Saliency Based Ulcer Detection using Wireless Capsule Endoscopy Diagnosis
Anjali S. Jadhav1, Smita. S. Ponde2
1Anjali S. Jadhav, ME-Student, Department of Computer Science and Technology, Deogiri Institute of Engineering & Management Studies, Aurangabad (Maharashtra). India.
2Smita. S. Ponde, Assistant Professor, Department of Computer Science And Technology, Deogiri Institute of Engineering & Management Studies, Aurangabad (Maharashtra). India.
Manuscript received on October 15, 2016. | Revised Manuscript received on October 23, 2016. | Manuscript published on November 05, 2016. | PP: 13-17 | Volume-6 Issue-5, November 2016. | Retrieval Number: E2921116516/2016©BEIESP
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Published By: Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Ulcer is one of the most common indications of many serious diseases in the human digestive tract. Especially for ulcers in the small intestine where other methods may not display properly, capsule endoscopy (WCE) is increasingly used in the diagnosis and clinical management. Since WCE generates lots of images of the entire inspection process, computer-aided detection ulcer is considered an essential relief for clinicians. In this work, a CAD system is proposed for fully automated computer in two stages to detect images WCE ulcer. In the first step, a detection method based on the effective prominence superpixel multilevel outline representation candidates proposed ulcer. To find the perceptual and semantically meaningful salient regions, the first image segment in multi-level superpixel segmentations. Each level corresponds to different initial sizes of super pixels region. Then the corresponding prominence according to the characteristics of color and texture of each level superpixel region is evaluated. At the end, we merge the salience maps of all levels together to obtain the final saliency map. The experiment results achieved promising accuracy 94.72% 94.63% sensitivity and, validating the effectiveness of the proposed method. Moreover, the results of the comparison show that our detection system outperforms the methods of prior art in the detection task of the ulcer.
Keywords: linear-town with limited coding (LLC), multilevel superpixel representation, prominence and the max-sharing method based on prominence.