Computational NeurosciencePhysiologically based arousal state estimation and dynamics
Introduction
The brain consists of a vast network of interacting elements, but exhibits large scale coordinated activity that is readily measurable. Understanding the relationship between properties of the individual components and the large scale properties of the system is crucial for understanding the operation of the brain. One readily accessible large scale measure is arousal level, which changes over the course of the sleep–wake cycle.
The arousal level of a subject is typically classified according to the Rechtschaffen and Kales (R&K) (Rechtschaffen and Kales, 1968) classification scheme, or the American Academy of Sleep Medicine (AASM) scheme (Iber et al., 2007). These schemes force the arousal level to be matched to one of a small selection of stages: wake (W); stage 1 sleep (called S1 in R&K, N1 in AASM), which corresponds to light sleep and is usually short in duration; stage 2 sleep (called S2 in R&K, N2 in AASM), which is a deeper stage of sleep marked by K-complexes (typically a large negative peak in the EEG, followed by a positive peak, similar to an evoked response) and sleep spindles (short bursts of activity at around 12–14 Hz); slow wave sleep (called S3 and S4 in R&K, N3 in AASM), which corresponds to deep sleep in which K-complexes and sleep spindles are sometimes present; and rapid eye movement (REM) sleep, which occurs during dreaming. In a typical night, sleep cycles between all the sleep stages several times.
Although the sleep stage can a provide a useful qualitative summary, it falls short in analyzing brain states, dynamics, and physiology for multiple reasons:
- (i)
Real brain states vary continuously (notwithstanding the transition between sleep and wake, which is rapid but continuous), and therefore cannot be accurately captured by discrete stages. Discrete stages necessarily group many different brain substates into each single sleep stage, with somewhat arbitrary boundaries.
- (ii)
The traditional sleep stages are motivated by the external appearance of the subject, which are then matched to a number of different markers for these stages including EEG. Because stages were not designed around the physiology that underlies brain states and associated EEG, there is no unique correspondence of EEG and other features with specific sleep stages.
- (iii)
Sleep stages are usually assigned to 30 s epochs based on the presence or absence of EEG features in that epoch; e.g., S1 epochs cannot contain sleep spindles or K-complexes, and S3 and S4 are distinguished by amount of time slow wave activity is present (Rechtschaffen and Kales, 1968). Therefore, the sleep stage that is assigned to the epoch can be quite sensitive to the precise timing of the epoch boundaries. Shifting the epochs forward or backward by only a few seconds can change the sleep stage assigned to otherwise identical data, by including or excluding features in the epoch, or by causing the percentage of slow wave activity in the epoch to be split between two epochs, or vice versa. This issue is particularly significant for short-lived stages like S1 that can last for less than a minute, and may therefore be significantly affected by the epoch timings.
- (iv)
Interobserver agreement with AASM staging is only 83% (Rosenberg and Van Hout, 2013), and even lower levels of agreement have been reported for R&K staging; e.g., interobserver agreement of just 73% was reported by Norman et al. (2000). These low levels of agreement arise because the classical sleep stages are defined via many criteria that are based partly on subjective classification.
- (v)
Classical sleep scoring forces a discretization of a continuum of brain states into a phenomenological classification scheme that provides a qualitative measure of the responsiveness of the subject but contributes little toward understanding the physiological differences between the states, and hinders the estimation and tracking of the continuous dynamics of arousal.
The above issues are illustrated schematically in Fig. 1. Note that throughout this study we use the term ‘state’ to refer to the physiological state of the brain at an instant in time, and the term ‘stage’ to refer to R&K or AASM classifications. We relate each state to a single set of underlying physiological parameters in our model. Discrete ‘transitions’ occur when the classified sleep stage changes, whereas brain states evolve continuously and are linked by ‘trajectories’.
