Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
MRI Acquisition Optimization: Active Learning of the Signal Sampling Strategy for Model- and Learning-Based Reconstruction
Usually, image reconstruction methods employ a predefined signal acquisition model. This project is aimed to optimize the signal acquisition directly to allow efficient reconstruction.
Keywords: image reconstruction, machine learning, linear algebra, reinforcement learning
Image reconstruction [1,2] algorithms provide an image estimate from the set of acquired measurements (i.e. slice projections in CT, spatial frequencies in MRI). Although such methods vary in their complexity and effectiveness, it is obvious that acquiring more informative measurements would help to improve reconstruction accuracy. Therefore, it is suggested to solve a two-level optimization problem: optimize signal acquisition in a way that an optimal reconstruction algorithm can resolve the ground truth in the most accurate way.
This opportunity encompasses a number of applications in cardiac MRI, for example:
- Optimization of k-space sampling trajectory for different modalities [3]
- Optimization of magnetic gradient waveforms for diffusion tensor imaging [4]
- Joint dictionary and encoding optimization for magnetic resonance fingerprinting, based on Bloch equations
Image reconstruction [1,2] algorithms provide an image estimate from the set of acquired measurements (i.e. slice projections in CT, spatial frequencies in MRI). Although such methods vary in their complexity and effectiveness, it is obvious that acquiring more informative measurements would help to improve reconstruction accuracy. Therefore, it is suggested to solve a two-level optimization problem: optimize signal acquisition in a way that an optimal reconstruction algorithm can resolve the ground truth in the most accurate way.
This opportunity encompasses a number of applications in cardiac MRI, for example: - Optimization of k-space sampling trajectory for different modalities [3] - Optimization of magnetic gradient waveforms for diffusion tensor imaging [4] - Joint dictionary and encoding optimization for magnetic resonance fingerprinting, based on Bloch equations
- Identify acquisition phenomena that have to be simulated (e.g. patient motion, eddy currents, noise)
- Identify physical constraints on optimization parameter (e.g. waveform energy, sampling budget)
- Define accuracy quantification protocol for model evaluation
- Implement differentiable acquisition simulation model in Tensorflow or Pytorch
- Joint optimization of acquisition and reconstruction model
- Application of findings in practical setup
- Write report and present results
- Identify acquisition phenomena that have to be simulated (e.g. patient motion, eddy currents, noise) - Identify physical constraints on optimization parameter (e.g. waveform energy, sampling budget) - Define accuracy quantification protocol for model evaluation - Implement differentiable acquisition simulation model in Tensorflow or Pytorch - Joint optimization of acquisition and reconstruction model - Application of findings in practical setup - Write report and present results
Dr. Valery Vishnevskiy vishnevskiy@biomed.ee.ethz.ch
Prof. Dr. Sebastian Kozerke
Dr. Valery Vishnevskiy vishnevskiy@biomed.ee.ethz.ch Prof. Dr. Sebastian Kozerke