Abstract
Longitudinal magnetic resonance imaging (MRI) is a key imaging technique for monitoring progression of diseases and evaluating the effects of treatment and therapy. Unfortunately, MRI is inherently slow, which is in particular disadvantageous for longitudinal follow-up. Acceleration of MRI scans can be achieved using state-of-the-art techniques such as compressed sensing, parallel imaging, and, more recently, deep learning. However, these techniques almost exclusively focus on the acceleration of cross-sectional MRI, leaving the temporal dimension for acceleration unexplored. Moreover, processing of longitudinal images is done in a multistep way (i.e., reconstruction of cross-sectional images followed by deformation estimation) with a high risk for error propagation. In this project, the longitudinal MRI workflow is revisited, and a radically new approach (Delta-MRI) is developed. Our novel approach relies on directly estimating anatomical changes (i.e., the 'delta') from a reference image and a strongly accelerated follow-up scan, thereby avoiding time inefficiencies and loss of accuracy caused by error propagation in the current multistep approach. Delta-MRI is envisaged to provide a five-fold reduction in acquisition time for follow-up scans without compromises in accuracy, which will leverage the clinical potential of longitudinal MRI.
Researcher(s)
Research team(s)
Project type(s)