General linear models for group studies in diffusion tensor imaging

DT-MRI has recently been reported as a relevant modality for the diagnosis and prognosis of neurodegenerative diseases [1]. An open challenge in diffusion tensor imaging is to take into account all information available in the tensor elements in group studies that compare cohorts of patients with healthy subjects according to neuroimaging and clinical data. Current methods do not use efficiently all this information coupling with clinical data. In this paper, we present a framework based on the general linear model that aims at conducting multivariate statistical tests to carry out group comparison or correlation analysis with respect to clinical scores, using the whole tensor information as opposed to using scalar indices only. We present successful tests on both simulated lesions and real images of patients suffering from neuromyelitis optica.