Abstract:
Magnification endoscopy with narrow-band imaging (ME-NBI) has been widely used for cancer diagnosis. However, there is a white opaque substance (WOS) whose main component is lipids, making some microstructures invisible. In such lesions, the morphological structure of lipids becomes another marker of tumor grade. In this paper, a lipids segmentation method is proposed. Firstly, the lipid image enhancement algorithm and the specular reflection correction algorithm are introduced. Then, in the framework of active contour model, we propose a segmentation method, which extracts local information from modified hue value, global information from intensity value, adaptively obtains weight factor to segment the lipid region based on the initial contour. The effectiveness of our method is verified by phantom experiment, which shows that the pixel accuracy, sensitivity, and Dice coefficient of our method are all higher than 90%. The proposed method can well reflect the shape of lipids to provide