Développement d’une approchehybride pour l’améélioration desimages CBCT pour laradiothérapie adaptative

Abstract

Recently, the integration of in-board cone beam computed tomography (CBCT)systems into linear accelerators has provided not only a powerful tool for improvingthe accuracy of patient positioning but also it makes it possible to adapt the treatmentplanning with the patient anatomy modifications throughout the entire treatment,known as adaptive radiation therapy (ART). During ART planning, after applyinga rigid registration between daily acquired CBCT images and planning CT (pCT)scans and getting information about patients structural changes and internal organsdeformations, a repeated acquisition of CT images (repCT) is needed for replanningif the patients anatomical changes are significative. Although repCT images providethe most up-to-date information for a human body with an accurate hounsfield unit(HU) distribution for dose calculation, their repeated acquisition may pose a healthconcern to patients due to the additional dose. Therefore, the aim of this thesis is toinvestigate the feasibility of daily CBCT images for dose calculation instead of acquiringrepCT images. The main interest is to propose a promising strategy to overcome thelimitations related to the high amount of scatter that reduces the quality of CBCTimages compared to the pCT.To this end, we proposed three complementary approaches that allow thecorrection of the HU distribution in the CBCT images. The first proposed approach isinspired from the fact that the CBCT and the pCT images have not the same gray-levelhistograms. Therefore, we applied the histogram matching (HM) process to resembletheir histograms and offer the possibility to distinguish the different tissue types inthe CBCT images. As a second approach, we proposed the application of multilevelthreshold (MLT) to improve the correctness of the HU distribution obtained with HM.Finally, in the third approach we proposed the generalization of the previously proposedapproaches to contribute in the minimization of the ART workload requirements. Themain idea is to include the notion of inter-patient deformable registration in order toinvestigate the pCT images available in the archive for the correction of CBCT imagesrather than acquiring new CT images for each patient.Results of the visual and statistical analysis showed that the proposed approachesii ABSTRACTcan improve the HU distribution in the CBCT images and generate corrected imagesin good agreement with the pCT.