3725 - Global Interobserver Variations of Clinical Target Volume (CTV) Delineation among International Experts in Predefined Gross Tumor Volume (GTV) and Organs at Risk (OAR) Volumes in Head and Neck Cancer

U. Selek¹, K. S. Hu², H. Tatli³, J. R. Palta⁴, and A. Jhingran⁵; ¹Koc University, School of Medicine, Department of Radiation Oncology, Istanbul, Turkey, ²Department of Radiation Oncology, NYU Langone Health, New York, NY, ³RTIMEA Education and Training Department, Elekta Instrument AB, Istanbul, Turkey, ⁴Department of Radiation Oncology, Virginia Commonwealth University, Richmond, VA, ⁵Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX

Purpose/Objective(s): To define the extent and pattern of interobserver variations of clinical target volume (CTV) delineation among international experts in predefined primary and nodal gross tumor volumes (GTV) and organs at risk (OAR) volumes in head and neck cancer Materials/

Methods: Experts were invited by email globally from ASTRO, ESTRO, IAEA, FARO, and TROD for this study, and DICOM images of two anonymized cases (T1N2M0 HPV+ right tonsillar cancer: OPX & T1N2M0 EBV+ left nasopharyngeal cancer: NPX) were delivered online via cloud large file share to each colleague. The data set included delineated GTV of primary tumor & nodal disease and OAR to decrease the discrepancy and interobserver variability for this step. Experts were asked to delineate the clinical tumor volumes (CTV1, higher dose; CTV2, intermediate dose; CTV3, lower dose) and planning tumor volumes (PTV1, higher dose; PTV2, intermediate dose; PTV3, lower dose) based on their institutional guidelines. We used a systematic process to create and analyze a consensus atlas among multiple observers facilitated by ProKnow DS Version 1.35.2 (Precision radiation medicine company, Golden Rule Version 1.2.8531.2131); forming 50% ground truth area common in all participants contours. The comparison metrics encompassed matching volume, missing volume, extra volume, Dice coefficient for similarity, and mean differences in displacement along the X, Y, and Z axes.
Results: 32 experts participated the study. NPX: Moderate similarity was observed in both CTV1 (mean DSC = 0.73±0.19) and CTV2 (mean DSC = 0.67±0.17). Additionally, both CTVs centers of mass exhibit slight displacement compared to the consensus CTVs centers of mass. While moderate similarity characterized PTV1 (mean DSC = 0.77±0.15) and PTV2 (mean DSC = 0.69±0.17), PTV3 demonstrated poor similarity (mean DSC = 0.45±0.21). PTV1 and PTV2 showed minor displacement compared to the consensus PTVs centers of mass, whereas PTV3 exhibited significant displacement in both X and Y directions. OPX: Moderate similarity is observed in both CTV1 (mean DSC = 0.72±0.19) and CTV2 (mean DSC = 0.66±0.17), while both CTVs show minor displacement of their centers of mass compared to the consensus CTVs centers of mass. Only PTV1 showed moderate agreement (mean DSC = 0.73±0.16) between the experts and the consensus, with minor displacement in centers of mass, while PTV2 and PTV3 had poor similarity. Dice scoring alone did not correlate well with displacements from the centers of masses in X, Y, Z axis in these comparison. While CTV1 in both cases showed moderate similarity among experts, agreement decreased on CTV2 with more variations, which accordingly diverged the PTVs.
Conclusion: Even in an environment of predefined primary and nodal GTV to decrease the discrepancy, interobserver variability is evident especially starting from CTV2, accordingly which can potentially effect and change the PTVs to lead into different OAR scenarios; appealing standardized peer review processes and more structured guidelines for CTV.