M.S. Thesis Defense - Dong Hong

Date: 20 Nov, 2008
Time: 10:00 AM
Location: 223 Atanasoff Hall
Topic: Semi-Automatic Approach for 3D Reconstruction of Colon Segments from Colonoscopy Images
Major Professor(s): Wallapak Tavanapong


Abstract:

We propose a new algorithm for reconstructing a 3D virtual colon segment from an individual image captured from colonoscopy. Colonoscopy is currently the preferred screening modality for colorectal cancer, which claims about 50,000 lives annually in the US. During colonoscopy, the camera at the tip of the colonoscope generates a video signal of the interior of the human colon, which is displayed on a monitor for real-time analysis by the endoscopist. The algorithm takes manually drawn contours of the colon folds of a colonoscopy image as input and calculates the depth of each fold from the camera. Next, it calculates the direction that each fold is facing based on the intensity of the contour pixels. Finally, the algorithm creates the colon wall via interpolation between two neighboring folds. The proposed algorithm is an important step toward 3D reconstruction of the virtual colon from video of an entire colonoscopy procedure. The reconstruction of the virtual colon is potentially useful for 1) understanding the colon structure of the patient and the endoscopistÂ’s navigation technique at the time of colonoscopy; 2) determining the estimated location and the area of the colon mucosa inspected by the endoscopist, which is an important quality indicator of the procedure; 3) improving effectiveness of automatic detection of abnormal lesions (e.g., polyps, diverticula) using the 3D information; and 4) bridging the gap between virtual colonoscopy and optical colonoscopy. No other similar techniques were reported in the literature. Existing 3D reconstruction techniques using colonoscopy images focus on reconstruction of the colon surface for image-guided automated colonoscopy or surgical planning. These techniques cannot reconstruct the shape of the colon segment. Experimental results on real colonoscopy images indicate that our proposed technique is promising. The generated virtual colon images look similar to the input colonoscopy images with a few exceptions.