Modeling dynamic characteristics of brain functional connectivity networks using resting-state functional MRI

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging Analysis

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) offers insights into brain dynamics for disease identification.
• Existing methods often overlook the temporal evolution of global brain network structures and separate feature extraction from classifier training.

Purpose of the Study:

• To develop an integrated framework for network-based brain disease identification using rs-fMRI time-series data.
• To address limitations in previous studies by modeling the evolution of global network structures over time.

Main Methods:

• Proposed a Temporal Dynamics Learning (TDL) method integrating network feature extraction and classifier training.
• Utilized overlapping sliding windows to segment rs-fMRI time series and construct longitudinally ordered functional connectivity networks.
• Introduced a group-fused Lasso regularizer to model temporal evolution patterns and an ℓ1-norm regularizer for network architecture, solved using the Alternating Direction Method of Multipliers (ADMM).

Main Results:

• The TDL model explicitly models evolving connectivity patterns of global brain networks over time.
• It captures unique characteristics of networks defined at each segment.
• Achieved superior results in identifying autism spectrum disorder (ASD) on three real-world datasets compared to state-of-the-art methods.

Conclusions:

• The TDL method provides an effective approach for automated brain disease identification using rs-fMRI.
• Integrating temporal dynamics and network analysis enhances diagnostic accuracy for neurological disorders like ASD.
• This unified framework advances the application of dynamic network analysis in clinical neuroscience.