Automatic Multi-Organ Segmentation from Abdominal CT Volumes with LLE-based Graph Partitioning and 3D Chan-Vese Model
Ping Tang,Yu-qian Zhao,Miao Liao
DOI: https://doi.org/10.1016/j.compbiomed.2021.105030
IF: 7.7
2021-01-01
Computers in Biology and Medicine
Abstract:This paper presents a fully automatic method for multi-organ segmentation from 3D abdominal CT volumes. Firstly, spines and ribs are removed by exponential transformation and binarization to reduce the disturbance to subsequent segmentation. Then, a Local Linear Embedding (LLE)-based graph partitioning approach is employed to perform initial segmentation for liver, spleen, and bilateral kidneys simultaneously, and a novel segmentation refinement scheme is applied composed of hybrid intensity model, 3D Chan-Vese model, and histogram equalization-based organ separation algorithm. Finally, a pseudo-3D bottleneck detection algorithm is introduced for boundary correction. The proposed method does not require heavy training or registration process and is capable of dealing with shape and location variations as well as the weak boundaries of target organs. Experiments on XHCSU20 database show the proposed method is competitive with state-of-the-art methods with Dice similarity coefficients of 95.9%, 95.1%, 94.7%, and 94.5%, Jaccard indices of 92.2%, 90.8%, 90.0%, and 89.5%, and average symmetric surface distances of 1.1 mm, 1.0 mm, 0.9 mm and 1.1 mm, for liver, spleen, left and right kidneys, respectively, and the average running time is around 6 min for a CT volume. The accuracy, precision, recall, and specificity also maintain high values for each of the four organs. Moreover, experiments on SLIVER07 dataset prove its high efficiency and accuracy on liver-only segmentation.