Unsupervised learning of functional network dynamics in resting state fMRI.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Systems Neuroscience

Background:

• Emerging evidence suggests that functional connectivity in resting-state functional magnetic resonance imaging (fMRI) is not static over time.
• Understanding these temporal dynamics is crucial for a comprehensive view of brain function.
• Existing methods may not fully capture the complex, time-varying nature of brain networks.

Purpose of the Study:

• To develop and validate a novel methodology for decoding functional connectivity dynamics in resting-state fMRI.
• To identify a temporal sequence of hidden brain network 'states' for individual subjects.
• To characterize these states by their unique covariance matrices derived from sparse basis networks.

Main Methods:

• Utilized a Hidden Markov Modeling (HMM) framework combined with sparse basis learning for positive definite matrices.
• The model generates covariance matrices from a common set of sparse basis networks, capturing co-variation patterns.
• Distinct hidden states are modeled as variations in the strengths of these underlying basis networks.

Main Results:

• Successfully recovered underlying basis networks and hidden states from simulated fMRI data.
• Application to a normative resting-state fMRI dataset revealed that brain activity comprises combinations of overlapping basis networks.
• Identified distinct temporal states characterized by different dominant basis networks influencing the covariance patterns.

Conclusions:

• The proposed HMM framework with sparse basis learning effectively models and decodes temporal non-stationarity in functional connectivity.
• This method provides insights into the dynamic interplay of brain networks during resting states.
• The findings highlight the dynamic composition of brain activity from fundamental network building blocks.