Fetal lung segmentation using texture-based boundary enhancement and active contour models

Abstract

Low signal noise ratio (SNR) of the ultrasound images makes the segmentation of fetal lung a difficult task. In this paper, a novel method using the texture-based boundary enhancement and active contour models is developed to semi-automatically segment the fetal lung from fetal chest ultrasound images. The texture-based boundary enhancement procedure is firstly proposed to enhance boundary regions by using multiple textural features. Then the Expectation Maximization (EM) algorithm followed by a morphological thinning process is applied to identify and obtain the interesting boundaries. Finally, three rectangular regions of interest (ROIs) are manually selected for the fetal chest, the fetal heart, and the fetal spine respectively. After initializing the deformation models, the vector field convolution model (VFC) extracts contours of the fetal chest, the fetal heart, and the fetal spine. The fetal lung is the region within the fetal chest but excluding the fetal heart and the fetal spine. Experiments on real clinical fetal chest ultrasound images demonstrate the feasibility of the proposed method.