Integrating Multimodal and Longitudinal Neuroimaging Data with Multi-Source Network Representation Learning.

Area of Science:

• Neuroimaging
• Network Neuroscience
• Computational Neuroscience

Background:

• Neuroimaging data is crucial for understanding brain mechanisms but is often costly, limiting sample sizes.
• Multimodal and longitudinal data offer complementary information but are challenging to integrate for brain network analysis.

Purpose of the Study:

• To propose a general fusion framework for multi-source learning of brain networks.
• To develop a method that simultaneously captures network properties from multimodal and longitudinal datasets.
• To enhance the prediction of psychometric scores using integrated neuroimaging data.

Main Methods:

• Proposed a multimodal brain network fusion with longitudinal coupling (MMLC) framework.
• Incorporated three layers of information: cross-sectional similarity, multimodal coupling, and longitudinal consistency.
• Utilized joint factorization of multimodal networks with rotation-based constraints and consensus factorization.

Main Results:

• Demonstrated MMLC's superior performance in predicting psychometric scores compared to state-of-the-art algorithms on two public datasets.
• Identified significant brain regions consistent with existing literature.
• Showcased the potential to boost statistical power in neuroimaging research.

Conclusions:

• The MMLC framework effectively integrates multimodal and longitudinal neuroimaging data for robust brain network representation.
• This approach offers a promising avenue for discovering neuroimaging network biomarkers for psychometric prediction.
• Highlights the synergistic potential of combining diverse data sources in neuroscience research.