As noted by Weisskoff (1996) in reference to fMRI experiments, ``MR scanners can have quite variable stability even when the `normal' stability measures are well within the instrumental norms.'' Consequently, measurement of instrument variability and methods which correct data for the systematic components of that variability are of considerable importance. Equally important in fMRI are methods which estimate and correct for the systematic component of variability due to the human subject. Here we present a visual tool to help assess the quality of existing (and to-be-developed) data correction methods.

The conversion of raw fMRI k-space data into images for subsequent statistical analysis typically encompasses many processing steps. For example, for echo planar (EPI) data acquisitions at 3.0 Tesla we routinely perform at least six distinct processing steps. Each of these steps is performed in the belief that it removes some systematic component of the underlying variability in the data. Each has associated with it some tuning parameters which have been set to predetermined values; the chosen values are believed to be the best possible choices. And, the order of the processing steps has been predetermined; again, the particular order is believed to be the best of the available choices.

The need for each processing step, the particular values of the tuning parameters for each step, and the order of the processing steps, can be assessed by a number of methods. One particularly intuitive choice is to look at the reduction in the variance within each voxel due to each processing step. The natural way to look at these variance reductions is in a spatial map. Because of the natural analogy to the analysis of variance in an experimental design we refer to these maps as a Visual ANOVA.

Back to Workshop Schedule