Reducing interscanner variability of activation in a multicenter fMRI study: controlling for signal-to-fluctuation-noise-ratio (SFNR) differences.

Abstract Variation in scanner performance will lead to variation in activation patterns in multicenter fMRI studies. The purpose of this investigation was to evaluate the effect of statistically covarying for scanner differences in signal-to-fluctuation-noise-ratio (SFNR) on reducing scanner differences in activation effect size as part of a multicenter fMRI project (FIRST BIRN). For SFNR, “signal” is typically the mean intensity over time and “fluctuation noise” is the temporal standard deviation. Five subjects were sent to 9 centers (10 scanners) and scanned on two consecutive days using a sensorimotor fMRI protocol. High-field (4 T and 3 T) and low-field (1.5 T) scanners from three vendors (GE, Siemens and Picker) were included. The effect size for the detection of neural activation during a sensorimotor task was evaluated as the percent of temporal variance accounted for by our model (percent of variance accounted for, or PVAF). Marked scanner effects were noted for both PVAF as well as SFNR. After covariate adjustment with one of several measures of SFNR, there were dramatic reductions in scanner-to-scanner variations in activation effect size. Variance components analyses revealed 75%–81% reductions in variance due to scanner with this method. Thus, controlling for scanner variation in SFNR may be an effective method to homogenize activation effect sizes in multicenter studies.