Internship / External MSc Project at Philips Research Hamburg: Loss functions for MRI reconstruction

Techniques from Deep Learning have become a central building block in the development of medical image reconstruction algorithms. For example, neural networks have been used to accelerate and improve MR and CT reconstructions, thereby allowing reduction of measurement duration or radiation dose. A central issue in the training of these networks is the choice of the objective functions, or loss functions, that do not generally reflect human visual perception. The aim of this Master’s Project is to study and develop suitable feature-based loss functions, which can be used for improving deep-learning based techniques in medical image reconstruction. An initial direction of study would be the unsupervised training of a feature loss based on a database of MR images and the subsequent application of the resulting loss in downstream tasks such as image reconstruction or contrast-to-contrast mapping.


  • Computer Science, Biomedical Engineering, Physics or similar background
  • Strong background in machine learning
  • Interest in medical imaging
  • Proficient in Python
  • Experience with ML frameworks, e.g., PyTorch / Tensorflow / Keras

We offer

  • a close and personal co-supervision between TUM and Philips Research Hamburg
  • a three month paid internship in the Philips Research Lab in Hamburg, taking place in the beginning of the project
  • to work in an interdisciplinary team
  • to collaborate with international experts in machine learning and MR image reconstruction from academia and industry

How to apply

Please send your application via email to

Kerstin Hammernik
Kerstin Hammernik
Research Scientist

My research interests include inverse problems, MRI and machine learning