Brain age prediction: A comparison between machine learning models using region- and voxel-based morphometric data.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Gerontology

Background:

• Brain morphology changes with age, making brain age prediction a tool for identifying aging abnormalities.
• Existing brain age prediction studies use diverse methods and data, creating uncertainty about optimal approaches.

Purpose of the Study:

• To compare machine learning model performance for brain age prediction using different structural MRI data types and preprocessing techniques.
• To identify the most accurate and generalizable methodological approach for brain age prediction.

Main Methods:

• Utilized UK Biobank data (N=10,824, age 47-73) to compare support vector regression, relevance vector regression, and Gaussian process regression.
• Evaluated whole-brain region-based and voxel-based structural MRI data, with and without principal component analysis (PCA) for dimensionality reduction.
• Assessed model performance using cross-validation and an independent test set.

Main Results:

• Models achieved mean absolute errors between 3.7 and 4.7 years.
• Brain age prediction models trained on voxel-level data combined with PCA demonstrated superior performance.
• Input data type (voxel-level vs. region-based) had a greater impact on performance than the choice of machine learning model.

Conclusions:

• Voxel-based structural MRI data with principal component analysis is a highly effective approach for accurate brain age prediction.
• The choice of machine learning model is less critical than the selection and preprocessing of neuroimaging data for brain age prediction.
• This research provides valuable insights and open-source code to advance future studies in neurodevelopmental and aging research.