MRI provides excellent soft tissue contrast, but is limited by extensive imaging times. We have developed the Recurrent Inference Machine (RIM), a physics-informed neural network for accelerating MRI. It has proven successful in learning domain invariant features. We hold a successful track record in participating in reconstruction challenges and have been winning in FastMRI knee and Calgary brain challenges, while being in the top three of the generalization track of the FastMRI Brain challenge.
Clinical decision making increasingly requires further quantification of the imaging data. We are developing end-to-end models for reconstruction and quantification, in parameter mapping and segmentation of structural and functional MRI.
70% of the world population has little or no access to 1.5T and 3T MRI facilities. Low-field MRI scanners find their application in specific clinical settings, are more affordable and require little support in installation and maintenance. Through deep learning techniques, we aim to accelerate the acquisition and efficiently denoise the images.
Together with the Netherlands Cancer Institute, we are developing methods for image guided radiotheraphy.