HighLine: High image quality for lines in MRI: from roots to angiograms
Funded by the Helmholtz Imaging Platform
Contact: Elisabeth Pfaehler, Hanno Scharr
The project HighLine has the objective to enhance 3D Magnetic Resonance (MR) images displaying line-like structures such as plant roots or vessels with modern deep learning methods. The images included in this project have very special characteristics: They yield very sparse information (i.e. thin roots) and display thin, line-like structures which are varying in thickness and length.
One focus of this project lies on image denoising of MR images. For MR images image quality is proportional to scan time, i.e. the longer a patient is scanned, the better is the image quality. However, to reduce the patient burden and to increase the patient throughput, it is desirable to reduce the scan time as much as possible. Modern deep learning architectures make it possible to reduce the scan time while still acquiring images with high image quality.
The second focus of the project is image super resolution, i.e. to display finer details than the original images. Increasing the resolution of our root or vessel images gives the opportunity to display finer details i.e. thiner objects. The accurate display of very fine structures yields important information as e.g. fine arteries might be blocked, or growth of very fine roots gives an early indication on the adaption of the plant to the soil.
To solve both tasks, modern deep learning methods are used. As the number of MR plant images is much larger than the number of MR images of vessels, we start with developing neural networks for MR root images and will transfer these networks to MR images of vessels.
Cooperation partners:
Daniel Pflugfelder, Forschungszentrum Jülich, IBG‐2: Plant Sciences
Tony Stöcker, Deutsches Zentrum für Neurodegenerative Erkrankungen (DZNE), Research group MR physics