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

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Gerontology

Background:

• Estimating biological brain age from MRIs helps understand aging deviations.
• Current deep learning models for brain age estimation may lack accuracy and interpretability.

Purpose of the Study:

• To introduce a novel convolutional neural network (CNN) for more accurate and interpretable brain age (BA) estimation.
• To map individual and cohort-level brain aging patterns, including sex differences and neurocognitive trajectories.

Main Methods:

• Trained a CNN on 4,681 cognitively normal (CN) participants' MRIs.
• Tested the CNN on an independent sample of 1,170 CN participants.
• Analyzed brain aging patterns in individuals with mild cognitive impairment (MCI) and Alzheimer's disease (AD).

Main Results:

• Achieved notably lower BA estimation errors compared to previous studies.
• Generated detailed anatomic maps of brain aging, revealing sex dimorphisms and neurocognitive trajectories.
• In MCI patients, BA significantly outperformed chronological age in assessing dementia severity, functional disability, and executive function.

Conclusions:

• The CNN framework provides a systematic way to map aging-related neuroanatomy changes in CN, MCI, and AD individuals.
• This approach can aid in the early identification of neuroanatomy changes and screening for Alzheimer's disease risk.
• Brain aging patterns correlate with cognitive decline and neuroanatomic changes, offering insights into disease progression.