Efficient use of uncertainty for medical procedures

"Efficient use of uncertainty for medical procedures" proposes to integrate uncertainty information about medical image processing methods completely into the clinical work flow. This requires to solve computational problems that arise when different algorithms try to monopolize the available hardware. These problems becomes even more demanding for algorithms that are accelerated by Graphics Processing Units (GPUs). To overcome the current shortcomings of recent programming methods for GPUs, the fellow will use a novel GPU-accelerated framework, which he has developed during his previous career together with his colleagues form Graz University of Technology as a base. During the proposed project, the fellow will first adapt this framework to make scheduled and prioritized execution of GPU-accelerated medical segmentation and registration algorithms feasible, subsequently apply and evaluate this framework for existing algorithms. To proof the flexibility and applicability of the proposed methods, the fellow will develop a novel Magnetic Resonance Imaging (MRI) scanner steering method, which is based on uncertainty information and which will allow to evaluate the organs of a moving foetus within the uterus. In contrast to Ultrasound-based imaging, the proposed new method will allow an earlier and more accurate diagnosis of malformations and foetal diseases and will enable treatments in time. Thereby, the fellow will draw on his rich experience with GPU accelerated three-dimensional and four-dimensional medical volume processing. During the return phase these results will be used for novel generic real-time visualization of uncertainties in medical scans, which are currently neglected during medical standard procedures. Thereby the fellow will develop unseen interactive visualization techniques for the proposed prenatal MRI sequence and other standard scan methods, as they are for example also used for adult subjects.