3373 - Automated Quality Assurance for Diagnosis of Common Contouring Errors in Radiotherapy Structure
Presenter(s)
Z. Yang1, X. Yang2, X. Huang3, S. Huang4, S. Wang5, and J. Zhu1; 1MedMind Technology Co., Ltd., Beijing, China, 2Department of Radiation Oncology, Sun Yat-sen University Cancer Center, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Guangzhou, China, 3State Key Laboratory of Oncology in South China, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Guangdong Provincial Clinical Research Center for Cancer, Department of Radiation Oncology, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China, 4Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China, 5MedMind Technology Co, Ltd, Beijing, China
Purpose/Objective(s):
This study aims to address uncertainties in automatic structure contouring in radiotherapy and potential inaccuracies during manual review. Despite increased efficiency from automatic contouring, algorithm inaccuracies require manual review and adjustment. Manual contouring can introduce errors due to fatigue or oversight, and current review processes lack precision and efficiency. Our goal is to develop and assess an automated quality assurance system for structure contouring to enhance accuracy and efficiency in radiotherapy planning.Materials/Methods:
We used a dataset of 220 CT cases (99 head and neck, 121 chest and abdomen) with corresponding automatically generated contours. Two workflows were established: manual review and automated quality control. Images and contours were sent to three radiology specialists and automated quality assurance software. The software detected six issues: scattered pixels, missing layers, cavities, overlaps, abnormal volumes, and empty ROIs. We compared expert review and software performance by measuring review time and detection accuracy, with cross-verification of results against expert feedback.Results:
The automated system reduced review time from 27 minutes (manual) to 3 minutes. Manual review identified 70 issues (41 faults, 9 overlaps, 15 voids, 5 abnormal volumes), while the automated system detected 460 issues across 220 cases. True Positive rates were >98% for bad points, faults, empty ROIs, and voids, but only 9% for overlaps and 6% for volume anomalies.Conclusion:
Automated systems significantly improve efficiency and coverage in radiotherapy structure reviews, especially for detecting overlaps and volume anomalies. However, high False Positive rates for these issues highlight the need to investigate causes and optimize the system. Future work will focus on improving algorithms to enhance accuracy and incorporating human expertise to improve overall system performance and clinical reliability.