Sparse temporally dynamic resting-state functional connectivity networks for early MCI identification

Area of Science:

• Neuroimaging
• Network Neuroscience
• Machine Learning in Medicine

Background:

• Conventional resting-state fMRI (R-fMRI) assumes static functional connectivity, potentially missing transient neural interactions crucial for understanding brain function.
• Dynamic changes in neural interactions, reflected in temporal network topology and correlation strength, may reveal subtle disruptions indicative of disease pathologies.

Purpose of the Study:

• To leverage dynamic temporal network properties derived from R-fMRI for enhanced classification performance in disease identification.
• To develop a framework that incorporates temporally dynamic R-fMRI information for improved diagnostic accuracy, particularly for mild cognitive impairment.

Main Methods:

• Employed a sliding window approach to generate overlapping R-fMRI sub-series and compute temporal networks representing neural interactions.
• Utilized a fused sparse learning algorithm to jointly estimate temporal networks, preserving temporal smoothness and encouraging similarity in topology and correlation strength.
• Designed a disease identification framework based on the estimated dynamic temporal networks.

Main Results:

• Group-level analysis revealed significant differences in network properties between controls and patients based on dynamic temporal connectivity.
• The proposed framework demonstrated improved classification accuracy by incorporating temporally dynamic R-fMRI scan information.
• The fused sparse learning approach effectively preserved temporal smoothness in the estimated networks.

Conclusions:

• Dynamic functional connectivity analysis in R-fMRI offers valuable insights beyond static connectivity assumptions.
• Incorporating temporally dynamic network properties significantly enhances the diagnostic accuracy for neurological conditions like mild cognitive impairment.
• The developed fused sparse learning method provides a robust approach for estimating dynamic brain networks from R-fMRI data.