Sparse representation of whole-brain fMRI signals for identification of functional networks

Overview of the computational pipeline of identifying functional brain networks via sparse representation of whole-brain fMRI signals. (a) An example of the learned sparse dictionary of 400 functional components (indexed by the horizontal axis). The vertical axis stands for the occurrence frequency of each component in over 40,000 fMRI BOLD signals in a whole brain. The three dictionary components highlighted by yellow, blue and red circles correspond to different functional networks. They are: (b) task related component in which the response well follows the external block-based task paradigm, (c) anti-task related component in which the response well follows the inverse of external block-based task paradigm, and (d) DMN component. In each component (b–d), the corresponding signals (colored curves) accompanied with the task stimulus (white curve) are shown in the top panels. Their spatial distributions are also back-projected onto the volumetric images in the lower panel. Each voxel is color-coded by the reference weight used in the sparse representation. [Display omitted] •A novel computational framework for sparse representation of whole-brain fMRI data.•Inference of a comprehensive collection of functional networks.•Characterize the inferred functional networks in spatial, temporal and frequency domains.•A unified framework for activation detection, de-activation detection, and network identification.•Novel insights into the functional brain architecture. There have been several recent studies that used sparse representation for fMRI signal analysis and activation detection based on the assumption that each voxel’s fMRI signal is linearly composed of sparse components. Previous studies have employed sparse coding to model functional networks in various modalities and scales. These prior contributions inspired the exploration of whether/how sparse representation can be used to identify functional networks in a voxel-wise way and on the whole brain scale. This paper presents a novel, alternative methodology of identifying multiple functional networks via sparse representation of whole-brain task-based fMRI signals. Our basic idea is that all fMRI signals within the whole brain of one subject are aggregated into a big data matrix, which is then factorized into an over-complete dictionary basis matrix and a reference weight matrix via an effective online dictionary learning algorithm. Our extensive experimental results have shown that this novel methodolog