Methods to assess changes in human brain structure across the lifecourse
View/ Open
Dickie2014.doc (91.03Mb)
Date
28/11/2014Author
Dickie, David Alexander
Metadata
Abstract
Human brain structure can be measured across the lifecourse (“in vivo”) with
magnetic resonance imaging (MRI). MRI data are often used to create “atlases” and
statistical models of brain structure across the lifecourse. These methods may define
how brain structure changes through life and support diagnoses of increasingly
common, yet still fatal, age-related neurodegenerative diseases. As diseases such as
Alzheimer’s (AD) cast an ever growing shadow over our ageing population, it is
vitally important to robustly define changes which are normal for age and those which
are pathological. This work therefore assessed existing MR brain image data, atlases,
and statistical models. These assessments led me to propose novel methods for
accurately defining the distributions and boundaries of normal ageing and
pathological brain structure.
A systematic review found that there were fewer than 100 appropriately tested
normal subjects aged ≥60 years openly available worldwide. These subjects did not
have the range of MRI sequences required to effectively characterise the features of
brain ageing. The majority of brain image atlases identified in this review were found
to contain data from few or no subjects aged ≥60 years and were in a limited range of
MRI sequences. All of these atlases were created with parametric (mean-based)
statistics that require the assumptions of equal variance and Gaussian distributions.
When these assumptions are not met, mean-based atlases and models may not well
represent the distributions and boundaries of brain structure.
I tested these assumptions and found that they were not met in whole brain,
subregional, and voxel-based models of ~580 subjects from across the lifecourse (0-
90 years). I then implemented novel whole brain, subregional, and voxel-based
statistics, e.g. percentile rank atlases and nonparametric effect size estimates. The
equivalent parametric statistics led to errors in classification and inflated effects by up
to 45% in normal ageing-AD comparisons. I conclude that more MR brain image
data, age appropriate atlases, and nonparametric statistical models are needed to
define the true limits of normal brain structure. Accurate definition of these limits will
ultimately improve diagnoses, treatment, and outcome of neurodegenerative disease.