Improved prediction of brain age using multimodal neuroimaging data

Area of Science:

• Neuroimaging
• Machine Learning
• Cognitive Neuroscience

Background:

• Brain age prediction using imaging and ML offers insights into cognition and mental disorders.
• A systematic comparison of ML models and imaging features is needed to optimize prediction accuracy.

Purpose of the Study:

• To evaluate 36 combinations of imaging features and ML models for brain age prediction.
• To investigate multimodal data integration (MRI, DTI, rs-fMRI) for improved accuracy.
• To propose a novel method for correcting systematic bias in brain age gap prediction and explore its link to behavioral performance.

Main Methods:

• Utilized single and multimodal brain imaging data (MRI, DTI, rs-fMRI) from 839 subjects.
• Evaluated 36 combinations of imaging features and ML models, including deep learning.
• Developed a bias-correction method for brain age gap considering gender, age, and interactions, and examined its relation to behavior.

Main Results:

• Identified optimal combinations of imaging features and ML models for brain age prediction.
• Demonstrated the effectiveness of the proposed bias-correction method for brain age gap estimation.
• Found that behavioral performance predicts brain age estimated from neuroimaging data.

Conclusions:

• Findings advance the optimization of analytical methodologies for brain age prediction.
• The study provides a framework for understanding the relationship between brain age, chronological age, and behavior.
• Results contribute to quantifying the practical implications of brain age prediction in clinical and research settings.