In Fig. 1(a), brain states are represented in terms of their underlying physiology, and continuous trajectories link one brain state to the next. Differences between individual subjects are reflected in the different trajectories taken. In Fig. 1(b), arousal stages have been identified by some criteria that correspond to the definitions of these stages provided in a scoring scheme, which are not expressed in terms of neural physiology. There is significant overlap between the assigned stages because each individual has different physiology, and a single combination of measurable or physiological parameters may correspond to more than one arousal stage. The ambiguity of the stages also requires scorers to make qualitative, subjective judgements that further contribute to the overlap in assignment of stages.
In Fig. 1(c), the arousal stages have been decoupled from the underlying physiology, and although overlap between the assigned stages is permitted, it can only be quantified by interobserver disagreement. Finally, Fig. 1(d) shows the current common usage of sleep staging, where each epoch of EEG is classified as belonging to one of the sleep stages, and the possibility of overlap between multiple stages is not considered because the classification schemes require that a single sleep stage be selected. Thus the end result is that the true continuous trajectories in Fig. 1(a) have been replaced by discrete jumps between artificially defined stages, thereby losing information about the physical processes underlying the change in brain state and resulting in inconsistency in assignment of stages.
The aim of this study is to find a continuous representation of brain states that reflects underlying physiology and can be observed simply and easily. We use EEG in this study as it is a readily accessible, noninvasive measure, although this does not preclude including other aspects of polysomnograms such as actigraphy, eye movement, or muscle tone, that might later expand the range of physiology that can be inferred.
Neural field modeling is a powerful technique for constructing relatively simple, physiologically based models of the brain that can predict large-scale measures of brain activity (Deco et al., 2008, Pinotsis et al., 2012). We have developed a neural field corticothalamic model (Robinson et al., 2001, Robinson et al., 2002, Robinson et al., 2004, Robinson et al., 2005, Rowe et al., 2004) that we have previously used to investigate the alpha rhythm (O’Connor and Robinson, 2004, Robinson et al., 2003b), age-related changes to the physiology of the brain (van Albada et al., 2010), evoked response potentials (Rennie et al., 2002), and many other phenomena. Among other measurable signals, the model accurately predicts EEG activity from physiologically based parameters that correspond to experimentally measurable quantities in the brain. Its predictions can be fitted to EEG spectra to estimate physiological parameters (Robinson et al., 2003a, Robinson et al., 2005, Rowe et al., 2004, van Albada et al., 2007, van Albada et al., 2010), and these estimates are consistent with a range of experimental EEG-related phenomena (Robinson et al., 2004, Rowe et al., 2004).
In this study, we first show that the model power spectrum is directly comparable to experimental EEG, and correlate each of the traditional sleep stages with model parameter values. This identifies the approximate regions of physiological parameters that are associated with the classical sleep stages across a population of subjects, and also quantifies the overlap in sleep stages in terms of the corresponding brain states, as shown in Fig. 1(b). Next, we approximate the trajectories between brain states, demonstrating that continuous, physiologically valid trajectories exist that can produce the observed states. Finally, we illustrate how the model can be fitted to experimental data to infer trajectories similar to those in Fig. 1(a), thereby revealing the underlying dynamics of brain states. Section 2 provides an overview of the model structure and equations. In Section 3 we discuss a systematic method for representing classical sleep stages in the model's parameter space, in order to provide backward compatibility with the more restrictive sleep staging approach. Finally, in Section 4 we show the correspondence between the model parameter space and classical sleep stages, and present illustrative examples of trajectories between brain states.
Section snippets
Theory
In this section we review the neural field model used in previous studies (Roberts and Robinson, 2012, Robinson et al., 1997, Robinson et al., 2002, Robinson et al., 2004, van Albada et al., 2007, van Albada et al., 2010).
Method
In this section, we develop our method for identifying the model parameter regimes associated with each of the traditional sleep stages. In previous work, we semiquantitatively associated the sleep stages with different X, Y, and Z (Robinson et al., 2002) values. In this study, we use new experimental data from a large number of subjects to quantitatively localize the sleep stages (including their boundaries) in terms of the model's full parameter space of νab values. We have also previously
Results and discussion
We first derive additional model-based constraints in Section 4.1. We then discuss the spectral features characterizing each sleep stage in Section 4.2. The model parameters corresponding to each sleep stage are presented in Section 4.3. Trajectories between sleep stages in the model and in experiment are presented in Section 4.4.
