Validation of MR-CT Deformable Registration Accuracy and MRI Fat-Fraction Quantification Using a Novel 3D-Printed Anthropomorphic Phantom
Abstract
Purpose
To evaluate a MR-CT deformable registration workflow using a novel 3D-printed anthropomorphic phantom that enables controlled, known lumbar spine deformation, with assessment of geometric accuracy and quantitative integrity of MR fat-fraction (FF).
Methods
A phantom was created to allow controlled deformation of the lumbar spine with known directions and magnitudes. Eight MRI datasets (MR0–MR7), acquired under different deformation configurations, were deformably registered to a baseline CT (CT0) using a convert local alignment (CLA) and contour-based (CB) algorithm. Geometric registration accuracy was assessed on contoured structures (L1–L5 vertebrae, sacrum, hips, and total bone) using Dice similarity coefficient (DSC), mean distance to agreement (MDA), Hausdorff distance (HD), and volume differences. The algorithms were compared using paired, two-sided Wilcoxon signed-rank test. Quantitative integrity was evaluated by comparing mean MR %FF values in inserts with known %FF before and after deformation.
Results
Mean DSCs coefficients for vertebral and sacral contours were 0.90 (CLA) and 0.93 (CB) with the CB method significantly (p<0.05) improving DSC and with corresponding MDA values 1mm. HD demonstrated greater variability, particularly for larger structures, with vertebral HD values < 10 mm and hip HD values exceeding 15 mm (p<0.05). Hip structures exhibited lower DSCs than spinal structures (mean DSC = 0.72-0.75 (CLA) and = 0.88-0.92 (CB); p<0.05). CB deformation preserved total bone volume more consistently than CLA. No correlation was observed between DSC and deformation magnitude. For quantitative evaluation, the mean difference in %FF before and after deformation was -0.60%FF with narrow limits of agreement indicating minimal perturbation of FF.
Conclusion
Using a deformable, anthropomorphic phantom with known lumbar spine displacement, this study demonstrates that the evaluated MR-CT deformable registration workflow maintains stable geometric accuracy for spinal anatomy and preserves quantitative MR FF measurements across the investigated deformation magnitudes, supporting its robustness under controlled deformation conditions.