MSc Thesis: Physics-Driven Self-Supervised Learning in MRI Reconstruction

Machine learning has evolved tremendously to accelerate the inherently low acquisition process of Magnetic Resonance (MR) images. However, it is challenging to obtain ground truth data for learning MRI reconstruction. The objective of this MSc thesis is to explore self-supervised learning for MRI reconstruction, where only the measurement (k-space) data and knowledge about the acquisition physics are available. The tasks are to get an overview of the field (literature review), to test existing methods, and to develop novel methods on MRI data.


  1. Computer Science, Biomedical Engineering or similar background
  2. Strong background in machine learning
  3. Interest in medical imaging
  4. Proficient in Python
  5. Experience with ML frameworks, e.g., PyTorch / Tensorflow / Keras (optional)

We offer

  1. a close, personal supervision
  2. to work in an interdisciplinary team
  3. to collaborate with international experts in machine learning and MR image reconstruction.
Kerstin Hammernik
Kerstin Hammernik
Research Scientist

My research interests include inverse problems, MRI and machine learning