Hengameh Mirzaalian1, Lipeng Ning1, Peter Savadjiev1, Ofer Pasternak1, Sylvain Bouix1, Oleg Michailovich2, Marek Kubicki1, Carl Fredrik Westin1, Martha E. Shenton1, and Yogesh Rathi1
1Harvard Medical School and Brigham and Women’s Hospital, Boston, USA., Boston, MA, United States, 2University of Waterloo, Toronto, ON, Canada
Diffusion MRI (dMRI) is increasing being used to study neuropsychiatric brain disorders. To increase sample size and statistical power of neuroscience studies, we need to aggregate data from multiple sites1. However this is a challenging problem due to the presence of inter-site variability in the signal originating from several sources, e.g. number of head coils and their sensitivity, non-linearity in the imaging gradient, and other scanner related parameters2. Prior works have addressed this issue either using meta analysis3, or by adding a statistical covariate4, which are not model free and may produce erroneous results.