Local Brain-Age: A U-Net Model.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Gerontology

Background:

• Current brain-age estimation methods primarily use a global approach, lacking spatial detail.
• Understanding localized brain aging is crucial for identifying subtle neurodegenerative changes.

Purpose of the Study:

• To develop and validate a deep learning framework for local brain-age estimation.
• To provide spatial maps of brain aging and assess their utility in differentiating healthy aging from neurodegenerative conditions.

Main Methods:

• A U-Net deep learning model was trained on 3,463 healthy brain MRI scans to predict brain age.
• A voxelwise method was developed to reduce age bias in local brain-age gap predictions.
• The model was validated on 692 healthy individuals and 267 participants with mild cognitive impairment or dementia (OASIS3 dataset).

Main Results:

• The framework achieved a median absolute error of 9.5 years in healthy individuals, with higher accuracy in the prefrontal cortex (around 7 years).
• Distinct local brain-age patterns were observed between healthy controls and individuals with mild cognitive impairment or dementia, particularly in subcortical regions.
• Significant differences in local brain-age were found between groups, with large effect sizes (Cohen's d > 1.5).

Conclusions:

• The proposed local brain-age framework offers spatial information for a more mechanistic understanding of brain aging.
• This approach has potential for early detection and characterization of neurodegenerative diseases by revealing localized aging patterns.