TY - JOUR AU - Tavanapong, Wallapak AU - Hong, Dong Ho AU - Wong, Johnny AU - de Groen, Piet C. AU - Oh, Jung Hwan PY - 2017/06/15 TI - Reconstruction of a 3D Virtual Colon Structure and Camera Motion for Screening Colonoscopy JF - Medical Research Archives; Vol 5 No 6 (2017): Vol.5 Issue 6, June, 2017 KW - Endoscopic image analysis, 3D reconstruction algorithms, visualization, screening colonoscopy N2 - Background/Purpose Optical colonoscopy is the gold standard for detection and prevention of Colorectal cancer. However, any lesion outside the field of view of the camera during colonoscopy is missed, and can develop from a polyp into a cancer or from a small cancer into a large, metastatic cancer. Hence, viewing the colon mucosa as much as possible is desirable. The ultimate goal is to use the reconstruction results to mark the area likely unobserved during colonoscopy where polyps may be hidden. Materials and Methods: This paper presents a novel method for reconstructing (i) a 3D virtual colon structure from a sequence of 2D colonoscopic images and (ii) the endoscope camera navigation path. Unlike existing work that focus on reconstruction of accurate colon surface for a computer-aided surgery, this work focuses on estimating the alignment of the colon haustral folds, thickness and protrusion of the folds, and the camera path for computer-aided screening colonoscopy to inspect as much as possible of the colon mucosa. Results: On endoscopy video of a synthetic colon model, we achieved at least 73% and 90% accuracy in estimating the directions of camera translation and rotation motions, respectively. The average percentage of the depth error is about 10% (8.5/85.8) of the average depth of all the folds seen in all the image sequences. The average percentage of the circumference error is about 10% (17/178.2) of the average circumference of all the folds seen in all the image sequences. The results are promising and give further insight to address this challenging problem. Conclusions : We present the work on reconstruction of a colon structure and endoscope motion from a colonoscopic image sequence for screening colonoscopy. We overcame several challenges such as the scale difference of consecutive reconstructed models, estimating closed fold contours from tracked fold edges, and registering the models. As future work, we will investigate a multi-frame registration approach to further reduce the depth and circumference estimation errors by registering nearby frames that are more similar in fold shape together. UR - https://esmed.org/MRA/mra/article/view/1351