Global-Local Transformer for Brain Age Estimation.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Deep learning models for brain age estimation often overlook local details in brain MRI.
• Existing methods primarily rely on global information extraction, limiting accuracy.

Purpose of the Study:

• To develop an advanced deep learning model, the global-local transformer, for precise brain age estimation.
• To improve brain age prediction by integrating both global context and local fine-grained details from brain MRI.

Main Methods:

• A novel global-local transformer architecture processing whole images and local patches.
• Utilizing an attention mechanism for fusing global and local feature information.
• Evaluation on 8 diverse datasets comprising 8,379 healthy brain MRIs.

Main Results:

• Achieved a mean absolute error of 2.70 years in age estimation.
• Increased the correlation coefficient between estimated and chronological age to 0.9853.
• Identified informative local regions crucial for accurate brain age prediction.

Conclusions:

• The global-local transformer significantly enhances brain age estimation accuracy and reliability.
• This model offers a more comprehensive approach by analyzing both global and local brain MRI features.
• Provides insights into regional importance for brain age prediction, aiding further research.