Visualizing Variance with Percent Change Threshold

This report proposes a standard method to characterize variability and sensitivity of functional neuroimaging experiments. Most functional neuroimaging analyses result in statistic images of signal-to-noise. However, it is essential to examine both signal and noise individually, to characterize sensitivity and to identify artifacts. Confidence intervals simultaneous convey signal and noise but are poorly suited to imaging. We describe an image which conveys the same information as a confidence interval when used with an image of estimated signal. This image is the half-length of a two-sided confidence interval, and is most easily seen as the minimum change required to obtain a significant activation at level (either corrected or uncorrected). If the estimated signal has units of percent change, then we call this image a level percent change threshold (PCT). We demonstrate the method on two fMRI datasets, showing that PCT can vary by an order of magnitude across the brain.