Mingwu Jin1, Rajesh Nandy2, Dietmar Cordes1
1Radiology, University of Colorado Denver, Aurora, CO, USA; 2Biostatistics and Psychology, UCLA, Los Angeles, CA, USA
There are no constraints of the spatial weights employed in conventional canonical correlation analysis (CCA) leading to model overfitting and a decrease of specificity. We propose two different constrained CCA (cCCA) methods to improve the detection power of fMRI data analysis and compare results with the GLM method without and with fixed Gaussian spatial smoothing. Quantitative results from pseudo-real data as well as qualitative results from real data show that both novel cCCA methods can detect activations more accurately for noisy fMRI data without losing specificity.