Deformation fields: a new source of information to predict brain age

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Accurate modeling of healthy aging requires methods to detect subtle changes.
• Current brain age models primarily use local volumetric information of brain tissues (grey matter, white matter, cerebrospinal fluid).
• Patterns of brain deformation remain an underexplored source of information for age prediction.

Purpose of the Study:

• To assess the predictive value of brain deformation fields compared to traditional volumetric measures.
• To investigate if deformation fields enhance the predictive power of grey matter volume in age estimation.
• To determine the robustness of deformation fields in BrainAge modeling.

Main Methods:

• Principal Component Analysis (PCA) for dimensionality reduction of brain images.
• Relevant Vector Regression (RVR) to learn age patterns from image components.
• Comparative analysis of models trained on deformation fields, grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF) volumes.
• Model fusion approach combining deformation fields and GM volume.
• Cross-validation and external dataset evaluation.

Main Results:

• Models trained with deformation patterns demonstrated higher predictive value than those using WM or CSF.
• Deformation fields showed significantly better performance on the test set with lower validation-test set discrepancies.
• Combining deformation patterns with GM volume yielded superior results compared to GM volume alone.

Conclusions:

• Brain deformation fields possess greater predictive power for age than WM and CSF volumes.
• Deformation fields are robust and invariant to confounding variables, making them reliable for age prediction.
• Deformation fields should be incorporated into future BrainAge modeling approaches.