Dietmar Cordes1, Mingwu Jin1, Tim
Curran2, Rajesh Nandy3
1C-TRIC & Dept. of
Radiology, University of Colorado-Denver, Aurora, CO, United States; 2Dept.
of Psychology & Neuroscience, University of Colorado-Boulder, Boulder,
CO, United States; 3Depts. of Biostatistic & Psychology, University
of California-Los Angeles, Los Angeles, CA, United States
An improved method to detect activations in fMRI uses local canonical correlation analysis (CCA) to encompass a group of voxels in a 3x3 pixel neighborhood. It is customary to assign the value of the test statistic to the center voxel of the local neighborhood. However, without spatial constraints such an assignment introduces smoothing artifacts in regions of strong localized activation, which we refer to as bleeding artifacts. To reduce this artifact we propose different spatial constraints in CCA to enforce dominance of the center voxel and introduce a method based on mixture modeling to further reduce this artifact.