Modified models and simulations for estimating dynamic functional connectivity in resting state functional magnetic
Maryam Behboudi1, Rahman Farnoosh2
1Department of Statistics, Science and Research Branch, Islamic Azad University, Tehran, Iran.
Statistics in Medicine
|February 28, 2020
Summary
This study introduces a novel copula-based method for simulating blood-oxygen-level dependent (BOLD) signals and proposes the KDCC model for dynamic functional connectivity (dFC) estimation, outperforming existing methods for both normal and nonnormal data.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Statistical Modeling
Background:
- Understanding brain networks via dynamic functional connectivity (dFC) is crucial.
- Existing blood-oxygen-level dependent (BOLD) signal simulation and dFC methods require improvement, especially for nonnormal data observed in fMRI.
- Current methods may not optimally detect the dynamic nature of functional connectivity (FC).
Purpose of the Study:
- To develop a copula-based method for simulating multivariate nonnormal BOLD signals.
- To introduce novel multivariate volatility models, KEWMA and KDCC, for dFC estimation.
- To evaluate and compare the performance of KDCC against existing models for dFC estimation in fMRI data.
Main Methods:
- A copula-based method was used to generate time-varying covariance matrices for simulating nonnormal BOLD signals.
- Two new models, Kendallized Exponentially Weighted Moving Average (KEWMA) and Kendallized Dynamic Conditional Correlation (KDCC), were introduced.
- Statistical tests confirmed bivariate normality in Iranian resting-state fMRI data, and KDCC was employed for dFC estimation.
Main Results:
- The KDCC model demonstrated superior performance in estimating conditional correlation compared to DCC, sDCC, EWMA, and sEWMA models.
- This improved performance was consistent across both multivariate normal and nonnormal data types.
- Novel Portmanteau and rank-based tests were proposed for examining conditional heteroscedasticity.
Conclusions:
- The KDCC model offers a significant advancement for accurately estimating conditional correlation in brain imaging data.
- The proposed copula-based simulation method facilitates robust model performance evaluation under nonnormal conditions.
- The study highlights the effectiveness of KDCC for dFC estimation in resting-state fMRI, particularly for data exhibiting nonnormal characteristics.
