基于窄带成像及放大内镜的脂质分割方法  

Lipid segmentation method based on magnification endoscopy with narrow-band imaging

作　　者：武治晟 邹鸿博 朱文武 齐伟明 王立强[1] 袁波[1] 杨青[1] 徐晓蓉 严蕙蕙[3] WU Zhi-sheng;ZOU Hong-bo;ZHU Wen-wu;QI Wei-ming;WANG Li-qiang;YUAN Bo;YANG Qing;XU Xiao-rong;YAN Hui-hui(College of Optical Science and Engineering,Zhejiang University,Hangzhou 310027,China;Zhejiang Center for Medical Device Evaluation,Hangzhou 310000,China;Department of Gastroenterology,the Second Affiliated Hospital,Zhejiang University School of Medicine,Hangzhou 310009,China)

机构地区：[1]浙江大学光电科学与工程学院,浙江杭州310027 [2]浙江省医疗器械审评中心,浙江杭州310000 [3]浙江大学医学院附属第二医院消化内科,浙江杭州310009

出　　处：《中国光学（中英文）》2024年第4期982-994,共13页Chinese Optics

基　　金：国家自然科学基金(No.T2293751);国家重点研发计划(No.2021YFC2400103)。

摘　　要：一种主要成分是脂质的白色不透明物质(WOS)会覆盖与癌症诊断有关的微观结构,但WOS的形态特征又与肿瘤分级有密切关系。为了给医生提供更多与脂质相关的可用信息,本文对脂质图像的分割方法进行了研究。首先,介绍了基于Retinex框架的脂质图像增强算法,并介绍了反光去除算法。然后,介绍了基于活动轮廓模型的脂质分割方法,该方法从校正后的色调值中提取局部信息,从强度值中提取全局信息,自适应地获得权重因子,并基于初始轮廓来分割脂质区域。最后,基于自研细胞内镜成像系统,设计了仿体实验来验证了该方法的有效性。实验结果表明,该分割方法的像素准确度、灵敏度、Dice系数均高于90%。该方法能够克服照明不均匀、反光等的影响,很好地反映脂质的形状,为医生提供可用的信息。Magnification endoscopy with narrow-band imaging(ME-NBI)has been widely used for cancer diagnosis.However,some microstructures are rendered invisible by a white opaque substance(WOS)com-posed mainly of lipids.In such lesions,the morphological structure of lipids becomes another marker of tu-mor grade.We propose a lipid segmentation method.First,the lipid image enhancement algorithm and the specular reflection correction algorithm are introduced.Then,in the framework of the active contour model,the proposed segmentation method extracts local information from modified hue value and global informa-tion from intensity value and adaptively obtains the weight factor to segment the lipid region based on the ini-tial contour.This method’s effectiveness is verified by a phantom experiment,which shows that it attained higher than 90%in several key measures:pixel accuracy,sensitivity,and Dice coefficient.The proposed method can accurately reflect the shape of lipids to provide available information for doctors.

关 键 词：窄带光成像 脂质 分割 活动轮廓模型 仿体 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]