Intersession variability is an important consideration in power calculations for the design of FMRI experiments. It is also a critical issue for the interpretation of studies that allow for only single observations, e.g., in many clinical applications of FMRI. Here we have provided quantitative data confirming that intersession variability in FMRI is not large relative to within-session variability. We also emphasise that inter-session variability should not be judged by apparent variability in thresholded activation maps.
There are several mechanisms by which intersession variability can be minimised. While considerable attention has been paid in the past to hardware and experimental design factors, we have shown here that additional benefits can come with optimisation of analysis methodology, as analysis methods add extra variance to the true intersession variance, causing apparent increase in intersession variance. It was found that with respect both to spatial alignment processing and time series statistics, FSL1.3 induced less error than SPM99b, i.e., was more efficient with respect to higher-level activation estimation.