Quantification of inter-subject variability in human brain: a validation framework for probabilistic maps

Probabilistic maps are useful in functional neuroimaging research for anatomical labeling and for data analysis. The degree to which a probability map can accurately estimate the location of the structure of interest in a new individual depends on many factors, including the variability in the morphology of the structure of interest over subjects, the registration (normalization procedure and template) applied to align the brains among individuals and the registration used to map a new subject's dataset to the frame of the probabilistic map. Here, we take Heschl's gyrus (HG) as our structure of interest, and explore the impact of different registration methods on the accuracy with which a probabilistic map of HG can approximate HG in a new individual. We compare three registration procedures; high-dimensional (HAMMER); template-free B-spline-based groupwise; and segmentation-based (SPM5); to each other and to a previously published (affine) probabilistic map of HG.1 We quantitatively evaluate the accuracy of the resulting maps using evidence-based diagnostic measures within a leave-one-out cross-validation structure, to demonstrate that maps created using either HAMMER or SPM5 have relatively high sensitivity, specificity and positive predictive value, compared to a map created using the groupwise algorithm or compared to the published map.