Premium
P3–100: Task‐free fMRI in clinical trials: Selection of covariates and numeric endpoints
Author(s) -
Schwarz Adam,
Pinsard Basile,
Perlbarg Vincent,
Lu Yuefeng,
Marais Lea,
Hill Derek,
Mangin JeanFrancois
Publication year - 2013
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2013.05.1170
Subject(s) - functional magnetic resonance imaging , covariate , default mode network , neuroimaging , alzheimer's disease neuroimaging initiative , biomarker , psychology , dementia , disease , artificial intelligence , computer science , medicine , neuroscience , machine learning , pathology , biology , biochemistry
Background: The purpose of the present study was to implement an automatic pipeline to obtainmeasures of fractional anisotropy (FA), using diffusion tensor imaging (DTI), in regions of interest corresponding to fiber bundles (ROI). To overcome partial volume effects, FA averages are weighted by fiber density. Thewithin and between-scanner reproducibility of our measurements are tested using a cross vendor harmonisation of the ADNI2 DTI protocol. Methods: Two young healthy volunteers participated in two scanning visits, in both a Siemens and in a Philips scanner, as part of a larger study designed to implement the DTI sequence from ADNI2 for multi-vendor clinical trials. Images were assessed for the presence of gross artifacts and preprocessed using BrainVISA/Connectomist-2.0. Mean FA measurements were performed on six ROIs (corpus callosum; cingulum; uncinate; superior longitudinal fasciculus; temporal white matter and internal capsule). In order to automatically define the ROIs, we propose a pipeline that includes whole brain tractography, obtained from a deterministic streamline method; clusteringbasedautomatic recognitionofdeepwhitematter bundlesusing amulti-subject atlas; computation of a tract density map for each bundle; denormalization of ROIs from ICBM DTI-81; and, finally, computation of the intersection between these ROIs and the associated density maps to focus the measurement on deep white matter. The final ROI allows, therefore, the computation of mean FA, weighted by the density of fibers in each voxel. Results: FA values were within the range expected for the regions for both subjects, and highly reproducible across sessions. There was a slight bias between scanners, with results from the Siemens scanner showing higher values compared to data from the Philips scanner. Conclusions: Our automatic pipeline provides a reliable estimation of potential endpoints for clinical trials. The use of fiber density maps to weight FA measurements overcomes the consequences of the partial volume effect that can occur at the periphery of the bundles. Since the density is higher in the core of the bundle, more weight will be given to the voxels without partial volume bias. Hence, FA measurements focus on the inner integrity of the bundle and are not contaminated by atrophy.