Paper Info
Title
Improved longitudinal gray and white matter atrophy assessment via application of a 4-dimensional hidden Markov random field model.
Abstract
SIENA and similar techniques have demonstrated the utility of performing “direct” measurements as opposed to post-hoc comparison of cross-sectional data for the measurement of whole brain (WB) atrophy over time. However, gray matter (GM) and white matter (WM) atrophy are now widely recognized as important components of neurological disease progression, and are being actively evaluated as secondary endpoints in clinical trials. Direct measures of GM/WM change with advantages similar to SIENA have been lacking. We created a robust and easily-implemented method for direct longitudinal analysis of GM/WM atrophy, SIENAX multi-time-point (SIENAX-MTP). We built on the basic halfway-registration and mask composition components of SIENA to improve the raw output of FMRIB's FAST tissue segmentation tool. In addition, we created LFAST, a modified version of FAST incorporating a 4th dimension in its hidden Markov random field model in order to directly represent time. The method was validated by scan–rescan, simulation, comparison with SIENA, and two clinical effect size comparisons. All validation approaches demonstrated improved longitudinal precision with the proposed SIENAX-MTP method compared to SIENAX. For GM, simulation showed better correlation with experimental volume changes (r=0.992 vs. 0.941), scan–rescan showed lower standard deviations (3.8% vs. 8.4%), correlation with SIENA was more robust (r=0.70 vs. 0.53), and effect sizes were improved by up to 68%. Statistical power estimates indicated a potential drop of 55% in the number of subjects required to detect the same treatment effect with SIENAX-MTP vs. SIENAX. The proposed direct GM/WM method significantly improves on the standard SIENAX technique by trading a small amount of bias for a large reduction in variance, and may provide more precise data and additional statistical power in longitudinal studies.
Year
DOI
Venue
2014
10.1016/j.neuroimage.2013.12.004
NeuroImage
Keywords
Field
DocType
Gray matter,White matter,Tissue-specific,Atrophy,Longitudinal,Segmentation,Direct method,4-Dimensional HMRF,Random field
Direct method,Random field,White matter,Hidden Markov random field,Segmentation,Psychology,Correlation,Statistics,Statistical power,Standard deviation
Journal
Volume
ISSN
Citations 
90
1053-8119
2
PageRank 
References 
Authors
0.42
27
3
Name
Order
Citations
PageRank
Michael G Dwyer1244.05
Niels Bergsland2193.56
Robert Zivadinov3203.24