Mucosal Region Detection and 3D Reconstruction in Wireless Capsule Endoscopy Videos Using Active Contours V. B. Surya Prasath 1 , Isabel N. Figueiredo 2 , Pedro N. Figueiredo 3,4 , K. Palaniappan 1 Abstract— Wireless capsule endoscopy (WCE) provides an inner view of the human digestive system. The inner tubular like structure of the intestinal tract consists of two major regions: lumen - intermediate region where the capsule moves, mucosa - membrane lining the lumen cavities. We study the use of the Split Bregman version of the extended active contour model of Chan and Vese for segmenting mucosal regions in WCE videos. Utilizing this segmentation we obtain a 3D reconstruction of the mucosal tissues using a near source perspective shape-from- shading (SfS) technique. Numerical results indicate that the active contour based segmentation provides better segmenta- tions compared to previous methods and in turn gives better 3D reconstructions of mucosal regions. I. INTRODUCTION Wireless capsule endoscopy (WCE) is an expanding imag- ing modality that is revolutionizing medical diagnostics of the human intestine system [1]. It helps to see inside of the human colon, small bowel and other parts of the digestive system that are normally inaccessible with little or no discomfort to the patient. Usually the captured images (on the order of 50,000 frames) are analyzed by doctors to find anomalies such as bleeding, ulcers, polyps, etc [2], [3], [4]. Recently automatic image processing techniques have been applied to speed up the diagnostic process [5], [6], [7]; see [8] for a recent review. Endoscopic images from WCE provide a circular view of the intestine with mucosa folds and lumen sections, which are two major anatomical constituents of interest. Identification of mucosa folds from each frame of the video is necessary before mucosal based anomalies can be determined. Automated segmentation and labeling greatly facilitates further analysis by enabling physicians to focus on the most diagnostically relevant regions present in WCE video. A variety of image segmentation techniques exist and a number of methods have been applied to WCE and other endoscopic procedures [9], [10], [11]. Unlike other endoscopic imaging modalities of the intestine, WCE images usually contain multiple folds. Also, gastrointestinal liquids and other materials such as digested food can be present, This work was partially supported by the research project UTAustin/MAT/0009/2008 of the UT Austin | Portugal Program (http://www.utaustinportugal.org/) and by CMUC and FCT (Portugal), through European program COMPETE/FEDER. 1 V. B. Surya Prasath K. Palaniappan are with the Department of Com- puter Science, University of Missouri-Columbia, Columbia MO 65211, USA {prasaths,palaniappank}@ missouri.edu 2 Isabel N. Figueiredo is with the Department of Mathematics, University of Coimbra, Portugal isabelf@mat.uc.pt 3,4 Pedro N. Figueiredo is with the Faculty of Medicine, University of Coimbra, Portugal and also with the Department of Gastroenterology, University Hospital of Coimbra, Portugal pnf11@sapo.pt (a) (b) (c) Fig. 1. WCE mucosa segmentation and 3D reconstruction. (a) Input image with the initialization of the fast active contour scheme. (b) Final segmentation curve (at time t = 80) laid on top of the original image for visualization. (c) Segmentation based 3D reconstruction of the mucosa part using the shape from shading technique. see Figure 1 (images obtained by Pillcam TM colon capsule, Given Imaging, Yoqneam, Israel). Most of the clustering based segmentation schemes that are based on intensity contrast can fail to identify the mucosa folds properly. The images are non-homogeneous (nonlinear contrast range) in nature, hence thresholding based schemes cannot segment the mucosa folds precisely. In this paper, we report on an approach developed for efficient mucosa detection us- ing a modified variational segmentation scheme based on extending the Chan and Vese active contour without edges model [12] and a 3D reconstruction step using a perspective, near source shape-from-shading (SfS) approach. Perspective and near light source shape from shading models can be studied in a variety of ways [13], [14], [15] and we consider a variational approach which gives robust reconstructions. Numerical results indicate the efficiency of the variational approach for mucosa region segmentation, and further 3D reconstruction results illustrate the potential diagnostic use- fulness of our approach. The rest of the paper is organized as follows. In section II, the active contours without edges method for segmentation and a near-lights perspective SfS scheme are introduced. Section III presents experimental results on a database of WCE videos and Section IV discusses the results and show comparison with other related segmentation schemes. II. MUCOSA SEGMENTATION AND 3D RECONSTRUCTION A. Active contour based mucosa detection To avoid the drawbacks of threshold type schemes we make use of the active contour without edges (ACWE) method [12] which gives better segmentation results. Cre- mers et al [16] extended the ACWE model to include the variance terms for Gaussian distributions of the regions. Suppose that the input image is u :Ω ⊂ R 2 → R, where 34th Annual International Conference of the IEEE EMBS San Diego, California USA, 28 August - 1 September, 2012 4014 978-1-4577-1787-1/12/$26.00 ©2012 IEEE