基于改进的CV-RSF模型的甲状腺结节超声图像自适应分割算法  

Adaptive segmentation algorithm for thyroid nodules based on improved CV-RSF model

作　　者：邵蒙恩 严加勇 崔崤峣[3] 于振坤[4] SHAO Mengen;YAN Jiayong;CUI Xiaoyao;YU Zhenkun(University of Shanghai for Science and Technology,Shanghai 200093;Zhoupu Hospital affiliated to Shanghai University of Medical & Health Science,Shanghai 201318;Institute of Biomedical Engineering and Technology,Chinese Academy of Sciences,Suzhou,Jiangsu Province 215163;Tongren Hospital of Nanjing,Nanjing 211102)

机构地区：[1]上海理工大学医疗器械与食品学院,上海200093 [2]上海健康医学院附属周浦医院,上海201318 [3]中科院苏州生物医学工程技术研究所,江苏苏州215163 [4]南京同仁医院,南京211102

出　　处：《北京生物医学工程》2020年第3期251-256,共6页Beijing Biomedical Engineering

基　　金：江苏省省级重点研发专项资金项目(BE2017601);上海市浦东新区科技发展基金民生科研专项医疗卫生项目(PKJ2017-Y41)。

摘　　要：目的甲状腺结节超声图像的精确分割对甲状腺结节的良恶性诊断尤为重要。目前,对于甲状腺结节超声图像的分割,有学者提出利用主动轮廓模型分割算法,但是由于活动轮廓分割算法需要手动设置迭代次数,未实现模型的自适应性。因此,本文提出了一种基于改进的无边缘主动轮廓-局部区域可控的拟合(Chan-Vese-region scalable fitting,CV-RSF)模型的甲状腺结节超声图像自适应分割算法。方法选取南京同仁医院12例患者的甲状腺结节超声图像用于实验。首先,在无边缘主动轮廓(Chan-Vese,CV)模型中,引入一个基于梯度的边缘引导函数,根据面积变化率,自适应地获取甲状腺结节的粗分割轮廓;然后,将粗分割轮廓作为局部区域可控的拟合(region-scalable fitting,RSF)模型的初始轮廓,并根据面积变化率,自适应地获取甲状腺结节最终分割结果。将改进模型分割的结果与CV模型、RSF模型分割的结果进行比较,并分析甲状腺结节边缘清晰度对分割结果的影响。结果本文模型算法分割结果的平均迭代次数、平均面积重叠率、平均Hausdorff分别达到了134、90.34%、9.77,均优于CV模型、RSF模型的分割算法。结论该算法有效地分割出边缘清晰和不清晰的甲状腺结节超声图像,并解决手动设置迭代次数的问题,从而实现甲状腺结节的有效、准确、自动分割。Objective The accurate segmentation of thyroid nodule ultrasound images is particularly important for the diagnosis of benign and malignant thyroid nodules.Currently,the segmentation of thyroid nodule ultrasound images have been proposed by some scholars who are using an active contour model segmentation algorithm.However,the active contour segmentation algorithm requires manual setting of the number of iterations so that model’s adaptability has not been achieved.Therefore,this paper proposes an adaptive segmentation algorithm for thyroid nodule ultrasound images based on an improved edgeless active contour-local region controllable fitting[Chan-Vese-Region scalable fitting(CV-RSF)model].Methods Ultrasound images of thyroid nodules from 12 patients in Nanjing Tongren Hospital are used for the experiment.Firstly,in a Chan-Vese(CV)model,a gradient-based edge guidance function is introduced to adaptively obtain the rough segmented contour of the thyroid nodule based on the area change rate.Then,the rough segmented contour is regarded as the initial contour of a region-scalable fitting(RSF)model.At last,the final segmentation results of thyroid nodules are adaptively obtained based on the area change rate.The results of the improved model segmentation are compared with the results of the CV model and RSF model segmentation,and the influence of thyroid nodule edge sharpness on segmentation results are analyzed.Results In this study,ultrasound images of thyroid nodules from 12 patients in Nanjing Tongren Hospital are used for the experiment.The proposed model is superior to the CV model and RSF model,which achieves an average number of iterations of 134,an average area overlap ratio of 90.34%and an average Hausdorff distance of 9.77.Conclusions This algorithm effectively segments thyroid nodule ultrasound images with clear and unclear edges.Meanwhile,it does not need to manually set the number of iterations to achieve effective,accurate and automatic segmentation of thyroid nodules.

关 键 词：CV-RSF模型 边缘引导函数 甲状腺结节 超声图像 自适应分割算法 

分 类 号：R318.04[医药卫生—生物医学工程]