Gradient non-linearity correction relocates normalised group activation hotspot

Introduction Gradient coils in MR scanners are assumed to produce linear magnetic field gradients, however high-speed gradients show nonlinear characteristics as a trade-off[1]. This fact creates structural deformations in MR images during reconstruction, as the gradients are assumed to be linear. A correction for these structural deformations has been shown to be imperative in multi-site structural studies, this issue has to our knowledge not been taken into account when dealing with fMRI data. Accurate localisation of activation is important for instance in presurgical planning where the location of activation may determine where and where not to perform surgery. Here we show that using the known information of the how the gradient field are distorted, it is possible to reveal significant differences in activation localisation even in situations that are supposed to be unaffected by these distortions, in particular in group data results where EPI-times series are realigned, normalised and smoothed before group inference is made.