Analysis of FMRI data with drift: modified general linear model and Bayesian estimator.

Area of Science:

• Neuroimaging
• Signal Processing
• Statistical Modeling

Background:

• Slowly varying drift is a significant challenge in functional magnetic resonance imaging (fMRI) data analysis.
• Noise in fMRI data often exhibits long memory fractional noise characteristics.
• Drift in fMRI data is typically associated with large-scale wavelet components.

Purpose of the Study:

• To develop a robust method for analyzing fMRI data by addressing the issue of slow drift.
• To improve the estimation of activation parameters in the presence of complex noise structures.
• To propose an effective model selection criterion for drift modeling in fMRI.

Main Methods:

• A modified General Linear Model (GLM) is examined in the wavelet domain within a Bayesian framework.
• The modified model estimates activation parameters across different scales of wavelet decomposition.
• A novel model selection criterion is proposed for drift modeling based on the modified GLM results.

Main Results:

• The proposed Bayesian estimator accurately captures the noise structure in fMRI data.
• Robust estimation of GLM parameters is achieved through the accurate noise modeling.
• The developed model selection criterion effectively removes drift from fMRI data.
• Simulated and real fMRI data analyses validate the proposed methods.

Conclusions:

• The Bayesian approach effectively handles long memory fractional noise and slow drift in fMRI.
• The proposed method enhances the reliability and accuracy of fMRI data analysis.
• The model selection criterion offers an efficient way to mitigate drift artifacts.