Understanding the contribution of neural and physiological signal variation to the low repeatability of emotion-induced BOLD responses☆

Previous studies have reported low repeatability of BOLD activation measures during emotion processing tasks. It is not clear, however, whether low repeatability is a result of changes in the underlying neural signal over time, or due to insufficient reliability of the acquired BOLD signal caused by noise contamination. The aim of this study was to investigate the influence of "cleaning" the BOLD signal, by correcting for physiological noise and for differences in BOLD responsiveness, on measures of repeatability. Fifteen healthy volunteers were scanned on two different occasions, performing an emotion provocation task with faces (neutral, 50% fearful, 100% fearful) followed by a breath-hold paradigm to provide a marker of BOLD responsiveness. Repeatability of signal distribution (spatial repeatability) and repeatability of signal amplitude within two regions of interest (amygdala and fusiform gyrus) were estimated by calculating the intraclass correlation coefficient (ICC). Significant repeatability of signal amplitude was only found within the right amygdala during the perception of 50% fearful faces, but disappeared when physiological noise correction was performed. Spatial repeatability was higher within the fusiform gyrus than within the amygdala, and better at the group level than at the participant level. Neither physiological noise correction, nor consideration of BOLD responsiveness, assessed through the breath-holding, increased repeatability. The findings lead to the conclusion that low repeatability of BOLD response amplitude to emotional faces is more likely to be explained by the lack of stability in the underlying neural signal than by physiological noise contamination. Furthermore, reported repeatability might be a result of repeatability of task-correlated physiological variation rather than neural activity. This means that the emotion paradigm used in this study might not be useful for studies that require the BOLD response to be a stable measure of emotional processing, for example in the context of biomarkers.