Spatial uncertainty associated with stereotaxic coordinates of neuroimaging results

Introduction: Results from functional neuroimaging studies are usually reported as coordinates for local maxima. Consequently, comparison across studies, either in discussions or quantitative meta-analyses, will is almost exclusively be based on those coordinates. The spatial uncertainty associated with stereotaxic coordinates, however, is yet largely unknown.

Materials & Methods: Empirical assessment of the spatial uncertainty associated with local maxima was performed based on fMRI data from 21 subjects imaging a simple motor task. Each subject was preprocessed and analysed 9 times, using different approaches for registration into MNI space. The maxima coordinates representing 16 functionally defined regions were identified in each individual analysis and the ensuing 9 group analyses. Between-subject and between-template variability was quantified by the mean Euclidean distance (ED) between corresponding maxima and parameterised by the full width of half maximum FWHM of a Gaussian variance model.

Results: Between-subject variability was comparable across normalisation procedures (11.0–12.1mm), and ranged from 7.6mm (caudate nucleus) to 17.6mm (prefrontal cortex). The average ED (across areas and normalisations) was 11.6mm. Converting into a Gaussian displacement and scaling by sample size yields an uncertainty of 17.1/sqrt(Nsubjects) mm FWHM. The EDs between coordinates for a particular area as obtained from the 9 group analyses (differing only by normalisation) ranged from 4.3mm to 8.4mm with a mean of 5.7mm. This corresponds to a Gaussian FWHM of 8.4mm for between-template variance.

Discussion: Combining these empirical estimates for between-subject and between-template variance indicates that the spatial uncertainty associated with local maxima reported in functional neuroimaging experiments is substantial. Assuming a Gaussian form of the uncertainty distribution, sample sizes common in functional neuroimaging translate into spreads of FWHMs between 10.8mm (8 subjects) and 9.0mm (30 subjects). Based on these estimates, we would conclude, that, while stereotaxic coordinates for local maxima represent valid point-estimators of the location of an effect, their associated uncertainty has to be considered before statements about a convergence or dissociation between results from different studiescan be made. Explicitly accommodating these variances on the other hand, should will enhance quantitative meta-analysis of neuroimaging data.