A Deep Spatio-Temporal Architecture for Dynamic ECN Analysis with Granger Causality based Causal Discovery.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Network Analysis

Background:

• Causal discovery methods are crucial for understanding brain connectivity but often neglect temporal dynamics and spatio-temporal data.
• Dynamic effective connectivity networks (dECNs) capture evolving directed brain activity and causal influences, aiding in identifying individual differences and understanding brain function.

Purpose of the Study:

• To propose a deep spatio-temporal fusion architecture for dynamic causality modeling in neuroimaging data.
• To effectively incorporate spatio-temporal information for more accurate inference of dynamic causal relationships within the brain.

Main Methods:

• A novel deep spatio-temporal fusion architecture was developed, comprising a dynamic causal deep encoder and a dynamic causal deep decoder.
• The encoder integrates spatio-temporal information for dynamic causality modeling, while the decoder validates the discovered causal relationships.

Main Results:

• The proposed method demonstrated superior performance in inferring dECNs using simulated data.
• Analysis of the Philadelphia Neurodevelopmental Cohort (PNC) data revealed the dynamic evolution of directed brain connectivity.
• Significant differences in dECNs were observed between children and young adults, indicating developmental changes.

Conclusions:

• The developed deep learning architecture accurately models dynamic causality and infers dECNs by integrating spatio-temporal information.
• Brain functional networks transition from undifferentiated to specialized systems during development, supporting enhanced cognitive abilities in young adults.
• This work provides insights into brain network modularization and adaptation throughout human development.