Abstract
BACKGROUND: Target volume definition of the primary tumor in esophageal cancer is usually based on computed tomography (CT) supported by endoscopy and/or endoscopic ultrasound and can be difficult given the low soft-tissue contrast of CT resulting in large interobserver variability. We evaluated the value of a dedicated planning [F18] FDG-Positron emission tomography/computer tomography (PET/CT) for harmonization of gross tumor volume (GTV) delineation and the feasibility of semiautomated structures for planning purposes in a large cohort. METHODS: Patients receiving a dedicated planning [F18] FDG-PET/CT (06/2011-03/2016) were included. GTV was delineated on CT and on PET/CT (GTV(CT) and GTV(PET/CT), respectively) by three independent radiation oncologists. Interobserver variability was evaluated by comparison of mean GTV and mean tumor lengths, and via Sørensen-Dice coefficients (DSC) for spatial overlap. Semiautomated volumes were constructed based on PET/CT using fixed standardized uptake values (SUV) thresholds (SUV30, 35, and 40) or background- and metabolically corrected PERCIST-TLG and Schaefer algorithms, and compared to manually delineated volumes. RESULTS: 45 cases were evaluated. Mean GTV(CT) and GTV(PET/CT) were 59.2/58.0 ml, 65.4/64.1 ml, and 60.4/59.2 ml for observers A-C. No significant difference between CT- and PET/CT-based delineation was found comparing the mean volumes or lengths. Mean Dice coefficients on CT and PET/CT were 0.79/0.77, 0.81/0.78, and 0.8/0.78 for observer pairs AB, AC, and BC, respectively, with no significant differences. Mean GTV volumes delineated semiautomatically with SUV30/SUV35/SUV40/Schaefer's and PERCIST-TLG threshold were 69.1/23.9/18.8/18.6 and 70.9 ml. The best concordance of a semiautomatically delineated structure with the manually delineated GTV(CT)/GTV(PET/CT) was observed for PERCIST-TLG. CONCLUSION: We were not able to show that the integration of PET/CT for GTV delineation of the primary tumor resulted in reduced interobserver variability. The PERCIST-TLG algorithm seemed most promising compared to other thresholds for further evaluation of semiautomated delineation of esophageal cancer.