Anatomically interpretable deep learning of brain age captures domain-specific cognitive impairment

Abstract

The gap between chronological age (CA) and biological brain age, as estimated from magnetic resonance images (MRIs), reflects how individual patterns of neuroanatomic aging deviate from their typical trajectories. MRI-derived brain age (BA) estimates are often obtained using deep learning models that may perform relatively poorly on new data or that lack neuroanatomic interpretability. This study introduces a convolutional neural network (CNN) to estimate BA after training on the MRIs of 4,681 cognitively normal (CN) participants and testing on 1,170 CN participants from an independent sample. BA estimation errors are notably lower than those of previous studies. At both individual and cohort levels, the CNN provides detailed anatomic maps of brain aging patterns that reveal sex dimorphisms and neurocognitive trajectories in adults with mild cognitive impairment (MCI, N = 351) and Alzheimer’s disease (AD, N = 359). In individuals with MCI (54% of whom were diagnosed with dementia within 10.9 y from MRI acquisition), BA is significantly better than CA in capturing dementia symptom severity, functional disability, and executive function. Profiles of sex dimorphism and lateralization in brain aging also map onto patterns of neuroanatomic change that reflect cognitive decline. Significant associations between BA and neurocognitive measures suggest that the proposed framework can map, systematically, the relationship between aging-related neuroanatomy changes in CN individuals and in participants with MCI or AD. Early identification of such neuroanatomy changes can help to screen individuals according to their AD risk.

Disentangling neurodegeneration from ageing in multiple sclerosis: the brain-predicted disease duration gap

“…The interpretability analyses showed that all measures were influenced by regional brain volumes, with the lowest effect sizes observed for the MS-age gap metric, revealing the lower capacity of the corresponding model to capture inter-subject variability. In line with what has been previously observed for healthy brain ageing models, 29,30 age and DD predictions were influenced by spatially distributed, rather than localized, variations in brain volume. Interestingly, the presence of lesions did not directly influence BAG values, while MS-age and, most prominently, DD predictions on unfilled scans were systematically higher than those obtained on lesion-filled counterparts, suggesting that multiple sclerosis-specific models can effectively measure disease-related phenomena that are not captured by the classical brain-age paradigm.…”

“…The interpretability analyses showed that all measures were influenced by regional brain volumes, with the lowest effect sizes observed for the MS-age gap metric, revealing the lower capacity of the corresponding model to capture inter-subject variability. In line with what has been previously observed for healthy brain ageing models, 29,30 age and DD predictions were influenced by spatially distributed, rather than localized, variations in brain volume.…”

Biomarkers of aging

“…Brain structural MRI data could also be applied to age prediction with the MAE of 4.16 years (Cole et al, 2017). More recently, a 3D CNN model (Yin et al, 2023) has been introduced with a more accurate MAE of 2.3 years. Their model can reveal neurocognitive trajectories in adults with MCI and AD and may serve as early indicators for AD.…”

The capacity of action observation to drag the trainees' motor pattern toward the observed model

“…Our sample is unbalanced in terms of participants' gender (76.39% of participants are female). Such unbalance might bias the results of our study, considering that males and females have different susceptibility to learning motor skills 58 , and brains under different sexes may have different salience or gene expression values in different brain regions 59 – 61 . While such a bias might affect the overall learning curves, it must be highlighted that we preserved an identical gender balance across the two groups, with AOT composed of 28 females—8 males and CTRL 27 females—9 males.…”