Statistical surface-based morphometry using a nonparametric approach

We present a novel method of statistical surface-based morphometry based on the use of nonparametric permutation tests. In order to evaluate morphological differences of brain structures, we compare anatomical structures acquired at different times and/or from different subjects. Registration to a common coordinate system establishes corresponding locations and the differences between such locations are modeled as a displacement vector field (DVF). The analysis of DVFs involves testing thousands of hypothesis for signs of statistically significant effects. We randomly permute the surface data among two groups to determine thresholds that control the family wise (type 1) error rate. These thresholds are based on the maximum distribution of the amplitude of the vector fields, which implicitly accounts for spatial correlation of the fields. We propose two normalization schemes for achieving uniform spatial sensitivity. We demonstrate their application in a shape similarity study of the lateral ventricles of monozygotic twins and nonrelated subjects.