If the therapy is meant to relieve pain, we ask the patient if it hurts less. When we try to restore consciousness, our question is what we should measure, since we can’t see the problem. How can we ask a question to a patient we are not able to communicate with? The scientists working at the Alarm Clock Clinic have to face these questions every day
Most people associate the Alarm Clock Clinic with children in a coma. They imagine white hospital rooms and little patients lying in small beds in silence…
“It’s far from that,” says Anna Duszyk*, a psychologist and a member of the interdisciplinary team conducting research on children in the state of impaired consciousness.
“People usually think that children at the Alarm Clock Clinic are in a coma. But that’s not true. This erroneous image comes from the lack of precise terminology in the Polish language describing children with disorders of consciousness. In reality, the patients at the Alarm Clock Clinic are referred to the facility only after coming out of a coma. Most of them are quadriplegic and spastic. Additionally, many display involuntary movements. Often they are so active that we find it difficult to run the tests.
Consciousness disorders include coma, vegetative state (also known as unresponsiveness), and state of minimal consciousness.
Coma is a state of deep consciousness impairment caused by severe brain injury, as a result of an accident, illness or poisoning leading to a brain damage, in which patients do not respond even to strong stimuli. Their brain metabolism is reduced to 30 percent. They are asleep. They do not open their eyes and nothing can wake them up despite our talking to them, touching them and stimulating them with pain inducing activity. Up to 5,000 children are estimated to fall into a coma in Poland every year, but most of them regain consciousness after a few days.
However, several hundred children do not awaken and progress to a vegetative or minimal consciousness state and remain that way for a number of years. The Alarm Clock Clinic was founded to help them by providing intensive rehabilitation and therapy for 12 to 15 months.
So far, the staff of the clinic have managed to bring 52 children back.
Can this kid hear and understand me?
Each of the children at the Alarm Clock Clinic are in a different condition. All of them open their eyes and some of them are capable of performing various movements. Yet, it is hard to tell who is reacting intentionally and who is displaying involuntary movements. Everyone is asking a question: “Can this kid hear and understand me? If so, to what extent? What are the chances for them to get better?” The doctors don’t always have a ready answer.
“Based on behavioral test results, we divide our patients into two groups,” says Anna Duszyk.
“The first group consists of patients in a vegetative state. They display reflexes only. If we clap our hands above their heads, they flinch. If we stimulate them with pain, they cringe. They respond automatically, but they stay unconscious.
Others are in a state of minimal consciousness. When we ask them to look at the ball, they do. They can follow the ball with their eyes. When we pinch them, they try to locate the source of pain.
But we can’t communicate with them. If we ask them to move their hand, they do it but if we tell them to move their hand to say “yes”, they are unable to follow our request. It seems as if they are aware of what’s happening around them but they can’t do more complicated tasks, for example answer questions.”
We are still not able to objectively estimate minor changes in brain activity that lead to the revival of normal brain functions. We need to develop a new diagnostic method
“The biggest challenge is to separate the former from the latter, the ones that are awake but unaware from the ones that are awake and partially aware.
Additionally we test other options of communication to determine if anyone diagnosed as being in a vegetative state may be fully aware and capable of communicating,” explains Duszyk.
It is estimated that as many as 40% of diagnoses are wrong.
“A patient with severe motor cortex damage is not capable of executing any tasks involving any movement,” explains Duszyk.
“If someone can’t see, he or she is not able to complete any task involving this particular sense. We have kids who can’t hear, we have children with nystagmus and aphasia, who don’t understand what we are saying to them. All this makes it extremely difficult to make a correct diagnosis.”
Thoughts leave electric traces
“If you want to treat your patients, you have to monitor the progress on an everyday basis. Otherwise, we won’t know if the treatment works,” says professor Piotr Durka**, Faculty of Physics, University of Warsaw. Since 2016 his team have been doing research on possibilities to assess the level of consciousness in the case of children at the Alarm Clock Clinic.
If the therapy is meant to relieve pain, we ask the patient if it hurts less. When we try to
restore consciousness, our question is what we should measure, since we can’t see the problem. How can we ask a question to a patient we are not able to communicate with?”
