Regional patch-based MRI brain age modeling with an interpretable cognitive reserve proxy.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarkers

Background:

• Brain age prediction from MRI is a key biomarker for brain health.
• Current deep learning models often lack anatomical specificity and clinical insight.

Purpose of the Study:

• To develop a regional, anatomically sensitive deep learning framework for brain age prediction.
• To create a cognitively informed proxy for cognitive reserve (CR-Proxy) using brain age and cognitive assessments.

Main Methods:

• A regional patch-based ensemble framework using 3D Convolutional Neural Networks (CNNs).
• Training on bilateral patches from ten subcortical structures.
• Combining ensemble predictions with cognitive assessments to derive the CR-Proxy.

Main Results:

• The framework achieves robust brain age prediction.
• The CR-Proxy effectively distinguishes between healthy controls, Alzheimer's disease, and mild cognitive impairment groups.
• The CR-Proxy identifies individuals with high or low cognitive reserve.

Conclusions:

• The proposed pipeline offers a scalable and clinically accessible tool for brain health monitoring.
• This method enhances anatomical sensitivity in brain age prediction.
• The CR-Proxy provides a practical and interpretable measure of resilience to age-related brain changes.