Evaluating Spatial Normalization Methods for the Human Brain

Cortical mapping (CSM) studies have shown cortical locations for language function are highly variable from one subject to the next. If individual variation can be normalized, patterns of language organization may emerge that were heretofore hidden. In order to uncover these patterns, computer-aided spatial normalization to a common atlas is required. Our goal was to determine a methodology by which spatial normalization methods could be evaluated and compared. We developed key metrics to measure accuracy of a surface-based (Caret) and volume-based (SPM2) method. We specified that the optimal method would i) minimize variation as measured by spread reduction between CSM language sites across subjects while also ii) preserving anatomical localization of all CSM sites. Eleven subject's structural MR image sets and corresponding CSM site coordinates were registered to the colin27 human brain atlas using each method. Local analysis showed that mapping error rates were highest in morphological regions with the greatest difference between source and target. Also, SPM2 mapped significantly less type 2 errors. Although our experiment did not show statistically significant global differences between the methods, our methodology provided valuable insights into the pros and cons of each method.