Reliability of a cortical surface-based analysis with subcortical regression in the identification of resting-state functional networks.

Abstract

Many methods exist for identifying brain networks in resting-state functional magnetic resonance imaging. During the last decade, there was a growing interest in functional connectivity using surface-based analysis. However, the advantages of this approach against volume-based analysis in a data-driven model are unclear. In this study, we propose an independent component analysis based method to extract the resting-state networks directly on the cortical surface. The components associated with the subcortical regions are identified by multiple linear regressions between the signals in subcortical voxels and independent components time courses. The accuracy and stability of our method were evaluated using resampling statistics calculated on 76 healthy male subjects and compared to those obtained with a similar volume-based approach. Seven of the most representative resting-state networks reported in previous studies were identified and used to compare both approaches. Our findings suggest that surface-based approach combined with subcortical linear regression is more sensitive and reproducible than similar volume-based approach for the extraction of resting-state networks.