Premium
IC‐P‐175: Hybrid Diffusion Imaging (HYDI) of White Matter Changes in Older Adults With Subjective Cognitive Decline (SCD): Assessment of Orientation Dispersion and Axonal Density
Author(s) -
Mustafi Sourajit Mitra,
Gandhi Pratik K.,
Risacher Shan L.,
West John D.,
Tallman Eileen F.,
O'Neill Darren P.,
Farlow Martin R.,
Unverzagt Frederick W.,
Apostolova Liana G.,
Saykin Andy J.,
Wu Yu-Chien
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.2016.06.206
Subject(s) - diffusion mri , white matter , cognitive decline , psychology , diffusion imaging , nuclear medicine , audiology , medicine , pathology , magnetic resonance imaging , radiology , dementia , disease
eventually spreading to most of the cortex. Since neurodegeneration in AD closely follows NFT pathology, granular MRI-based measures of change in MTL subregions can be more sensitive biomarkers for treatment evaluation than conventional markers such as hippocampal volume, particularly in preclinical disease, when effects are subtle and largely contained to the MTL. Several large studies, including ADNI, collect T2-weighted MRI scans of the MTL that offer higher resolution and significantly better contrast for visualizing hippocampal and MTL subregion boundaries than conventional 1mm isotropic T1w MRI. We previously developed an automatic multi-atlas segmentation technique “ASHS” that extracts MTL subregion volume and thickness measures in 3T and 7T MRI scans, and showed that it is can reliably reproduce manual segmentation. Until now, however, ASHS required access to a high-performance computing cluster. Through extensive optimization, we have accelerated ASHS by more than one order of magnitude, making it possible to use on commodity computers without compromising reliability. Methods: ASHS uses deformable registration (Avants et al., 2008) to warp 20-30 expert-labeled example scans called atlases to a new subject’s T2w MRI scan, and combines the deformed segmentations using intelligent label fusion. Registration accounts for >95% of computational requirements of ASHS. We optimized the registration in ASHS using separable one-dimensional algorithms for computing the normalized cross-correlation metric of image similarity, as well as approximate deformation field regularization. We evaluate ASHS using ten-fold cross-validation on 29 3T expert-labeled scans. Results: Optimized ASHS runs 16 times faster than the published ASHS algorithm on the same hardware. Processing for one subject takes 24 minutes on an 8-core MacBook laptop. Segmentation accuracy is not statistically different from previously published results. Conclusions:Accelerated ASHS will enable a broader range of researchers to derive quantitative measures of MTL subregions in T2w-MRI scans, which are increasingly commonplace. Unlike the automatic hippocampal subfield measures from T1w-MRI (Iglesias et al., 2015), T2w-MRI measures derived by ASHS have been explicitly validated against manual segmentation and have high reliability.