Aging as an active player in Alzheimer's Disease Classification: Insights from feature selection in BrainAge Models

Area of Science:

• Neuroscience
• Biomarker Discovery
• Computational Biology

Background:

• BrainAge models estimate biological brain age using neuroimaging or clinical data.
• These models are promising for studying neurodegenerative diseases like Alzheimer's disease.
• Age-related biases in neuroimaging features can confound Alzheimer's disease diagnosis.

Purpose of the Study:

• To investigate the impact of age bias on BrainAge models for Alzheimer's disease diagnosis.
• To determine if feature selection can mitigate confounding effects between aging and disease pathology.
• To compare the utility of BrainAge deltas versus direct feature classification for Alzheimer's disease staging.

Main Methods:

• Ranked neuroimaging and neuropsychological features by correlation with age and discriminative power.
• Trained BrainAge models using feature subsets optimized for age prediction or disease classification.
• Evaluated BrainAge delta error and classification performance across clinical groups.
• Compared BrainAge delta classification with logistic regression on raw features.

Main Results:

• Neuroimaging features correlate more with aging; neuropsychological features better discriminate Alzheimer's disease.
• Age-optimized BrainAge models yield suboptimal disease classification deltas.
• Disease-optimized BrainAge models show reduced age prediction accuracy, indicating a trade-off.
• BrainAge offers a continuous measure across stages, unlike discrete classification labels.

Conclusions:

• Aging significantly impacts BrainAge-based Alzheimer's disease classification.
• Task-specific feature selection is critical to mitigate age bias in BrainAge models.
• Model design must ensure appropriate application of BrainAge in neurodegenerative disease research.