Influence of experience and qualification on PET-based target volume delineation : When there is no expert-ask your colleague

C. Doll, V. Duncker-Rohr, G. Rücker, M. Mix, M. Macmanus, D. Ruysscher, W. Vogel, J. Eriksen, W. Oyen, A. Grosu, W. Weber and U. Nestle

Radiation Oncology Department, University Medical Center Freiburg, Robert-Koch-Str. 3, 79106, Freiburg/Breisgau, Germany.
Mar, 2014



The integration of positron emission tomography (PET) information for target volume delineation in radiation treatment planning is routine in many centers. In contrast to automatic contouring, research on visual-manual delineation is scarce. The present study investigates the dependency of manual delineation on experience and qualification.A total of 44 international interdisciplinary observers each defined a [(18)F]fluorodeoxyglucose (FDG)-PET based gross tumor volume (GTV) using the same PET/CT scan from a patient with lung cancer. The observers were "experts" (E; n = 3), "experienced interdisciplinary pairs" (EP; 9 teams of radiation oncologist (RO) + nuclear medicine physician (NP)), "single field specialists" (SFS; n = 13), and "students" (S; n = 10). Five automatic delineation methods (AM) were also included. Volume sizes and concordance indices within the groups (pCI) and relative to the experts (eCI) were calculated.E (pCI = 0.67) and EP (pCI = 0.53) showed a significantly higher agreement within the groups as compared to SFS (pCI = 0.43, p = 0.03, and p = 0.006). In relation to the experts, EP (eCI = 0.55) showed better concordance compared to SFS (eCI = 0.49) or S (eCI = 0.47). The intermethod variability of the AM (pCI = 0.44) was similar to that of SFS and S, showing poorer agreement with the experts (eCI = 0.35).The results suggest that interdisciplinary cooperation could be beneficial for consistent contouring. Joint delineation by a radiation oncologist and a nuclear medicine physician showed remarkable agreement and better concordance with the experts compared to other specialists. The relevant intermethod variability of the automatic algorithms underlines the need for further standardization and optimization in this field.