Improved state change estimation in dynamic functional connectivity using hidden semi-Markov models.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Functional brain network (FBN) analysis is rapidly advancing.
• Static functional connectivity (FC) analyses are common, but time-varying FC is gaining interest.
• Hidden Markov models (HMMs) identify brain states but assume a geometric distribution for time spent in each state, which can be inaccurate.

Purpose of the Study:

• To propose and evaluate a hidden semi-Markov model (HSMM) for inferring time-varying brain networks from fMRI data.
• To address the limitation of inaccurate sojourn time estimation in traditional HMMs.
• To improve the modeling of brain state durations in functional connectivity analyses.

Main Methods:

• Developed a hidden semi-Markov model (HSMM) approach for fMRI data analysis.
• Applied HSMMs to estimate brain states and their associated network graphs.
• Explicitly modeled the sojourn distribution (time spent in a state) for each brain state.
• Validated the approach using simulation studies and real fMRI data (task-based and resting-state).

Main Results:

• HSMMs provide a more accurate estimation of sojourn times compared to standard HMMs.
• The choice of model significantly impacts the estimation of time spent in brain states.
• Analysis revealed distinct brain network dynamics in both task-based and resting-state fMRI data.
• The proposed HSMM approach successfully identified probable series of network states and associated graphs.

Conclusions:

• Accurate modeling of sojourn distributions is critical for understanding time-varying functional brain networks.
• HSMMs offer a more robust framework for analyzing dynamic brain connectivity from fMRI.
• This method has potential for advancing the study of healthy and diseased brain mechanisms through improved dynamic network analysis.