Affiliations: [a] Department of Electronic and Information Engineering, Seoul National University of Science and Technology, Seoul, Korea | [b] Department of Electronic & IT Media Engineering, Seoul National University of Science and Technology, Seoul, Korea
Corresponding author: Gilwon Yoon, Department of Electronic & IT Media Engineering, Seoul National University of Science and Technology, 232 Gongneung-ro, Nowon-gu, Seoul 139-743, Korea. Tel.: +82 2 970 6419; Fax: +82 2 979 7903; E-mail: [email protected].
Abstract: This paper presents an image processing algorithm for the diagnosis of ulcers, which is a lesion occurring in the digestive tract, based on endoscopic images. In general, ulcers are visually distinguishable from normal tissues owing to the defective state in the mucosal membrane, cornea or skin tissue. Based on this characteristic, we used different colors to distinguish between ulcer and normal tissues in the proposed method. First, image luminance was adjusted to ensure similar luminance distribution values through a preprocessing stage in which the captured images were normalized to achieve uniform intensity distribution for each channel. Then, we selected distinctive elements for the detection of ulcer tissues with distinct image-associated chromatic characteristics. Because image luminance can affect detection even after preprocessing, we selected elements that were distinguishable from normal tissues based on the distribution of values displayed by ulcers from both RGB and HSV bands. Moreover, most of the digestive tract ulcers occur on the mucosal surface and tend to cluster together to form a specific zone. This implies that a detected ulcer pixel is more likely to be surrounded by ulcer tissue than normal tissue. Therefore, we used the intensity of each image channel as an additional detection element and performed ulcerative zone detection. An additional advantage of the zone detection process is the exclusion of errors caused by image-emanated random impulse noise. For performance evaluation of the image processing algorithm, we used fifty sheets of endoscopic images and conducted ulcer detection experiments. Finally, we validated our algorithm as showing 91.05% sensitivity and 98.64% specificity.