0112 AUTOMATIC DETECTION OF SLEEP SPINDLES IN SIMULTANEOUSLY ACQUIRED EEG-FMRI DATA

Abstract

Sleep spindles play an important role in memory and cognition, and have complex spatial properties that are not well understood. Simultaneous EEG-fMRI has the potential to identify local neural activity underlying sleep spindles. Past work has developed automated spindle detectors that can compare to a “gold standard,” i.e. human experts, to enable quantitative study of sleep spindles. However, these methods have been used primarily on EEG datasets, as it is difficult to detect spindles in combined EEG-fMRI experiments due to artifacts induced by the MR environment. In this study, we compare an automated method for detecting sleep spindles in EEG and combined EEG-fMRI data. Raw and sleep-scored EEG data were obtained from 6 subjects in the preexisting DREAMS Sleep Spindles Database. 256-channel EEG-fMRI data was collected from 2 subjects in a 3T scanner and then sleep scored. Stage 2 EEG data was filtered using a 10–15 Hz bandpass filter and subjected to a Hilbert transform to estimate instantaneous amplitude. Spindles were detected by locating peaks that were above a set threshold (mean plus 1 standard deviation), and then compared to spindles detected manually. We found that sleep spindles in the DREAMS database were detected with 55% sensitivity and 43% precision. Using EEG-fMRI data, we detected sleep spindles with 54% sensitivity, but only 26% precision. When the threshold was increased to mean + 2 s.d., sensitivity decreased to 30% while precision increased to 31%, demonstrating that precision was lower even when using a more stringent threshold. While the sensitivity of the spindle detector was not affected in EEG-fMRI data, its precision decreased, suggesting that MR-induced artifacts may increase 10–15 Hz power and cause spurious EEG events. Future work will address artifact cleaning approaches to improve these results. Since the detector was able to find spindles in EEG-fMRI data, it can next be used to find neural activity correlated with spindles. This work was supported by the Athinoula A. Martinos Center for Biomedical Imaging, NIH grants P41-EB015896, S10-RR023403, S10-RR020948, and S10-OD010759, the Harvard Society of Fellows and a William F. Milton Fund Award.