Fast Automated Liver Delineation from Computational Tomography Angiography

Accurate liver segmentation is essential for surgery planning and diagnosis of liver abnormality with algorithms. We propose and validate a multi-atlas segmentation approach with local decision fusion for fast automated liver (with/without abnormality) segmentation on computational tomography angiography (CTA). Thirty-five patients were enrolled in this study. A co-registered segmented CTA atlas is constructed with 20 CTA scans, normal and abnormal subjects with wide range of body-mass index (BMI). Liver segmentation candidates are achieved by a multi-atlas registration algorithm which propagates the segmentation label on each atlas image to the test image by image registration. The final segmentation result is calculated by applying local decision fusion weights to each propagated candidate segmentation. We applied our algorithm on the rest 15 patients and compared them with manual segmentation by two expert readers. Voxel overlap by Dice coefficient between the algorithm and expert readers was 0.93 (range 0.89 - 0.94). The mean surface distance and Hausdorff distance in millimeters between manually drawn contoursand the automatically obtained contours were 1.1 ± 0.9 mm and 5.9 ± 1.7 mm respectively. Using our approach, physicians can accurately segment liver from CTA without tedious manual tracing. Our automated algorithm for liver segmentation achieved accurate segmentation with/without abnormality.