Advanced Brain Age Prediction Using Multi-Head Self-Attention: A Comparative Analysis of Western and Middle Eastern...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Brain age estimation is crucial for detecting neurodegenerative diseases.
• Current AI models are limited by Western-centric training data, impacting diverse populations.
• Brain aging patterns may differ across ethnic groups.

Purpose of the Study:

• To develop and evaluate an AI model for brain age estimation.
• To assess the model's performance across different ethnic populations.
• To investigate the impact of diverse datasets on model generalizability.

Main Methods:

• Trained a deep learning model integrating multi-head self-attention and residual connections.
• Utilized a large dataset of 4,635 healthy individuals (40-80 years) from ADNI, OASIS-3, Cam-CAN, and IXI.
• Tested the model on Western (n=935) and Middle Eastern (n=107) datasets, evaluating with Mean Absolute Error (MAE).

Main Results:

• Achieved state-of-the-art accuracy (MAE = 1.99 years) on the Western test set with a lightweight architecture (approx. 3 million parameters).
• Demonstrated significantly lower performance on the Middle Eastern dataset (best MAE = 4.35 years).
• Bias correction did not improve performance on the Middle Eastern data.

Conclusions:

• Population-specific differences in brain aging exist, impacting AI model generalizability.
• Current AI models for brain age estimation require diverse training data.
• Further research into cross-population adaptation techniques is necessary for equitable AI in neuroimaging.