(A) locally adaptive region growing algorithm to segment tree-like tubular structures in 3-D medical images

Abstract

With the rapid development of 3-D imaging modalities and very powerful computer hardware, medical visualization systems have been developed for wide ranges of applications. However, one of the current limitations in building such systems is that the time-consuming preprocessing step to segment various organs is indispensable for accurate quantitative analysis. Especially, an accurate description of vessel structures is very important in most clinical applications. Compared to other human organs, vascular structures have some discernable characteristics as follows. First they are usually composed of thin narrow pipes to construct a tree-like organism. However, the detection of abnormal shapes is thought to be more important than that of normal shapes in a clinical sense. Finally, it should be noticed that the intensity values inside a single vascular structure are not constant due to various factors, such as density changes of contrast agent. To segment vascular structures in 3-D CTAIMRA images, a new region-growing algorithm on the basis of local cube tracking is proposed. In this method, a small local cube is adopted to segment a vessel s egment, and the following local cubes are determined b ased o n its s egmentation result. T his procedure is repeated until whole segmentation is completed. Inside a local cube, the proposed method performs the minimum path analysis and cost-histogram analysis to extract the two kinds of seed areas; one for vessel of interest and the other for non-vessel objects. Using the extracted two kinds of seed areas, a competitive region-growing algorithm is applied to produce more precise vessel boundaries. By confining all the above processes to the inside of each local cube, a robust result can be obtained even in a tubular structure having steadily changing intensities. Through the experiment on a phantom dataset, it is proved that the proposed method can segment an object of interest quite well even in a very noisy environment. Co...