Error and t Images Depend on ANOVA Design and Anatomical Standardization in PET Activation Analysis

In a positron emission tomography activation analysis with task replications within subject, a number of analysis of variance (ANOVA) designs are applicable with different definitions of t and error. The characteristics of t (and z) maps and error images and how they depend on the ANOVA design and on the anatomical standardization method have been investigated. Six subjects underwent measurement of regional cerebral blood flow with [ 15 O]water under resting and while thinking of verbs associated with auditorily presented nouns, three times for each. The images were anatomically standardized with LINear, SPM95, or HBA. ANOVA was performed pixel by pixel to compute t statistics for the task main effect (verb vs rest) in four different ANOVA designs: (i) two way (subject and task) (2W), (ii) two way with interaction (2WI), (iii) two way with interaction, except that the “subject” was considered a random factor (2WI-RF), and (iv) three way (subject, task, and replication). The left frontal cortex extending from Broca's area to the premotor cortex was activated by the verb generation. The foci localization in the z images depended both on the anatomical standardization method and on the ANOVA design, and the variation ranged from 1 to 3 cm. SPM tended to present a higher peak z than LIN and HBA. The z images of 2W and 2WI looked alike, but 2WI-RF and 3W each presented a different z map within the activated area. The peak z score by 2WI-RF was lower than the others. The error images for 2W, 2WI, and 3W were heterogeneous, being high in gray and low in white.