An improvement on local FDR analysis applied to functional MRI data

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biostatistics

Background:

• Effective brain connectivity discovery is a key area of neuroscience research.
• Vector autoregressive (VAR) models are used for exploratory structural modeling of brain networks.
• Accurate identification of non-zero VAR coefficients is crucial for interpreting effective connectivity.

Purpose of the Study:

• To propose an improved False Discovery Rate (FDR) analysis procedure tailored for functional MRI (fMRI) data.
• To address challenges with non-symmetric VAR coefficient estimates common in fMRI data.
• To enhance the correct discovery rate in effective connectivity analysis.

Main Methods:

• Developed a two-step procedure to estimate the null distribution of VAR coefficient estimates.
• The method accommodates non-symmetric distributions and non-zero modes in coefficient estimates.
• Validated through theoretical arguments, simulations (ROC curves), and real fMRI data analysis.

Main Results:

• The proposed FDR analysis method demonstrated superior performance compared to standard methods in simulation experiments.
• Outperformed existing methods in terms of true positive rate and performance consistency across various sample sizes and VAR model dimensions.
• Successfully applied to real fMRI data, yielding interpretable results in cognitive neuroscience.

Conclusions:

• The novel FDR analysis procedure is more suitable for fMRI data analyzed with VAR models, especially when coefficient estimates are asymmetric.
• The method offers improved accuracy and reliability for discovering effective brain connectivity.
• Provides a valuable tool for advancing computational neuroscience and understanding brain function.