Improving current normalization approaches to detect longitudinal changes in gray and white matter using DTI

JACOBACCI, F; BORÉ, A; JOVICICH, J; AMARO, E; HIDALGO-MARQUES, M; DELLA-MAGGIORE, V; ARMONY, J; LERNER, G; TAVERNA CHAIM, K; DOYON, J

Suntec city

Congreso; 24TH ANNUAL MEETING OF THE ORGANIZATION FOR HUMAN BRAIN MAPPING; 2018

INTRODUCTION● Scalar DTI measures such as MD and FA are increasingly being used to evaluate longitudinal changes in brain microstructure induced by learning 1,2, development 3,4 or neurodegenerative disease 5,6.● These studies require pre-processing scalar DTI volumes to align them to a template in standard stereotaxic space (a process called normalization).● There is no unanimity regarding the optimal registration approach to draw valid conclusions from voxelwise analysis conducted on multisession DWI data.OBJECTIVE In this work, we used a voxel-based approach (VBA) to compare different normalization pipelines based on ANTs.Our goal was to look for a volumetric normalization pipeline that optimized detection of longitudinal changes in Diffusion Tensor Images (DTI).We aimed at minimizing across-session test retest reproducibility error in the following traits: 1) Normalization tool (FSL vs ANTs) 2) Normalization target (FMRI B58 FA template vs MNI152 T1 template)3) Moving image to bring to standard space (MD, FA or B0)4) Normalization strategy (direct normalization vs intermediate template ) COLCUSIONS 1) Using ANTs improves registration reproducibility for longitudinal studies. 2) Using the MNI152 T1 template improves reproducibility of MD and FA in WM. It is the best option if one aims at evaluating changes in FA,but is of particular importance if one is interested in assessing changes in MD or wishes to study cortical gray matter.3) Using FA as moving image yields the best reproducibility of FA in GM. Combining FA and b0 gives the lowest reproducibility error of MD in GM but it does not significantly improve reproducibility over using FA alone. Using MD asmoving image deteriorates reproducibility.4) Normalizing via an individual FA Template yields the best reproducibility.