Predicting Brain Age Using Structural Neuroimaging and Deep Learning

Abstract

Early detection of age-related diseases will greatly benefit from a model of the underlying biological aging process. In this paper, we develop a brain-age predictor by using structural magnetic resonance imaging (SMRI) and deep learning and evaluate the predicted brain age as a marker of brain-aging in Alzheimer's disease. Our approach does not require any domain knowledge in that it trains end-to-end on the SMRI image itself, and has been validated on real SMRI data collected from elderly subjects. We developed two different models by using convolutional neural network (CNN) based regression and bucket classification to predict brain ages from SMRI images. Our models achieved root mean squared errors (RMSE) of 5.54 and 6.44 years in predicting brain ages of healthy subjects. Further analysis showed that there is a substantial difference between the predicted brain ages of cognitively impaired and healthy subjects with similar chronological ages.

Quantifying uncertainty in brain-predicted age using scalar-on-image quantile regression

“…This undesirable effect arises from the simple fact that by construction the residuals (which become the objects of interest when we want to explore the relationship with other variables such as disease conversion) in a regression model are uncorrelated with respect to the predicted values, but not with the observed ones. Similar problems (underpredictions for older subjects, overpredictions for younger ones) are also reported in the deep learning approaches to brain age prediction (Cole et al, 2017; Varatharajah et al, 2018). The work by Smith et al (2019) identifies potential reasons for this phenomenon and proposes some solutions.…”

“…The results from the analysis of ADNI data are encouraging: the point (median) prediction performances in terms of MAE and RMSE for the control subjects are comparable with the literature on the topic - even with deep learning approaches applied on bigger ADNI datasets (Varatharajah et al, 2018) - while being also more principled and interpretable. The correlation between chronological and predicted age results to be lower than the one found with other methods.…”

“…Cole et al (2019) collects a larger number of studies dealing with brain age prediction conducted from 2007 to 2018 with different imaging modalities and pathologies. Many of them adopt support vector regression (as the ones listed in Franke et al, 2012; Franke and Gaser, 2019 or Sone et al, 2019) or more recently Gaussian processes and convolutional neural networks (Cole et al, 2017; Cole, 2017; Varatharajah et al, 2018; Wang et al, 2019). A comparison between the predictive performances of these methods is difficult due to the use of different datasets and different age ranges, but according to Cole et al (2019) the choice of the algorithm does not seem to play a fundamental role.…”

Using deep learning to predict brain age from brain magnetic resonance images and cognitive tests reveals that anatomical and functional brain aging are phenotypically and genetically distinct

“…We report the feature importances in Table S2. We found that the nine most predictive scalar features are all related to speed of answering questions and include, in order: (1) time to answer, (2) mean time to correctly identify matches, (3) duration to complete alphanumeric path (trail #2), (4) time to complete test, (5) duration to complete numeric path (trail #1), ( 6) time to complete round (second round), (7) time to complete round (first round), (8) mean time to solve puzzles and (9) mean time to press snap-button. These nine features were associated with older age based on their elastic net (R 2 =33.0) regression coefficients.…”

Cardiometabolic risk factors associated with brain age and accelerate brain ageing

“…Higher brain age has been associated with poorer cognitive functioning in healthy individuals (Richard et al, 2018) and people with cognitive impairment (Varatharajah et al, 2018), mild cognitive impairment (MCI), dementia (Kaufmann et al, 2019), and mortality in elderly people (Cole et al, 2018). Larger BAGs have also been reported among patients with psychiatric and neurological disorders, including schizophrenia, bipolar disorder, multiple sclerosis (Høgestøl et al, 2019; Kaufmann et al, 2019; Tønnesen et al, 2020), depression (Han et al, 2020), and epilepsy (Pardoe et al, 2017; Sone et al, 2019).…”