Constrained Canonical Correlation Analysis for fMRI Analysis Utilizing Experimental Paradigm Information.

Data-driven methods have been successfully used in functional magnetic resonance imaging (fMRI) data analysis, utilizing no explicit prior information. However, in analysis of task fMRI, incorporating prior information about paradigms would be useful to improve the power in detecting desired activations. In this paper, we incorporated the experimental paradigm information into canonical correlation analysis (CCA) model and proposed a temporal constrained CCA approach. Comparing to noise and artifact signal, the response BOLD signal from activated regions changes only after stimulus begin. By incorporating the difference before and after stimulus, the spatial patterns which respond well to stimulus-occurrence become manifest, and the activations can be detected more accurately. Comparisons on simulated data indicated that incorporating prior information about paradigm can improve the accuracy of CCA on activation detection. The proposed method obtained more accurate and robust results than conventional CCA and showed improved power in activation detection than pure data-driven methods.