Experience of Implementing Deep Learning-Based Automatic Contouring in Breast Radiation Therapy Planning: Insights from Over 2,000 Cases

Byung min Lee MD, Jin Sung Kim PhD, Yongjin Chang MS, Seo Hee Choi MD, Jong Won Park MD, Hwa Kyung Byun MD, PhD, Yong Bae Kim MD, PhD, Ik Jae Lee MD, PhD, Jee Suk Chang MD, PhD
Radiation Oncology - Biology - Physics (IJROBP)
Related Product


Date Published

This paper presents the findings from implementing a deep learning-based automatic contouring system in breast radiation therapy planning, analyzing over 2,000 cases. Introduced in 2019, the study assessed the system's impact and clinical utility by comparing auto-contours with final contours, adjusted manually, in 2,428 adjuvant breast radiation therapy patients. The evaluation utilized the Dice similarity coefficient (DSC) and the 95% Hausdorff distance (HD95) for comparison, analyzing 22,215 structures. Results indicated that final contours were generally larger, with significant improvements in DSC and reduced HD95 for organs-at-risk (OAR) post-implementation, except for the lungs. Target volumes also showed improved outcomes, albeit less pronounced than OARs. The study highlights the auto-contouring system's utility and the increased reliance on automated settings, raising concerns about automation bias. It suggests the necessity of stringent risk assessments and quality management strategies to optimize the use of such systems, ensuring patient safety and treatment effectiveness.


제품, 인재 채용, 투자 관련 또는 기타 문의사항이 있으신 경우 편하신 방법으로 연락주시기 바랍니다