TY - GEN
T1 - Sparse representation of higher-order functional interaction patterns in task-based FMRI data
AU - Zhang, Shu
AU - Li, Xiang
AU - Lv, Jinglei
AU - Jiang, Xi
AU - Zhu, Dajiang
AU - Chen, Hanbo
AU - Zhang, Tuo
AU - Guo, Lei
AU - Liu, Tianming
PY - 2013
Y1 - 2013
N2 - Traditional task-based fMRI activation detection methods, e.g., the widely used general linear model (GLM), assume that the brain's hemodynamic responses follow the block-based or event-related stimulus paradigm. Typically, these activation detections are performed voxel-wise independently, and then are usually followed by statistical corrections. Despite remarkable successes and wide adoption of these methods, it remains largely unknown how functional brain regions interact with each other within specific networks during task performance blocks and in the baseline. In this paper, we present a novel algorithmic pipeline to statistically infer and sparsely represent higher-order functional interaction patterns within the working memory network during task performance and in the baseline. Specifically, a collection of higher-order interactions are inferred via the greedy equivalence search (GES) algorithm for both task and baseline blocks. In the next stage, an effective online dictionary learning algorithm is utilized for sparse representation of the inferred higher-order interaction patterns. Application of this framework on a working memory task-based fMRI data reveals interesting and meaningful distributions of the learned sparse dictionary atoms in task and baseline blocks. In comparison with traditional voxel-wise activation detection and recent pair-wise functional connectivity analysis, our framework offers a new methodology for representation and exploration of higher-order functional activities in the brain.
AB - Traditional task-based fMRI activation detection methods, e.g., the widely used general linear model (GLM), assume that the brain's hemodynamic responses follow the block-based or event-related stimulus paradigm. Typically, these activation detections are performed voxel-wise independently, and then are usually followed by statistical corrections. Despite remarkable successes and wide adoption of these methods, it remains largely unknown how functional brain regions interact with each other within specific networks during task performance blocks and in the baseline. In this paper, we present a novel algorithmic pipeline to statistically infer and sparsely represent higher-order functional interaction patterns within the working memory network during task performance and in the baseline. Specifically, a collection of higher-order interactions are inferred via the greedy equivalence search (GES) algorithm for both task and baseline blocks. In the next stage, an effective online dictionary learning algorithm is utilized for sparse representation of the inferred higher-order interaction patterns. Application of this framework on a working memory task-based fMRI data reveals interesting and meaningful distributions of the learned sparse dictionary atoms in task and baseline blocks. In comparison with traditional voxel-wise activation detection and recent pair-wise functional connectivity analysis, our framework offers a new methodology for representation and exploration of higher-order functional activities in the brain.
KW - dictionary learning
KW - GES
KW - higherorder interaction
KW - sparse coding
KW - task-based fMRI
UR - http://www.scopus.com/inward/record.url?scp=84894607786&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-40760-4_78
DO - 10.1007/978-3-642-40760-4_78
M3 - 会议稿件
C2 - 24505814
AN - SCOPUS:84894607786
SN - 9783642407598
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 626
EP - 634
BT - Medical Image Computing and Computer-Assisted Intervention, MICCAI 2013 - 16th International Conference, Proceedings
T2 - 16th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2013
Y2 - 22 September 2013 through 26 September 2013
ER -