Spatially Controllable Region Growing for Segmenting Heart Chambers

Abstract

Segmentation of computed tomography (CT) images has provided promising methods of constructing precise 3-dimensional heart models. However, the process is labor intensive, because heart regions such as cardiac chambers and blood vessels have similar intensities and exist within a small space. In this paper, we present a tool to efficiently segment cardiac chambers and blood vessels. We extend traditional region growing to be spatially controllable. A user places multiple seeds, each having a bounding area and a threshold, and our tool “grows” regions around each seed independently within its bounding area. To efficiently specify the bounding area, we propose two types of seeds (i.e., sphere and cylinder). We also provide a negative seed that generates fixed background to avoid over-extraction errors. We compared our tool with a traditional scissor tool and confirmed that ours significantly reduced the time required for a segmentation task. We also present segmentation results of CT images of hearts having congenital diseases to illustrate the feasibility of our tool.