A Graph Gaussian Embedding Method for Predicting Alzheimer's Disease Progression With MEG Brain Networks

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early detection of Alzheimer's disease (AD) is crucial for intervention, but subtle changes in brain networks often precede clinical symptoms.
• Understanding the pathological cascade of AD requires characterizing functional brain network alterations.
• Current methods may struggle to capture the heterogeneity and subtlety of these network changes.

Purpose of the Study:

• To develop a novel deep learning method for characterizing functional brain networks in Alzheimer's disease.
• To enable early prediction of AD and progression from mild cognitive impairment (MCI).
• To identify specific brain regions exhibiting network alterations related to MCI and AD.

Main Methods:

• Development of the multiple graph Gaussian embedding model (MG2G), a deep learning approach.
• Mapping high-dimensional resting-state brain networks into a low-dimensional latent space.
• Utilizing latent distribution-based embeddings for quantitative characterization of brain connectivity patterns.

Main Results:

• MG2G successfully detected intrinsic latent dimensionality in magnetoencephalography (MEG) brain networks.
• The model demonstrated capability in predicting MCI to AD progression.
• Identification of brain regions with network alterations associated with MCI was achieved.

Conclusions:

• The MG2G model provides a powerful tool for quantitative characterization of brain connectivity.
• This method facilitates early diagnosis and prediction of Alzheimer's disease progression.
• MG2G aids in identifying specific neurobiological markers for MCI and AD.