A variance components model for statistical inference on functional connectivity networks.

Area of Science:

• Neuroimaging
• Statistical modeling
• Brain connectivity

Background:

• Functional magnetic resonance imaging (fMRI) data exhibits temporal autocorrelation and subject heterogeneity.
• Existing statistical methods for functional connectivity networks often ignore these factors, potentially reducing statistical power.

Purpose of the Study:

• To propose a novel variance components linear modeling framework for statistical inference on functional connectivity networks.
• To account for temporal autocorrelation and subject heterogeneity in fMRI data.
• To characterize resting-state networks related to reading processes by comparing typical and reading-impaired adults.

Main Methods:

• Developed a nonparametric, subject-specific estimation of autocorrelation structure.
• Employed iterative least squares to estimate variance due to subject heterogeneity.
• Applied the model to resting-state fMRI data from typical and reading-impaired young adults.
• Compared the proposed model against methods that do not account for autocorrelation or heterogeneity using simulated data.

Main Results:

• The proposed model provides more powerful statistical inference for functional connectivity networks.
• Accounting for temporal autocorrelation and subject heterogeneity enhances the detection of significant connections.
• The study identified specific resting-state networks associated with reading processes.

Conclusions:

• The novel variance components linear modeling framework effectively addresses temporal autocorrelation and subject heterogeneity in fMRI.
• This approach leads to more accurate and powerful statistical inference in functional connectivity analysis.
• The findings offer insights into the neural basis of reading by characterizing relevant resting-state brain networks.