Smoothing and cluster thresholding for cortical surface-based group analysis of fMRI data

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Analysis

Background:

• Cortical surface-based analysis offers advantages over 3D volumetric methods for fMRI.
• Adapting statistical methods from 3D to surface analysis is crucial for advanced neuroimaging.
• Accurate multiple comparisons correction is essential for reliable fMRI findings.

Purpose of the Study:

• To implement a surface-based cluster size exclusion method for fMRI multiple comparisons correction.
• To develop a novel method for generating regions of interest on the cortical surface.
• To empirically validate theoretical models of surface smoothing in neuroimaging.

Main Methods:

• Implemented a surface-based cluster size exclusion method within the FreeSurfer software.
• Developed a sliding threshold cluster exclusion and growth method for region of interest generation.
• Estimated cluster size limits using random field theory and Monte Carlo simulations.
• Assessed intrinsic data smoothness using simulated noise fMRI data.
• Empirically determined Gaussian kernel width for iterative surface smoothing.

Main Results:

• A surface-based cluster size exclusion method was successfully implemented.
• A new region of interest generation technique was developed.
• Random field theory and Monte Carlo simulations provided estimates for cluster size limits.
• Simulated noise data analysis estimated intrinsic smoothness of group analysis statistics.
• Significant disparities were observed between predicted and actual Gaussian kernel widths in surface smoothing.

Conclusions:

• Surface-based analysis provides a robust framework for fMRI statistical methods.
• The developed methods enhance the accuracy of multiple comparisons correction and region of interest definition.
• Empirical validation revealed limitations in current theoretical models of surface smoothing, necessitating further research.