Image Registration for the Management of Organ Motion and Deformation in Image Guided Radiotherapy

Slagmolen P., ''Image registration for the management of organ motion and deformation in image guided radiotherapy'', Proefschrift voorgedragen tot het behalen van het doctoraat in de ingenieurswetenschappen, K.U.Leuven, September 2010, Leuven, Belgium.Cancer has evolved to one of the leading diseases in the Western world today. The treatment of cancer usually consists of a mix of radiotherapy, chemotherapy and surgery. Radiotherapy is therein an important part as it has been proven to be very efficient in a large number of indications. In the last decades, radiotherapy has undergone a huge evolution with treatment machines allowing the delivery of much more conformal, 3D dose distributions. Additionally, the availability of image guidance during planning, treatment and follow-up, allows to further increase conformality thereby decreasing normal tissue dose and increasing tumour dose. However, these new techniques become increasingly demanding in terms of image acquisition and image processing. Often, a prerequisite to use image guidance is the availability of accurate image registration, i.e. the establishment of point-by-point correspondence between images to allow fusion of images with different, complementary information. In this work, we investigate the challenges posed by image guided radiotherapy and the possible answers provided by using image registration techniques. First, we assess the use of nonrigid registration for multimodal planning. To accurately irradiate the target, it is important to know the location of the malignant cells. This is not always straightforward as the image modality required for dose planning, CT, doesn’t always provide sufficient soft-tissue contrast to properly see the tumour. This is for example the case in rectal cancer, where other imaging modalities, such as MRI or PET, shouldbe preferred for target delineation as they provide better soft-tissue anatomy contrast or functional information about the tissue, respectively. For dose planning, however, these additional images need to be fused with the planning CT. As the rectum is a deformable organ, susceptible to variations in rectal and bladder filling, mere rigid registration techniques will not suffice. Therefore, we explore the use ofnonrigid registration techniques to align these image modalities fortreatment planning (or additionally for treatment follow-up). We show that, while multimodal treatment planning is feasible, it is very challenging in the pelvic region due to the uncertainties that arise in the images between different acquisitions. Next, we show how to calculate treatment margins which account for deformation in rectal cancer. At treatment planning, when the target is known, it is required to take uncertainties into account which are related to either microscopic tumour invasion (invisible on conventional imaging), organ motion and deformation or machine inaccuracies. Usually, the target is enlarged by a margin to account for these uncertainties. Among these uncertainties, organ motion and deformation contribute significantly and while it has been shown how to take organ motion into account to determine margins, this has not yet been done for organ deformation. We introduce a novel methodology which calculates margins that take the deformation of the target into account. We apply this methodology to rectalcancer patients who would benefit from an additional boost to the mesorectum. We show that our non-uniform, 3D margins are likely much more suited than the commonly used uniform margins. Considering currenttechnical limitations, we simplify our 3D margins and give recommendations on the size of margins for clinical use. Next, we want to decrease the motion of the target with respect to the treatment beam by using image guidance. Initially, we look at the use of implanted fiducial markers for aligning a patient between different treatment fractions. We describe an automatic 2D/3D registration which aligns the patient based on the marker positions as automatically detected in the planning CT and on the marker positions as manually detected in orthogonal kV and MV electronic portal images. We compare our automatic registration, which also includes prostate shrinkage and prostate rotation,to manual registration in terms of translations and show that it is nearly equivalent. Furthermore, we show that prostate rotation can be significant and that prostate shrinkage occurs due to hormone therapy combined with radiotherapy. We conclude that automatic 2D/3D registration is a useful technique for replacing manual registration. Furthermore, it might be useful to perform weekly replanning of these patients to take the prostate shrinkage into account. Next, we look at the use of 3D CBCT imaging for patient alignment. As some patients are not eligible for implantation with fiducial markers, they require an alternative positioning as the initial laser positioning is known to suffer from a series of inaccuracies. We propose to use on-board CBCT images based on the soft-tissue information to align the patient with respect to the planning CT. We use a methodology based on masked CBCT images and mutual information based rigid registration and validate using fiducial markers as ground-truth. We show that our method accurately recovers left-right translation and is a strong improvement for both AP and IS translation. However, the method is currently not able to recover prostate rotations. Our method allows to reduce margins in comparison to traditional laser-based positioning. Finally, we lookat the detection of intrafraction motion using automatically detected fiducial markers. While interfraction motion has been well described, the impact of intrafraction motion is just now being investigated. Using images acquired during the delivery of a treatment beam in IMRT treatment, we are able to observe the implanted fiducial markers during treatment. We have developed a method that automatically detects these markers in 2D images using a series of image filtering and discrete optimization. We show how these detected markers lead to the actual intrafraction motion by assuming a shortest path between the treatment beams. Using this methodology, we show that intrafraction motion is largest at the start of treatment, between the initial kV/MV positioning and the first treatment beam. Furthermore, we show that intrafraction motion occurring during treatment itself (between the first andlast treatment beam) is fairly small but increases significantly as the treatment duration increases. This leads us to the conclusion that treatment time should be as short as possible and that initial positioning should be as close to the first beam as possible. Overall, wehave shown the use of image registration for optimizing every step of the radiotherapy treatment process. While in some cases we provided a definite answer to a certain problem, some challenges still need to be overcome in the future. Mainly in terms of image acquisition and the reduction of uncertainties by proper patient preparation, a lot of gain can be expected.status: publishe