Echo Planar Imaging (EPI) sequences used for acquiring fMRI time series data have a high temporal resolution but are also highly sensitive to the magnetic field inhomogeneity resulting in geometric distortions¹. These distortions can lead to misalignment of the functional data to the anatomical reference resulting into an inaccurate spatial localization of brain activity, and also to errors in the statistical group analysis. Many different solutions have been proposed for the problem, including the Fourier and the Point Spread Function (PSF) methods² . In this work an alternative approach to account for EPI distortions in fMRI sequences is proposed. Our method takes advantage of a non-distorted T2-weighted (T2W) sequence as intermediate step between the acquired fMRI data and the anatomical image. This strategy allows to use non-linear registration functions, which should be used only to register images with the same tissue contrast.Data from 13 right-handed healty subjects (mean age ± SD=31.3±8.1 yo) were acquired on a Philips Achieva 3T with a 32 channel head coil. The acquisition protocol included a fMRI sequence (Fast Echo EPI, TR = 2 sec, TE = 30 ms, voxel resolution = 2.5 x 2.5 x 3.5 mm3, 178 dynamic scans) during a finger-tapping task, a 3D T1W sequence (T1TFE, TR = 8 sec, voxel size = 1 mm3) and a T2W sequence(TSE, TR = 3 sec, voxel size 1.5 x 1.5 x 1.5 mm3 ). Preprocess of data was performed using different common imaging tools (FSL³ for the realignment of functional data and Advanced Normalization Tools⁴ for all registrations). Single and group statistical analysis were performed using SPM12⁵. The key step for the proposed pipeline is the co-registration between functional data and the T2W volume of the same subject: we performed a two-steps registration using ANTs (Affine + Non-linear Symmetric Normalized registration) with the cross-correlation metric. The non-linear part of the registration was restricted to deform mainly in the phase-encoding direction. We compared two pipeline: a traditional one (fMRI data realignment, co-registration to the T1W, normalization to the MNI space) and the proposed one (fMRI data realignment, co-registration to the T2W, rigid registration of the T2W to the T1W, normalization to the MNI space). Single-subject GLM analyses were performed in the native fMRI space with 3 levels of smoothing (FWHM = 2,5,8 mm). Contrast maps where then projected on the MNI space with only one interpolation. Group analysis of the normalized T-maps was performed using a 1-group t-test to evaluate the average effect of the task. The percentage of the active voxels that fell in the gray matter over the overall active voxels was computed to evaluate the performances of the two pipelines. Voxels belonging to the gray matter were chosen on the basis of the segmentation of the MNI template using FSL.Fig. 1 shows the percentage of activate voxels in GM for different t-values (from T = 1 to T= 10, with the last one almost corresponding to the t-value for the Bonferroni correction). A paired t-test was computed between t-values belonging to the pipelines. For all 3 smoothing levels the proposed pipeline shows significant (p<0.0001) higher percentage of active voxels in gray matter. This is due to the fact that a better alignment of the functional/anatomical data (see Fig. 2) influences the statistical result in the group analysis. We then compared the activation maps obtained with the two pipelines: in Fig 3 we can see that most of the significant (p<0.001 uncorrected) higher activations for the traditional pipeline are mainly located in the white matter (thus being false positive activations), while the proposed pipeline shows significant higher values in the cerebellar cortex, a region of expected activation for the finger-tapping task. Focusing on the cerebellum we can see that the standard pipeline shows a monolateral widespread activation (p<0.05 Bonferroni corrected), while the proposed pipeline highlight a cortical activation region in both hemispheres (Fig. 4).In this work, we proposed a pipeline to account for susceptibility distortion of EPI acquisition using a T2W volume. The acquisition of a non-distortedT2W volume takes only 2 minutes, but significantly improves the registration performances. The advantage of a reduced mismatch between different subjects in the standard space let us to use a lower level of smoothing, reducing the number of false positive activations in group studies, and potentially decreasing the number of false negative activations.