Abstract
In living-donor liver transplantation, the safety of the donor is critical. In addition, accurately measuring the liver volume is significant as the amount that can be resected from living donors is limited. In this paper, we propose an automated segmentation and volume estimation method for the liver in computed tomography imaging based on a deep learning-based segmentation network. Our framework was trained using the data of 191 donors, achieved a dice similarity coefficient of 0.789, 0.869, 0.955, and 0.899, respectively, in the segmentation task for the left lobe, right lobe, caudate lobe, and whole liver. Moreover, the R^2 score reached 0.980, 0.996, 0.953, and 0.996 in the volume estimation task. We demonstrate that our approach provides accurate and quantitative liver segmentation results, reducing the error in liver volume estimation. Therefore, we expected to be used as an aid in estimating liver volume from CT volume data for living-donor liver transplantation.