TESTS FOR COMPARING IMAGES BASED ON RANDOMIZATION AND PERMUTATION METHODS

Tests comparing image sets can play a critical role in PET research, p roviding a yes-no answer to the question ''Are two image sets differen t?'' The statistical goal is to determine how often observed differenc es would occur by chance alone. We examined randomization methods to p rovide several omnibus test for PET images and compared these tests wi th two currently used methods. In the first series of analyses, normal ly distributed image data were simulated fulfilling the requirements o f standard statistical tests. These analyses generated power estimates and compared the various test statistics under optimal conditions. Va rying whether the standard deviations were local or pooled estimates p rovided an assessment of a distinguishing feature between the SPM and Montreal methods. In a second series of analyses, we more closely simu lated current PET acquisition and analysis techniques. Finally, PET im ages from normal subjects were used as an example of randomization. Ra ndomization proved to be a highly flexible and powerful statistical pr ocedure. Furthermore, the randomization test does not require extensiv e and unrealistic statistical assumptions made by standard procedures currently in use.