|
Activity Number:
|
467
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #304456 |
|
Title:
|
Application of Randomized Singular Value Decomposition for Partial Least Squares Analysis: Multimodality Neuroimaging Data
|
|
Author(s):
|
Bedda L. Rosario*+ and Lisa A. Weissfeld and William E. Klunk and Chester A. Mathis and Julie C. Price
|
|
Companies:
|
University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
|
|
Address:
|
200 Lothrop Street , Pittsburgh, PA, 15213,
|
|
Keywords:
|
partial least squares ; randomized singular value decomposition ; positron emission tomography ; neuroimaging ; Alzheimer's disease ; multimodality imaging
|
|
Abstract:
|
Partial least squares (PLS) is an useful tool for the analysis of high-dimensional data. However, PLS requires the computation of the singular value decomposition (SVD) of the cross-correlation matrix, which is not feasible for extremely large matrices. We address this issue by applying randomized singular value decomposition (RSVD, Drineas et al. 2004) as an approximation of the SVD. We apply PLS-SVD and PLS-RSVD to determine the relationship between multimodality PET data (FDG metabolism and PIB amyloid measures) in terms of region-of-interest measurements and functional voxel-based image data acquired in 3 groups: control, Alzheimer's disease and mild cognitive impairment. In addition, we assess group differences in the PET measures. Simulation studies showed that the RSVD method provides a good approximation of SVD and therefore a good approximation of the PLS summary scores.
|
