Impact of vigilance fluctuations on resting state fMRI functional connectivity analysis

During the acquisition of resting state fMRI which by definition excludes the performance of a cognitive task, vigilance fluctuations up to the level of sleep stage 1 even deeper sleep may resemble a significant confound. This is hypothesized as meanwhile a wealth of data exists that described vigilance related FC changes. If vigilance is not monitored directly throughout the acquisition through EEG or eye tracking, functional connectivity analysis with relation to e. g. genomic information or behavioural data may then not be truly representative for the waking state. Here we use a support vector machine algorithm (previously developed on two independent fMRI/EEG datasets) to predict vigilance fluctuations in a third resting state fMRI dataset (N > 225, yound healthy subjects). By this, subjects with overly presence of non-wakefulness can be identified. We then perform functional connectivity analysis of the limbic system using seed analysis anterior cingulate as a model region) and ICA to investigate the influence of “sleepy subjects” in a large scale FC analysis by two appraoches: first, subjects with overly non-wakefulness exceeding a defined amount of the recorded 5 minutes are excluded completely; second, by using a sliding window technique to determine vigilance more continuously, subparts of the recording with lower vigilance are excluced to preserve a maximal amount of data. Both approaches are compared qualitatively and quantitatively.