Unfortunately, we are still not able to objectively estimate minor changes in brain activity that lead to the revival of normal brain functions. We need to develop a new diagnostic method that would allow to monitor changes in patient’s condition during rehabilitation and to find the best treatment program.
Professor Durka also works on brain-computer interfaces (BCI), i.e. systems of communication that is based on brain activity readings and that bypasses muscles. Professor Durka is a pioneer of the research in that field in Poland. In 2008, for the first time in history, he organized a public presentation of a brain-computer interface and demonstrated how it worked in practice. In his studies, he uses machine learning algorithms.
It turns out that certain mathematical methods developed for this purpose can also be used to analyze the EEG recordings of patients suffering from consciousness disorders.
Why EEG? There are many methods for testing brain activity, but most of them are expensive and difficult to use with children with impaired consciousness. For example, functional nuclear magnetic resonance imaging requires a patient to be transported to a specialized facility, and, in the case of patients in states of consciousness impairment, to be put to sleep during the examination. The EEG device is much cheaper and portable; it can be placed right next to a child’s bed.
“EEG is an electric trail of a thought, registered at the head’s surface,” clarifies Durka. “It allows us to read user’s conscious intentions, which the BCI translates into computer controlling commands,” he explains.
For example if we have YES and NO answers available on the computer screen, we can choose the one we want and concentrate our efforts on it. The BCI system will use EEG signal to read our intention.
“The goal of our research at the Alarm Clock Clinic is to look for certain indicators that are characteristic for the state of consciousness and that distinguish it from the state of unconsciousness,” says Durka. “The most important indicator is a conscious reaction to stimuli. We are trying to record characteristic patterns of brain’s activity responsible for conscious perception. Even if it is not demonstrated in the behavior, it may still go on in the brain. We „just” need a way to read it.
New AI pathways
Unfortunately, the interpretation of a registered EEG signal is extremely difficult. That’s where some mathematical methods come in handy. They are similar to those used for reading our intentions with the use of EEG to control brain computer interfaces.
Does AI also help in diagnosing?
“It is a difficult question and a delicate subject,” says Durka. “In recent years AI has been a popular and overly used term. Of course, similarly to other science fields, machine learning algorithms are also used in EEG analysis in order to evaluate data to find correlations that stay hidden for classic statistics. But the results of such new methods have to be interpreted with caution – especially when the result is to be used to diagnose patients in a critical condition. That is why we are trying to base our research, aimed at supporting clinical diagnostics, on proven and stable statistical methods and to use machine learning mainly in innovative BCI university studies,” spells out professor Durka.
We are touching on one of the unsolved mysteries – the essence of consciousness. But we are driven by much more than scientific curiosity. A progress in these studies may lead to huge improvements in diagnostics, patient care and rehabilitation planning for patients with impaired consciousness
AI cannot be used without limitations
“To effectively apply machine learning we need enormous sets of uniform, consistent and correctly marked data,” explains Durka. “Recognizing cats in pictures is relatively easy because the internet is full of images of those animals and a cat is a concept that is fairly well defined.”
What is simple for the facial or animal recognition technology gets more complicated in medicine. “EEG of various persons is much less homogenous than cats’ images. Besides, EEG of the same person on each „picture” (i.e. on consecutive recordings) may look different, which makes searching for subtle differences coming from a particular medical conditions much more complicated,” stresses Durka.
“Our last article describes an example of application of machine learning. We were searching for entirely new indicators that could be identified on the basis of EEG, hoping they would be helpful in the state of consciousness assessment,” professor says.
“Instead of analyzing traces of conscious reactions to stimuli we know from BCI, we studied the network of connections, independent of the task performed, called Default Mode Network (DMN). This is a fairly new and unexplored concept, so we don’t know exactly how the accurate functioning of this network should manifest itself in EEG. We calculated the strength of these connections between EEG electrodes, hoping we would see some differences in the structure of the connections between healthy persons and patients with various degrees of consciousness impairment. These connections can be calculated between different electrode groups and we don’t know prior which of those would be characteristic for various patient groups. At first and even second glance we couldn’t see any clear patterns, so we applied machine learning algorithms, which brought back interesting results.
