First Validation of Semi-automatic Liver Segmentation Algorithm

Abstract

Selective internal radiotherapy (SIRT) is a treatment modality in advanced liver cancer. A careful treatment planning including an angiographic evaluation of the hepatic blood supply and determination of liver and tumor volume by imaging (e.g. contrast-enhanced MRI) is part of the pre-therapeutic workflow. Treatment is performed by an intra-arterial angiographic application of ⁹⁰Y-labeled microspheres. Particularly, the segmentation of the liver and calculation of tumor burden are time-consuming activities, but they are a prerequisite of state-of-the-art dosimetry models. In this study, we validated an interactive software tool, which provides a semi-automatic segmentation of co-registered image data (e.g. contrast-enhanced abdominal CT, SPECT/CT, and MRI). A reader experienced in manual liver segmentation for SIRT work-up employed dosimetry software to segment abdominal CT and MRI scans and documented the completion time. A first database contains ten cases (contrast-enhanced CT data) from the Liver Tumor Segmentation (LiTS) Challenge. Results from software-guided semi-automatic segmentation were compared to the published ground truth using Sørensen-Dice coefficient (mean score = 92.9 ± 1.5%). The average time for semi-automatic liver segmentation was 3.8 times faster compared to the manual slice-by-slice delineation. Main deviations between both approaches were observed in the areas of the gall bladder, vena cava inferior and hepatic vascular network. Furthermore, a second analyzed database consists of SPECT/CT and MRI datasets (or CT in case of contraindication for MRI) from pre-therapeutic imaging in 20 patients scheduled for SIRT. All 20 datasets were segmented manually and by using the semi-automatic algorithm of the analyzed software. The examination of the second database does not show any significant difference between both applied methodologies. Using the present tool, segmentations of the liver can be achieved with high accuracy while speeding up the physician’s workflow.