Estimating the total variance explained by whole-brain imaging for zero-inflated outcomes

Area of Science:

• Neuroimaging
• Statistical Modeling
• Behavioral Science

Background:

• Statistical models often fail to capture the distributed nature of whole-brain imaging signals for predicting neurobehavioral phenotypes.
• Neurobehavioral data, like questionnaire responses, frequently exhibit zero-inflated and highly skewed distributions, complicating analysis.
• Existing methods struggle to adequately model the total signal from whole-brain imaging features in the presence of such data characteristics.

Purpose of the Study:

• To develop a novel statistical approach for characterizing the total signal from whole-brain imaging features in the context of zero-inflated outcomes.
• To introduce a Variational Bayes algorithm designed to handle the complexities of neuroimaging data and skewed behavioral measures.
• To enhance the analysis of brain-behavior relationships by providing a method that accounts for zero-inflated data.

Main Methods:

• Development of a novel Variational Bayes algorithm, termed the zero-inflated variance (ZIV) estimator.
• The ZIV estimator quantifies the fraction of variance explained (FVE) and the proportion of non-null effects (PNN).
• Application and simulation studies comparing ZIV performance against traditional linear models using large-scale neuroimaging datasets like the ABCD Study.

Main Results:

• The ZIV estimator demonstrated superior performance compared to other linear models in simulation studies.
• Whole-brain imaging features explained a larger fraction of variance (FVE) for externalizing behaviors than for internalizing behaviors in the ABCD Study data.
• The ZIV estimator successfully identified key neurocircuitry associated with specific neurobehavioral traits by focusing on features contributing to the proportion of non-null effects (PNN).

Conclusions:

• The zero-inflated variance (ZIV) estimator is the first specialized method for analyzing zero-inflated neuroimaging data.
• This novel method improves the ability to capture the total signal from whole-brain imaging features for predicting complex behavioral outcomes.
• The ZIV estimator holds significant promise for advancing future research on brain-behavior relationships and understanding neurobehavioral disorders.