Premium
[P3–390]: WHITE MATTER LOSS IN THE HEALTHY ELDERLY BRAIN INDICATIVE OF IMPENDING COGNITIVE DECLINE
Author(s) -
Zhang Linda,
Long Christopher,
Medina Miguel,
Strange Bryan
Publication year - 2017
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.2017.06.1606
Subject(s) - white matter , magnetic resonance imaging , atrophy , medicine , cohort , alzheimer's disease neuroimaging initiative , voxel , cognition , psychology , audiology , cognitive impairment , disease , neuroscience , radiology
methods were FreeSurfer 5.3.0 both cross sectional and longitudinal, FreeSurfer 6.0 beta longitudinal, MAPS-HBSI and FSL/FIRST 5.0.8. BTB manual segmentations of 75 patients from a previous study over 1 year were included to check the Gaussianity of manual BTB differences. The ratio of the standard deviation of the BTB differences (SDBTBD) over the median of absolute value of the BTB differences (MDBTBD) was compared to 1.3654 for Gaussian. Results:All methods including manual violated the Gaussianity assumption with all methods having at least one BTB difference distribution with p < 0.0001. The table in Figure 1 provides the SDBTBD/MDBTBD ratio for each of the distributions and their p-values for the Gaussian null hypothesis. The ratios were too large for a Gaussian distribution indicating the distributions had large shoulders containing many outlying BTB differences. The scatter plots in Figures 2 and 3 of the BTB differences for the fully automatic methods, alongside and simulated Gaussian distribution, provide visual confirmation of the large number of outlying BTB differences for the fully automatic methods. Conclusions:All of the atrophy segmentation methods examined had BTB difference distributions of their segmentation noises that violated the Gaussian assumption because of too many outlying BTB differences. Most of the statistical analysis of the segmentation noise published in the literature used parametric measures. Therefore, statistics of the BTB difference distributions that used parametric statistic, such as noise and group size, should be recalculated with nonparametric statistics unless the Gaussian distribution is confirmed.