Deep learning to quantify the pace of brain aging in relation to neurocognitive changes

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Gerontology

Background:

• Brain age (BA) from MRI estimates cumulative neuroanatomic aging but doesn't capture recent aging trends.
• Existing methods for brain aging pace (P) often use DNA methylation, which is separated from brain cells by the blood-brain barrier.

Purpose of the Study:

• To introduce a novel, noninvasive method using longitudinal MRI to estimate the pace of brain aging (P).
• To validate the model's accuracy and compare its performance against cross-sectional methods.
• To explore regional variations in brain aging rates and their association with cognitive function and neurocognitive status.

Main Methods:

• A three-dimensional convolutional neural network (3D-CNN) was developed as a longitudinal model (LM) to estimate P from longitudinal MRI data.
• The LM was trained on 2,055 cognitively normal (CN) adults and validated on 1,304 CN adults.
• The model was further applied to an independent cohort of 104 CN adults and 140 Alzheimer's disease (AD) patients.

Main Results:

• The LM estimated P with a mean absolute error (MAE) of 0.16 years (7% error), significantly outperforming cross-sectional models (MAE of 1.85 years, 83% error).
• The model identified regional variations in brain aging rates associated with sex, decade of life, and neurocognitive status.
• LM estimates of P showed significant associations with changes in cognitive functioning across domains.

Conclusions:

• The developed longitudinal model (LM) provides an accurate and noninvasive method to estimate the pace of brain aging (P) from MRI.
• This approach captures the relationship between neuroanatomic aging and neurocognitive aging, offering insights into age-related cognitive changes.
• The findings complement existing Alzheimer's disease risk assessment strategies by providing a measure of individual brain aging rates.