Brief overview I wrote on the plane to a conference largely about anesthesia and methods to monitor consciousness in human patients. I went to the conference to learn about pragmatic approaches to measuring consciousness and to get a feel for how the field thinks. This is unpolished, and I added anecdotes from clinicians and researchers I talked to + some takeaways about the field.
There are a lot of theories of consciousness circulating in academic discourse — a recent count places it at over 30, with the more popular including Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNW). Yet, neurologists and anesthesiologists are faced with the problem of inferring consciousness in their patients every day, and the majority of them attempt to do so with much simpler tools than these theories require.
Background
Measures of consciousness are usually tested first on healthy patients in varying states (waking, various depths of sleep, anesthetized, etc), and then extended to patients with disorders of consciousness (DoCs). Several of these have been developed to monitor anesthetic depth during medical procedures or to evaluate consciousness and potential for recovery in patients with DoCs. I mostly focus on neurological techniques, excluding simple physiological measures like pupil dilation, body temperature, heart rate, etc.
DoCs are defined pretty broadly — the NHS describes a DoC as a state where consciousness has been affected by damage to the brain. So, this could range from a coma in which a patient is totally unresponsive to split brain patients in which precepts presented to left and right sense organs are only perceived by one half of the brain.
DoCs are tricky because self-report is often compromised. Measuring consciousness typically depends on comparing neurological data to self-report or behavior. A clinician or researcher may look at EEG data for example and compare that to the patient’s behavior or verbal description of their state. But, there are cases where a patient appears unconscious, but their cognition isn’t compromised (usually called locked-in syndrome). This is where (imo) the interesting research happens in the field — what do we do when we can’t depend on self-report? For more on how to think through this question, I’d check out No-Report Paradigms: Extracting the True Neural Correlates of Consciousness and my favorite experimental paper on this topic A new no-report paradigm reveals that face cells encode both consciously perceived and suppressed stimuli.
Perturbational Complexity Index
It seems like the Perturbational Complexity Index (PCI) may be the most effective existing measure of consciousness in patients with DoCs. The clinicians I spoke to at this conference weren’t familiar with PCI although several researchers were. As far as I know, PCI hasn’t been actually translated to the clinic (except perhaps in the cases of clinicians who participate in PCI research) although I think the creators are pushing for that.
The basic idea is that brain activity is thought to be less complex during unconsciousness. PCI works by perturbing the brain (typically the superior frontal and/or parietal cortex) with transcranial magnetic stimulation (TMS), recording the resultant electrical activity, and then calculating its normalized Lempel-Ziv complexity (a popular measure of compressibility). In the standard clinical setting, doctors use non-invasive methods, so activity is typically measured with EEG.
PCI is optimized for cases in which patients appear unconscious and cannot self-report as it doesn’t depend on sensory or motor function which the creators emphasize as an advantage.
Importantly, it’s estimated that up to 20% of patients who present UWS symptoms may be “misclassified” with existing measures of assessing their consciousness as their PCI scores them at levels of complexity on par with healthy, waking patients. This could mean they’re having an internal experience that’s akin in “complexity” to a healthy control, but they just can’t report it.
How well does PCI work?
From Measuring Consciousness in the Intensive Care Unit, a paper published in April arguing for better integrating PCI into the ICU:
Over the past 10 years, PCI values above an empirically derived threshold (PCI>0.31) have identified consciousness with 100% specificity and 100% sensitivity in a validation data set of 48 conscious patients with brain injury and 102 healthy study participants across a broad range of behavioral states, including resting wakefulness, anesthesia, slow-wave sleep, and REM sleep
PCI outperforms other clinical approaches:
TMS-EEG measurements of PCI detect high complexity in approximately 95% of patients with severe brain injury who have recovered behaviorally to a minimally conscious state [20], as compared with a ~60% detection rate with task-based fMRI and EEG motor imagery paradigms
PCI doesn’t use IIT explicitly to produce its consciousness score, but Giulio Tononi, the creator of IIT, was involved in the development of PCI and has published on it multiple times. It’s sort of like a cousin of IIT?
There’s also some evidence that PCI can predict functional recovery from brain injuries like stroke.
Bispectral index
The bispectral index (BIS) combines multiple measures to produce a single number from 1-100 for anesthesiologists to monitor anesthetic depth, and it’s the most common method for monitoring anesthetic depth in the clinic. BIS is a product sold to hospitals, and its algorithm is proprietary although Christopher Conner recently reverse engineered the algorithm and published it (he gave a pretty awesome/funny talk on this at the conference).
