Factorized binary search: Change point detection in the network structure of multivariate high-dimensional time...

Area of Science:

• Neuroscience
• Data Science
• Computational Biology

Background:

• Understanding dynamic brain connectivity from functional magnetic resonance imaging (fMRI) time series data is crucial in neuroscience.
• Existing models often struggle with the high dimensionality and complex dynamics of whole-brain fMRI data.
• Accurate change point detection and network estimation are needed to analyze these intricate brain mechanisms.

Purpose of the Study:

• To introduce a novel method, FaBiSearch, for accurate change point detection in the network structure of high-dimensional fMRI data.
• To develop a new network estimation technique for analyzing brain data between detected change points.
• To investigate dynamic functional connectivity in resting-state and task-based fMRI experiments.

Main Methods:

• Developed FaBiSearch, a method combining non-negative matrix factorization (NMF) and a novel binary search algorithm for multiple change point detection.
• Proposed a new network estimation method for multivariate time series data between change points.
• Applied these methods to resting-state and task-based fMRI datasets, including a reading task ('Harry Potter').

Main Results:

• FaBiSearch effectively identifies multiple change points in the network structure of high-dimensional fMRI data.
• Analysis of resting-state data revealed test-retest behavior of dynamic functional connectivity.
• Task-based fMRI analysis explored network dynamics during reading, correlating change points with narrative events and identifying dynamic hub nodes.

Conclusions:

• FaBiSearch provides a robust framework for understanding large-scale brain dynamics and network changes.
• The methods offer insights into both resting-state and task-evoked brain activity patterns.
• The FaBiSearch R package and associated experiments are publicly available for further research.