3375 - A Quality Control Framework for AI-Based Contouring in Radiotherapy: Clinical Implementation and Validation

12:45pm - 02:00pm PT

Hall F

Screen: 33

POSTER

Presenter(s)

Te Zhang, MD - Department of Radiation Oncology, Xijing Hospital, Fourth Military Medical University, Xi'an, Shannxi

T. Zhang¹, and L. Zhao²; ¹Department of Radiation Oncology, Xijing Hospital, Fourth Military Medical University, Xi'an, Shaanxi, China, ²Department of Radiation Oncology, Xijing Hospital, Air Force Medical University, xi an, shan xi, China

Purpose/Objective(s):

While deep learning algorithms have significantly improved the efficiency of automatic contouring for organs-at-risk (OARs) and target volumes in radiotherapy, the inherent trade-off between efficiency and hidden contour inaccuracies remains a critical challenge. This study investigates the necessity of quality control (QC) for AI-generated OAR and target contours, proposes a multi-dimensional QC framework, and evaluates its clinical efficacy.

Materials/Methods:

A self-developed DICOM-RT structure set validation module was implemented to detect contouring errors, including ROI outlier detection, slice discontinuity, ROI overlap, cavity detection, and abnormal volume alerts. A tri-dimensional QC system integrating "geometric-anatomical-dosimetric" criteria was established. We analyzed 300 AI-contoured cases with artificially introduced errors through manual audits.

Results:

Results demonstrated that the DICOM-RT validation module effectively intercepted 86.7% of AI-generated hidden contour errors. Utilizing a closed-loop feedback mechanism, the rate of clinically unacceptable errors decreased from 7.2% to 2.4% (p<0.001), while the average contour correction time was reduced from (23.4 ± 6.7) minutes with traditional manual review to (8.2 ± 3.1) minutes (p<0.001). Post-implementation, the clinical acceptance rate improved from 82.4% to 96.7%.

Conclusion:

This study confirms that systematic QC frameworks effectively balance the efficiency of AI-driven contouring with clinical safety, providing a quantifiable solution for quality assurance in intelligent radiotherapy.