One of the ways to check how this method would classify a „new” patient is cross validation. Two data subsets, a learning subset and a testing subset, are created on the basis of all data available. The learning set is to create („teach”) a classifier. Correct operation of the classifier is checked with the testing set, i.e. the data that have not been used for learning,” professor explains.
“However, it needs to be emphasized that these particular tests are at this point far from clinical applications. In this case we are testing an unknown phenomenon (DMN in consciousness disturbances) with complex machine learning methods we understand much less than classic statistics tools which have been used for decades. As far as the potential diagnostic value of these methods is concerned, we need to err on the side of caution.
A mystery of consciousness
Testing at the Alarm Clock Clinic started in 2016. Since then 30 children aged 7 and up have been tested multiple times. How does the testing look like in practice?
“We place a cap with EEG electrodes on the child’s head and explain what the task to be performed by the child consists in,” says Anna Duszyk. “Performance of the task must translate into a readable EEG signal. The task doesn’t require any behavioral reaction. All the patient has to do is to concentrate on a stimulus applied or to think about performing the movement he or she has been asked to make.
Unfortunately, most of the times the signal is of poor quality as it is difficult for kids to lie steadily in one position. Additionally, every two months, children are tested in their sleep. A healthy brain works differently when we are asleep and when we are awake. We can observe this in EEG recordings. Night time recordings help us look for the traces of daily rhythms and structures characteristic for the sleep as their presence is considered a sign of correct brain functioning.”
Despite multiple obstacles, sometimes the researchers are positively surprised with the test results.
“I remember a boy who turned out to be conscious although he wouldn’t give any behavioral signals. It was amazing,” says Duszyk. “The boy was 7 years old, he had almost drowned. We asked him to count pictures, to move his arm and leg. We didn’t see any reaction during the tests, but the EEG signal led to the conclusion he had understood the task and tried to complete it. That happened two or three times. This is the best proof of how important these tests are and how often we are wrong about patient’s consciousness assessments based on behavioral tests.
All the tests touch on one of the biggest unsolved mysteries, which is the essence of consciousness. But there is more to what we do than scientific curiosity,’ adds Durka. “The progress in our studies could allow us to profoundly improve diagnosing, care and rehabilitation planning for patients with consciousness impairment. Provided the algorithms detect traces of conscious responses in a patient’s EEG, it will be possible to communicate with them through BCI.”
*Anna Duszyk is a graduate of cognitive neuroscience studies at SWPS University of Social Sciences and Humanities, and of musicology at Jagiellonian University. She completed interdisciplinary PhD Studies at SWPS. She specializes in the dynamics of cognitive control processes and in psychophysiological issues connected with the construction of brain-computer interfaces. Since 2009 she has been collaborating with the Biomedical Physics Division of the University of Warsaw. She served an internship in Leibniz Institute for Neurobiology in Magdeburg and in Le Centre hospitalier universitaire de Liège, where she worked with the Coma Science Group led by professor Steven Laureys. Since 2016 she has been involved in professor Durka’s project, drawing on her Belgian experience in behavioral and electrophysiological studies to help patients with impaired consciousness and in BCI use to communicate with brain damage patients.
**Professor Piotr Durka – a physicist and neuroinformatics specialist, the author of multiple publications and five books. Since the beginning of his professional career he has been inextricably linked with the Faculty of Physics at the University of Warsaw. His field of study is focused on EEG analysis, brain-computer interface (BCI) and assessment of the state of patients with impaired consciousness (recently in cooperation with the Alarm Clock Clinic). In 2008 he organized the first public BCI presentation. Four years later, the system he had designed proved to be the fastest BCI at the international CeBIT fair. In 2009, at the Faculty of Physics at the University of Warsaw, he launched the first complete neuroinformatics studies program in the world. In 2012 he founded BrainTech, a firm which three years later implemented an open and free software system for the disabled with extremely limited communication capabilities (Polish Interactive Alternative Communication System PISAK). Currently, BrainTech is introducing innovative brain-computer interface hardware and software. For more information, please visit http://durka.info and https://braintech.pl.