Determining interobserver variability in prostate bed CTV target delineation using MRI

Abstract

Purpose or Objective Up to a third of patients who have a radical prostatectomy for localised prostate cancer will have a rising PSA post operatively and salvage radiotherapy is a curative treatment option in this setting. Accurate delineation of the target remains the ‘weakest link’ in the radiotherapy planning process. Pelvic anatomy is better visualised on MR compared to CT due to a higher soft tissue resolution and may facilitate CTV delineation, yet it’s use in prostate bed radiotherapy planning is limited. Current guidelines for prostate bed CTV delineation are primarily based on CT and can differ considerably. This is a multicentre international contouring study which aims to assess the degree of interobserver variability for prostate bed CTV delineation on MRI amongst experienced clinical and radiation oncologists. Materials and Methods In total, 22 observers, predominantly experienced radiation/clinical oncologists, were invited to take part in the study. Each participant was provided with T2W Elekta Unity MR-linac scans for three patients post prostatectomy. A clinical vignette containing the patient’s history, diagnostic MRI report, histopathology report and the baseline prostate mpMRI scan were provided. Observers were asked to outline the prostate bed CTV using the information provided as per their current clinical practice, on ProKnow, a cloud-based contouring software. On completion, the contours were imported into the ADMIRE research version 2.0 (Elekta AB, Stockholm, Sweden) program. A Simultaneous Truth and Performance Estimate (STAPLE) was created for each case and interobserver comparisons were then carried out by comparing each observer’s individual contour to the STAPLE structure. Results A total of 19 observers partook in this study; 17 radiation/clinical oncologists, 1 urological surgeon and 1 radiologist, from a total of 11 institutions in 7 countries. Only the contours drawn by the oncologists were used to create the STAPLE. A totally of 51 contours were analysed, of which three were not included for technical reasons. Figure 1 demonstrates the median results for four comparison metrics. Figure 2 demonstrates visually the variability seen for one of the cases. On visual inspection of the volumes, the greatest areas of disagreement appear to be at the inferior and superior extent and the cranial aspect of the anterior border. Conclusion Despite minimal experience in MR contouring for prostate bed, the agreement between the radiation/clinical oncologists as demonstrated by the median DICE metric is reasonable. The DICE metric however tends to be more generous for the same absolute error for larger volumes as opposed to smaller volumes. The distance metrics show greater variability amongst observers whilst the volume measurement shows the greatest variability. Outliers are not uncommon. As a group, we have reviewed and discussed the contours with contribution from radiology and surgical colleagues and will be creating consensus guidelines for contouring the prostate bed CTV on MR.