Investigating systematic bias in brain age estimation with application to post‐traumatic stress disorders

Abstract

Brain age prediction using machine-learning techniques has recently attracted growing attention, as it has the potential to serve as a biomarker for characterizing the typical brain development and neuropsychiatric disorders. Yet one long-standing problem is that the predicted brain age is overestimated in younger subjects and underestimated in older. There is a plethora of claims as to the bias origins, both methodologically and in data itself. With a large neuroanatomical dataset (N = 2,026; 6-89 years of age) from multiple shared datasets, we show this bias is neither data-dependent nor specific to particular method including deep neural network. We present an alternative account that offers a statistical explanation for the bias and describe a simple, yet efficient, method using general linear model to adjust the bias. We demonstrate the effectiveness of bias adjustment with a large multi-modal neuroimaging data (N = 804; 8-21 years of age) for both healthy controls and post-traumatic stress disorders patients obtained from the Philadelphia Neurodevelopmental Cohort.

Improved prediction of brain age using multimodal neuroimaging data

“…This study was a follow‐up to the initial work (Liang et al, ) to combine three modalities of T 1 ‐weighted MRI, DTI and resting‐state fMRI features in brain age prediction. In ML, noninformative features often deteriorate prediction performance on an independent validation set.…”

“…In accordance with previous studies, a positive brain age gap referred to advanced brain development. To account for the systematic bias in brain age prediction (Liang et al, ), the nonlinear brain development trajectory (Dosenbach et al, ), as well as potential gender difference (Goyal et al, ), we extended the linear model (Liang et al, ) by using the nonlinear formula below to control for the confounding effects of chronological age and gender. Importantly, to fit the formula below, brain age was estimated from an independent test set using brain imaging data and ML methods. …”

“…Specifically, brain age tended to be underestimated for older subjects and overestimated for younger ones, leading to systematic bias of the estimated brain age gap (Aycheh et al, 2018;Cole & Franke, 2017;Pardoe & Kuzniecky, 2018). The potential reasons for this bias are under debate, including regression toward the mean (Liang, Zhang, & Niu, 2019), and nongaussian distribution of subject ages (Smith, Vidaurre, Alfaro-Almagro, Nichols, & Miller, 2019). This systematic bias tends to introduce a confounding effect of chronological age when the brain age gap is examined as a biomarker for a mental disease or cognition development.…”

“…The multimodal brain imaging data including T 1 -weighted MRI, DTI, and resting-state fMRI features has been used in brain age prediction (Liang et al, 2019). In this work, we follow up with the initial study (Liang et al, 2019) with more subjects.…”

“…Different ML algorithms have been used for brain age prediction, which include support vector regression (SVR) (Dosenbach et al, 2010;Erus et al, 2015;Franke et al, 2010;Koutsouleris et al, 2014;Liem et al, 2017), Gaussian process regression (GPR) (Aycheh et al, 2018;Cole et al, 2015, relevance vector regression (RVR) (Franke & Gaser, 2012;Gaser et al, 2013;Mwangi et al, 2013), ridge regression (Chung et al, 2018) and elastic net (Khundrakpam, Tohka, & Evans, 2015). Overall, these ML models showed comparable performance on age prediction (Liang et al, 2019). It has been shown that SVR, RVR, ridge regression and elastic net performed similarly on the prediction of the motor and cognitive scores with resting-state functional connectivity features (Cui & Gong, 2018).…”