-
Research
Brain research with artificial intelligence
Research
Brain research with artificial intelligence
Jülich scientists show that it is possible to obtain information from brain scans about personality traits and mental illness.
show
Jülich scientists show that it is possible to obtain information from brain scans about personality traits and mental illness.
Our brain is highly complex. In everything we do, different brain regions work together in so-called functional networks. Prof. Simon Eickhoff, Director at the Institute of Neurosciences and Medicine (INM-7), deals with these networks and their activity patterns. The neuroscientist has a big goal: he wants to assess the extent to which these patterns have changed in people suffering from depression, schizophrenia or Parkinson’s disease. He hopes that, based on this information, the further progression of a disease can be predicted individually.
Image above: Simon Eickhoff wants to elicit information from the human brain about the future course of mental and neurological diseases.
Eickhoff and his team have investigated how artificial intelligence can be used to obtain information from brain scans. For this purpose, the researchers trained a self-learning software to predict the personality traits of people on the basis of their brain scans. The programme delivered promising forecasts for three out of five features: openness, agreeableness and emotional stability. Using the software, the researchers also found that certain functional networks in the brains of people suffering from schizophrenia or Parkinson’s are disturbed. effzett has spoken with Mr. Eickhoff.
To what extent could people with mental or neurological diseases benefit from your research in the future?
Let’s take depression: up to 30 per cent of all people who have recovered from severe depression fall ill again later. For many patients, it is very important to know if indeed this will happen to them. However, no doctor can give them a reliable prediction at present. We think that our methods have the potential to give an individual prognosis, that is, provide the individual probability of a relapse.
Another example is Parkinson’s disease. For example, many patients want to know whether they will develop dementia in the foreseeable future – a question that is of great interest to relatives as well. I see a very important field of application for prognostic methods here as well.
How is that going to work?
The idea can be explained with the help of two patients whose current state of disease appears to be the same, but for whom our algorithms indicate differences based on brain scans. They suggest that the condition of one patient should actually be better, while the condition of the other patient should be worse. This means that the algorithms may indicate different courses of the disease even before the symptoms can be recognised by the doctor. Such differences in the functional networks could be made visible from MRI images by new methods of machine learning. In the best case, doctors could deduce the further course of the disease from this and treat patients accordingly.
I would like to emphasise that we are still a long way from the clinical application of our methods. However, especially the approaches taken for network characterisation and individual prediction are already progressing very promisingly.
But there will certainly be people with depression or Parkinson’s who don’t want to know their future …
Yes, and I am an advocate of the right not to know. On the other hand, however, as a doctor, I have learned that the question as to what comes next is the most important and urgent question for most patients and their relatives. The reason for that is that many psychiatric and neurological disorders progress in phases, such as schizophrenia, or they progress slowly, such as dementia.
Could your results also change the treatment of mental and neurological illnesses in the future?
I very much hope so. Another problem is that not every medication helps every patient. Not every depressive person benefits from every antidepressant, for example. The situation is similar with schizophrenia. In the case of unsuccessful treatment, something new has to be tried out, which is very unsatisfactory. Our goal would be to find signatures in the brain that tell us that we should best treat these patients with drug A and not with drug B. However, we are still at the beginning in this.
You use machine learning methods for your approach. There is criticism and concern about the use of artificial intelligence and the possible misuse of patient data. What is your opinion?
In any case, these fears must be taken seriously. US insurance companies, for example, are currently considering evaluating brain scan data. They want to calculate the risk of illness for individual people and I fear that they want to exclude risk cases with foresight. Or think of the “Social Credit System”, which is already being tested in China: the state evaluates the behaviour of citizens with the help of artificial intelligence.
However, the consequence cannot be that we in Germany stop research on these topics. Here in Jülich, we have the opportunity to shape developments in the healthcare sector and to make the possibilities and limits of these technologies transparent. The research centres of the Helmholtz Association are committed to society: We do not do research to maximise profits, but to find answers to major societal challenges. We should seize this opportunity.
Interview: Frank Frick
From brain scan to prognosis
1
Identifying functional networks
What is it?
Functional networks consist of different brain regions that work together when we perform complex tasks. For example, when we recognise faces or memorise something. From the individual characteristics of such networks, conclusions can be drawn about mental illnesses as well as cognitive performance or personality traits of a person.What have the Jülich researchers done?
They evaluated several thousands of studies from all over the world in which the brain activities of test subjects were examined using functional magnetic resonance imaging. This enabled the Jülich researchers to precisely identify functional networks in the brain – the basis for determining future connection patterns of these networks individually and using them to predict personality traits or diseases such as Parkinson’s and schizophrenia.2
Brain scans with the help of functional MRI
What is it?
Functional magnetic resonance imaging (fMRI) produces special images of the brain. These make brain regions visible that are active at the moment of exposure. Different regions are active depending on what we do – whether we are moving an arm or simply lying still.What have the Jülich researchers done?
They evaluated fMRI images of hundreds of subjects who gave free rein to their thoughts during the scans. The researchers determined the brain activities and interactions in the previously identified functional networks.3
Self-learning software calculates prognoses
What is it?
Self-learning software is not programmed to solve a problem, but is trained with the help of data – to recognise patterns in fMRI data, for example.What have the Jülich researchers done?
They used such a software to assign certain personality traits to activity patterns in the functional networks. These characteristics were derived from personality tests that the test persons had previously completed. The researchers trained the software on data sets in which personality traits or diagnoses were known. The software then adapted its mathematical model and was finally able to predict these characteristics in new people.In another study, it was possible to determine from fMRI images whether a person suffered from schizophrenia or Parkinson’s disease or whether he or she was mentally healthy.
© 2022 Forschungszentrum Jülich