Development and validation of a robust deformable dose accumulation approach in magnetic resonance-guided radiation therapy

Abstract

Cancer is a common disease worldwide and the second cause of death in Europe. In addition to chemotherapy and surgery, radiation therapy (RT) is used to kill cancer cells by delivering a high dose of radiation. Today, innovative adaptive RT techniques, such as image-guided RT (IGRT), enable precise treatment. However, IGRT, such as cone beam computed tomography, often does not account for deformations between fractions and volume variations, leading to uncertainties in dose delivery. In recent years, developments in RT, including the combined magnetic resonance linear accelerator (MR-LINAC) system, have revolutionized adaptive RT. Online adaptive magnetic resonance-guided RT (OA-MRgRT) allows daily adaptation of the treatment plan based on the anatomy of the respective day. Despite these technological innovations, such as daily anatomical imaging and plan adaptation, the accurate calculation of the total dose delivered to the patient during treatment remains a challenge due to the deformations between each fraction; this underlines the urgent need for robust deformable dose accumulation (DDA) solutions to accurately calculate the total dose delivered to the target volume and surrounding tissue.

The first objective of this thesis was to investigate the applicability and use of dose mapping and accumulation (DMA) in RT. This first phase involved the identification and classification of current DMA applications in RT to develop a comprehensive landscape, including the description of strengths and limitations. We then identified and analyzed the significant barriers to clinical implementation and broad use for dose accumulation. This phase included a thorough review of existing literature, case studies, and implementations to identify the current use of DMA. We closely examined the identified barriers, which included technical, methodological, and biological issues, to understand their causes and their impact on the applicability of DMA. This detailed analysis led to the development of comprehensive guidelines and requirements tailored to researchers, healthcare providers and manufacturers. These guidelines address the identified barriers by providing practical solutions and best practices to overcome them.

Deformable dose accumulation includes deformable image registration, corresponding dose mapping, and dose summation. Different mathematical models are available for the first two applications. The second study of this dissertation involved the development and investigation of a robust DDA solution, along with a comparison the DDA implementations from other research groups. This was achieved by conducting a multi-center study evaluating several solutions in different anatomical areas. The study evaluated a gold standard case that served as known ground truth and analyzed five clinical cases treated with online adaptive MRgRT. The study showed significant agreement between the different implementations, but also revealed differences that depend on specific cases and algorithms and may have significant consequences in the therapeutic context. The study also showed that our proposed DDA solution worked consistently and reliably, producing results equivalent to those of the other algorithms.

Adaptive RT, such as IGRT and OA-MRgRT, aims to account for inter-fraction anatomic variations that differ from situation during treatment planning. Fractionated irradiation requires the use of DDA to compare different adaptive approaches in terms of the total dose delivered to the tumor and surrounding tissue. In the third objective of this thesis, we used the deformed dose accumulation solution from the second objective of this thesis to investigate the accumulated doses in a clinical study treating patient with prostate cancer with adaptive, moderately hypo-fractionated RT. The total deformably accumulated doses of two adaptive treatment strategies - conventional IGRT (conv-IGRT) and OA-MRgRT - and the treatment plan generated during simulation were compared dosimetrically to investigate the hypothesis that OA-MRgRT allows more effective sparing of organs-at-risk compared to reference planning and conv-IGRT. All techniques, including reference planning, conv-IGRT, and OA-MRgRT, showed similar mean doses in the target volume. Although all accumulated approaches resulted in lower maximum organ-at-risk doses compared to the treatment plan, OA-MRgRT in particular showed encouraging results compared to conv-IGRT. However, the dosimetric differences did not translate into clinical relevance according to normal tissue complication probability models. Overall, the study demonstrated feasibility and transferability of the established DDA method and workflow.

This thesis provided a comprehensive methodological basis for the development of a dedicated DDA approach and applied it for the first time to a clinical study dataset using DDA in fractionated prostate cancer RT. Clinical evaluation and further research focusing on dose differences and their effects are currently being developed to initiate the next phase of OA-MRgRT, which may include online or real-time dose accumulation in order to further personalize cancer treatment with RT.