A new Graph Gaussian embedding method for analyzing the effects of cognitive training.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Early identification of cognitive impairment is crucial for Alzheimer's disease (AD) diagnosis.
• Quantifying functional brain connectomic changes in amnestic mild cognitive impairment (aMCI) during interventions is challenging due to complex network features.

Purpose of the Study:

• To develop and validate a quantitative method for analyzing functional brain networks from fMRI data.
• To identify potential biomarkers for monitoring cognitive alterations in aMCI patients undergoing non-pharmacological interventions.

Main Methods:

• Utilized the multi-graph unsupervised Gaussian embedding (MG2G) method, a neural network-based model.
• Learned low-dimensional Gaussian distributions from high-dimensional sparse functional brain networks.
• Employed Wasserstein distance to measure probabilistic changes in brain networks.

Main Results:

• Identified significant variations in brain regions within the default mode network, somatosensory/somatomotor hand, fronto-parietal task control, memory retrieval, and visual/dorsal attention systems during non-pharmacological training.
• The MG2G method demonstrated the ability to capture subtle, individual-specific changes before and after short-term interventions.
• Discovered potential distinct biomarkers for fine-grained monitoring of cognitive alterations in aMCI.

Conclusions:

• The MG2G method provides a quantitative approach to analyze functional brain networks and monitor cognitive changes.
• The identified brain region variations may serve as valuable biomarkers for aMCI and early AD detection.
• The MG2G method has broader applicability in studying various neurological disorders like Parkinson's disease and traumatic brain injury (TBI).