Effects of four normalizing methods on data analytic results in functional brain imaging

Functional brain imaging data may contain large individual differences in information about whole brain and regional levels of activity, and it is common to remove these differences using arithmetic transformation (normalization) prior to statistical analysis. As no single transformation is widely accepted, we examine the effects of four normalizing methods (ratioing, residuals from regressions on global cerebral blood flow, Z scores, and subject residual profiles) on 1) profile shape, 2) correlations between regions, 3) correlations between subjects, and 4) analysis of variance results. These effects are evaluated using an empirical data set consisting of regional cerebral blood flow values from 22 regions of interest in 46 depressed adults and 48 age-matched normal controls obtained by ¹³³Xe single photon emission computed tomography. Results show that the normalization method has substantial but different effects on characteristics of the data and statistical results. The ratioing method appears to be an optimal choice for most analyses.