|
Activity Number:
|
436
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Learning and Data Mining
|
| Abstract - #303704 |
|
Title:
|
Spatial Voxel Co-Occurrence Matrices and Correction Functions for Multiple Magnetic Resonance Imaging Scanners
|
|
Author(s):
|
Arpad G. Kelemen and Yulan Liang*+
|
|
Companies:
|
University of Maryland, Baltimore
|
|
Address:
|
655 West Lombard Street, Baltimore, MD, 21201,
|
|
Keywords:
|
Spatial voxel co-occurrence matrix ; Threshold methods ; Correction function ; Model averaging ; Magnetic Resonance Imaging ; GLM
|
|
Abstract:
|
Persistent differences among MRI images obtained with different MRI scanners is a major challenge in image studies. With millions of measurements of voxels arises another key statistical challenge in order to obtain correction functions. In this article, we propose a novel three-stage statistical approach to construct Spatial Voxel Co-occurrence Matrices for dimension reduction and to estimate correction functions with generalized linear models and model averaging technique in order to obtain a more general disease diagnosis and prediction tool. A working example is given for illustration. Accurate estimation of important MRI statistics and construction of efficient correction functions was performed. It also becomes feasible to create "artificial MRI images" for given MRI scanners by applying the obtained correction functions to the scans that were taken at different MRI scanners.
|
