The Constrained Network-Based Statistic: A New Level of Inference for Neuroimaging

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Network Analysis

Background:

• Neuroimaging research requires robust inferential procedures for brain network organization analysis.
• Current network-based statistic (NBS) methods leverage local dependencies but overlook shared membership in large-scale brain networks.
• Accurate inference is crucial for advancing reproducible discovery in neuroscience.

Purpose of the Study:

• To introduce a novel inferential method, the Constrained Network-Based Statistic (cNBS), that pools information within predefined large-scale brain networks.
• To evaluate the sensitivity and specificity of cNBS compared to standard NBS and threshold-free NBS.
• To determine if cNBS enhances the validity of inference in neuroimaging studies.

Main Methods:

• Proposed the Constrained Network-Based Statistic (cNBS) for neuroimaging data analysis.
• Evaluated cNBS, standard NBS, and threshold-free NBS using resampling of task-based fMRI data from the Human Connectome Project.
• Assessed sensitivity, specificity, and family-wise error rate (FWER) control for all methods.

Main Results:

• Constrained Network-Based Statistic (cNBS) demonstrated higher sensitivity to effect sizes below medium, capturing the majority of ground truth effects.
• Threshold-free NBS was more sensitive to larger effect sizes.
• Observed grouping of effects within large-scale networks supported the relevance of the cNBS approach; all methods maintained intended FWER control.

Conclusions:

• Constrained Network-Based Statistic (cNBS) represents a promising advancement in neuroimaging inference by incorporating large-scale network structure.
• cNBS enhances sensitivity for detecting subtle effects, contributing to more valid and reproducible scientific discovery in neuroscience.
• The findings underscore the importance of considering network membership in statistical inference for brain imaging data.