Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Segmentation and in-vitro Evaluation of 5D Flow MRI
5D Flow MRI offers novel ways of assessment of cardiovascular diseases through insights in blood flow hemodynamics in vessels such as the aorta. Emerging techniques include non-invasive estimation of turbulence and motion states[1] and improved estimation of pressure drop[2] in stenotic flows.
4D Flow MRI[3] evolved from Phase-Contrast MRI which uses motion-sensitizing magnetic field gradients to image blood flow over time non-invasively. The latest extension 5D Flow MRI offers the possibility to encode rich information about the blood flow in the acquired 3D volume, including mean velocities, information about turbulence (in stenotic jets arising from heart valve diseases for example) and the breathing motion over time. As Flow MRI has become a recommended feature in diagnosing heart valve diseases[4], to satisfy clinical needs, Compressed Sensing (CS) techniques[5] are employed to reduce the time duration needed for Flow MRI examinations. Specialized trajectories are described for correct quantification of peak flows with even higher undersampling factors[6]. For reconstruction, machine learning techniques are used to further improve image quality and practicality[7]. In order to describe the encoding process theoretically, we conduct Computational Fluid Dynamics (CFD) simulations, simulate the MRI experiment in-silico and perform in-vitro experiments using flow phantoms[8].
At the Cardiovascular Magnetic Resonance Group (CMR, Institute for Biomedical Engineering, ETH Zurich) we focus on the development of fast sampling strategies for 5D Flow MRI, improved reconstruction techniques using machine learning and simulations mimicking the encoding process in Flow MRI. The proposed projects play an essential role in validating and evaluating 4D Flow MRI data in future research projects of our group.
**Segmentation**
Quantification of 5D flow MRI often requires 3D delineation of the vessel wall, called segmentation. This is usually done on static, time-averaged images. Some 5D flow MRI parameters however, are sensitive to the exact vessel wall location, which can be a problem as the aortic root moves about 8 mm within each heartbeat[9]. This is when time-resolved 3D segmentations can become necessary and even useful in order to derive additional parameters such as vessel wall elasticity, a parameter directly related to cardiovascular disease. Usually 3D segmentations are done using manual or semi-automatic tools. Time-resolved 3D segmentations have therefore been very limited or impossible, as such a task can take up to 13 hours of manual work. For segmentation machine learning methods have proven essential. Convolutional neural networks (CNN), like for example the UNet[10], allow for voxel-based classification and localization of tissue. Similar approaches can be tested in this project to achieve fast and accurate time-resolved 3D segmentations of 4D flow MRI data.
**Flow Phantoms**
In order to validate theoretical and in-silico results, we employ flow phantoms in in-vitro experiments. Phantom geometries range from simple straight stenotic tubes[8] to realistic, compliant aorta geometries[11], depending on the focus of the investigation, to which steady and pulsatile inlet conditions can be applied.
4D Flow MRI[3] evolved from Phase-Contrast MRI which uses motion-sensitizing magnetic field gradients to image blood flow over time non-invasively. The latest extension 5D Flow MRI offers the possibility to encode rich information about the blood flow in the acquired 3D volume, including mean velocities, information about turbulence (in stenotic jets arising from heart valve diseases for example) and the breathing motion over time. As Flow MRI has become a recommended feature in diagnosing heart valve diseases[4], to satisfy clinical needs, Compressed Sensing (CS) techniques[5] are employed to reduce the time duration needed for Flow MRI examinations. Specialized trajectories are described for correct quantification of peak flows with even higher undersampling factors[6]. For reconstruction, machine learning techniques are used to further improve image quality and practicality[7]. In order to describe the encoding process theoretically, we conduct Computational Fluid Dynamics (CFD) simulations, simulate the MRI experiment in-silico and perform in-vitro experiments using flow phantoms[8].
At the Cardiovascular Magnetic Resonance Group (CMR, Institute for Biomedical Engineering, ETH Zurich) we focus on the development of fast sampling strategies for 5D Flow MRI, improved reconstruction techniques using machine learning and simulations mimicking the encoding process in Flow MRI. The proposed projects play an essential role in validating and evaluating 4D Flow MRI data in future research projects of our group.
**Segmentation**
Quantification of 5D flow MRI often requires 3D delineation of the vessel wall, called segmentation. This is usually done on static, time-averaged images. Some 5D flow MRI parameters however, are sensitive to the exact vessel wall location, which can be a problem as the aortic root moves about 8 mm within each heartbeat[9]. This is when time-resolved 3D segmentations can become necessary and even useful in order to derive additional parameters such as vessel wall elasticity, a parameter directly related to cardiovascular disease. Usually 3D segmentations are done using manual or semi-automatic tools. Time-resolved 3D segmentations have therefore been very limited or impossible, as such a task can take up to 13 hours of manual work. For segmentation machine learning methods have proven essential. Convolutional neural networks (CNN), like for example the UNet[10], allow for voxel-based classification and localization of tissue. Similar approaches can be tested in this project to achieve fast and accurate time-resolved 3D segmentations of 4D flow MRI data.
**Flow Phantoms**
In order to validate theoretical and in-silico results, we employ flow phantoms in in-vitro experiments. Phantom geometries range from simple straight stenotic tubes[8] to realistic, compliant aorta geometries[11], depending on the focus of the investigation, to which steady and pulsatile inlet conditions can be applied.
**Segmentation Project tasks:**
- Train CNN on existing 4D flow MRI data and corresponding manual segmentations (static)
- Combine 5D flow MRI segmentations and high resolution 2D cine MRI data for better results in time-resolved segmentations
**Flow Phantom Project tasks:**
- CFD simulation in OpenFOAM
- Planning and construction of the phantom in our workshop
- Flow MRI measurements of the phantom
[1] doi: 10.1186/s12968-019-0549-0.
[2] doi: 10.1002/mrm.27437.
[3] doi: 10.1002/jmri.23632.
[4] doi: 10.1186/s12968-015-0174-5.
[5] doi: 10.1109/MSP.2007.914728.
[6] doi: 10.1186/s12968-019-0582-z
[7] doi: 10.1038/s42256-020-0165-6.
[8] doi: 10.1002/mrm.28236.
[9] doi: 10.1148/radiology.218.2.r01ja07548
[10] doi:10.1007/978-3-319-46723-8_49
[11] doi: 10.1007/s00348-012-1371-8.
**Segmentation Project tasks:**
- Train CNN on existing 4D flow MRI data and corresponding manual segmentations (static) - Combine 5D flow MRI segmentations and high resolution 2D cine MRI data for better results in time-resolved segmentations
**Flow Phantom Project tasks:**
- CFD simulation in OpenFOAM - Planning and construction of the phantom in our workshop - Flow MRI measurements of the phantom
Please contact peper@biomed.ee.ethz.ch and dillinger@biomed.ee.ethz.ch if you are interested in the topic. The projects’ focuses can be adjusted to the student’s interest and experience and their extent will cover the corresponding thesis (bachelor, master, semester thesis).
Interested students are asked to send a CV and a transcript of records.
For students from abroad, we also need to know how they plan to organize funding for their stay.
Please contact peper@biomed.ee.ethz.ch and dillinger@biomed.ee.ethz.ch if you are interested in the topic. The projects’ focuses can be adjusted to the student’s interest and experience and their extent will cover the corresponding thesis (bachelor, master, semester thesis). Interested students are asked to send a CV and a transcript of records. For students from abroad, we also need to know how they plan to organize funding for their stay.