An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling

Abstract

The deviation between chronological age and age predicted from neuroimaging data has been identified as a sensitive risk marker of cross-disorder brain changes, growing into a cornerstone of biological age research. However, machine learning models underlying the field do not consider uncertainty, thereby confounding results with training data density and variability. Also, existing models are commonly based on homogeneous training sets, often not independently validated, and cannot be shared because of data protection issues. Here, we introduce an uncertainty-aware, shareable, and transparent Monte Carlo dropout composite quantile regression (MCCQR) Neural Network trained on N = 10,691 datasets from the German National Cohort. The MCCQR model provides robust, distribution-free uncertainty quantification in high-dimensional neuroimaging data, achieving lower error rates compared with existing models. In two examples, we demonstrate that it prevents spurious associations and increases power to detect deviant brain aging. We make the pretrained model and code publicly available.

Linking Brain Age Gap to Mental and Physical Health in the Berlin Aging Study II

“…To our knowledge, pre-trained and publicly available brain age models are still very scarce. One reason for this is that the employed models often contain data that allow partial reconstruction of the training dataset and, thus, they pose privacy issues that hamper data sharing among researchers (Hahn et al, 2022). Shielding individual privacy is a major challenge in the field of machine learning and artificial intelligence.…”

“…Another advantage to be noted is that using custom models-and thereby introducing some degree of methodological heterogeneityfurther strengthens the robustness and generalizability of results derived in this field of research. Still, we believe that future research will tremendously benefit from establishing shareable age estimation models as proposed by Hahn et al (2022), as they facilitate research in small datasets, promote consensus-building and increase interpretability of results.…”

Age-level bias correction in brain age prediction

“…Our simple correction method and its efficacy in experiments suggest that uncertainty analysis and measurement is a potential approach. Sample-level correction method ( Hahn et al, 2022 ) based on uncertainty analysis demonstrates its superiority. Besides, although some sample-level correction methods train prediction models by adding the correlation between the corrected PAD and the chronological age into the objective function or constraints, how to execute age-level bias correction during the model training is unknown.…”

Brain age gap in neuromyelitis optica spectrum disorders and multiple sclerosis

“…The uncertainty and distributional pattern of predicted brain age is an important field of research that has attracted little attention. A recent study modeled brain-age uncertainty with a single-layer neural network that addressed aleatoric uncertainty with quantile regression and epistemic uncertainty with the Monte Carlo drop-out technique 24 . In contrast to other studies that utilize quantile regression, the novel method in our study renders aleatoric uncertainty a natural derivative since the model output itself is a distribution instead of the point estimate used in previous studies 4 .…”