The critical regularization value: Incorporating spatial smoothness to enhance signal detection in highly noisy fMRI data

Comparing serially acquired fMRI scans is a typical way to detect functional brain changes in different conditions. However, this approach introduces additional variation on physical and physiological conditions, which results in substantial noise. To improve sensitivity and accuracy of signal detection in such highly noisy fMRI data, potentially important information should be incorporated. Here we propose a new significance indicator, the critical regularization value (CR-value), which detects significantly changed voxels by taking both the magnitude of the voxel-wise signal variation and spatial smoothness into account. The CR-value allows voxels that survive in a stronger sparse constraint to be considered as more significant. We demonstrate our method using a simulation dataset and a real fMRI dataset collected from the previous study. The results show that CR-value more accurately detects the true activation than GLM P-value, Posterior Probability Maps (PPM) and the Threshold Free Cluster Enhancement (TFCE) in noisy datasets.