Brain Age Prediction: A Comparison between Machine Learning Models Using Brain Morphometric Data.

Area of Science:

• Neuroimaging
• Machine Learning
• Aging Research

Background:

• Brain structural morphology changes with age, making brain age prediction vital for identifying atypical aging patterns.
• Neuroimaging-derived brain age quantifies brain health by measuring deviation from normative aging trajectories.
• Machine learning (ML) offers potential for accurate brain age prediction but faces challenges due to algorithm diversity.

Purpose of the Study:

• To compare the performance of various ML models in estimating brain age from structural MRI morphometric data.
• To evaluate the effectiveness of different algorithms across independent datasets.
• To identify computationally efficient ML models for reliable brain age prediction.

Main Methods:

• Evaluated 27 ML models using structural MRI data from three independent cohorts: HCP (n=1113), Cam-CAN (n=601), and IXI (n=567).
• Assessed model performance via cross-validation and an independent test set within each dataset.
• Calculated mean absolute error (MAE) and Pearson's correlation coefficient (r) to measure prediction accuracy.

Main Results:

• Model performance varied significantly, highlighting the impact of algorithm choice on brain age prediction accuracy.
• Achieved MAE ranging from 2.75-3.12 years (HCP), 7.08-10.50 years (Cam-CAN), and 8.04-9.86 years (IXI).
• Regularized linear regression models demonstrated performance comparable to nonlinear and ensemble methods across all datasets, with lower computational cost.

Conclusions:

• Regularized linear regression algorithms are as effective as more complex ML models for brain age prediction.
• The choice of ML model significantly influences brain age prediction outcomes.
• Findings provide a quantitative reference for selecting efficient ML models for brain morphometric data analysis in aging research.