1450 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 21, NO. 12, DECEMBER 2002 Automatic Centerline Extraction for Virtual Colonoscopy Ming Wan, Zhengrong Liang*, Qi Ke, Lichan Hong, Ingmar Bitter, and Arie Kaufman Abstract—In this paper, we introduce a concise and concrete def- inition of an accurate colon centerline and provide an efficient au- tomatic means to extract the centerline and its associated branches (caused by a forceful touching of colon and small bowel or a deep fold in twisted colon lumen). We further discuss its applications on fly-through path planning and endoscopic simulation, as well as its potential to solve the challenging touching and colon collapse prob- lems in virtual colonoscopy. Experimental results demonstrated its centeredness, robustness, and efficiency. Index Terms—Centerline extraction, distance mapping, flight-path planning, virtual colonoscopy (VC). I. INTRODUCTION V IRTUAL endoscopy is an integration of medical imaging and computer graphics technologies, leading to a com- puter-based alternative to the traditional fiberoptic endoscopy for examining the interior structures of human organs [17], [19], [24]. It has many advantages compared with the traditional en- doscopy procedures, such as being noninvasive, cost-effective, highly accurate, free of risks and side effects (e.g., perforation and infection), and easily tolerated by the patient. Therefore, many prototype systems have been developed for a variety of clinical applications, including virtual colonoscopy (VC), vir- tual bronchoscopy, virtual angioscopy, and others. We have been developing a three-dimensional (3-D) VC system [12], [13] to- ward a fast and accurate computer-aided screening modality for early detection of colonic polyps, which are the major cause ( 90%) of colon cancer, the second leading cause of cancer deaths in the USA. Our VC system takes a spiral computed tomography (CT) scan of the patient’s abdomen after the colon is cleansed and Manuscript received June 6, 2001; revised August 14, 2002. This work was supported by the National Cancer Institute under Grant NIH CA82402, by the National Science Foundation under Grant MIP9527694, and by Viatronix Inc. The Associate Editor responsible for coordinating the review of this paper and recommending its publication was W. Higgins. Asterisk indicates corresponding author. M. Wan is with The Boeing Company, Seattle, WA 98124-2207 USA (e-mail: Ming.Wan@Boeing.com). *Z. Liang is with the Department of Computer Science and the Department of Radiology, StateUniversity of New York at Stony Brook, Stony Brook, NY 11794-8460 USA (e-mail: jzl@clio.rad.sunysb.edu). Q. Ke and I. Bitter are with the Department of Computer Science, State Uni- versity of New York at Stony Brook, Stony Brook, NY 11794 USA (e-mail: qke@cs.sunysb.edu; ingmar@cs.sunysb.edu). L. Hong is with the Palo Alto Research Center, Palo Alto, CA 94304 USA (e-mail: hong@parc.com). A. Kaufman is with the Department of Computer Science and the Department of Radiology, StateUniversity of New York at Stony Brook, Stony Brook, NY 11794 USA (e-mail: ari@cs.sunysb.edu). Digital Object Identifier 10.1109/TMI.2002.806409 distended with room air or CO gas. Several hundred high-res- olution slice images are rapidly acquired during a single breath hold by currently available CT technology, forming a volumetric abdomen dataset. A model of the entire colon is then extracted from the abdomen dataset, where the tagged (by contrast so- lutions) residual stool and fluid are electronically removed by image segmentation algorithms from the dataset [6], [14], [16]. A centerline of the colon model is then determined, where a po- tential field is built within the colon lumen. The colon model can be viewed by an automatic planned navigation following the centerline for a general overview [12], or by an interactive navigation for a more flexible and detailed study on suspicious regions [13], [26]. The navigation is based on volume rendering [25] and executed in real time on a personal computer (PC) plat- form [8]. A crucial component of our VC system is the extraction of the centerline, which not only provides a compact colon shape description, accurate colon geometry measurement, and pre- cise polyp registration with fiberoptic colonoscopy or between supine and prone CT scans, but also supports automatic path planning for both planned and interactive navigations. An accu- rate and fast extraction of the centerlines from patient datasets has been a challenging problem, due to the complex structure of the colon. In Section II, we will review the centerline concepts and related algorithms and introduce our centerline definition and its extraction algorithm. II. THEORY The concept of centerlines (also known as medial or sym- metric axes) was first introduced by Blum [3]. In a tubular ob- ject like the colon, there is normally only one centerline that spans it. In more general cases, an object may have a more complicated shape—such as airways in the lungs—and, there- fore, there is a set of centerlines attaching to each other through the object, forming a topology similar to a skeleton. Hence, the set of connected centerlines in an object is also called a skeleton. (Skeleton and centerlines are used interchangeably in this paper.) A concise definition of “skeleton” was given in [3] as the locus of centers of maximal disks (in two dimensions) or balls (in three dimensions) contained in the shape. Extracting the skeleton has been a very challenging task in various appli- cations, resulting in various modifications on the centerline def- inition, so that it can be extracted and utilized efficiently. In the following, we will focus on the colon object. Based on previous reports [1], [2], [5], [7], [10]–[13], [18], [22], [23], an adequate colon centerline definition and extraction should meet the following requirements. 0278-0062/02$17.00 © 2002 IEEE