Multi-Stage Liver Segmentation in CT Scans Using Gaussian Pseudo Variance Level Set

No single technology can be rich enough to segment accurately due to the challenges of liver segmentation, which include low contrast with neighboring organs and the presence of pathology as well as highly varied shapes between subjects. This paper presents a Multi-stage framework for location and segmentation. First, Faster RCNN is employed to locate the liver region. Then, the Gaussian mixture model-based signed distance function is proposed to increase the flexibility of shape prior models. To reach the long and narrow ravine liver regions, the Gaussian pseudo variance level set is applied. Experimental results demonstrate the efficiency of the proposed method. More specifically, the proposed method is evaluated on 40 CT scan images, which are publicly available on three databases: SLIVER07, 3Dircadb, and LiTS. Our method has a slightly superior performance compared with other newly published methods.

Automated summary

This study presents a multi-stage framework for liver location and segmentation, using Faster RCNN for liver region localization and a Gaussian mixture model-based signed distance function to enhance shape prior flexibility. Experimental results demonstrate the efficacy of the proposed method on 40 CT scan images.