Drowsiness increases slow oscillations in rs-fMRI signal before sleep onset
Poster No:
2025
Submission Type:
Abstract Submission
Authors:
Ivan IGOR GAEZ1, Elpidio Attoh-Mensah2, Clement Nathou1, Lydie Vincent1, Marc Joliot3, Luc Brun4, Mikaël Naveau5, Olivier Etard1
Institutions:
1Normandie Université, UNICAEN, INSERM, COMETE, CYCERON, CHU Caen, Caen, France, 2Univ. Limoges, HAVAE, UR 20217, Limoges, France, 3GIN, IMN, UMR5293, CEA , CNRS, Université de Bordeaux, Bordeaux, France, 4Normandie Université, UNICAEN, ENSICAEN, CNRS, GREYC, Caen, France, 5Normandie Université, UNICAEN, CNRS, INSERM, UAR3408 CYCERON, Caen, France
First Author:
Co-Author(s):
Introduction:
Sleep can significantly affect the functional MRI (fMRI) signal as reported in previous studies where subjects were allowed to fall asleep during the examination (Fukunaga et al., 2006, Duyn 2019). Fluctuation in the MRI signal has been associated with the increase in functional connectivity observed during light slow wave sleep (Tagliazucchi et al., 2014). Nevertheless, a major concern arises regarding the limited understanding of the evolution of fMRI signals as a function of drowsiness, when subjects are instructed to remain awake, as typically prescribed in standard resting-state exploration (rs-fMRI). This knowledge gap primarily results from the challenge of quantifying such vigilance-related events during fMRI, as they mostly occur within the complex period of wakefulness-to-sleep transition (Ogilvie, 2001). In this study, we used the percentage of eye closure (PERCLOS index (Wierwille et al., 1994)) as a proxy of drowsiness to explore its effect on the evolution of the rs-fMRI signal.
Methods:
A 45 min T2* gradient echo sequence (MRI GE 3T SIGNA Premier, TR = 1 s, TE = 30 ms, flip angle = 62°, 2700 volumes) was performed in 50 healthy subjects (26 Females, 24 Males, age = 23.3 ± 4.45) with instructions not to move, not to do constructed mental tasks and to keep their eyes open without sleeping. Subjects were asked to sleep for only 5 hours the night before and all examinations were carried out at 1 p.m.
Drowsiness was assessed using video from the MRI surveillance camera allowing to automatically compute an eye closure degree called Eye Aspect Ratio (EAR). Afterwards, PERCLOS index was calculated based on proportion of the time spent with eyes closed for at least 80% of their size in a 60s window. The drowsiness state was identified according to the following thresholds of the index (awake: [0.0, 08]; probably drowsy: [0.08, 0.15]; drowsy: [0.15, 1]; sleep: eyes closed > 5s). Lastly, all signals were resampled to MRI's sampling rate (1s).
MRI signals were preprocessed using fMRIPrep 20.2.6 with each brain region being extracted through python's nilearn library 0.10.1, using AAL3 v1 1mm atlas. We considered the subject's specific mask and filtered out confounds computed in the preprocessing stage (low frequency drift and motion parameters). Signals were analyzed in a time-frequency manner using a Taper spectrogram (window length: 60s, window step: 1s, time half-bandwidth: 2.5, number of tapers: 4). Each column of spectrogram was classified using our drowsiness scale then grouped by state. Afterwards, mean amplitude by frequency band was calculated and finally projected on an inflated surface of the left hemisphere of the brain.
Drowsiness was assessed using video from the MRI surveillance camera allowing to automatically compute an eye closure degree called Eye Aspect Ratio (EAR). Afterwards, PERCLOS index was calculated based on proportion of the time spent with eyes closed for at least 80% of their size in a 60s window. The drowsiness state was identified according to the following thresholds of the index (awake: [0.0, 08]; probably drowsy: [0.08, 0.15]; drowsy: [0.15, 1]; sleep: eyes closed > 5s). Lastly, all signals were resampled to MRI's sampling rate (1s).
MRI signals were preprocessed using fMRIPrep 20.2.6 with each brain region being extracted through python's nilearn library 0.10.1, using AAL3 v1 1mm atlas. We considered the subject's specific mask and filtered out confounds computed in the preprocessing stage (low frequency drift and motion parameters). Signals were analyzed in a time-frequency manner using a Taper spectrogram (window length: 60s, window step: 1s, time half-bandwidth: 2.5, number of tapers: 4). Each column of spectrogram was classified using our drowsiness scale then grouped by state. Afterwards, mean amplitude by frequency band was calculated and finally projected on an inflated surface of the left hemisphere of the brain.
Results:
Figure 2 shows the evolution of the power spectral density of the rs-fMRI signal as a function of the drowsiness states. We can observe that the amplitude of slow oscillations, particularly in the ]0.047-0.063] Hz band, increases overall at the first sign of drowsiness. This increase clearly predominates within primary cortices (motor, somesthetic, visual and auditory), as demonstrated by a drowsiness effect in the generalized linear models computed in each 165 brain regions and corrected for Bonferroni multiple comparisons test (Wald's χ2 ranging from 21,96 to 334,1). Conversely, the deep gray structures, notably the thalamus, did not undergo the same changes.
Conclusions:
This study is the first attempt in exploring rs-fMRI signals while subjects struggle against sleep by considering 4 drowsiness states. The increase in slow oscillations observed, which may underlie an increase in functional connectivity and which predominate in the primary cortices, is paradoxical given the progressive isolation of the cortex induced by the thalamus during the descent into sleep and requires further investigation.
Modeling and Analysis Methods:
Methods Development
Task-Independent and Resting-State Analysis 1
Novel Imaging Acquisition Methods:
BOLD fMRI 2
Physiology, Metabolism and Neurotransmission :
Neurophysiology of Imaging Signals
Keywords:
Sleep
Somatosensory
Other - drowsiness
1|2Indicates the priority used for review
Provide references using author date format
Fukunaga, M. (2006) ‘Large-amplitude, spatially correlated fluctuations in BOLD fMRI signals during extended rest and early sleep stages’, Magnetic Resonance Imaging, 24(8), pp. 979–992
Ogilvie, R.D. (2001) ‘The process of falling asleep’, Sleep Medicine Reviews, 5(3), pp. 247–270
Tagliazucchi, E. (2014) ‘Decoding Wakefulness Levels from Typical fMRI Resting-State Data Reveals Reliable Drifts between Wakefulness and Sleep’, Neuron, 82(3), pp. 695–708
Wierwille, W. (1994) Research on vehicle-based driver status/performance monitoring; development, validation, and refinement of algorithms for detection of driver drowsines. Technical report, DOT-HS-808-247. Virginia Tech Transportation Institute