Computer Vision News - May 2022

50 Congrats, Doctor! a quantitative reconstruction (i.e., maps with physical T1 and T2 relaxation times) of MR data, as well as to facilitate the workflow of dynamic cardiac MRI. Using such quantitative and dynamic MRI instead of the conventional qualitative MRI is beneficial for a thorough analysis of tissues (e.g., differentiation of normal and pathological tissues), as well as for the simultaneous evaluation of the function of moving tissues like the heart. In a first project, DL for the reconstruction of so-called MR Fingerprinting (MRF) data was applied. MRF acquires multiple image contrasts of the same tissue with strong undersampling by modifying the parameters for each acquisition. The aim is to generate characteristic signals (so- called fingerprints) for different tissues. The State-of-the-Art (SOTA) reconstruction compares the measured fingerprints with a presimulated dictionary to derive the underlying quantitative maps of T1 and T2 values. This is a highly non-efficientmethod due to the exhaustive search. To overcome theselimitations, recurrentneuralnetworks (RNNs) for the MRF reconstruction [ 1 ] were introduced, which simultaneously handle spatial and temporal correlations Elisabeth Preuhs (née Hoppe) recently completed her PhD at the Pattern Recognition Lab at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) in close collaboration with the Magnetic Resonance (MR) predevelopment of Siemens Healthineers. Her research focused on novel methods for multidimensional MR data to enable an extensive and quantitative analysis of, e.g., brain tissues or the beating heart. For her pioneer work in this field, she was awarded as an Artificial Intelligence Newcomer of 2019 by Deutsche Gesellschaft für Informatik, as well as with the Innovator Award 2020 from the FAU. Currently, she is working in the innovation department of Siemens Healthineers for minimally invasive interventions in the scope of new robotic-assisted workflows. Congrats, Doctor Elisabeth! The application of deep learning (DL) increased tremendously in recent years. In my PhD work I made use of these technique for the improvement of medical MR Imaging (MRI), especially to enable