Exploring undersampled MRI reconstruction using deep learning with images from non-healthy subjects

Description
In the recent years many works have been published demonstrating the capabilities of deep learning based methods for MR image reconstruction from undersampled k-space data, however mostly using images from healthy subjects and/or focusing on accurate reconstruction of healthy images. Given that the main pupose of imaging is for non-healthy subjects, working with such data is essential while exploring deep learning solutions to the problem. Furthermore, working with data from non-healthy subjects incorporating both healthy and non-healthy structures in the images allows for asking questions about more than only accurate reconstruction of structures.

In this project we need a Master student to work on MR image reconstruction related questions. The student will have the chance to work on a dataset, which is currently not publicly available and explore reconstruction in such a context.

In this project the student will:

- familiarize themselves with the related literature from our group as well as others
- implement diferent ideas in a deep learning framework such as Tensorflow, PyTorch etc..
- work on a non-public dataset
- asses/simprove the accuracy of existing methods for the dataset and extend the methods for further anlysis
- if successful, publish their findings.

Supervisors:
Kerem Tezcan (tezcan@vision.ee.ethz.ch), Prof. Ender Konukoglu (kender@vision.ee.ethz.ch)