Impact of an atlas on radiographer inter-observer contour variation in prostate radiotherapy

Abstract

Purpose or Objective The clinical implementation of MR-guided adaptive radiotherapy (MRgART) requires fast, accurate delineation of target volumes and organs at risk (OARs) to mitigate the potential for contouring inaccuracies and therefore treatment errors. To support the introduction of radiographer (RTT)-led real-time MRgART on the MR Linac (MRL), an MR contouring atlas for prostate radiotherapy delineation was developed. The impact of its use on inter-observer variation is reported in this work. Materials and Methods A contouring atlas was developed through the multi-disciplinary consensus of 3 MRL radiographers, 2 clinical oncologists, and 1 diagnostic radiographer for the prostate, seminal vesicles, bladder and rectum. Contours were defined on T2-weighted three-dimensional turbo spin echo sequences, axial images (T2w 3D Tra) acquired on the MRL. To assess the guide’s utility, 8 RTTs with varying MRL experience contoured MR images from each of 5 fractions for 5 patients in the Monaco treatment planning system (Monaco 5.40.01), from 5 patients (n= 25 image data sets ). Each RTT contoured the prostate, seminal vesicles, bladder, and rectum on MR data sets before and after the introduction of the atlas. The ‘after’ contours were generated > 21 days following the ‘before’ contours to minimise the effects of repetition. Intra- and inter-observer contour variations (measured in volume), time to contour and observer contouring confidence were determined at both time-points using a 5-point Likert scale (1 being not confident and 5 being extremely confident). Results There were no statistically significant differences in any volumes (prostate or OARs) pre- and post- atlas introduction, with high correlations between all observers (R≥ 0.89) for all structures. Although the variance in volume definition decreased for all structures among all observers post intervention and the greatest in the prostate (mean variance before atlas 7.37 mm3 to 3.47 mm3 after) the change was not statistically significant. Contour overlap regions using Boolean structures are currently being calculated. Over all structures, the mean contouring time for all observers was reduced by 50%, from 53 to 27 minutes (range for no atlas 33 to 82 minutes and with atlas 22 to 55 minutes, p=0.01) following the introduction of the atlas (Figure 1). The mean contouring times for the prostate, bladder and rectum individually were reduced significantly from 14 to 7 minutes, (p=0.02) 14 to 7 minutes (p=0.002), 14 to 7 minutes (p=0.04) respectively. For all structures, the observer mean contouring confidence increased from 2.3 to 3.5 out of 5 (p≤0.02) Conclusion No statistically significant improvements were observed in contour variance amongst observers following the introduction of a consensus based atlas, the guidance suggesting decent baseline image interpretation. However, use of the atlas facilitated improved observer contour confidence and speed; key factors for a real-time RTT-led adapt-to-shape workflow.