P2–196: Predicting prodromal Alzheimer's disease in people with mild cognitive impairment using multicenter diffusion‐tensor imaging data and machine learning algorithms

Alzheimer's disease (AD) patients show significant changes in white matter (WM) structural integrity. Diffusion tensor imaging (DTI) is a neuroimaging technique that allows in vivo assessment of WM fiber tract integrity and could improve the detection of AD at the predementia stage. We used machine learning (ML) methods to automatically detect AD specific structural WM changes in patients with amnestic mild cognitive impairment (aMCI). Within the framework of the European DTI study on Dementia (EDSD) we collected multicenter DTI and structural MRI data from 132 subjects with aMCI, 137 patients with clinically probable AD and 143 healthy controls (HC). aMCI subjects with available CSF information were stratified into amyloid-β42 negative and positive subjects using previously established thresholds. A support vector machine (SVM) classifier was used to automatically determine group membership based on DTI-derived fractional anisotropy (FA) and mean diffusivity (MD) maps, as well as volumetric gray matter (GM) and white matter (WM) maps derived from the structural MRI. Discriminative voxels were preselected using the information gain criterion and performance was validated using a tenfold cross validation. We achieved an accuracy of 93% for FA and 94% for MD for the discrimination of amyloid-β42 positive aMCI subjects and HCs. For the tissue density maps we obtained 90% for WM and 90% for GM. In contrast, first analyses showed modest group separation between amyloid-β42 positive and negative aMCI subjects. Further tuning of the models will likely increase the diagnostic accuracy. Multicenter acquisition of DTI data in combination with multivariate ML approaches shows promising results which are comparable to diagnostic accuracies reported from earlier monocenter DTI studies.