BrainAGE: Revisited and reframed machine learning workflow

Abstract

Since the introduction of the BrainAGE method, novel machine learning methods for brain age prediction have continued to emerge. The idea of estimating the chronological age from magnetic resonance images proved to be an interesting field of research due to the relative simplicity of its interpretation and its potential use as a biomarker of brain health. We revised our previous BrainAGE approach, originally utilising relevance vector regression (RVR), and substituted it with Gaussian process regression (GPR), which enables more stable processing of larger datasets, such as the UK Biobank (UKB). In addition, we extended the global BrainAGE approach to regional BrainAGE, providing spatially specific scores for five brain lobes per hemisphere. We tested the performance of the new algorithms under several different conditions and investigated their validity on the ADNI and schizophrenia samples, as well as on a synthetic dataset of neocortical thinning. The results show an improved performance of the reframed global model on the UKB sample with a mean absolute error (MAE) of less than 2 years and a significant difference in BrainAGE between healthy participants and patients with Alzheimer's disease and schizophrenia. Moreover, the workings of the algorithm show meaningful effects for a simulated neocortical atrophy dataset. The regional BrainAGE model performed well on two clinical samples, showing disease-specific patterns for different levels of impairment. The results demonstrate that the new improved algorithms provide reliable and valid brain age estimations.

Keywords: Alzheimer's disease; Gaussian process regression; UK Biobank; brain age; machine learning; mean absolute error; pre-processing; schizophrenia; structural MRI.

FIGURE 1

(a) Various single models differing in spatial resolution (R: 4 mm/8 mm), smoothing kernel (S: 4 mm/8 mm), and tissue segments (GM, WM, NGM: non‐linearly registered modulated GM) were estimated and combined. The results of model estimations combined by averaging are presented as age‐bias corrected MAEs and the Pearson's correlation coefficients between predicted‐ and chronological age for analyses run on a UKB 1 subsample. The comparison of the ML algorithms (RVR/GPR), as well as the effects of dimension reduction and the choice of brain tissues or their concatenation are shown. The results in bold represent the same combined model. (b) Eight single models were ensembled by averaging, weighted averaging, or GPR stacking. The results are presented for the subsample and the full UKB sample.

FIGURE 2

BrainAGE scores of the simulated neocortical thinning dataset with 0 and 1 mm atrophy examples on the right side.

FIGURE 3

(a) Mean BrainAGE scores for healthy individuals and patients with differing levels of neurocognitive impairment from the ADNI sample. (b) Mean BrainAGE scores for healthy control participants and groups of patients with varying symptoms from the SZ sample.

FIGURE 4

(a) Regional BrainAGE scores in healthy control subjects (HC), patients with stable MCI (sMCI), progressive MCI (pMCI), and AD patients from the ADNI dataset. (b) Regional BrainAGE scores for control subjects and three subgroups of SZ patients.