褚晶辉, 王星宇, 吕卫
AuthorsHTML:褚晶辉, 王星宇, 吕卫
AuthorsListE:Chu Jinghui, Wang Xingyu, Lü Wei
AuthorsHTMLE:Chu Jinghui, Wang Xingyu, Lü Wei
Unit:天津大学电气自动化与信息工程学院，天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:针对乳腺磁共振图像序列的肿瘤分割问题, 提出一种基于超像素和改进C-V模型的三维全自动分割方法.该方法利用磁共振图像序列的帧间相关性, 约束相邻帧图像的分割轮廓.采用超像素算法提取肿瘤的大致轮廓, 再用改进的C-V水平集算法对可疑区域边缘进行优化, 使其更接近肿瘤的实际边缘. 将该方法及3种对比方法应用于89例乳腺MRI序列图像. 以手动分割的轮廓为基准, 该方法得到的平均重叠率为87.84% , 相比于C-V模型的58.90% 、超像素和水平集结合的76.36% 、K均值+C-V的83.62% , 有明显提升. 实验结果表明, 该方法的全自动分割结果对于肿瘤起始和终止帧图像具有较高的分割精度.
Abstract_English:An automatic 3D segmentation method for tumor mass in breast magnetic resonance image(MRI) image sequence which uses superpixel segmentation and modified C-V model is proposed in this paper. The proposed method explores the inter-frame correlation in the MRI image sequence to constrain the contour in the adjacent frames. The SLIC0 superpixel method is adopted to extract the outline of masses，and then the modified C-V level set algorithm is adopted to refine the border of the suspicious area to approximate the real border. In experiments，the proposed method and three competitive methods were applied to 89 cases of breast MRI sequential images. Compared with the contour of manual segmentation，the average overlap rate of the proposed method is 87.84%，and it is 58.90% for C-V model，76.36% for superpixel combined with level set and 83.62% for K-means +C-V. It is also demonstrated that the automatic segmentation results of the proposed method have a higher accuracy for the start and the end frame of the breast mass.
Keyword_Chinese:乳腺磁共振图像; 病灶分割; 帧间相关性; 超像素; 水平集
Keywords_English:breast magnetic resonance image; lesion segmentation; inter-frame correlation; superpixel; level set

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=5873