Artificially intelligent consciousness meter

Project supervisors

Dr Levin Kuhlmann, Faculty of Information Technology (Main supervisor)
Prof Nao Tsuchiya, Faculty of Medicine, Nursing and Health Sciences

PhD project abstract

Worldwide over 230 million people undergo surgery involving anaesthesia every year. Despite the fact that pain is perceived in the brain, commercial monitors for determining depth of anaesthesia and consciousness levels during surgery are not broadly adopted. This is because they are based on a simplistic view of how consciousness emerges in the brain and do not reliably estimate consciousness levels for the full variety of anaesthetics used in clinical practice. Therefore, there is massive scope to develop an artificially intelligent consciousness meter that monitors the brain during general anaesthesia and surgery. This project will bring together the leading theory in consciousness science, Integrated Information Theory, and cutting edge graph neural network deep learning techniques and apply them to multivariate brain imaging data obtained during delivery of different anaesthetics in humans and sourced from top international labs. The result will be an anaesthetic depth/consciousness level meter that: works reliably for the full gamut of anaesthetics; leads to less cases of awareness during surgery by ensuring drug levels are high enough; reduces the occurrence of post-surgery cognitive deficits (occurs in up to 20% of the young and elderly) by ensuring drug levels are not too high; and minimises the use of anaesthetics that are rare or sometimes in short supply. Given the number of people undergoing anaesthesia during surgery every year, such an artificially intelligent consciousness meter also has high potential for significant commercial impact.

Areas of research

Digital Health, Artificial Intelligence, Data Science, Machine Learning, Integrated Information Theory, Consciousness Science, Depth of Anaesthesia Monitoring

Project description

This project will leverage the current understanding of consciousness science and state-of-the-art graph neural network deep learning methods to develop an artificially intelligent consciousness meter. Using multivariate brain imaging data (Electroencephalography – EEG; Magnetoencephalography – MEG) recorded during the intake of a variety of different anaesthetics in different patient groups, the consciousness meter will be designed to gauge levels of consciousness during general anaesthesia and surgery. Such a meter can be used to make sure patients are given enough anaesthetic so that they do not awaken during surgery, but also not too much anaesthetic that could lead to post-surgical cognitive deficits, or even in some very rare cases brain damage or death. Current commercial technology based on monitoring brain activity during anaesthesia is not widely adopted as it fails to work well for all anaesthetics and general anaesthesia is still strongly associated with post-surgery cognitive deficits in the young and elderly in around 20% of cases. A more robust consciousness meter is needed that works for different anaesthetics and drug combinations, and ensures that optimal drug levels can be delivered.

Current advances in consciousness science have put forward integrated information theory which proposes the level of consciousness can be captured by simultaneously maximising the integration and discriminability of information, and is referred to as Ф. While the exact theoretical definition of this measure is challenging to calculate for the whole brain, various approximate measures of Ф have been defined and applied to multivariate brain imaging data. These approximate measures have shown promise as candidate measures of consciousness level in different conditions such as anaesthesia, sleep, disorders of consciousness and epileptic seizures. The different measures are not always consistent with each other and have differing levels of efficiency and accuracy when it comes to approximating Ф. Thus a more universal approach to estimating Ф is required that is applicable in all scenarios to provide a more reliable measure of consciousness level.
With a view of developing a universal consciousness meter capable of tracking consciousness level during general anaesthesia with different kinds of drugs, this project will consider two different approaches: (1) The time series of different approximate measures of Ф will be calculated for diverse multivariate brain imaging datasets and graph neural networks will be employed to combine these measures at different spatiotemporal scales to derive a single measure of consciousness level; and (2) current graph neural network approaches will be modified such that the raw multivariate data is fed into the network and a single measure of consciousness level is derived at the output by designing the cost function and topology of the network such that the core concepts underlying Ф of simultaneously maximising the integration and discriminability of information are embedded within the process of how the network learns.

