Deriving a multi-subject functional-connectivity atlas to inform connectome estimation.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging Analysis

Background:

• Estimating functional connectivity from neuroimaging data is crucial for understanding brain diseases and developing biomarkers.
• Challenges include low signal-to-noise ratio and limited data, hindering accurate connectome mapping.
• Increasing availability of public neuroimaging datasets offers opportunities for improved estimation methods.

Purpose of the Study:

• To propose a novel learning scheme for functional connectivity estimation.
• To minimize estimation bias by separating model support and coefficient estimation.
• To develop a computationally efficient method for incorporating new data into connectome mapping.

Main Methods:

• Utilized sparse Gaussian graphical models for functional connectivity estimation.
• Implemented a strategy to separate the estimation of model support from coefficients.
• Applied the method to the Human Connectome Dataset (46 subjects).

Main Results:

• Developed a novel learning scheme for functional connectivity.
• Successfully minimized bias induced by regularization in estimation.
• Demonstrated the physiological relevance of the learned prior as a functional connectivity atlas.

Conclusions:

• The proposed method offers an improved approach to functional connectivity estimation.
• The technique effectively addresses challenges of low signal-to-noise and data paucity.
• The learned functional connectivity atlas has significant physiological relevance and aids in connectome mapping.