Investigating the generalisation of an atlas-based synthetic-CT algorithm to another centre and MR scanner for prostate MR-only radiotherapy

There is increasing interest in MR-only radiotherapy planning since it provides superb soft-tissue contrast without the registration uncertainties inherent in a CT-MR registration. However, MR images cannot readily provide the electron density information necessary for radiotherapy dose calculation. An algorithm which generates synthetic CTs for dose calculations from MR images of the prostate using an atlas of 3 T MR images has been previously reported by two of the authors. This paper aimed to evaluate this algorithm using MR data acquired at a different field strength and a different centre to the algorithm atlas. Twenty-one prostate patients received planning 1.5 T MR and CT scans with routine immobilisation devices on a flat-top couch set-up using external lasers. The MR receive coils were supported by a coil bridge. Synthetic CTs were generated from the planning MR images with (sCT(1V)) and without (sCT) a one voxel body contour expansion included in the algorithm. This was to test whether this expansion was required for 1.5 T images. Both synthetic CTs were rigidly registered to the planning CT (pCT). A 6 MV volumetric modulated arc therapy plan was created on the pCT and recalculated on the sCT and sCT1V. The synthetic CTs' dose distributions were compared to the dose distribution calculated on the pCT. The percentage dose difference at isocentre without the body contour expansion (sCT-pCT) was Delta D-sCT = (0.9 +/- 0.8)% and with (sCT(1V)-pCT) was Delta D-sCT1V = (-0.7 +/- 0.7)% (mean +/- one standard deviation). The sCT(1V) result was within one standard deviation of zero and agreed with the result reported previously using 3 T MR data. The sCT dose difference only agreed within two standard deviations. The mean +/- one standard deviation gamma pass rate was Gamma(sCT) = 96.1 +/- 2.9% for the sCT and Gamma sCT(1V) = 98.8 +/- 0.5% for the sCT(1V) (with 2% global dose difference and 2 mm distance to agreement gamma criteria). The one voxel body contour expansion improves the synthetic CT accuracy for MR images acquired at 1.5 T but requires the MR voxel size to be similar to the atlas MR voxel size. This study suggests that the atlas-based algorithm can be generalised to MR data acquired using a different field strength at a different centre.