Probabilistic thresholding of functional connectomes: Application to schizophrenia

Area of Science:

• Neuroscience
• Graph Theory
• Medical Imaging

Background:

• Functional connectomes are typically analyzed as sparse graphs derived from neurophysiological signal cross-correlations.
• Common thresholding methods (absolute weight or edge density) risk including false positives or excluding true positives.
• Existing methods may not adequately control for statistical errors in graph construction.

Purpose of the Study:

• To introduce and apply novel wavelet-based methods for constructing probabilistically thresholded functional brain graphs.
• To investigate differences in functional connectomes between patients with schizophrenia and healthy controls using these new methods.
• To evaluate the impact of probabilistic thresholding on graph properties and cross-participant consistency.

Main Methods:

• Application of wavelet-based methods for probabilistic graph construction with type I error control.
• Analysis of resting-state functional MRI (fMRI) data from 56 schizophrenia patients and 71 healthy controls.
• Comparison of connectomes thresholded by edge-specific P-values versus traditional correlation-based density thresholding.

Main Results:

• Schizophrenia patients exhibited more "dysconnected" functional connectomes (lower edge density, more disconnected components) compared to controls.
• Many participants' connectomes could not reach conventional densities (5-30%) when controlling for type I error.
• Probabilistically thresholded connectomes demonstrated reduced randomness and enhanced consistency across participants compared to correlation-based methods.

Conclusions:

• Probabilistic thresholding offers a more statistically rigorous approach to functional connectome analysis.
• Traditional density-based thresholding may introduce artifacts, potentially explaining prior findings of topological randomization in schizophrenia.
• The developed methods improve the reliability and interpretability of graph theory applications in neuroimaging, particularly with heterogeneous data.