基于变分法的超声乳腺肿瘤分割  

Ultrasonic Breast Tumor Segmentation Based on Variational Method

作　　者：许珊 陈科[1] 林江莉[1] XU Shan;CHEN Ke;LIN Jiangli(School of Materials Science and Engineering,Sichuan University,Chengdu 610065 China)

机构地区：[1]四川大学材料科学工程学院,四川成都610065

出　　处：《西华大学学报（自然科学版）》2021年第3期15-22,共8页Journal of Xihua University:Natural Science Edition

基　　金：国家自然科学基金(81571697);四川省应用基础研究(2019YJ0055)。

摘　　要：乳腺肿瘤区域的选择性分割是乳腺肿瘤计算机辅助诊断中的关键一步。由于变分法灵活且计算简单,因此文章采用改进的Mumford-Shah(MS)变分模型分割超声乳腺肿瘤,以获得较准确的肿瘤边界。首先利用医生标记的4个点获得近似椭圆,再自动选取椭圆边界的4个点,共8个点作为目标区域的标记点;然后利用边缘函数和距离函数构造加权函数,并将加权函数与MS模型结合形成加权选择性分割函数,该函数在目标区域周围有更大的值,其余区域有更小的值,从而实现精确分割。实验结果表明:该方法减少了MS模型人工取点的数量;乳腺肿瘤区域分割准确率均值为90%,交并比(IOU)均值为85%。The selective segmentation of breast tumors is a key step in the computer-aided diagnosis of breast tumors.The variational method is flexible and simple to calculate.In order to accurately obtain the breast tumor area,this article used a Mumford-Shah(MS)model to segment ultrasound breast tumors.First,the approximate ellipse is obtained by using the four points marked by the doctor,and then the four points of the elliptical boundary are automatically selected as the mark points of the target region.Then the weighted function is constructed by using the edge function and the distance function,and the weighted function is combined with the MS function to form the weighted selective segmentation function.As a result,the resulting restriction function has a larger value around the target area,and a smaller value in the rest of the area.And select extra points on the ellipse to selectively segment the breast tumor area.Experiments were carried out and the results show that this method greatly reduces the number of manually marked points for the MS model,and the accuracy rate of tumor segmentation is 90%and the intersection-over-union(IOU)is 85%.

关 键 词：图像识别 图像分析 超声乳腺肿瘤 选择性分割 椭圆拟合 MUMFORD-SHAH模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] R737.9[自动化与计算机技术—计算机科学与技术]