Estimating the Total Variance Explained by Whole-Brain Imaging for Zero-inflated Outcomes

Area of Science:

• Neuroscience
• Statistics
• Behavioral Science

Background:

• Zero-inflated outcomes are prevalent in behavioral data, complicating brain-behavior relationship assessments.
• Existing statistical models struggle to characterize whole-brain imaging signals for these distributions.

Approach:

• Developed a novel variational Bayes algorithm for modeling zero-inflated outcomes using whole-brain imaging data.
• Introduced the zero-inflated variance (ZIV) estimator to quantify variance explained and non-null effects.

Key Points:

• ZIV demonstrated superior performance over other linear prediction algorithms in simulations.
• Analysis of the Adolescent Brain Cognitive Development (ABCD) Study data revealed greater variance explained by whole-brain imaging for externalizing behaviors.
• The ZIV estimator effectively localizes neurocircuitry associated with human behavior, particularly when applied to focal sub-scales.

Conclusions:

• The ZIV estimator provides a robust method for analyzing brain-behavior relationships with zero-inflated data.
• Whole-brain imaging features explain more variance in externalizing behaviors compared to internalizing behaviors.
• This approach enhances the localization of neurobiological correlates of behavior.