Maximilian Riehl

Thesis Topic: Model-based Deep Learning Reconstruction of Cardiac T1 and T2 mapping

Supervisors: Marc Vornehm, Florian Knoll

Description

Cardiac T1- and T2-mapping can serve as a valuable non-invasive tool for diagnosing a variety of cardiac pathologies but are challenged by limited acquisition time due to cardiac motion. Undersampling k-space offers a solution to increase resolution despite short acquisition times. Deep learning-based methods combined with physics models have shown promising results in quantitative MRI reconstruction. In this work, the incorporation of physics-based models into deep learning-based reconstruction methods for cardiac T1- and T2-mapping is evaluated utilizing the publicly available CMRxRecon 2023 dataset. A T1 inversion recovery signal model and T2 relaxation model are incorporated into an unrolled network architecture for parametric mapping from undersampled k-space. Furthermore, relaxation model-based loss functions are explored.