MATERIALSAND METHODS DataAcquisitionand Collection

human brain function for the past decade (see Roland [1] for review). Intersubject analysis is essential (2) because physiologi cal changes, such as altered regional cerebral blood flow, are usually too slight compared with noise caused by random decay ofnucide to allow statistically significant inference with a single thtaset. It also aims at determining the average and common findings applicable to a group of subjects instead of describing findings for each individual subject Spatial normalization meth ods developed with voxel-by-voxel analysis, which have taken the place of the region of interest (ROI) analysis, are now indispensaMe in current PET activation studies. The first spatial normalization method was proportional adjustment of the brain size after realignment of the brain into the three-dimensional coordinates (Talairach space) (3,4) using skull radiographs (5). Some methods have improved the precision with nonlinear as well as linear transformations at the cost of computational labor and complexity. Some have become more convenient by automa tization of procedures. These spatial normalization methods do not necessarily mean anatomic normalization because the PET imagesare not morphological imagesbut rather are functional images. In fact, some spatial normalization methods, such as statistical parametric mapping (5PM) (6,7), even dispense with morphological images. However, it is veiy intriguing how precisely the spatially normalized brains anatomically coincide with each other because brain function and morphology obvi ously havea closerelationshipandthe terminology usedto infer brain functions are anatomic denominations based on brain atlases. In this study, we validate the anatomic precision of the spatial normalization method with linear parameters alone (linear method) (8), SPM95 (7) and human brain atlas (HBA) (9).