“…A patient CT dataset that best represented the average anatomy of the cohort served as common template and mapping target. Non-rigid registration was performed with the 3DSlicer SlicerRT Plastimatch B-spline deformable registration module ( 17 , 19 , 20 ). After a first rigid registration step (subsampling 2 × 2 × 1, maximum of 100 iterations), a 3-stage B-Spline deformable registration (stage 1: subsampling 2,2,1, grid 25 mm, regularization 0.01, landmark penalty 0.005, maximum iterations 100, stage 2: subsampling 2,2,1, grid 10 mm, regularization 0.01, landmark penalty 0.005, maximum iterations 100, stage 3: subsampling 1,1,1, grid 2 mm, regularization 0.01, landmark penalty 0.005, maximum iterations 100) with Mean Squared Error as cost function empirically provided the best results and was used in all cases.…”