Structural Connectome Analysis using a Graph-based Deep Model for Prediction of Non-Imaging Variables.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Brain connectivity analysis is crucial for understanding neurological health.
• Predicting age and cognitive status from brain structure can aid in early diagnosis and personalized medicine.

Purpose of the Study:

• To develop and validate a novel machine learning model for predicting age and Mini-Mental State Examination (MMSE) scores using structural brain connectivity.
• To introduce a new graph convolutional network (GCN) architecture with a connectivity attention module for enhanced feature representation.

Main Methods:

• Utilized diffusion magnetic resonance imaging (dMRI) to derive structural brain connectivity graphs.
• Developed a parallel GCN model with multiple branches to disentangle node and graph features.
• Incorporated a connectivity attention module for graph-level attention and embedding representation.

Main Results:

• The proposed GCN model demonstrated superior performance in age prediction compared to classical and deep learning methods.
• Experiments on PREVENT-AD and OASIS3 datasets validated the model's effectiveness.
• Ablation studies confirmed the contribution of individual model components.

Conclusions:

• The novel GCN model offers a promising approach for predicting age and cognitive function from brain connectivity.
• The connectivity attention module enhances the model's ability to learn informative graph embeddings.
• This work contributes to a better understanding of brain variations in health and disease.