Anatomical changes (such as weight loss and disease regression) during radiotherapy for head and neck cancer patients are common. These changes often result in the need to update the treatment plan, a strategy known as adaptive radiotherapy. Adaptive radiotherapy can be offline (between treatment fractions) or online (during the treatment fraction). Online adaptive radiotherapy relies on rapid recontouring of organs and targets on images taken just before delivery of a treatment fraction, for example on cone beam computed tomography (CBCT) or magnetic resonance images. Contour propagation, based on non-rigid registration, offers a solution. However, it is not clear whether errors in contour propagation result in clinically significant differences in dose for the patient. In this thesis two main investigations have been performed to assess the accuracy of propagated contours with emphasis given to the dosimetric reliability. For both studies, data from ten head and neck patients was used. Data included the planning CT, target, and organ contours and five CBCTs acquired throughout treatment. CBCTs were corrected to improve dose calculation accuracy. Plans with 65.1 Gy to the primary CTV and 54 Gy to the intermediate risk CTV in 30 fractions were created in RayStation (RaySearch, Sweden) using the clinical contours for the target and organs at risk. Then, organs at risk were propagated to the CBCTs via five commercial packages; Mirada (Mirada Medical, UK), RayStation, Pinnacle (Philips, Netherlands), ADMIRE (Elekta, UK) and ProSoma (Medcom, France). Additionally, two sets of gold standard contours were generated: a clinical gold standard based on re-contouring on CBCT by a single Oncologist, and a geometric gold standard based on a consensus contour generated from the propagated contours. In the first study, the propagated contours were compared to the gold standards, both geometrically and dosimetrically. For the geometric comparison, the common geometric metrics were calculated. These metrics demonstrated generally acceptable accuracy (for example, mean distance to agreement for all 0.78 for most contours. However, the spinal cord showed moderate reliability (ICC>0.66) for the clinical gold standard, likely due to steep dose gradients. In the second study, the suitability of propagated contours for direct replanning on CBCT images was assessed. Using the scripting interface of RayStation, a revised plan was re-optimised on each CBCT for each set of propagated and gold standard contours (a total of 350 plans) and DVH statistics extracted for the organs at risk using the gold standard contours. The DVH statistics for the plans created with the gold standards and the propagated contours were compared. For example, differences up to 3.7 Gy for the brainstem and 2.4 Gy for the spinal cord were found. From these comparisons, thresholds for planning DVH statistics were determined which allow the operator to flag structures that require (or do not require) review for accuracy. For instance, if the propagated contour is with 3.46 Gy of tolerance for the brainstem clinical goal, the contour should be reviewed. To conclude, the results presented in this thesis show the utility of propagated contours for use in adaptive radiotherapy workflows for head and neck cancer, alongside recommendations for safe use to maximise patient benefit.
- deformable image registration
- head and neck cancer
- contour propagation