GabrielAll the best,Many thanks for your help,Is there any mathematical explanations or have I done something wrong ?I went back to the original paper (Ashburner and Ridgway 13 Fr Neuroscience) and got quite clearly the benefit from preprocessing through the longitudinal pipeline but it is still not very clear to me why deformation based morphometry provide contraction of the voxels but volumetric approach suggest increased volumes.When extracting the brain volumes from the same clusters and subtracted from FU to BL, I find increased brain volumes.In additional analysis, I have preprocessed both the Baseline and the FU scans using the similar steps (segmentation etc...) cross-sectionnally and warped the images to the midpoint template I have created during the longitudinal pipeline, using DARTEL. Smoothing was the same in the two procedure as well as resolution.Dear Spm experts,Just in case someone has an answer and forgot about the email, I am reposting my question:I am running the SPM12 longitudinal pairwise pipeline and followed the recipe suggested by John Ashburner. In the final steps, I normalise the GM x JD maps by preserving concentration (unmodulated data). I find 4 clusters correlated with my phenotype. Extracting the values from the GM x JD maps provide small but consistent shrinkage in my four clusters (consistent with the litterature given my age at BL= 14 years and FU=19 years).