Conclusions
We have fitted the predictions of an established physiologically based neural field model to EEG spectra to track the physiology of individual brain states. In this framework, the temporal evolution of brain states is represented as a continuous trajectory in the physiological parameter space, thereby extending our previous work in waking states. Our approach addresses a number of serious shortcomings with traditional sleep staging by avoiding the need for discrete stages and establishing an
Acknowledgements
This work was supported by Neurosleep, the Australian Research Council, the National Health and Medical Research Council (through the Center for Integrated Research and Understanding of Sleep), the Westmead Millennium Institute, and Brain Resource Ltd. We acknowledge the support of the Brain Resource International Database (under the auspices of Brain Resource; www.brainresource.com) for data acquisition and processing. All scientific decisions are made independently of any BR commercial
References (56)
- et al.
Prediction and verification of nonlinear sleep spindle harmonic oscillations
J Theor Biol
(2014) - et al.
Experimental observation of a theoretically predicted nonlinear sleep spindle harmonic in human EEG
J Clin Neurophysiol
(2014) - et al.
A new EEG biomarker of neurobehavioural impairment and sleepiness in sleep apnea patients and controls during extended wakefulness
Clin Neurophysiol
(2013) - et al.
Sleep-related variations of membrane potential in the lateral geniculate body relay neurons of the cat
Brain Res
(1983) - et al.
Dynamic causal modeling with neural fields
Neuroimage
(2012) - et al.
Estimation of neurophysiological parameters from the waking EEG using a biophysical model of brain dynamics
J Theor Biol
(2004) - et al.
Neurophysiological changes with age probed by inverse modeling of EEG spectra
Clin Neurophysiol
(2010) - et al.
Cortical firing and sleep homeostasis
Neuron
(2009) - et al.
Central sleep apnea in stable methadone maintenance treatment patients
Chest
(2005) - et al.
Modeling and investigation of neural activity in the thalamus
J Theor Biol
(2007)
Role of the ferret perigeniculate nucleus in the generation of synchronized oscillations in vitro
J Physiol
Cortex: statistics and geometry of neuronal connectivity
A unifying explanation of primary generalized seizures through nonlinear brain modeling and bifurcation analysis
Cereb Cortex
The dynamic brain: from spiking neurons to neural masses and cortical fields
PLoS Comput Biol
The synaptic organization of the brain
Contrasting roles of neural firing rate and local field potentials in human memory
Hippocampus
Integrative neuroscience: the role of a standardized database
Clin EEG Neurosci
Arm immobilization causes cortical plastic changes and locally decreases sleep slow wave activity
Nat Neurosci
Local sleep and learning
Nature
The AASM manual for the scoring of sleep and associated events: rules, terminology and technical specification
Modulation of cortical activation and behavioral arousal by cholinergic and orexinergic systems
Ann N Y Acad Sci
Encoding of natural scene movies by tonic and burst spikes in the lateral geniculate nucleus
J Neurosci
Intracortical connectivity of pyramidal and stellate cells: estimates of synaptic densities and coupling symmetry
Netw Comput Neural Syst
Direct evidence for wake-related increases and sleep-related decreases in synaptic strength in rodent cortex
J Neurosci
Effects of sleep and arousal on the processing of visual information in the cat
Nature
Slow waves, synaptic plasticity and information processing: insights from transcranial magnetic stimulation and high-density EEG experiments
Eur J Neurosci
Lateral geniculate nucleus unitary discharge in sleep and waking: state- and rate-specific aspects
J Neurophysiol
Equation of state calculations by fast computing machines
J Chem Phys
Cited by (58)
Method for quantifying arousal and consciousness in healthy states and severe brain injury via EEG-based measures of corticothalamic physiology
2023, Journal of Neuroscience Methods