BELdART: Belgian dosimetry audits in radiotherapy

Abstract

Purpose: Contemporary external radiotherapy requires the outmost precision in patient positioning. This can be achieved through 2D/3D image acquisition by the on-board imager (OBI) integrated into the linear accelerator. In this study, OBI 2D image acquisition protocols were assessed and optimized with regard to dose in an effort to reduce overall patient dose without degrading kV image quality. Materials and Methods: A phantom for high/low spatial resolution assessment was placed at the center of appropriately compiled PMMA plates to serve as cephalic (19cmX17cm) and pelvic (24cmX37cm) phantoms. Anterior-posterior and lateral scanning of the phantoms was carried out using the predefined/default acquisition protocols to assess image quality and measure the entrance dose utilizing a low dose rate probe. Anterior-posterior and lateral images were then acquired for all available kV/mAs configurations and corresponding entrance dose were recorded. An experienced user rejected all protocols that produced images with quality inferior to that achieved by the default protocol. The optimized head and pelvis protocols were selected as those corresponding to the lowest doses. A Rando phantom was scanned with both default and dose-optimized protocols, while proposed protocols were used for intra-session imaging in 10 patients. A physician employed a 5-point scale to subjectively assess quality of Rando and patient images. Effective doses were estimated in an average-sized anthropomorphic model for default and optimized acquisitions. Results: Low spatial resolution exhibited a strong dependence on kV/mAs settings, in contradistinction to high spatial resolution. Optimized head protocols for anterior-posterior/lateral scans employed 80/60 kVp and 6.4/5.0 mAs, whilst proposed pelvis protocols for anterior-posterior/lateral acquisitions utilized 80/100 kVp and 6.25/64.0 mAs. In comparison to the default parameters, the implementation of optimized settings resulted in 49% and 29% entrance dose reduction in the cephalic anterior-posterior and lateral scans, respectively, and in 29% and 27% entrance dose reduction in the pelvic anterior-posterior and lateral acquisitions, respectively. Quality of Rando images acquired with proposed protocols was equivalent to that of default protocols. Quality of patient images obtained with optimized settings was graded at least as acceptable, indicating that proposed settings may be routinely used. Optimized protocols were found to decrease effective dose by 27.5% and 49% in pelvic and cephalic acquisitions, respectively. Conclusions: Considering the relatively large number of radiotherapy sessions and consequently of image acquisitions, it is postulated that dose-optimized kV protocols may contribute towards a significant reduction in the overall patient radiation dose without compromising positioning accuracy. P101 Implementation and clinical validation of an atlas based auto segmentation method of head and neck radiotherapy volumes of interest Purpose: To implement and validate an atlas based segmentation method for the automatic contouring of organ at risks (OARs) in the frame of head and neck radiotherapy planning. In this context, an accurate and standardized delineation of OARs is required to face the challenge posed by the location of many critical organs in close proximity to target volumes and the high focalized dose distributions. The atlas can help to speed up the planning workflow, partially replacing manual segmentation, and reduce the intra-and inter-operator dependence improving the standardization process. Materials and Method: In this study, an auto-segmentation method with a "Medium-Atlas" approach was implemented. The atlas consisted of twenty datasets of head and neck radiotherapy patients. The optimization and validation of the method were carried out by comparing the results of the automatic segmentation with those of the manual one. The comparison was carried out on a group of ten datasets belonging to the same institution from which the atlas was built and on a group of ten patients provided by another center. The comparison was made using the following quantitative metrics: Dice similarity index (DSC), average distance to agreement (MDC), sensitivity and positive predictive value (PPV). Results: The number of the ATLAS datasets used to build the medium atlas significantly affects the quality of the segmented contours on the test images (ANOVA: F = 2,96, p <0,05): the optimal number for the accuracy of delineation is N=5. The results of the analysis showed that the automatic segmentation method was accurate for OAR volumes higher than 10 cc: DSC=0,82±0,01, sensitivity=0,74±0,01, PPV=0,92±0,01 and MDC=2,77±0,06 mm. For volumes less than 10 cc the agreement between automatic and manual segmentation worsened and were deemed unsatisfactory. Time consuming was reduced by 50% compared to the manual segmentation. The atlas also showed good performances in its multi-institutional application: for volumes greater than 10 cc: DSC=0,77±0,04, sensitivity=0,83±0,04, PPV=0,75±0,05 and MDC=2,62±0,06 mm; the method does not show statistically significant differences with its mono center application (p=0.43, =0,05). Conclusions: The implemented atlas-based auto-segmentation, in both tested applications, was found to be an efficient and accurate method in the delineation of OARs with volumes greater than 10 cc.