Brain age prediction reveals aberrant brain white matter in schizophrenia and bipolar disorder: A multi-sample diffusion tensor imaging study

Abstract

Word count abstract: 249/250 | Main text: 3810/4000 Number of Figures 2 | Tables: 1 | Supplementary information: 1 document Abstract (249/250 words)Background: Schizophrenia (SZ) and bipolar disorders (BD) share substantial neurodevelopmental components affecting brain maturation and architecture. This necessitates a dynamic lifespan perspective in which brain aberrations are inferred from deviations from expected lifespan trajectories. We applied machine learning to diffusion tensor imaging (DTI) indices of white matter structure and organization to estimate and compare brain age between patients with SZ, BD, and healthy controls across 10 cohorts.Methods: We trained six cross-validated models using different combinations of DTI data from 927 healthy controls (HC, 18-94 years), and applied the models to the test sets including 648 SZ (18-66 years) patients, 185 BD patients (18-64 years), and 990 HC (17-68 years), estimating brain age for each participant. Group differences were assessed using linear models, accounting for age, sex, and scanner. A meta-analytic framework was applied to assess the heterogeneity and generalizability of the results.Results: 10-fold cross-validation revealed high accuracy for all models. Compared to controls, the model including all feature sets significantly over-estimated the age of patients with SZ (d=-.29) and BD (d=.18), with similar effects for the other models. The meta-analysis converged on the same findings. Fractional anisotropy (FA) based models showed larger group differences than the models based on other DTI-derived metrics.Conclusions: Brain age prediction based on DTI provides informative and robust proxies for brain white matter integrity. Our results further suggest that white matter aberrations in SZ and BD primarily consist of anatomically distributed deviations from expected lifespan trajectories that generalize across cohorts and scanners.

A history of previous childbirths is linked to women's white matter brain age in midlife and older age

“…In line with recent studies demonstrating high age prediction accuracy based on diffusion‐weighted imaging data (Beck et al, 2021; Cole, Marioni, et al, 2019; Richard et al, 2018; Smith et al, 2020; Tønnesen et al, 2020), the WM prediction showed higher accuracy compared to the GM model, of which the accuracy corresponded to our previous UK Biobank studies (de Lange et al, 2019; de Lange, Barth, Kaufmann, Anatürk, et al, 2020; de Lange, Barth, Kaufmann, Maximov, et al, 2020). Importantly, we found unique contributions by both models, suggesting that the diffusion‐based WM model may pick up variance not explained by the T1‐based GM model.…”

A history of previous childbirths is linked to women’s white matter brain age in midlife and older age

“…In line with recent studies demonstrating high age prediction accuracy based on diffusion imaging data [37,63,89,90,91], the WM prediction showed higher accuracy compared to the GM model, of which the accuracy corresponded to our previous UK Biobank studies [10,11,54]. Importantly, we found unique contributions by both models, suggesting that the diffusion-based WM model may pick up variance not explained by the T1-based GM model.…”

Mind the gap: performance metric evaluation in brain-age prediction

“…While age prediction models have been used for more than a decade to generate imaging-based biomarkers [20], the approach continues to be developed and extended (see for example [3,9,56,31,49,58]). Although not covered in the current study, an increasingly common scenario involves combining data from various cohorts and scanners, which poses additional challenges related to site-and scanner-dependent variance [25]. Improving methods for site/scanner adjustments [59,60], or incorporating uncertainties into the predictions [61,62], represent promising avenues for further developing robust and valid biomarkers for brain health and disease.…”

“…[42,40,38]). In other words, it provides the same results as regressing out chronological age from brain age delta and using the residuals [41,6,9,25,12], or including chronological age as a covariate in regressions/correlations between brain age delta and other variables of interest [41,42,49,31]. The orange line shows the linear fit applied to model the age bias.…”

“…Prediction models can then be applied to unseen data, providing an estimate of brain-predicted age for each individual in the dataset. The difference between an individual's brain-predicted and chronological age (brain age delta) provides a proxy for deviations from expected age trajectories, and has been associated with clinical risk factors [10,11,23] as well as neurological and neuropsychiatric conditions [2,9,13,20,24,25,26,27,28,29]. Brain age delta estimates have also been linked to biomedical variables and lifestyle factors in healthy population cohorts [3,22,11,12,30,31,32], and the overall evidence supports the use of brain-predicted age as a surrogate marker for brain integrity and health [21].…”