Dosimetric impact of interobserver variability in MRI-based delineation for cervical cancer brachytherapy.

PURPOSE AND BACKGROUND To study the dosimetric impact of interobserver delineation variability (IODV) in MRI-based cervical cancer brachytherapy. MATERIALS AND METHODS MR images of six patients were distributed to 10 experienced observers worldwide. They were asked to delineate the target volumes and the organs at risk (OARs) for each patient. Two types of reference contours were created (Expert Consensus - EC and Simultaneous Truth and Performance Level Estimation - STAPLE). Optimised plans based on both EC- and STAPLE-contours were prepared. These plans were transferred to each of the observer contour sets and the resulting DVH parameters (D(90) and D(2cc)) were calculated. For each patient the standard deviation (SD) for the 10 observers was calculated. RESULTS A mean relative SD of 8-10% was found for GTV and High Risk CTV (HR-CTV) D(90) analysing one single fraction. For rectum and bladder the mean relative SD for D(2cc) was 5-8% while sigmoid was at 11%. For the whole treatment the IODV in HR-CTV caused an uncertainty of ±5 Gy(α/β=10) (1SD). The corresponding figure for OARs was ±2-3 Gy(α/β=3). The results were not sensitive as to which structure set was used for the optimisation. CONCLUSIONS For the target volumes the dosimetric impact of IODV was smallest for the GTV and HR-CTV, while IODV had an even smaller impact on the bladder and rectum.