Advances in MRI-guided radiotherapy: Acquisition, analysis and evaluation methods

Sofie Rahbek

Research output: Book/ReportPh.D. thesis


In radiotherapy the main goal is to kill tumor cells with ionizing radiation, typically high-energy photons, while sparing the surrounding healthy tissue. The precision of the radiation dose delivery is crucial for the efficacy of radiotherapy and imaging is utilized for tumor and normal tissue delineation and for guidance at the moment of treatment, leading to the concept image-guided radiotherapy (IGRT). The most recent advancement here is the hybrid MRI – linear accelerator system (MR-linac), which enables high soft-tissue contrast images for adapting the dose plan to the anatomy-of-the-day.
The MR-linac also makes it feasible to obtain daily advanced MRI which may potentially be used for revealing functional features of the tumor informative of local radio-sensitivities in the tumor and predictive of the overall treatment outcome. Utilizing MRI for such purposes requires high image quality and advanced data processing that involves analysis of multi-contrast measurements, such as multi-echo T2-weighted measurements, to extract characteristics of the tumor tissue. A model-based method is most commonly used, but the resulting quantitative maps are often prone to partial volume effects and may be biased or uninformative if the chosen model does not properly fit the data. Here, monotonous slope non-negative matrix factorization (msNMF) is proposed as a novel data-driven method. This extended version of the NMF decomposes the data under monotony constraints that fit many types of MRI data. A demonstration of the method showed its ability to extract interpretable components related to the underlying tissue micro-structure, and applications was also exemplified by estimation of edema water fractions in spinal cord white matter.
MR-linacs enable longitudinal imaging data series, which requires dedicated handling of the covarying time-resolved measurements to investigate the tumor dynamics during the course of fractionated radiotherapy. This work also proposes a prediction framework as a tool to search for biomarkers using longitudinal MRI data. The framework relies on an initial data-driven decomposition and includes fitting over time to capture therapy-induced tumor changes that may be predictive of the outcome. Its feasibility was demonstrated using example datasets and the msNMF for decomposition, and results indicated a value of early T2-relaxation changes for predicting tumor response.
Diffusion-weighted imaging (DWI) is an interesting multi-contrast technique due to the lower diffusivity of many tumor types that is altered when the ionizing radiation induces microstructural changes. Unfortunately, standard DWI suffers from geometric distortions, incompatible with radiotherapy purposes which includes both target delineation and response prediction. To address this, the current work includes optimization of the single-shot split acquisition for fast spin-echo (SPLICE) sequence. Contrary to standard diffusion-weighted echo-planar imaging, the SPLICE sequence is based on a fast spin-echo readout and results in geometrically robust images, but with a relatively poor voxel shape due to signal modulation during readout. The suggested optimization method maximizes the signal-to-noise ratio (SNR) for a controlled pointspread-function by varying the refocusing flip angles. A clear SNR gain, which also improved the accuracy of apparent diffusion coefficient (ADC) estimates, was seen for a healthy subject brain. In summary, the acquisition and analysis strategies developed during this research project may provide directions for future radiotherapy studies and can advance the usage of MRI for both treatment planning and evaluation.
Original languageEnglish
PublisherDTU Health Technology
Number of pages136
Publication statusPublished - 2022


Dive into the research topics of 'Advances in MRI-guided radiotherapy: Acquisition, analysis and evaluation methods'. Together they form a unique fingerprint.

Cite this