Liver Segmentation Using a Statistical Shape Model Based on Learning Local Objective Functions

Abstract

This Model-based segmentation approach plays an important role in organ segmentation from 3D CT volumes, but the accuracy is not enough of organs which are of the large variations in the shapes and the intensity pattern inside and along the boundaries. However, the methods based statistical shape models (SSMs) have a better performance among the approaches. But the automatic liver segmentation is still a challenging work. In this paper, a novelty method is proposed..which is based on statistical shape model integrated a new intensity model..combining the edge information with region intensity information for liver segmentation..First, edge information is extracted from the distance image obtained by distance translation. Second, we analyze region intensity information and construct a weighted Gaussian mixture model. Then an objective function is learned from the training data to fit the ideal objective function, which dynamically adjust the weights of the different features in image segmentation procedure. The experiment results evaluated on the test data, demonstrate the approach not only produces excellent segmentation accuracy, but also increases the robustness.