Deep neural networks learn general and clinically relevant representations of the ageing brain

Abstract

The discrepancy between chronological age and the apparent age of the brain based on neuroimaging data — the brain age delta — has emerged as a reliable marker of brain health. With an increasing wealth of data, approaches to tackle heterogeneity in data acquisition are vital. To this end, we compiled raw structural magnetic resonance images into one of the largest and most diverse datasets assembled (n=53542), and trained convolutional neural networks (CNNs) to predict age. We achieved state-of-the-art performance on unseen data from unknown scanners (n=2553), and showed that higher brain age delta is associated with diabetes, alcohol intake and smoking. Using transfer learning, the intermediate representations learned by our model complemented and partly outperformed brain age delta in predicting common brain disorders. Our work shows we can achieve generalizable and biologically plausible brain age predictions using CNNs trained on heterogeneous datasets, and transfer them to clinical use cases.

Linking brain maturation and puberty during early adolescence using longitudinal brain age prediction in the ABCD cohort

“…T1-weighted images were acquired with real time motion correction and imaging parameters harmonized for three 3T scanner platforms ( Siemens Prisma, General Electric (GE) 750 and Philips) (Casey et al, 2018) and minimally processed (skull stripping, reorientation, and normalization) as described in detail in Leonardsen et al, (2022).…”

“…The estimated brain age for each participant was calculated using a CNN trained and validated in minimally processed T1-weighted MRI data (n=53542, 5-93 years) from 21 publicly available datasets (Leonardsen et al, 2022). The model architecture is a regression variant (SFCN-reg) of the PAC2019-winning SFCN model (Peng et al, 2021).…”

“…In the current study, we calculated brain age using a recent deep learning approach based on convolutional neural networks (CNNs) in a large training set comprised of minimally processed T1-weighted MRI data from >50,000 individuals aged 5 to 93 years (Leonardsen et al, 2022). We applied the model to predict brain age in the Adolescent Brain Cognitive Development (ABCD) cohort, including 7459 baseline scans and 2384 scans from the first MRI follow-up two years later.…”

Mind the gap: Performance metric evaluation in brain‐age prediction

“…A number of recent studies show that ML algorithms can predict age based on MRI data with high accuracy, for example, (Couvy‐Duchesne et al, 2020 ; Gong, Beckmann, Vedaldi, Smith, & Peng, 2021 ; Han et al, 2020 ; Kaufmann et al, 2019 , Leonardsen et al, 2021 ). However, in addition to differences in feature sets included (Cole, 2020 ; de Lange, Anatürk, et al, 2020 ; Jollans et al, 2019 ), training and test set characteristics such as size and age range (de Lange, Anatürk, et al, 2020 ; Jollans et al, 2019 ) can lead to considerable variation in model performance metrics across studies.…”

Longitudinal brain age prediction and cognitive function after stroke

“…To increase feasibility in a clinical context future work may implement novel approaches allowing for accurate estimations based on e.g. deep neural networks and minimally processed MRI data (Leonardsen et al, 2021). Next, study attrition at 36 months was considerable, leading to reduced statistical power and possible biases.…”