Abstract

In photon radiotherapy of lung cancer, respiratory-induced motion introduces systematic and statistical uncertainties in treatment planning and dose delivery. By integrating magnetic resonance imaging (MRI) in the treatment planning process in MR-guided radiotherapy (MRgRT), uncertainties in target volume definition can be reduced with respect to state-of-the-art X-ray-based approaches. Furthermore, MR-guided linear accelerators (MR-Linacs) offer dose delivery with enhanced accuracy and precision through daily treatment plan adaptation and gated beam delivery based on real-time MRI. Today, the potential of MRgRT of moving targets is, however, not fully exploited due to the lack of time-resolved four-dimensional MRI (4D-MRI) in clinical practice. Therefore, the aim of this thesis was to develop and experimentally validate new methods for motion characterization and estimation with 4D-MRI for MRgRT of lung cancer. Different concepts were investigated for all phases of the clinical workflow - treatment planning, beam delivery, and post-treatment analysis. Firstly, a novel internal target volume (ITV) definition method based on the probability-of-presence of moving tumors derived from real-time 4D-MRI was developed. The ability of the ITVs to prospectively account for changes occurring over the course of several weeks was assessed in retrospective geometric analyses of lung cancer patient data. Higher robustness of the probabilistic 4D-MRI-based ITVs against interfractional changes was observed compared to conventional target volumes defined with four-dimensional computed tomography (4D-CT). The study demonstrated that motion characterization over extended times enabled by real-time 4D-MRI can reduce systematic and statistical uncertainties associated with today’s standard workflow. Secondly, experimental validation of a published motion estimation method - the propagation method - was conducted with a porcine lung phantom under realistic patient-like conditions. Estimated 4D-MRIs with a temporal resolution of 3.65 Hz were created based on orthogonal 2D cine MRI acquired at the scanner unit of an MR-Linac. A comparison of these datasets with ground truth respiratory-correlated 4D-MRIs in geometric analyses showed that the propagation method can generate geometrically accurate estimated 4D-MRIs. These could decrease target localization errors and enable 3D motion monitoring during beam delivery at the MR-Linac in the future. Lastly, the propagation method was extended to create continuous time-resolved estimated synthetic CTs (tresCTs). The proposed method was experimentally tested with the porcine lung phantom, successively imaged at a CT scanner and an MR-Linac. A high agreement of the images and corresponding dose distributions of the tresCTs and measured ground truth 4D-CTs was found in geometric and dosimetric analyses. The tresCTs could be used for post-treatment time-resolved reconstruction of the delivered dose to guide treatment adaptations in the future. These studies represent important steps towards a clinical application of time-resolved 4D-MRI methods for enhanced MRgRT of lung tumors in the near future.