Description
The purpose of this study is to optimize Deformable Image Registration (DIR) and dose accumulation in High-Dose-Rate (HDR) cervical brachytherapy. The conventional brachytherapy dose accumulation relies on the assumption that the hot spot region of the Organs At Risk (OAR) remain anatomically consistent each fraction. However, two sources of uncertainty introduce inaccuracies: 1) inter-fractional shifts in patient organ geometry and brachytherapy applicator placements; 2) inconsistencies in OAR contouring in the high dose regions. Although DIR has been considered as a tool to correct the fluctuations, the intensity- based algorithm is prone to errors if high-intensity structures, such as the brachytherapy Applicator Region (AR), are involved. In this study, 23 HDR cervical brachytherapy fractions from 6 patients treated with Tandem & Ovoids applicator were included. Firstly, a 3D random walks-based Matlab algorithm was used to semi-automatically segment the AR from all images. Subsequently, artificial intensity masks were applied on the OAR, AR, skeleton, and treatment couch, to prioritize the OAR and minimize the influence of high-intensity structures during DIR. It should be noted that inter-fractional inconsistencies in OAR contouring act as a major limiting factor in this method, since masking removes indications of inconsistency within contours. Cumulative OAR dose was calculated from Dose-Volume Histograms, generated from: 1) the reference contours; 2) each fraction’s deformed contours. AR segmentation produced consistent results for differently sized ARs. Contour masking was utilized in cases where nearby high intensity structures interfered with the algorithm. After DIR, each image and contour was cropped to better represent only the high-dose regions of the OAR. Compared to regular DIR, intensity-masked DIR resulted in increased Dice Similarity Coefficient (DSC) values in all patients, ranging: 0.067(s=0.021)~0.416(s=0.226) for bladder, and 0.046(s=0.014)~0.321(s=0.170) for rectum. Dose accumulation results were inconclusive due to dose uncertainties introduced by high-dose-region OAR contouring inconsistencies. The DSC analyses indicate that intensity mask-guided DIR results in optimized deformable registration in high-dose regions. However, inconsistencies in OAR contouring is a major source of uncertainty in dose accumulation and image registration. Root Mean Square Distance (RMSD) analysis suggests a positive correlation between contouring inconsistency and image registration inaccuracy.
