Robust Identification of Sleep States and Their Transitions Using Whole-Night fMRI Data
Poster No:
2570
Submission Type:
Abstract Submission
Authors:
Nils Yang1, Dante Picchioni1, Jacco de Zwart1, Yicun Wang1, Peter van Gelderen1, Jeff Duyn1
Institutions:
1National Institutes of Health, Bethesda, MD
First Author:
Co-Author(s):
Introduction:
Sleep staging based on electroencephalography (EEG) and polysomnography (PSG) has been widely used to characterize sleep, but cannot track the detailed spatial and temporal features of brain activity across a night's sleep. For example, brain activity patterns involved with transitions or changes within and between sleep stages are not captured with common sleep staging approaches. Recent investigations have turned to functional magnetic resonance imaging (fMRI) to address this shortcoming1,2. Nevertheless, because of the sleep-adverse MRI environment, these studies have been limited to brief (~1 hour) sleep periods, limiting the ability to capture the full range of sleep stages and their cycling. Overcoming these challenges may provide valuable insights into conditions marked by disrupted sleep transitions, e.g. insomnia and various mental disorders.
Methods:
Data acquisition was conducted as part of a previously described well-controlled sleep experiment3 with two consecutive nights of concurrent fMRI-EEG recording from 23:00 to 7:00 for each night. Twelve participants (18-35 years old, 8 females) had at least one complete sleep cycle, encompassing all four sleep stages (N1-3 and REM), during both Night 1 and 2. Throughout each night, the fMRI experiments were intermittently disrupted by either acoustically stimulated or spontaneous awakenings. Detailed fMRI (with global signal regression), EEG, and photoplethysmography (PPG) preprocessing steps can be found elsewhere3,4. We employed an unsupervised learning approach - Hidden Markov Model (HMM)5,6 to analyze fMRI timecourses extracted from 300 ROIs7. See Figure 1. The HMM, trained on Night 2 data, encompassed 21 whole-brain states based on various model evaluation indexes. Each state was characterized as a multivariate Gaussian distribution including a mean fMRI activation distribution and a functional connectivity (FC) matrix. Furthermore, the HMM featured a transition probability matrix that detailed the likelihood of transitioning between states.
Results:
Modular analysis of this transition probability matrix revealed five modules, each containing a set of states that had similar preferences to occur during a specific sleep stage or part thereof (See Figure 2). None of the states clearly preferred N1. Remarkably, the trained HMM predicted brain sleep states during Night 1 that had highly similar sleep stage preference (r = 0.94, p < 0.001). For mean activation, Wake-related HMM states 16 and 20 demonstrated the classic opposite activation pattern between the default-mode network (DMN) and frontal-parietal network (FPN). In contrast, during N3-related HMM states 8 and 10, DMN and FPN showed the same activation direction. For FC, similar patterns were found. In Wake-related HMM states 16 and 20, the FCs between DMN and Salience Network (SAL) / Control Network (CON) were negative, while during N3-related HMM states 8 and 10, these FCs were positive.
Conclusions:
HMM analysis of overnight fMRI sleep data identified 21 brain states and their transition probability. A strong preference for specific sleep stages was observed that was highly reproducible across successive overnight scan sessions. Furthermore, analysis of the transition preference between states naturally grouped the states into the classical sleep stages. These observations suggest that HMM may provide valuable information about sleep-state-specific brain activity patterns that extend well beyond the information provided by classical sleep staging. For example, cursory inspection of the HMM results suggests a subdivision of N2 in two groups of multiple states with different transition probabilities to Wake; analysis of brain activation and FC patterns of HMM states indicated that the connections between DMN and SAL/CON may play a critical role in the transition from wake to light sleep and, to deep sleep. A more detailed analysis of the HMM states and their FC patterns is currently underway.
Novel Imaging Acquisition Methods:
BOLD fMRI 2
Perception, Attention and Motor Behavior:
Sleep and Wakefulness 1
Keywords:
FUNCTIONAL MRI
NORMAL HUMAN
Sleep
Other - EEG
1|2Indicates the priority used for review
Provide references using author date format
1. Stevner, A. B. A. et al. Discovery of key whole-brain transitions and dynamics during human wakefulness and non-REM sleep. Nat. Commun. 10, 1035 (2019).
2. Zhou, S. et al. Dynamic functional connectivity states characterize NREM sleep and wakefulness. Hum. Brain Mapp. 40, 5256–5268 (2019).
3. Moehlman, T. M. et al. All-night functional magnetic resonance imaging sleep studies. J. Neurosci. Methods 316, 83–98 (2019).
4. Picchioni, D. et al. Autonomic arousals contribute to brain fluid pulsations during sleep. NeuroImage 249, 118888 (2022).
5. Vidaurre, D., Smith, S. M. & Woolrich, M. W. Brain network dynamics are hierarchically organized in time. Proc. Natl. Acad. Sci. 114, 12827–12832 (2017).
6. Vidaurre, D. et al. Discovering dynamic brain networks from big data in rest and task. NeuroImage 180, 646–656 (2018).
7. Seitzman, B. et al. A set of functionally-defined brain regions with improved representation of the subcortex and cerebellum. NeuroImage 206, 116290 (2019).