215 error margins and/or to provide single or two-fraction treatment. Multiple solutions have been proposed to improve the speed of current MR-Linac workflows. These solutions are aimed at specific steps of these workflows, such as the steps displayed and discussed in chapter 2 and chapter 3 of this thesis. In general, adaptive workflows consists of several basic components. These include: 1. MRI acquisition (multiple times during the workflow for treatment preparation and position verification purposes); 2. Image registration, contour propagation, manual editing of contours, and approval of contours; 3. Treatment planning, QA, and approval; 4. Treatment delivery. With respect to MRI acquisition, efforts are being made to obtain high-quality images in the shortest time possible. Current 3D T2-weighted MRI sequences that are used in clinical practice, including the ones used at the UMC Utrecht, generally take 2-4 min to acquire. This makes them prone to motion artefacts and significantly influences the time needed for treatment preparation.16 Nowee et al.17 discussed the impact of acquisition time on the image quality and the registration and delineation steps. Although image quality (a subjective assessment by two expert radiologists) was significantly worse for the shortest sequence (1.3 min), there was no significant difference in target delineation results as assessed through median 95% Hausdorff distance.17 Whether or not the image quality is good enough depends on the intended use of the MR scan. For example, for position verification purposes, quality can be a bit lower as long as the MRI still provides enough information on (large) shifts that may have occurred. On the other hand, for accurate (manual) contouring of the target and organs-at-risk (OARs), high-quality images are required. NewMRI acquisition approaches, such as compressed sensing, will provide high-quality MR images with less (motion) artefacts in shorter acquisition times.18 These promising innovations will allow faster workflow cycle times and could potentially limit the effect of intrafraction motion and deformations. MRI acquisition is just one of the components that contributes to the overall treatment time with MRI-guided radiotherapy. In chapter 3, we have looked in more detail at component number 2, which is currently the most time-consuming component of the online adaptive workflow.14 The contours that are used on a daily basis can be obtained through several ways. For example, contours can be propagated from pre-treatment planning MRI scans to the daily treatment MRI scan. With the current clinical Unity software, contours can be propagated using deformable or rigid image registration, depending on the type of workflow (Adapt-to-Shape [ATS] versus Adapt-to-Position [ATP]).19 With rigid contour propagation or ATP, the contours are not changed in shape, but simply shifted to the new location to obtain the best possible fit. With deformable contour propagation or ATS, the contours are deformed based on the deformation vector fields that are obtained through the deformable image registration (DIR) algorithm. The ATS workflow is aimed at creating a treatment plan that neatly fits the daily anatomy and thus provides the most accurate treatment plan. However, as mentioned previously, the contours provided by the DIR algorithm often need (extensive) manual editing. With editing times of over 10 min, this significantly affects the overall treatment time. Improvements in the DIR algorithm could improve contour quality and thereby 11 General discussion and future perspectives
RkJQdWJsaXNoZXIy MTk4NDMw