These novel approaches are expected to help solve the current problems with commercial monitors of brain activity for general anaesthesia which are based on too simple an understanding of how consciousness emerges in the brain. Given that the approach is based on world leading consciousness science, the method will also potentially hold promise for application of monitoring patients in the areas of sleep medicine, disorders of consciousness and epilepsy where the presence, absence, or alteration of consciousness plays an important role.

To achieve the aims of the project various datasets will be resourced from our labs in the Department of Data Science and AI and the School of Psychological Sciences and international collaborators from the University of Auckland (Suresh Muthkumaraswamy), Michigan University (George Mashour, UnCheol Lee), Ghent University (Michel Struys) and University of Milan (Marcello Massimini). This data includes EEG/MEG studies of humans during delivery of different anaesthetics including propofol, xenon, nitrous oxide, ketamine, isoflurane and sevoflurane. This set of anaesthetics covers a broad range of anaesthetics used during surgery that either have the effect of enhancing inhibition of the brain or suppressing excitation. Thus the data provides a strong base to develop a universal consciousness meter that works for different drugs.

The proposed project will bring together two world leading labs in Monash working in this area. The Kuhlmann run brAIn lab works on developing AI-based brain monitoring methods for anaesthesia and epilepsy and the Tsuchiya run Monash Neuroscience of Consciousness lab, works at the frontier of consciousness science with a focus on empirically validating the ideas of integrated information theory. By bringing together science and applied algorithms we will achieve more than we have on our own and deliver an artificially intelligent consciousness meter that will have a broad impact across the 230 million major surgical procedures being carried out under anaesthesia worldwide every year as recognised by the World Health Organisation. The novel meter will lead to less cases of awareness during painful surgery, reduced incidence and severity of cognitive deficits following surgery, optimised use of, sometimes rare, anaesthetics, and a meter that works as it is supposed to for different kinds of anaesthetics. Both of our labs are supported by ARC, NHMRC and other funding and contribute to the broader research community at Monash, in particular the MDFI theme of AI and Data Science in the Health Sciences and Monash Biomedical Imaging. We also have close ties to global giant Medtronic Public Limited and Australian depth of anaesthesia monitoring device company Cortical Dynamics Pty Ltd which may potentially assist with commercial and clinical translation.

PhD student role description

The student is expected, as described above, to take already existing multivariate brain imaging data and implement algorithms in the form of the two different approaches to combining the frontier consciousness science of integrated information theory and state-of-the-art graph neural networks to develop a reliable artificially intelligent consciousness meter that works for a variety of different anaesthetics. Algorithms will be trained using all data from all anaesthetics and compared to the situation where algorithms are trained on individual anaesthetics to determine if there are benefits to training on all anaesthetic types or if an ensembling approach that combines models trained separately on different anaesthetic types would be more optimal. To enable prediction of consciousness level from the raw brain imaging data, consciousness level/anaesthetic depth will be associated with the surrogate measure of responsiveness level determined from behavioural and responsiveness tasks that were obtained during the recordings of the brain imaging data. The student will be guided by the project supervisors who have made exciting discoveries in consciousness science, AI-based brain monitoring and depth of anaesthesia monitoring, and also have the chance to collaborate with, and potentially visit, world leading groups at Monash and globally as mentioned above. This project will provide a diverse set of opportunities and skillset for the student involved to branch their career into different directions in psychological science, data science and AI, digital health or biomedical imaging and engineering in academia and/or industry. The skillset covered will be a deepening of the recommended skills listed below. Maybe even one day the artificially intelligent consciousness meter the student develops for the brain could be modified to develop a method to decide the consciousness level of an artificial intelligence.

Required skills and experience

Recommended skills span but do not necessarily require all of the following: Python, R or Matlab Programming; Tensor flow; PyTorch; Artificial Intelligence, Machine Learning, Data Science, Deep Learning, Graph Neural Networks, Statistical Inference, Time Series Analysis, Signal Processing, Information Theory, Neuroscience, Brain Imaging.