Impacts of volumetric vs. anatomical accuracies in GTV segmentation for nasopharyngeal carcinoma: a consensus-guideline-based study.

BACKGROUND: Consensus guidelines for target volume segmentation have been issued for nasopharyngeal carcinoma (NPC) radiotherapy, yet prior studies have largely overlooked anatomical accuracy in this context. METHODS: Sixteen NPC patients were enrolled. GTVnx and prophylactic clinical target volume of primary tumor (CTVp) were independently contoured by three radiation oncologists and one senior radiologist to obtain observer segmentations (OSs) and reference GTVnx. Original CTVp of OS were adjusted according to reference GTVnx. CTVp consistency was evaluated with Dice Similarity Coefficient (DSC), 95th Hausdorff Distance (HD95) and Intraclass Correlation Coefficient (ICC). Impacts of volumetric and anatomical accuracy were evaluated with precision and true positive rate (TPR), dosimetric parameters and tumor control probability (TCP). RESULTS: Anatomical accuracy was strikingly high (median anatomical TPR: 89.5%-100%; precision: 80%-92%), volumetric accuracy was substantially lower (median volumetric TPR: 63.88%-73.92%; precision: 47.08%-70.29%). Tiny differences were observed in CTVp consistency between original and adjusted CTVp (Median DSC: 0.84 vs. 0.85, Median HD95: 6.53 vs. 6.27 mm, ICC: 0.98 vs. 0.99), which were both higher than that of GTVnx consistency across OSs (Median DSC: 0.63, Median HD95: 8.21 mm, ICC: 0.92). Treatment plans based on OS resulted in decreased D95, EUD and TCP (average decrease of D95, EUD and TCP: 7.38 Gy, 5.24 Gy and 17.13%). CONCLUSIONS: In NPC, primary tumor segmentation inaccuracies are predominantly volumetric rather than anatomical. Low volumetric accuracy may potentially reduce TCP. Under consensus guidelines, high anatomical accuracy in GTVnx segmentation may contribute to consistent CTVp segmentation.