Abstract Book

S869

ESTRO 37

Material and Methods Five patient datasets, each incorporating CT and T2- weighted MR scans, were circulated to 13 volunteers for contouring. Guidelines and a sample patient dataset with expert defined contours were also distributed. Volunteers consisted of radiation oncologists, radiotherapists, and medical physicists, each covering a range of clinical experience. Structures to be contoured on CT were CTV (prostate), seminal vesicles, rectum, distal colon, bowel bag, bladder, and peri-rectal space. T2-MR structures included CTV, trigone, membranous urethra, penile bulb, neurovascular bundle (left and right), and multiple pelvic floor muscles (internal anal sphincter, external anal sphincter, puborectalis muscle, and levator ani muscles). Post processing of bowel bag contours required subtraction of bladder, CTV, and seminal vesicles outlines. Contouring variations were calculated by pairwise analysis using DSC, volume similarity, Hausdorff distance, mean absolute surface distance, sensitivity, specificity, and centroid distances. Results Spread in contoured volumes are shown in figure 1, where volumes have been normalised with respect to the median structure volume for each patient. Pairwise analysis of DSC between observer contours are shown in figure 2. Mean DSC for CTV (CT: 0.83, MR: 0.88), bladder (0.93), and rectum (0.81) were comparable to values observed in the literature. Distal colon and bowel bag contours recorded poor overlap, however minimal volume spread suggest that slice selection, dependent on the superior slice contoured for rectum, was the most significant factor for contouring variation for these structures. Variations in volume and DSC were more substantial for MR structures. Multiple neurovascular contours had minimal to no overlap. Average Hausdorff and mean absolute surface distance for CT structures (17.669 mm, 3.342 mm respectively) were comparable to MR structures (13.275 mm, 3.954 mm), suggesting that localisation of the structure, as opposed to identification of the structure’s boundary, was the largest contribution to variations in observer’s MR contours.

Median comparisons for each imaging type, delineation times and imaging scores are summarised in Table 1. Overall there is good agreement between radiographers and between radiographers and the gold standard. T2*W shows significantly reduced IOV and significantly higher agreement with the gold standard compared to CT, for all comparison metrics. In addition, there is significantly decreased IOV for prostate contours delineated using T2*W compared to T2W MRI (DSC and mean distance) and significantly improved accuracy (DSC, Cohen’s kappa and mean distance) when comparing to the gold standard (Figure 1). Greater quality images and confidence in contouring were reported for T2*W MRI, reflected in the shorter time to complete contours. Conclusion Radiographer prostate contours are more accurate, show less IOV and are more confidently and quickly outlined using a T2*W MR sequence compared to T2W or CT imaging. EP-1614 Inter-observer contouring variation of multiple pelvic structures on CT and MR for prostate cancer D. Roach 1,2 , M.A. Ebert 3,4,5 , M. Jameson 2,6 , J.A. Dowling 1,4,7,8 , A. Kennedy 3 , P. Greer 7,9 , L. Holloway 1,2,4,6 1 University of New South Wales, Faculty of Medicine, Sydney, Australia 2 Ingham Institute for Applied Medical Research, Medical Physics, Liverpool, Australia 3 Sir Charles Gairdner Hospital, Radiation Oncology, Nedlands, Australia 4 University of Wollongong, Centre for Medical Radiation Physics, Wollongong, Australia 5 University of Western Australia, School of Physics and Astrophysics- Faculty of Science, Crawley, Australia 6 Liverpool and Macarthur Cancer Therapy Centres, Department of Radiation Oncology, Liverpool, Australia 7 University of Newcastle, School of Mathematical and Physical Sciences, Newcastle, Australia 8 Royal Brisbane Hospital, Australian e-Health Research Centre- CSIRO, Herston, Australia 9 Calvary Mater Newcastle Hospital, Radiation Oncology, Newcastle, Australia Purpose or Objective Prostate cancer toxicity studies often have smaller pelvic structures contoured by a single expert clinician. Presently limited information is available regarding variability of clinician contouring of these structures. This study sought to document inter-observer contouring variation for a range of pelvic structures on both CT and MR.

Figure 1: Contoured volume variation for CT (blue) and MR (green) structures, normalised to median structure volume for each patient. Outlier volumes for neurovascular bundle (ranging from 4 – 7) were omitted for clarity.