Rapid quality control method for automated organ-at-risk contouring in head and neck radiotherapy
10.13491/j.issn.1004-714X.2026.02.006
- VernacularTitle:头颈部危及器官自动勾画的快速质量控制方法
- Author:
Xiaoyu YANG
1
;
Kaining YAO
2
;
Yichen PU
2
;
Shun ZHOU
2
;
Ruoxi WANG
2
;
Haizhen YUE
2
;
Hao WU
3
Author Information
1. Institute of Medical Technology, Peking University Health Science Center, Beijing 100191, China.
2. Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiotherapy, Peking University Cancer Hospital & Institute, Beijing 100142, China.
3. Institute of Medical Technology, Peking University Health Science Center, Beijing 100191, China;Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiotherapy, Peking University Cancer Hospital & Institute, Beijing 100142, China.
- Publication Type:OriginalArticles
- Keywords:
Adaptive radiotherapy;
Automated contouring evaluation;
Geometric consistency metrics;
Quality assurance
- From:
Chinese Journal of Radiological Health
2026;35(2):193-199
- CountryChina
- Language:Chinese
-
Abstract:
Objective To address the need for rapid review of automated organ-at-risk (OAR) contouring for head-and-neck cancer in online adaptive radiotherapy, this study used dosimetric indices as the reference standard to evaluate the suitability of geometric similarity metrics and to develop a rapid screening model that balances the risk of missed errors with review efficiency. Methods This retrospective study included 29 patients with head-and-neck radiotherapy, yielding 243 pairs of OAR contours (auto-generated vs. manual). Geometric similarity metrics, including the Dice similarity coefficient (Dice), the 95th percentile Hausdorff distance (HD95), and the maximum Hausdorff distance (HD100), were computed between the two contour sets. Using the dose distributions from clinical treatment plans, dosimetric differences between the two contour sets were calculated for key indices, including mean dose (Dmean) and maximum dose (Dmax). Correlations between geometric metrics and dosimetric differences were analyzed. Dosimetric discrepancy events were defined as Y3Gy using a threshold of 3 Gy. Within a univariable logistic regression triage framework, each geometric metric was used as an independent variable to estimate the probability of Y3Gy. Discriminative performance was assessed using receiver operating characteristic (ROC) curves and precision-recall (PR) curves, with the area under the curve (AUC) including ROC-AUC and PR-AUC. Practical operating thresholds were determined via threshold sweeping. Results Geometric similarity metrics showed weak-to-moderate correlations with dosimetric differences; compared with maximum dose, correlations with mean dose differences were more consistent. Among the evaluated metrics, HD95 achieved the best classification performance for Y3Gy. Threshold sweeping suggested that an HD95 threshold in the range of 4-9 mm can balance the risk of missed discrepancies with the efficiency of automated contour quality assurance (QA). Conclusion Several geometric similarity metrics demonstrated only weak-to-moderate associations with dosimetric differences. For head-and-neck OAR contour QA in this study, HD95 provided the best discriminative performance and can support rapid triage of automated contours, with a practical operating threshold range of 4-9 mm, potentially improving the efficiency of adaptive radiotherapy workflows.
