An Explainable Unified Framework of Spatio-Temporal Coupling Learning With Application to Dynamic Brain Functional
IEEE Transactions on Medical Imaging
|September 25, 2024
Summary
This study introduces an explainable deep learning framework to model spatio-temporal coupling in brain data, revealing developmental changes in functional connectivity. The model captures dynamic brain activity patterns and enhances understanding of neural mechanisms.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Machine Learning
Background:
- Time-series data like fMRI and MEG contain crucial spatio-temporal information.
- Existing deep learning models often fail to capture intrinsic spatio-temporal coupling and lack explainability.
Purpose of the Study:
- To develop an explainable deep learning framework for modeling spatio-temporal coupling in neuroimaging data.
- To improve the understanding of biological mechanisms by analyzing dynamic functional connectivity (dFC).
Main Methods:
- A deep learning network was constructed based on spatio-temporal correlation.
- The framework integrates time-varying coupled relationships and explores spatio-temporal evolution.
- The model was applied to analyze brain dynamic functional connectivity (dFC).
Main Results:
- The framework effectively captures variations in dFC during brain development and resting-state evolution.
- Two distinct developmental functional connectivity patterns were identified: decreased emotional regulation connectivity and increased cognitive activity connectivity.
- Cyclic fluctuations in resting-state dFC were observed in children and young adults.
Conclusions:
- The proposed framework offers enhanced explainability for spatio-temporal analysis in neuroscience.
- It provides novel insights into brain development and resting-state dynamics through dFC analysis.
- The findings highlight significant shifts in functional brain networks during development.
