Improving the reliability of functional localizers.

A critical assumption underlying the practice of functional localization is that the voxels identified by functional localization are essentially the same as those activated in the main experiment for a particular anatomical area. Violations of this assumption bias the resulting analyses and can dramatically increase the likelihood of both Type I and Type II errors. Here we investigated how the amount of data affects the reliability of a set of common functionally-defined regions-of-interest (fROIs). Four participants were scanned ten times each to functionally localize extrastriate regions sensitive to visually presented words, objects and faces. A within-subject random-effects analysis was used as the "gold standard" for identifying the fROIs and the results were compared to within-subject, fixed-effect analyses typically used for functional localization. By varying the quantity of data included in the analyses, we empirically assessed the amount needed to ensure reliable identification of the fROIs. The results demonstrated that the most consistent fROIs were based on either stringent statistical thresholding (Z>5.0) of large quantities of data or on lenient thresholding (Z>2.3) of a modest amount of data, with both methods yielding 70-80% overlap between the functional localization results and the "gold standard." Stringent statistical thresholds on typical quantities of localizer data led to the poorest reliability (<20% overlap). These findings suggest that the most reliable and cost-efficient method for functional localization involves collecting a relatively small amount of data (~10 min) and using a lenient statistical threshold to identify all voxels in a given region that are sensitive to the process-of-interest.
(Copyright © 2011 Elsevier Inc. All rights reserved.)