Affiliations: [a] Department of Psychiatry, University of Rostock, Germany | [b] Institute of Psychology, University of Osnabrueck, Germany | [c] Department of Radiology, University of Munich, Germany | [d] Department of General Psychiatry, University of Heidelberg, Germany | [e] Discipline of Psychiatry, School of Medicine & Trinity College Institute of Neuroscience (TICN), Laboratory of Neuroimaging & Biomarker Research, The Adelaide and Meath Hospital incorporating the National Children's Hospital (AMNCH), Trinity College, University of Dublin, Ireland | [f] Alzheimer Memorial Center, Department of Psychiatry, University of Munich, Germany | [g] Deutsches Zentrum für neurodegenerative Erkrankungen (DZNE), Germany
Correspondence:
[*]
Correspondence to: Uwe Friese, University of Osnabrueck, Institute of Psychology, Seminarstrasse 20, D-49076 Osnabrueck, Germany. Tel.: +49 (0) 541/969 6213; Fax: +49 (0) 541/969 4470; E-mail: uwe.friese@uni-osnabrueck.de.
Abstract: We report evidence that multivariate analyses of deformation-based morphometry and diffusion tensor imaging (DTI) data can be used to discriminate between healthy participants and patients with Alzheimer's disease (AD) with comparable diagnostic accuracy. In contrast to other studies on MRI-based biomarkers which usually only focus on a single modality, we derived deformation maps from high-dimensional normalization of T1-weighted images, as well as mean diffusivity maps and fractional anisotropy maps from DTI of the same group of 21 patients with AD and 20 healthy controls. Using an automated multivariate analysis of the entire brain volume, widespread decreased white matter integrity and atrophy effects were found in cortical and subcortical regions of AD patients. Mean diffusivity maps and deformation maps were equally effective in discriminating between AD patients and controls (AUC =0.88 vs. AUC=0.85) while fractional anisotropy maps performed slightly inferior. Combining the maps from different modalities in a logistic regression model resulted in a classification accuracy of AUC=0.86 after leave-one-out cross-validation. It remains to be shown if this automated multivariate analysis of DTI-measures can improve early diagnosis of AD in predementia stages.
