Optimising Region of Interest Characteristics and Standardisation Methodologies for Lumbar Magnetic Resonance Imaging Vertebral Bone Quality Scoring: Reproducibility, Precision, and Validation for Thoracolumbar Fragility Fractures.

STUDY

Design

Retrospective multi-cohort methodological and validation study.

Objective

To evaluate the impact of region of interest (ROI) size, placement, and calibration on MRI-derived vertebral bone quality (VBQ), and to assess the performance of VBQ methodologies for fragility fracture discrimination. SUMMARY OF

Background

DATA VBQ has emerged as a promising opportunistic biomarker of bone quality on routine MRI, with potential applications in preoperative risk stratification. However, lack of methodological standardisation, particularly regarding ROI placement and signal normalisation, limits reproducibility and contributes to variable diagnostic performance.

Methods

Three independent cohorts were analysed: (1) an optimisation cohort (n=50) evaluating ROI size, placement, and calibration, (2) a scan-rescan cohort (n=30) assessing reproducibility, and (3) a validation cohort of 100 fragility fracture patients and 100 matched controls. VBQ metrics, including anterior (AVBQ), posterior (PVBQ), and calibrated variants (cVBQ, cPVBQ), were compared. Reproducibility was assessed using intraclass correlation coefficients and precision metrics. Diagnostic performance was evaluated using ROC analysis, with comparisons performed using DeLong's test. Multivariable logistic regression models were constructed for each VBQ metric.

Results

ROI size had no significant effect on SI (P=0.14), whereas ROI placement significantly influenced measurements (P<0.001), with posterior regions demonstrating higher SI. Calibration improved reproducibility, with cVBQ showing superior precision (RMS SD 0.157 vs. 0.346; %CV 7.1% vs. 14.1%). In validation, discriminatory performance improved progressively from VBQ (AUC 0.718) to PVBQ (0.734), cVBQ (0.800), and cPVBQ (0.829). cPVBQ outperformed conventional VBQ (P=0.015). cPVBQ demonstrated the strongest independent association with fragility fractures (OR 4.65 per SD).

Conclusion

VBQ measurement is highly sensitive to methodological variation. Approaches incorporating posterior sampling and signal calibration may improve reproducibility and diagnostic performance but require validation in unselected clinical populations.