MR - Guided Radiotherapy
One of the latest innovations in radiotherapy has been the clinical introduction of integrated MR and linac devices (MR-linac), where a linear accelerator (linac) is integrated with a magnetic resonance imaging (MRI) scanner. MR-linacs enable in-room MR imaging before and during radiotherapeutic treatment and allow treatment adaptation based on the anatomy of the day, as well as gating based on real-time images of the moving anatomy. The precise localization of the tumor and surrounding healthy tissue promises improved treatment outcomes, especially for entities that are highly affected by motion caused by breathing or digestion, such as the lung or the pancreas. The Department of Radiation Oncology of the University Hospital of the LMU Munich started patient treatment with the ViewRay MRIdian MR-linac in January 2020.
Time resolved MR imaging
One research focus in MR-guided radiotherapy is on the use of MRI-based motion information for improved treatment planning and delivery to fully exploit the potentials of this innovative machine. In a retrospective patient study, a definition of the internal target volume (ITV) for lung tumors based on real-time 4D-MRI data was investigated. Using the probability-of-presence of the target volume, a higher robustness against interfractional changes of the proposed margin concept compared to today’s 4D-CT-based workflow was found. Another research focus lies on validation measurements of motion models proposed in literature for improved lung tumor localization during beam delivery. For this purpose, a porcine lung phantom is employed that allows the simulation of realistic patient-like data with high reproducibility. Ex-vivo porcine lungs are mounted in the phantom and artificial agar nodules can be injected into the lungs to simulate target lesions. The lungs are inflated and periodically moved to simulate the breathing motion of patients. The anatomy of the moving lung is then estimated in 4D (temporally resolved 3D images) based on cine-MRI data and different published motion estimation methods. The ex-vivo porcine lung phantom is provided by Prof. Julien Dinkel and his group from the LMU Department of Radiology. These research projects are funded by the German Research Foundation (DFG) within the Research Training Group GRK 2274 (www.grk2274.de) and the Friedrich-Baur-Stiftung. Our department has a research agreement with ViewRay Inc.
- Rabe M, Thieke C, Düsberg M, Neppl S, Gerum S, Reiner M, Nicolay NH, Schlemmer HP, Debus J, Dinkel J, Landry G, Parodi K, Belka C, Kurz C, Kamp F. Real‐time 4DMRI‐based internal target volume definition for moving lung tumors. Medical Physics. 2020.
Deep learning based pseudoCT generation from MR images
The basis for planning a radiation treatment is the 3D image data of the patient, which allows localization and determination of the target volume (tumor) and the organs-at-risk next to the tumor. Nowadays the prior imaging data for treatment planning is a computed tomography (CT) scan, supported by an overlay of a magnetic resonance (MR) scan of the patient. The CT scan is required to obtain an electron density map and to calculate the dose distribution in the patient, while the MR image offers superior soft tissue contrast, facilitating precise delineation of tumor and organ outlines. This project aims to convert MR images to CT scans using artificial intelligence and thereby removes the necessity to acquire the CT scan (MR-only). A U-shaped convolutional neural network (U-net) was trained on a set of CT and corresponding MR images. The trained model is used to convert new MR images to CT images and thereby enables dose calculations for MR images on these so-called pseudoCTs. The quality of the artificial pseudoCT images is verified by calculating dose distributions for photons and protons on the generated image data and a comparison to the dose calculations on the corresponding original CT scans.
- Neppl S., Landry G., Kurz C., Hansen D.C., Hoyle B., Stöcklein S., Seidensticker M., Weller J., Belka C., Parodi K., Kamp F. Evaluation of proton and photon dose distributions recalculated on 2D and 3D Unet-generated pseudoCTs from T1-weighted MR head scans. Acta Oncologica. 2019 July 4;58(10):1429-1434.