AUCs than the simplified age-and-sex clinical baseline. Decision-curve analysis showed positive net benefit for both imaging-based approaches, but between-modality differences were small and should be interpreted cautiously because confidence intervals overlapped. Conclusion Pelvic MRI-derived IDEAL-IQ and radiomics markers showed modest and comparable performance for predicting lumbar QCT-defined bone status. Their discrimination was mainly driven by separation between normal bone status and osteoporosis, whereas osteopenia remained poorly distinguished. These findings support exploratory opportunistic risk flagging rather than complete three-class risk stratification.