Session Item

Numerous delineation guidelines exist, often focusing exclusively on target structures and a single diagnosis. Furthermore, the organs at risk (OAR) are described briefly and with great variation, including whether or not constraints should be absolute or relative. Standardized contouring of pelvic OARs would improve data reporting, clinical trials and assessment of late toxicity. No previous studies have assessed the impact of inter-observer variation when contouring OARs in the pelvis. Thus, this study aimed to assess the inter-observer variation when contouring OAR in the pelvis across pelvic diagnoses.

Consensus atlas: Nine Danish pelvic multidisciplinary cancer groups were invited to a contouring seminar involving experienced clinicians and radiographers, with representatives from all Danish oncology centres. Before the seminar, all participants reached a consensus on relevant OAR delineation. Next, each participant contoured a female and a male audit case. All contours were evaluated, a consensus was reached, and the participants contoured the two cases again. The consensus cases were evaluated at the workshop, and a final contouring atlas was established based on these two cases. For the primary workshop, 7 OARs were selected  (see figure) based on the most frequently used OARs across the multidisciplinary cancer groups.

Inter-observer variation: Nineteen workshop participants contoured 20 randomly selected cases. CT (3mm slice thickness) and T2-MR scans were available for ten males and ten females. The interobserver variation was assessed pairwise in terms of the dice similarity coefficient (DSC) and mean surface distance (MSD).

In total, 843 OAR were contoured across the 20 patients, with a median of 6 (range 4-8) observers per organ. The DSC and MSD were best for the high-contrast bladder and femoral heads. The DSC was high for the bowel bag due to the large volume; however, the median MSD was 3.4 mm (interquartile range: 2.8-4.2 mm), showing considerable inter-observer variation. For the anal canal and rectum, the transition between the two OAR introduced inter-observer variance of 2.8 mm (2.2-3.5 mm) and 1.9 mm (1.5-2.6 mm), respectively (figure). The gender-specific OARs revealed a large variance.

A clinical relevant inter-observer variance was observed despite contouring experts agreeing on consensus guidelines and performing audits before the workshop. OAR with less soft-tissue contrast had a higher variance. The multiple contours of the 20 patients will now be used to validate automated contouring algorithms.