Purpose: In resin Yttrium-90 (Y-90) selective internal radiation therapy (SIRT) of metastatic liver cancer, liver volume sizes are needed for determining Y-90 activity with the body-surface-area method, which are obtained from liver contours delineated by physicians. The aim is to evaluate Eclipse Smart Segmentation, a knowledge-based auto-segmentation module, for potential application in liver delineation for resin Y-90 SIRT.
Methods: CT images of 10 patients treated with resin Y-90 were studied. Auto-segmentation of liver was performed with the Smart Segmentation in an Eclipse treatment planning system (Version 15.6). For each case, an expert case in the built-in library, which had the highest similarity to the studied case as indicated by the system, was selected for auto-segmentation. Liver contours generated with the Smart Segmentation were compared with physician manually-delineated contours. The latter were taken as the standard. To evaluate the auto-segmentation accuracy, Dice similarity coefficient (DSC) was calculated and liver volume sizes were compared.
Results: The average DSC was 0.77 (standard deviation: 0.10; range: 0.56-0.9). The average relative volume difference was 12% (standard deviation: 9%; range: 0.4%-22.6%).
Conclusion: The Smart Segmentation can be used to generate initial liver contours, which need to be reviewed and edited by physicians. Application of the Smart Segmentation will save contouring time and improve the efficiency of the workflow in resin Y-90 SIRT.