ATLAS-BASED SEGMENTATION IMPROVES CONSISTENCY AND DECREASES TIME REQUIRED FOR CONTOURING POSTOPERATIVE ENDOMETRIAL CANCER NODAL VOLUMES

Purpose: Accurate target delineation of the nodal volumes is essential for three-dimensional conformal and intensity-modulated radiotherapy planning for endometrial cancer adjuvant therapy. We hypothesized that atlas-based segmentation ("autocontouring") would lead to time savings and more consistent contours among physicians. Methods and Materials: A reference anatomy atlas was constructed using the data from 15 postoperative endometrial cancer patients by contouring the pelvic nodal clinical target volume on the simulation computed tomography scan according to the Radiation Therapy Oncology Group 0418 trial using commercially available software. On the simulation computed tomography scans from 10 additional endometrial cancer patients, the nodal clinical target volume autocontours were generated. Three radiation oncologists corrected the autocontours and delineated the manual nodal contours under timed conditions while unaware of the other contours. The time difference was determined, and the overlap of the contours was calculated using Dice's coefficient. Results: For all physicians, manual contouring of the pelvic nodal target volumes and editing the autocontours required a mean standard deviation of 32 9 vs. 23 7 minutes, respectively (p = .000001), a 26% time savings. For each physician, the time required to delineate the manual contours vs. correcting the autocontours was 30 +/- 3 vs. 21 +/- 5 min (p = .003), 39 +/- 12 vs. 30 +/- 5 min (p = .055), and 29 +/- 5 vs. 20 +/- 5 min (p = .0002). The mean overlap increased from manual contouring (0.77) to correcting the autocontours (0.79; p = .038). Conclusion: The results of our study have shown that autocontouring leads to increased consistency and time savings when contouring the nodal target volumes for adjuvant treatment of endometrial cancer, although the autocontours still required careful editing to ensure that the lymph nodes at risk of recurrence are properly included in the target volume. (C) 2011 Elsevier Inc.