Introduction to BrainAge

This lecture provides a foundational overview of the "Brain Age" concept—a neuroimaging-derived metric that estimates an individual's biological brain age based on structural and functional characteristics, offering a potential proxy for overall brain health. By training machine learning (ML) models on chronological age data from healthy individuals, researchers calculate a "Brain Predicted Age Difference" (or "brain age gap"), where a discrepancy between predicted and actual age can serve as a biomarker for neurodegenerative risk or age-related disease. The lecture highlights how the combination of increased open-source data availability and advanced machine learning techniques has enabled this field to evolve into a tool for personalized medicine, while also detailing the methodological workflow—from model training and validation to statistical analysis—and cautioning that multimodal approaches must be chosen carefully to avoid masking specific, tissue-level clinical associations.

Learning objectives

By the end of this lecture, students will be able to:

• Understand how brain age is estimated through neuroimaging and what the "brain age gap" represents as a biomarker
• Learn the difference between traditional machine learning (pattern recognition) and deep learning (neural network-based feature discovery) in the context of age prediction
• Recognize key statistical measures for model validation, such as the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), and their sensitivity to sample size
• Evaluate the trade-offs of using multiple imaging modalities (T1, T2, DTI, functional connectivity) in modeling, including the risk of masking unique biomarkers