Multiband Group Independent Component Analysis: Unveiling Frequency-Dependent Dynamics of Functional Connectivity in Group-Level fMRI Analyses

Group independent component analysis (ICA) offers a method for decomposing fMRI data from multiple subjects into spatially independent maps and associated time courses. Traditionally, group ICA is applied to full-band fMRI data, with typically sampling rates between 0.25-1 Hz. In this paper, we introduce a novel approach known as "multiband group ICA." This method involves the application of bandpass filters to segment the fMRI data into distinct subbands, followed by the application of blind group ICA to all subbands of multisubject fMRI data, followed by back-reconstruction of individual subband information.To assess the feasibility and efficacy of our method, we utilized a bandpass filter to divide the fMRI data into two specific subbands: low frequency and high frequency. Our results not only showcase a substantial distinction in spatial maps and time courses for task-related components but also provide noteworthy insights into network-specific functional network connectivity (FNC) patterns, particularly within the visual network. Notably, motor regions exhibit predominant power in the lower frequency range, while visual regions span both low and high frequencies. Additionally, certain regions, such as the bilateral insular regions, display exclusive engagement in the high-frequency domain. Furthermore, low-frequency task-related components exhibit anticorrelation, contrasting with the strong correlation observed in high-frequency task-related components. This approach enhances our understanding of the frequency-dependent dynamics of functional connectivity in group-level fMRI analyses, offering valuable insights for future studies.