Sometimes a donor kidney is rejected years after it is transplanted. Researchers at Radboudumc teaching hospital in Nijmegen are developing a computer model that can predict this process, so that those cases can be treated properly. They are working closely with a number of partners in the Netherlands and abroad.
A glass plate with slices of tissue from biopsies of a donor kidney lies on the table. These biopsies were taken for further investigation after the patient’s kidney function declined. Senior lecturer and computer scientist Jeroen van der Laak and PhD student Meyke Hermsen, both of Radboudumc’s Pathology department, explain that up to now, pathologists would always study this kind of thin-section under a microscope. They demonstrate the fact that this will soon be a thing of the past in Nijmegen by showing me an image on a laptop. Kidney biopsies digitalised using a special scanner appear on the screen. ‘Next year our department is going completely digital’, says Van der Laak. ‘This will have a major impact on our infrastructure and procedures. One change will be a bigger role for computational pathology.’
Researchers in Van der Laak’s computational pathology group are attempting to analyse and interpret digital cross-sections of biopsies using computer algorithms to aid diagnosis and treatment. With support from the ERACoSysMed programme, an international network of grant providers which includes ZonMw (see box), they are focusing on chronic rejection processes after kidney transplants, which are associated with interstitial fibrosis and tubular atrophy (IFTA), in which vital kidney tissue is replaced by non-functioning tubuli and scar tissue. The project has been named SysMIFTA because of its use of systems medicine to decipher the processes that lead to IFTA.
Kidney transplants have become steadily more successful over the years thanks to better matching between patient and donor, and the advent of more effective medication. Hermsen says that this means that acute rejection is now less common. ‘Unfortunately, we know less about the incipient processes that lead to IFTA. It can cause a donor kidney to be rejected years after transplant, meaning the patient has to be added to a waiting list that is already long. In SysMIFTA we are trying to get a picture of the rejection process. We have developed neural networks that help us analyse kidney tissue. Use of these models to teach a computer is known as deep learning.’
In deep learning, a computer algorithm is trained by being exposed to many examples of recognisable components. For SysMIFTA the model had to learn all the tissue structures in a kidney biopsy. It is now so successful that Van der Laak and Hermsen are also investigating whether the model can provide support for pathologists. Besides recognising structures, the focus in the SysMIFTA project is now also on detection of macrophages and T-helper lymphocytes which might together play a key role in the emergence of IFTA.
Van der Laak says that this study has only been possible thanks to close collaboration with international partners, because the mathematical model receives input from so many angles to help it make a good prediction. ‘SysMIFTA is a European project. We are working with centres in Germany, France and Italy. A centre in Israel is also participating. We in Nijmegen can use our own analytical techniques to study what macrophages are present in a transplant biopsy, where they are and how this relates to the outcome with the donor kidney. Other groups are looking, for example, at cell culture experiments, studying the polarisation of macrophages under certain conditions, like low oxygen resulting from the transplant. The results from the different groups are brought together in the mathematical model, the idea being that together we can offer more information than each discipline separately. So we complement each other nicely.’ The collaboration has now extended beyond ERACoSysMed. TU Eindhoven, Amsterdam UMC and the Mayo Clinic in Rochester, USA are also working with Nijmegen. ‘The idea is for our models to be universally applicable,’ says Van der Laak, ‘so that all patients all over the world can benefit.’
‘But we’re not there yet’, Hermsen points out. ‘The models can already recognise a lot but there is room for improvement. And we will have to consult with pathologists first to look at where our models can have most benefit. The algorithm can quantify a lot of structures, but we could for example remove structures not needed for a good diagnosis from the analysis.’ She is convinced it could eventually save pathologists a lot of time. ‘Pathologists will not only be able to work more efficiently, they will also make more objective diagnoses. And of course we hope that SysMIFTA will enable us to identify donor kidney rejection sooner and prevent it from happening. That is our ultimate goal.’
ERACoSysMed (ERA-network Cofunding of Systems Medicine) is a network that organises funding rounds to promote systems medicine in Europe. Systems medicine uses mathematical models and large quantities of clinical and biological data to enable more personalised care. Tailor-made treatment will be more effective and should have fewer side effects. To this end, ERACoSysMed encourages clinical practitioners, biologists, bioinformaticians, geneticists and pharmacologists to work together on an international basis. ZonMw is closely involved with this integrated approach, organising meetings and workshops, for example, to make participants more aware of the potential and usefulness of systems medicine.