SU-E-J-169: Process for QA of Normal Tissue Delineation and Reported Doses in Clinical Trials




To create and evaluate a process of using auto-segmentation for quality assurance of doses reported to normal structures in clinical trials.


Multiple atlases for normal tissue segmentation were created for the esophagus, trachea/large bronchus, and brachial plexus. These atlases were used to independently delineate 21 patients who were enrolled in a phase III study of accelerated hypofractionated IGRT for patients with stage II-III NSCLC. The results of the multi-atlas segmentation were visually evaluated for acceptability. The dose volume limits and the maximum point doses specified by the protocol were compared for the originally delineated and auto-segmented contours. It was noted whether these differences affected whether plans met or failed the protocol constraints.


On visual inspection, the auto-segmented contour was acceptable with the exception of 8 cases (38%) of the esophagus. In these cases, the anatomy of the esophagus was notably different from that of atlases so that the autosegmentation failed. On the other hand, the process using auto-segmentation identified two cases where not all contours of the structure had been delineated in the original plan (one trachea/large bronchus case and one brachial plexus case). The average difference in maximum dose to the esophagus, trachea, and brachial plexus were 0.2±4.4Gy (range: – 13.6:4.7Gy), 0.2±1.1Gy (−3.9:0.8Gy), and 3.6±10.7Gy (−1.2:34.1Gy), respectively. In one esophagus case and one trachea case, the plan became unacceptable when calculated using the auto-segmented contour, but in both cases the dosimetric difference was very small.


Two cases (10%) of contouring with insufficient coverage (trachea/large bronchus and brachial plexus) in the original plan were identified using the proposed process. However, patient-to-patient variations in the position of the esophagus meant that our current auto-segmentation is inadequate to evaluate this organ, and more work is needed to optimize the process.

Work funded by Cancer Prevention Research Institute of Texas (CPRIT)