Estimation of brain age delta from brain imaging.

Smith SM, Vidaurre D, Alfaro-Almagro F, Nichols TE, Miller KL. · 2019

Read Estimation of brain age delta from brain imaging. by Smith SM, Vidaurre D, Alfaro-Almagro F, Nichols TE, Miller KL. (2019) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6711452.txt.

Introduction

Smith and co-authors examined how best to compute brain age delta. Brain age is how old a brain scan makes someone look, judged against a large reference group. Delta is the gap between that estimate and a person's real age. A positive gap may signal faster aging, a negative one resilience.

The note shows that the common way of computing delta is biased. The authors test fixes on simulated and real data. They then check corrected delta against 5,792 non-imaging measures in about 19,000 UK Biobank participants. Their method gives cleaner links to health, life factors, and thinking test scores.

Catalog record

CREATORSmith SM, Vidaurre D, Alfaro-Almagro F, Nichols TE, Miller KL.
YEAR2019
COLLECTIONPAPERS
FORMATTXT
FILE SIZE49 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC6711452

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Estimation of brain age delta from brain imaging By Stephen M Smith, Diego Vidaurre, Fidel Alfaro-Almagro, Thomas E Nichols, Karla L Miller Published in: Neuroimage DOI: 10.1016/j.neuroimage.2019.06.017 ABSTRACT Abstract It is of increasing interest to study “brain age” - the apparent age of a subject, as inferred from brain imaging data. The difference between brain age and actual age (the “delta”) is typically computed, reflecting deviation from the population norm. This therefore may reflect accelerated aging (positive delta) or resilience (negative delta) and has been found to be a useful correlate with factors such as disease and cognitive decline. However, although there has been a range of methods proposed for estimating brain age, there has been little study of the optimal ways of computing the delta. In this technical note we describe problems with the most common current approach, and present potential improvements. We evaluate different estimation methods on simulated and real data. We also find the strongest correlations of corrected brain age delta with 5,792 non-imaging variables (non-brain physical measures, life-factor measures, cognitive test scores, etc.), and also with 2,641 multimodal brain imaging-derived phenotypes, with data from 19,000 participants in UK Biobank. Keywords: Brain aging, Brain imaging, UK biobank INTRODUCTION Brain imaging (and other sources of relevant data) can be used to predict “brain age” - the apparent age of individuals, when comparing their data against a population dataset spanning a range of ages. The difference between brain age and actual age (the “delta”) is often then computed, providing a measure of whether a subject's brain appears to have aged more or less than the population average for their actual chronological age. For example, looking at structural MRI data, a high degree of atrophy (e.g., caused by disease) would cause a subject's brain to appear older than a normal age-matched brain [Franke et al., 2010; Cole et al., 2017; Cole and Franke, 2017]. The approach typically taken is to use one or more imaging modalities, for example, acquiring a T1-weighted structural image from each subject. The data then receives some level of preprocessing, e.g., alignment to standard space and tissue type segmentation. The imaging data then becomes “features” for predicting brain age - for example, from voxelwise maps of grey matter partial volume estimates, the voxelwise values themselves can be the features. Alternatively, a smaller number of more highly-condensed features, such as volumes of grey matter within multiple distinct brain regions of interest, may be generated. The entire dataset of multiple subjects' features, and their true ages, are fed into a supervised-learning algorithm (e.g., regression, support vector machine, deep learning), which learns to predict the subjects' ages from their brain imaging features. The hope is that a given subject's predicted brain age will deviate from their true age according to a meaningful delta, as long as this training is not badly overfitting. While there have been a range of methods proposed for estimating brain age (i.e., choice of imaging-derived features to use, and choice of supervised-learning approach), there has been very little study of the optimal ways of then computing the delta. Presumably this is in part because this seems like a very simple calculation (delta equals brain age minus age). However, there are frequently various sources of bias in estimating brain age delta, which can give rise to significant false positives and false negatives when looking for associations between delta and other measures [Le et al., 2018]. Here we describe some important problems with this most common approach, and present potential improvements via explicitly laid out (albeit simple) mathematical frameworks. We study these effects in simulated and real data, and suggest simple models for removing bias and increasing the accuracy of delta estimation. This includes models for correcting underestimation of brain aging, removing the resulting dependency of delta on age, modelling nonlinear dependence of brain aging (as a function of age), and studying non-additive brain age delta. One of the causes of bias (regression dilution, see below) has recently also been discussed in depth in [Le et al., 2018], where the same linear correction as Eq. (4) below was proposed. A closely-related linear correction was also recently proposed in [Liang et al., 2019], although these authors suggest that regression towards the mean is the only source of bias, and we find empirically that non-Gaussian distribution of subject ages can be a major cause of bias. METHODS AND RESULTS In order to present the clearest description of the gradual development of models throughout the paper, we do not separate out Methods and Results sections, but intermix discussions of model improvements with simulation and real-data results. Common approach 2.1. We start with some basic definitions, and outline the typical approach for brain age delta estimation. Actual age is Y (an Nsubjects×1 vector), brain age is YB and the delta is δ=YB−Y. The imaging data matrix is X, which has Nsubjects rows and D columns; the columns are features from the imaging data, and might be different voxels, or different IDPs (imaging-derived phenotypes - summary measures of brain structure and function).1 It is common to try to predict YB from X: [formula: (1)YB=Y+δ=f(X).] Although many alternatives have been proposed for the form of f() and the fitting of its model parameters, the issues explored in this paper generally hold, irrespective of these choices (e.g., most brain age literature shows underestimation of brain age for old subjects, and overestimation in young subjects). We start with a very simple linear model, with f(X)=Xβ, a multiple regression with parameters β (a D×1 vector). We therefore have: [formula: (2)Y=Xβ1−δ1,] where we have added subscripts to differentiate some variables in this model from later variants. Clearly δ1 is the brain age delta being estimated, and the noise term in this formulation. This multiple regression can be solved with standard simple methods,2 for example, setting β1=X+Y using the pseudo-inverse X+=(X'X)−1X'. This gives the initially predicted brain age YB1=XX+Y, and brain age delta: [formula: (3)δ1=Xβ1−Y=XX+Y−Y=(XX+−I)Y.] Two complications are immediately clear: δ1 will be orthogonal to the imaging data matrix X (though there is no reason to assume that this is desirable), and it will not be orthogonal to age (Y). This latter issue is likely to be problematic, given the simplest conception of the delta as being the difference between brain age and age, as a useful (objective, non-changing) description of a subject's brain-health that is not dependent on their current age. In the extreme case of X being an entirely useless model for brain aging, Xβ1≈0 and δ1 will just be −1 times the actual age. In addition, there are several factors that almost always cause the estimated β1 to be biased towards zero (Fig. 1), resulting in under-fitting to Y and hence “moving” age dependence into δ1: - 1.It is typical for more sophisticated fitting methods (e.g., sparse/regularised regression) to result in underestimation of β1 [Grosenick et al., 2013]. - 2.It is common for datasets to not have a Gaussian distribution (across all subjects) for age, often because of hard limits on age in the study design. In the linear model framework, it does not generally matter what the distribution of the predictors is (here, X), but a non-Gaussian distribution in the independent variable (here Y)

Read the complete text