Voxel-Based Morphometry

All participants underwent a structural MRI scan. Imaging was performed on a 3T Philips Achieva TX scanner at IRM Québec-Mailloux in Québec City. High-resolution T1-weighted structural images were acquired with a volumetric magnetization prepared rapid gradient echo (MP-RAGE) sequence (repetition time = 8.2 ms, echo time = 3.7 ms, field of view = 250 mm, flip angle = 8°, 256 × 256 matrix, 180 slices/volume, slice thickness = 1mm, no gap).

We performed VBM analyses using the Statistical non-Parametric Mapping toolbox⁴ for SPM12 (Welcome Department of Imaging Neuroscience, London, UK⁵). The structural images were first segmented into GM, white matter and cerebrospinal fluid. We further created a specific template for this study using the diffeomorphic anatomical registration through an exponentiated Lie algebra algorithm (DARTEL; Ashburner, 2007). All grey matter images were warped to the custom template and then spatially normalised into Montreal Neurological Institute (MNI) space. To compensate for the effect of spatial normalisation, the spatially normalized grey matter was adjusted by multiplying its relative volume before warping. The modulated images were then smoothed with a Gaussian kernel of 8 mm.

For each of the two error types of interest (regularization errors for exception words and errors with complex OPM during regular word reading), we entered the number of errors of each participant as a covariate of interest in multiple regression statistical models with age, gender and group (svPPA, AD and healthy control participants) included as nuisance covariates. We followed the same procedure for the accuracy in pseudoword reading as a control condition, since we do not expect to find a significant correlation between reading performance of this word type and GM volume in the left ATL. We entered smoothed GM images of all participants as a single group in these statistical models. We set specific contrasts to identify the brain regions whose GM volume correlated with the number of regularizations and errors with complex OPM and accuracy in pseudoword reading. We tested the correlations with a [-1] t-contrast for number of errors and a [1] t-contrast for pseudoword reading accuracy, postulating that these variables of interest would correlate in a negative and positive way, respectively, with GM volume. Whole brain analyses were conducted using a statistical threshold of p < 0.001 uncorrected for multiple comparisons.