The project is part of a European collaboration between Sweden, the Netherlands, Belgium and Turkey, under the auspices of the European R&D&I cluster ITEA4. Sweden’s part in the ASSIST project is coordinated by researchers in the Department of Biomedical Engineering (IMT), led by Associate Professor Anders Eklund and Senior Lecturer Evren Özarslan. Anders Eklund points out that the research group applied for grants for this particular project for several reasons.
– It fits in well with our work, and we have been working since 2015 with similar projects and companies. So it seemed logical to continue when the other countries wanted to continue. The process has taken a long time, but it’s fortunate that we were awarded the grant.
Evren Özarslan adds that their success in the recently-completed project called IMPACT could have contributed to this decision.
– We were able to show that sophisticated methods in data analysis can make a difference in health care, for example, by revealing high-quality information from shortened magnetic resonance imaging sessions.
IMT’s part in ASSIST focusses on brain tumours and how queues in the medical care system can be reduced with the aid of computers. This is particularly relevant in the light of the huge care backlog created by the pandemic.
– We hope to achieve a reduction using several methods within machine learning (also known as deep learning). Instead of placing certain tasks onto a physician, you can train a computer to do them, and more rapidly. This allows the physician to focus on other matters, says Anders Eklund.
Five companies: RaySearch, Inovia, Spectronic, Scaleout and Eigenvision, are involved in the Swedish part of the project. This is a requirement imposed by ITEA4. The researchers must produce in collaboration with the companies what is known as a proof-of-concept, which must, in turn, be linked to the selected area. In LiU’s case, this is the brain.
– They want to see cases in which we show that the various parties have done something together, but the parties are then relatively free to work along other paths and with other parts of the body.
A major part of the project concerns what is known as federated learning. A general problem with machine learning in image analysis is that a large amount of training data is required. The process is made more difficult in the medical field by the need to follow ethical legislation and GDPR. This is why solutions based on federated learning are being developed. In brief, this means that the images never leave the hospital, which makes it easier for companies to gain access to the data they need to develop their products. Another line of research will look at synthetic images. Here, a computer is to be trained to create completely new images that do not belong to any specific person. This means that they can be distributed without risk of violating ethical rules.
– The principal goal is to save time, and you could say that the two tracks are different ways of achieving this, says Anders Eklund.
Yet another track focuses on MRI.
– By modelling the tissue structure and developing new measurement methods, we can more effectively use MRI images, which could obviate biopsy, says Evren Özarslan.
Anders Eklund finds working in the project gratifying since it will be relatively easy to determine clearly whether they have managed to give medical personnel more time for other activities or not. “Saving time is something that most people can relate to, and something that really makes a difference for many patients”, he continues.
– If we can help patients return to their normal life more quickly – well – it’s easy to see the benefits of that.
ASSIST will continue for three years, and the Swedish part has a budget of SEK 65 million. The project has received SEK 32.5 million from Vinnova, SEK 9.4 million of which will go to IMT.
Translated by George Farrants