NORCE has worked with medical imaging for many years, with a main focus on image recording, segmentation and data-driven flow simulations based on medical imagery. Among other things, image registration has been used to develop physically motivated deformation models to recognize kidney fibrosis. Segmentation has, among other things, focused on multimodal MR images of gynecological cancer, where we recognize primary tumor in MR images using deep learning. Both of these focus areas aim to develop improved and more personalized models for diagnosis and choice of treatment method.
We also work with blood flow simulations in blood vessels in the brain, where we have developed a precise model for perfusion that is mathematically valid on multiscale simulations. We simulate the model of the geometry of an MRI image, in which we recognize arteries, veins and brain tissue, and thus achieve a more realistic model than would otherwise be the case. The aim of this model is to link it with programs for planning operations, and also to do parameter estimation in a better way than is done today. The work on parameter estimation is ongoing in NORCE and is very exciting in a clinical perspective to achieve better image markers used in diagnostics.