Premium
IC‐P‐040: Using White Matter Seed Regions Produces Stronger and More Complex Structural Networks in Healthy Elderly Subjects and Subjects with Alzheimer’s Disease
Author(s) -
Zajac Lauren,
Koo Bang-Bon,
Killiany Ronald
Publication year - 2016
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.2016.06.050
Subject(s) - region of interest , diffusion mri , white matter , artificial intelligence , voxel , computer science , pattern recognition (psychology) , neuroimaging , magnetic resonance imaging , neuroscience , medicine , psychology , radiology
Background: Whole-brain structural networks derived from MRI data are built from fiber tracts, which are generated using measurements of the orientation of water motion embedded in diffusion tensor imaging (DTI) or high angular resolution diffusion imaging (HARDI) data. These tracts need starting points, or seeds, and ending points, or targets. We investigated how using gray matter (GM) or white matter (WM) regions of interest (ROIs) as seeds impacts structural networks derived from DTI and HARDI data. Methods:Data from healthy elderly controls and individuals with Alzheimer’s disease (AD) were selected from DTI datasets in the ADNI database and HARDI datasets collected at the Center for Biomedical Imaging. These data were used to construct two whole-brain networks for each subject. Probabilistic tracts were formed usingWM-ROIs or GM-ROIs as seeds and GM-ROIs as targets. GM-ROIs were defined using the Desikan-Killiany FreeSurfer atlas. WM-ROIs were defined as a strip of voxels one voxel below the gray/white border of each GM-ROI. Comparisons were performed using the Network Based Statistic (NBS) toolbox. Results: In the DTI data, networks derived using WM-ROIs had approximately 400 more probabilistic connections than those derived from GM-ROIs. GM-ROI networks shared 94% of their connections with WM-ROI networks. Within-group NBS revealed a stronger subnetwork in WM-ROI networks compared to GM-ROI networks (p<0.001) in both controls and AD. In the HARDI data, WM-ROI networks had approximately 530 more probabilistic connections than GM-ROI networks. Again, GM-ROI networks shared 94% of their connections with WM-ROI networks. Within-group NBS revealed a stronger subnetwork in WM-ROI networks compared to GM-ROI networks (p1⁄40.029) in both groups. Between-group NBS performed on networks generated fromDTI data revealed a stronger subnetwork in controls compared to ADwhen usingWM-ROI networks (p1⁄40.004) and GM-ROI networks (p1⁄40.0326). Conclusions:Generating robust networks is an important prerequisite for network analyses because it prevents artifactual findings that can arise when networks with few and/or weak connections are used. Our analyses using DTI data show that WM-ROI networks are more sensitive to differences between control and AD, and ongoing analyses are exploring whether the same is true in HARDI data.