Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Learning Deep Priors for Image Reconstruction
Dynamic Magnetic Resonance (MR) imaging offers exquisite views of cardiac anatomy and function. The objective of this project is to develop and implement methods that allow learning a data model from large sets of training data to be used in nonlinear data recovery from highly undersampled MR data u
Keywords: Magnetic Resonance imaging, cardiac imaging, image reconstruction, inverse problems, sparsity, machine learning, optimization, deep learning, tensorflow
Many methods for image reconstruction (MRI, ultrasound) and inverse problems can be implemented via iterative numerical scheme. These iterations can be explicitly unfolded and treated as a deep (tunable) network with tunable parameters. Such approach allows to learn optimal prior information for reconstruction.
Many methods for image reconstruction (MRI, ultrasound) and inverse problems can be implemented via iterative numerical scheme. These iterations can be explicitly unfolded and treated as a deep (tunable) network with tunable parameters. Such approach allows to learn optimal prior information for reconstruction.
The objective of the present project is to develop deep variational network for undersampled cardiac MR reconstruction. Simulated and in-vivo MR data of the heart are available to test and validate the method.
The project entails:
-Implementing unfolded gradient descent iterations for MR reconstruction with Tensorflow framework;
-Training parameters of the unfolded scheme;
-Reconstruction of highly undersampled clinical imaging data of the heart.
Requirements:
- programming experience in Matlab and Python (numpy),
- basic knowledge of direct and iterative methods for solving system of linear equations,
- basic knowledge of gradient descent optimization method.
The objective of the present project is to develop deep variational network for undersampled cardiac MR reconstruction. Simulated and in-vivo MR data of the heart are available to test and validate the method. The project entails: -Implementing unfolded gradient descent iterations for MR reconstruction with Tensorflow framework; -Training parameters of the unfolded scheme; -Reconstruction of highly undersampled clinical imaging data of the heart.
Requirements: - programming experience in Matlab and Python (numpy), - basic knowledge of direct and iterative methods for solving system of linear equations, - basic knowledge of gradient descent optimization method.
Supervisors: Dr. Valery Vishnevskiy (vishnevskiy@biomed.ee.ethz.ch); Prof. Dr. Sebastian Kozerke (kozerke@biomed.ee.ethz.ch)
Supervisors: Dr. Valery Vishnevskiy (vishnevskiy@biomed.ee.ethz.ch); Prof. Dr. Sebastian Kozerke (kozerke@biomed.ee.ethz.ch)