Single-trial classification of awareness state during anesthesia by measuring critical dynamics of global brain activity

www.nature.com/scientificreports OPEN Received: 5 June 2018 Accepted: 1 March 2019 Published: xx xx xxxx Single-trial classification of awareness state during anesthesia by measuring critical dynamics of global brain activity Leandro M. Alonso Guillermo A. Cecchi 1,2 6 , Guillermo Solovey3, Toru Yanagawa4, Alex Proekt5, & Marcelo O. Magnasco1,2 In daily life, in the operating room and in the laboratory, the operational way to assess wakefulness and consciousness is through responsiveness. A number of studies suggest that the awake, conscious state is not the default behavior of an assembly of neurons, but rather a very special state of activity that has to be actively maintained and curated to support its functional properties. Thus responsiveness is a feature that requires active maintenance, such as a homeostatic mechanism to balance excitation and inhibition. In this work we developed a method for monitoring such maintenance processes, focusing on a specific signature of their behavior derived from the theory of dynamical systems: stability analysis of dynamical modes. When such mechanisms are at work, their modes of activity are at marginal stability, neither damped (stable) nor exponentially growing (unstable) but rather hovering in between. We have previously shown that, conversely, under induction of anesthesia those modes become more stable and thus less responsive, then reversed upon emergence to wakefulness. We take advantage of this effect to build a single-trial classifier which detects whether a subject is awake or unconscious achieving high performance. We show that our approach can be developed into a means for intra-operative monitoring of the depth of anesthesia, an application of fundamental importance to modern clinical practice. Over the last century the invention of general anesthesia transformed modern medicine by enabling highly invasive surgeries and diagnostic procedures to be performed while the patient is rendered unconscious. Since the 1930’s it has been proposed that the anesthetized state can be monitored using electroencephalography (EEG)1. Exactly how the EEG ought to be monitored and interpreted to assure that, on the one hand, the patient is not awake during surgery and, on the other hand, not overdosed on anesthetics is still not clear. Many different measures have been applied to quantify the effects of anesthetics on brain activity. Most commonly deployed measures include the spectral characteristics of the EEG2. Indeed, many anesthetics such as propofol elicit the canonical slow waves, also associated with slow wave sleep1 and anteriorization of alpha oscillations2–4. Yet, other anesthetics, most notably ketamine, are not reliably associated with slowing of the EEG5 or the anterior shift in alpha oscillations6. Furthermore, even at a fixed anesthetic concentration, spectral characteristics of local field potentials recorded from the thalamus and cortex fluctuate stochastically among different discrete states7. Other EEG-based measures of anesthetic depth include the bispectral index8. Yet, bispectral index is also not reliably altered by ketamine9 and other anesthetics8. More recent attempts at quantifying the effects of anesthetics on brain activity focused on functional connectivity between different brain areas. This promising approach identified that frontoparietal connectivity is suppressed by mechanistically distinct anesthetics that include ketamine, propofol and sevoflurane6,10. Indeed, as consciousness is thought to be an emergent phenomenon arising out of the interactions between different brain areas, it seems likely that a robust and theoretically 1 Laboratory of integrative neuroscience, The Rockefeller University, New York, NY, 10065, USA. 2Present address: Volen Center for Complex Systems, Department of Biology, Brandeis University, Waltham, MA, 02454, USA. 3 Instituto del Cálculo, FCEyN, Universidad de Buenos Aires, (C1428EGA), Buenos Aires, Argentina. 4Laboratory for Adaptive Intelligence, Brain Science Institute, RIKEN, Saitama, 351-0198, Japan. 5Anesthesiology and Critical Care, University of Pennsylvania, Philadelphia, PA, 19104, USA. 6IBM, Thomas J. Watson Research Center, Yorktown Heights, NY, USA. Correspondence and requests for materials should be addressed to L.M.A. (email: lalonso@ brandeis.edu) or G.S. (email: ) Scientific Reports | (2019) 9:4927 | https://doi.org/10.1038/s41598-019-41345-4 1 www.nature.com/scientificreports/ www.nature.com/scientificreports sound measure of anesthetic depth should take into account the interactions between signals emitted by different parts of the cortex. Consistent with this line of reasoning Massimini and colleagues demonstrated that loss of consciousness associated with sleep11,12, anesthesia13, and brain injury14 result in decrease in the complexity of responses elicited by transcranial magnetic stimulation. Yet, it is unclear what dynamical features of brain activity result in the disruption of functional connectivity and loss of complexity of evoked responses. One possible explanation for loss of connectivity and complexity of responses observed in the unconscious state is that in order to exhibit consciousness the brain must operate in a critical regime similar to phase transitions in physics, given several computational desirable features of such states represented by the statistics of the thermodynamic variables15. Evidence for statistical criticality is based on the observation that various aspects of neuronal activity such as avalanches observed in local field potentials and action potentials in tissue preparations and in animal models16,17, as well as magneto-encephalography (MEG) and electro-corticography (ECoG) in human subjects18,19, exhibit long tailed-distributions well approximated by power laws. More recently, the dynamical aspect of criticality has been brought into focus, as a similarly desirable feature not fully captured by steady-state statistics such as avalanche size distributions20–22; a perturbation in an extended dynamical system that is close to a critical point will neither decay nor explode, thus allowing for long range communication across the entire system. This will manifest as increase in functional connectivity and the complexity of responses. In contrast, if the system is far from criticality (therefore stable), perturbations damp out and no information integration takes place beyond the characteristic time scale which characterize the damping. This will result in the apparent loss of functional connectivity and loss of complexity of responses. The dynamically critical regime provides important functional benefits; quantities such as dynamic range and information transmission are optimized near criticality23. If indeed dynamical criticality is a useful feature of brain activity, stability of neuronal dynamics ought to be modulated by the behavioral state of the subject. When the brain is awake and displaying complex sta (...truncated)