Publicado 4 números por año
ISSN En Línea: 2162-3511
Fast Constrained Canonical Correlation Analysis for FMRI
SINOPSIS
Local canonical correlation analysis (CCA) methods can increase the detection power in functional magnetic resonance imaging (fMRI) data analysis. Spatial constraints are usually needed to overcome model overfitting and loss of spatial specificity. The resulting constrained CCA (cCCA) method suffers from a marked increase of computation time, even though a computationally efficient branch-and-bound method (cCCA-BB) was developed that avoids an exhaustive search. In this study, a fast implementation of cCCA using a novel region growing strategy (cCCA-RG) is proposed. This method can significantly shorten the computational time without compromising accuracy. Using simulated and real fMRI data, the performance (accuracy and speed) between cCCA-RG, cCCA-BB and the conventional general linear model (GLM) methods (GLM with or without fixed Gaussian smoothing) is compared. Results demonstrate that cCCA-RG achieves similar detection power to cCCA-BB, and both cCCA methods have a higher detection power than any of the GLM methods. The cCCA-RG method is at least 12 times faster than the cCCA-BB method. The proposed fast implementation using cCCA-RG enables a broad application of cCCA for fMRI data analysis.
-
Zhuang Xiaowei, Yang Zhengshi, Curran Tim, Byrd Richard, Nandy Rajesh, Cordes Dietmar, A family of locally constrained CCA models for detecting activation patterns in fMRI, NeuroImage, 149, 2017. Crossref