Author: Wookjin Choi, Jun Li 👨🔬
Affiliation: Thomas Jefferson University 🌍
Purpose: Resin Yttrium-90 (Y-90) selective internal radiation therapy (SIRT) is a radioembolization procedure which uses Y-90 microspheres to treat metastatic liver cancer. In the procedure, liver volume size is needed for Y-90 activity calculations, which is obtained from liver structure delineated by physicians on CT or MR images. TotalSegmentator is an open-source, deep-learning based auto-segmentation software for organ delineations. The aim of the study was to explore the feasibility of applying the auto-segmentation tool for liver delineation in SIRT, to improve the workflow efficiency of the procedure.
Methods: The study was conducted using CT images obtained from 12 SIRT patients. The TotalSegmentator software was installed on a server computer equipped with high-performance GPUs and was deployed using DICOM communication. The CT images were exported from clinical database to the TotalSegmentator’s DICOM location, where a monitoring software detected the incoming data and automatically ran the auto-segmentation. The results were returned to the original DICOM location where the CT images were stored. The auto-segmented liver contours were compared with physician manually-delineated contours which were taken as the ground truth. Dice similarity coefficient (DSC) and mean distance to agreement (MDA) were calculated. Ratio of volume (RV) and ratio of Y-90 activity (RA, Y-90 activity calculated using an auto-segmented liver contour to the accurate activity calculated using a manually-delineated contour), were evaluated.
Results: The average DSC and MDA are 0.94±0.01 (range: 0.91-0.95) and 2.1±0.5 mm (range: 1.5-3.3 mm), respectively. The average RV is 1.021±0.041 (range: 0.951-1.077), and the average RA is 0.999±0.004 (range: 0.988-1.006).
Conclusion: The auto-segmented liver contours have good agreement with the manually-delineated liver contours. The Y-90 activities calculated using the auto-segmented liver contours are close to the accurate activities. The auto-segmentation tool can be applied in SIRT.