Clinical evaluation of atlas-based auto-segmentation for contouring pelvic CTVs in the treatment of anal cancer with FDG-PET-positive lymph node involvement

Max Bieder^1,2*Markus Böhm^3,4Marciana-Nona Duma^1,5,6Andrea Wittig-Sauerwein^1,7

• ¹Department of Radiation Therapy and Radiation Oncology, University Hospital Jena, Jena, Germany
• ²Department of Neurology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
• ³Abbott Rapid Diagnostics Jena GmbH, Jena, Germany
• ⁴Institute of Medical Statistics, Computer and Data Sciences, University Hospital Jena, Jena, Thuringia, Germany
• ⁵Department of Radiation Oncology, Helios Clinics Schwerin, Schwerin, Germany
• ⁶Department for Human Medicine, MSH Medical School Hamburg, Hamburg, Germany
• ⁷Department of Radiation Oncology, University Hospital Würzburg, Würzburg, Germany

Current evidence on atlas-based auto-segmentation (ABS) in radiotherapy primarily addresses organs at risk, whereas its application for clinical target volume (CTV) delineation remains insufficiently explored. Additionally, the optimal number of datasets required for ABS atlases is debated. This study investigates ABS performance for automated CTV (aCTV) segmentation in anal cancer patients with ^18F-fluorodeoxyglucose positron emission tomography/computed tomography (^18F-FDG PET-CT)-positive lymph node (LN) metastases, using varying atlas sizes.A retrospective analysis was conducted on 51 anal cancer patients who underwent ^18F-FDG PET-CT-based treatment planning between 2009 and 2018. Patients with FDG-positive LN metastases were identified. Manual CTV (mCTV) delineation was performed in accordance with the UK National Guidance for IMRT in Anal Cancer. The resulting 51 mCTV datasets were integrated into a single ABS atlas, which was used to generate aCTVs for the 27 patients with FDG-positive LN metastases. For each of these 27 patients, five different atlas sizes (n = 10, 20, 30, 40, 50) were evaluated using a leave-one-out approach.Automated and manual CTVs were compared using the Dice Similarity Index (DSI), the percentage of FDG-positive LNs adequately covered, and volumes either erroneously included (mistakenly contoured volume, MCV) or omitted (not contoured volume, NCV) by the ABS process. Of the 51 patients, 27 (52.9%) had FDG-positive LN metastases. The mean DSI for atlas sizes of n = 10, 20, 30, 40, and 50 were 0.73, 0.78, 0.79, 0.79, and 0.80, respectively. A DSI ≥ 0.7 was achieved in 24 patients (88.9%) across all atlas sizes. The increase in DSI between n = 10 and n = 40 was statistically significant (Bonferroni-adjusted p < 0.05). Mean relative NCV and MCV ranged from 21.8-23.9% and 17.7-19.5% of the respective mCTV volume, with decreasing trends as atlas size increased. Segmentation inaccuracies predominantly occurred in the upper mesorectal and lower ischiorectal regions.In conclusion, ABS facilitates the delineation of CTVs in anal cancer patients and improves contouring efficiency. However, manual correction by radiation oncologists remains necessary.