Neuroimaging of emotion: empirical effects of proportional global signal scaling in fMRI data analysis

Global variations of BOLD-fMRI signal are often considered as nuisance effects. This unwanted source of variance is commonly eliminated using proportional global signal scaling (PGSS). However, application of PGSS relies on the assumption that global variations of BOLD signal and experimental conditions are uncorrelated. It has been shown for cognitive tasks that the unjustified application of PGSS might greatly distort statistical results. The present study examined this issue in the domain of emotion research. Specifically, fMRI data were obtained in a block-design, while 21 subjects passively viewed high and low emotionally arousing pleasant, unpleasant, and neutral pictures. Violations of the orthogonality assumption were found for analyses of emotional pictures high in arousal, causing dramatically different outcomes when compared to analyses performed without PGSS. Application of PGSS was associated with attenuated emotional activation in visual cortical areas, insensitivity to emotional activations in limbic and paralimbic regions, and widely distributed artificial deactivations. In contrast, the orthogonality assumption was not violated for low arousing emotional materials. Thus, the validity of using PGSS varied as a function of the emotional arousal of the stimuli. Taken together, the unwarranted use of PGSS might contribute to conflicting results in affective neuroscience fMRI studies, in particular with respect to limbic and paralimbic structures.