P. Matthies, J. Wuestemann, J. Neba, F. Pinto, O. S. Grosser First validation of semi-automatic liver segmentation algorithm World Congress on Medical Physics and Biomedical Engineering, Prague, Czech Republic, June 2018 (bib)
	Selective internal radiotherapy (SIRT) of liver malignancies requires a careful evaluation of patient's condition including liver size, tumour burden, hepatic arterial vessel supply, and liver lung shunt. The segmentation of the liver is a time-consuming step to the physician. In this work, we validated an interactive software tool which provides a semi-automatic segmentation of anatomical 3-dimensional datasets, in this case contrast-enhanced abdominal CT scans. Methods: A medical assistant with experience in manual liver segmentation for SIRT dosimetry employed dosimetry software (DosePlan, SurgicEye GmbH, Germany) to segment the contrast-enhanced CT datasets for ten patients and documented the completion time. The software provides an interactive segmentation based on user-defined initial seeds of the foreground and background regions of several slices in orthogonal views. The segmented region is then computed based on these seeds by a graph-based segmentation algorithm. Ten out of the 130 public datasets from the "Liver Tumor Segmentation Challenge" were used, including the provided ground truth segmentation of the liver which was acquired by a board-certified radiologist and verified by another expert. While no timings are provided for these datasets, timing data based on 35 manual liver segmentations yielded 32:48 +/- 6:54min. Segmentation quality vs. the ground truth was evaluated using the Sørense-Dice coefficient. Results: The average Sørense-Dice score for the reader was 92.9%, with a standard deviation of 1.5%. The average time to segment the entire liver using the interactive algorithm was 8:45 +/- 1:13min, a factor of 3.8 times fast than the manual slice-by-slice delineation. The main differences between the segmentations were in the areas of the gall bladder, vena cava and hepatic vascular network. Conclusion: Using the semi-automatic segmentation algorithm employed, segmentations of the liver can be achieved with a very high accuracy of >92% in under 9min, greatly speeding up the physician's workflow.
	This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each authors copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.