BIS appears pretty imperfect, but it seems like as long as anesthesiologists stay between a score of 40-60 the probability of inadequately anesthetizing a patient is very low. That said, about 1 in 1000 patients will not be sufficiently anesthetized. People seem generally unexcited about BIS with the general reaction being “it’s what we have”. I was told that clinicians often use the BIS’s display system to watch for changes in EEG activity (i.e. increased power of waves in x frequency range) rather than make decisions solely on the number provided.
There are other measures I didn’t get into like CSI which also uses EEG with a different algorithm and is thought to have near-equivalent performance to BIS.
Slow-wave saturation
Warnaby et al. found that slow-wave saturation (when slow-wave power (particularly in the 0.5 to 1.5 Hz range) increases then plateaus) coincides with induction of anesthesia and loss of response to sensory stimuli. “Slow-waves” refers to data collected with EEG measures, so a large increase in slow-wave power would = a large population of cortical neurons syncing up. Regarding sensory stimuli, they measured changes in neural activity in response to sounds and lasers.
Warnaby et al. hypothesize that slow wave activity coincides with the isolation of thalamocortical activity from external stimuli. Using fMRI, at slow-wave saturation, they observed a loss of thalamic and primary cortical activation specifically in response to the sensory stimuli —
Although the behavioral transition (LOBR) was associated with a significant reduction in activation in several cortical areas relevant to auditory and nociceptive inputs, significant activity persisted in the thalamus and primary cortical processing regions of the now unresponsive subjects (Fig. 4C, left and middle). Moreover, the EEG transition to SWA saturation was specifically associated with loss of thalamic and primary cortical activation (Fig. 4C, right). However, beyond this thalamocortical isolation, the brain was neither inactive nor unresponsive. Both auditory and noxious stimuli were associated with evoked BOLD signal changes. This activation was not within the sensory thalamocortical system but involved a network of regions: the precuneus and the posterior parietal and prefrontal cortices.
In later work, they further confirmed the slow-wave saturation effect using EEG in 92% of participants across four studies (n = 393).
Task-based fMRI and EEG
Task-based fMRI and EEG measure patients’ neural activity in response to varying stimuli such as listening to music or being prompted to imagine something. While these can both be used in the ICU or a research context, fMRI is considerably more challenging. Edlow et al. 2023 claim TMS-EEG (the method used to calculate PCI) is superior to task-based fMRI and EEG because TMS-EEG “applies a perturbational rather than an observational approach, bypasses sensory systems, which can be impaired in patients with traumatic, hypoxic, and other forms of brain injury, bypasses motor systems and is thus not dependent on a behavioral output, requires no cognitive effort.”
The isolated forearm technique
(This one freaked me out) The Isolated forearm technique is a method where a patient is paralyzed except for one arm and then they are asked a question like “clench your fist if you can hear me”. A clinician at the conference told me 5% of people are thought to respond, and I later found this study that supports their claim. Paralytics are often used in surgeries where any subtle movements could jeopardize the surgery.
Why does monitoring anesthetic depth matter clinically?
Besides the obvious (let’s not operate on an awake person who can feel pain), better monitoring anesthetic depth, such that doctors can provide the minimally required anesthetic dose, could reduce adverse effects in patients — especially older patients who are more likely to experience postoperative delirium and cognitive decline. Lighter anesthetic depth has been found to reduce the risk of postoperative delirium and cognitive decline.
Other stuff
- Measuring L5 pyramidal burst firing via optogenetic stimulation: When the distal apical dendrites of layer 5 pyramidal cells are optogenetically depolarized, there is spiking at the cell body during awake conditions and no spiking during anesthesia. During anesthesia, layer 5 pyramidal neurons become synchronized, and this synchronization coincides with the loss and recovery of consciousness.
- Finding the starter motor for the engine of consciousness
The brainstem arousal nuclei showed an unexpectedly transient increase in coupling amongst themselves that occurred around the point of regaining responsiveness to verbal command. But this period of coordination between these nuclei lasted less than 2.5 min. It would appear that these nuclei are the starter motors, not the main engine of consciousness. Once the engine is running, these nuclei have no need for continued synchrony, so they become more functionally separate and revert to their ongoing, somewhat independent, roles.
- Temporal irreversibility of neural dynamics as a signature of consciousness
- Monitoring nociception during anesthesia.