Anmerkungen:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Beschreibung:
MRI is arguably the most versatile imaging modality for clinical use available today. In recent years, the development of advanced deep-learning-based reconstruction methods has significantly accelerated data acquisition in MRI, especially in standard scenarios where large amounts of training data are available. However, for many advanced applications, such as for quantitative MRI or for real-time imaging, the availability of training data is limited. Especially in the case of cardiac real-time imaging, the motion of the heart makes it difficult or impossible to acquire ground truth references for supervised learning. This thesis addresses these challenges by exploiting strong physical signal models for deep learning reconstructions in quantitative MRI and by developing calibrationless self-supervised learning strategies for real-time cardiac MRI. Moreover, a flexible and computationally efficient deep learning framework for reproducible MRI reconstruction is developed, on which the research in this thesis is based. ; 2025-05-21