MDFNet: a multi-dimensional feature fusion model based on structural magnetic resonance imaging representations for...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Brain age estimation is crucial for understanding aging and neurodegenerative diseases.
• Current methods often rely on limited data or complex imaging techniques.

Purpose of the Study:

• To develop a unified Multi-Dimensional Feature Fusion Network (MDFNet) for enhanced brain age estimation.
• To integrate diverse structural MRI features including whole brain, gray matter volume, and brain connectivity networks.

Main Methods:

• MDFNet integrates whole-brain, tissue-level, node-based graph convolution, edge-based graph path convolution, and demographic data channels.
• Validated on 1872 healthy subjects across four datasets and applied to an Alzheimer's Disease cohort.
• Included interpretability analysis and normative modeling for robust validation.

Main Results:

• MDFNet achieved superior performance with a Mean Absolute Error of 4.396 years and a Pearson Correlation Coefficient of 0.912.
• Significantly greater brain age gap observed in Alzheimer's Disease patients compared to healthy controls.
• Interpretability analysis confirmed the model's reliability at group and individual levels.

Conclusions:

• The MDFNet effectively enhances brain age estimation using structural MRI.
• Multi-dimensional feature integration is key to improving accuracy and reliability in brain age prediction.