Magnetic resonance imaging (MRI)-guided radiotherapy planning (MRIgRT) is feasible and in some instances may be superior to computed tomography (CT)-guided planning owing to its improved soft tissue contrast. However, because MR images do not communicate tissue electron density information necessary for dose deposition modeling they must either be co-registered to CT or algorithmically converted into a synthetic CT. No robust quality assessment of commercially available MR-CT registration algorithms is yet available. Head and neck non-contrast CT and T2 MRI scans acquired with standard treatment immobilization techniques were mined from 15 patient records. Per scan, 35 anatomic regions of interest (ROIs) were manually segmented. MRIs and ROI segmentations were propagated to CT by rigid registration (RIR) and by three commercial deformable registrations (DIR). Dice similarity coefficient (DSC), Hausdorff mean (HD mean) and max (HD max) metrics assessed the concordance between post-registration MRI and CT segmentations. Per-ROI and per-metric each DIR algorithm was compared to DIR using nonparametric Steel tests with RIR as control (n = 105 tests). Per-metric, DIR algorithms were compared to RIR for all ROIs in aggregate (n = 3 tests). The influence of tissue type on registration accuracy was assessed using nonparametric Wilcoxon pairwise tests between ROIs grouped by tissue type, per metric and per registration method (n = 12 tests). Bonferroni corrections were applied for multiple comparisons. No DIR algorithm improved the registration quality over RIR for any ROI (p values > 0.05) nor for all ROIs in aggregate (p values > 0.05). Nevertheless, registration quality varied by tissue type for all methods. Muscle and gland ROIs registrations were significantly more conformal than vessel and bone (median DSC, respectively: 0.66 and 0.75 vs. 0.47 and 0.61; p values < 0.0001) and had better HD distance metrics (median HD max (in mm): 10.23 and 10.42 vs. 15.77 and 12.18, p values < 0.005; median HD mean (in mm): 0.94 and 0.60 vs. 2.05 and 1.29, p values < 0.0001). The table reports median results from one DIR algorithm, RIR, and all methods in aggregate. For MR and CT images acquired by radiotherapy immobilization protocols, our results question the utility of commercially available DIR for MR-CT registration rather than RIR alone. The poor overall performance also questions the utility of translating tissue electron density information to MRI by MR-CT registration at all, rather than resolving this need by synthetic CT generation.Abstract 3615; Table 1BoneCordGlandMuscleVesselDSCDIR 10.610.670.750.680.50RIR0.580.630.740.650.50Agg.0.610.650.750.660.47HD max (mm)DIR 112.526.3710.4210.0415.9RIR12.248.3910.7410.0815.53Agg.12.187.5610.4210.2315.77HD mean (mm)DIR 11.270.710.590.882.06RIR1.390.960.630.941.93Agg.1.290.840.600.942.05 Open table in a new tab