Brain Research and Artificial Intelligence – What Brain Scans Tell Us
Feature article by Prof. Dr. Simon Eickhoff
Your brain is truly unique, dear reader – just like your physical appearance, or your psychological makeup. Once it has been scanned in a conventional magnetic resonance imaging (MRI) scanner, any doctor can use reference images to show you certain distinctive features in the structure of your brain. Not all differences can be detected from such scans, however, even by a radiologist’s or brain researcher’s trained eye. Yet artificial intelligence (AI) is capable of doing so – and this capability has far-reaching implications.
Hirnscan verrät Alter, Geschlecht, Persönlichkeit
Differences that are not directly apparent include the networks formed by different brain regions when a complex task has to be solved – when you recognize a face, for instance, or make a mental note to remember something. We analysed MRI scans performed on hundreds of volunteers and determined their brain activity within functional networks. We then fed this data to self-learning AI software. We also provided the software with information about the people who had been scanned, such as their age, their gender, and the results of a personality test they had taken. We then trained the software to link the activity from the functional networks with the other data provided. The software learned from this training and was subsequently able to make predictions about further volunteers on the basis of their brain scans alone. In this way, we succeeded in using AI to determine volunteers’ ages based only on the scans, with a maximum margin of error of five years. We could also successfully predict how a volunteer would perform in psychological tests assessing their openness, agreeableness, and emotional stability.
What this boils down to is: show the AI your brain, and it’ll show you who you are. This research has particular significance for medical applications.
Hope for the future: AI could predict course of illnesses
One example is depression: up to 30 % of those who recover from severe depression will suffer a relapse. Many patients in recovery are therefore very anxious to find out if they are likely to have another episode. At the moment, however, no doctor can reliably predict this. Our hope for the future is that it will be possible to provide individual predictions using brain scans and AI, i.e. to calculate the probability that a particular person will relapse. Another example relates to predictions for neurodegenerative diseases such as Parkinson’s. How will the individual’s mobility be affected? Will they suffer from dementia? To explain our idea for this approach, imagine the following scenario: outwardly, two patients appear to be in the same condition, but AI detects differences in their brain scans. These differences suggest that one patient’s condition should actually be better than it appears, while the other’s should be worse. This could indicate – before the symptoms become apparent to doctors – that the disease will take a different course in each patient. In the best-case scenario, this would allow doctors to adapt their treatment accordingly at an early stage.
Broad range of applications
Although our research focuses on clinical application, the processes we are developing are fundamental in nature: the outcome of the predictions is limited only by the training data available. As a result, brain scans could also one day replace the conventional tests and questionnaires that are used for recruitment processes in assessment centres, for example. And it’s precisely because of this enormous potential that we take very seriously any criticism or concerns about potential misuse by insurance companies or authoritarian states.
However, the consequence of this cannot be that we in Germany, and particularly at publicly funded institutions like Forschungszentrum Jülich, stop all research on topics like these. Instead, we must aim to shed light on the possibilities and limitations of AI.
Further information: