Figure 5.

Processing times for non-linear spatial normalization of one T1 volume of size 192 × 192 × 144 voxels to a MNI template (1 and 2 mm³ resolution) for the different software packages, averaged over 198 T1 volumes. The error bars represent the standard deviation. Note that AFNI uses a combination of cubic, quintic, and sinc interpolation as default, while SPM, FSL, and BROCCOLI all use linear interpolation as default. A linear registration was first applied to achieve a good starting point for the non-linear registration. BROCCOLI running on a GPU can perform non-linear normalization to a 1 mm³ template in 4–6 s, and still provide a satisfactory result. BROCCOLI running on a CPU is also significantly faster than FSL and AFNI OpenMP, even if a single CPU core is used.