MRI胸腔图像中肺组织的自动分割  

Automatic segmentation of lung tissue in thoracic MRI images

作　　者：杨维萍[1] 罗洪艳[1] 张绍祥[2] 吴毅[2] 胡南[1] 潘进洪 

机构地区：[1]重庆大学生物工程学院,重庆400044 [2]第三军医大学基础医学部解剖学教研室,重庆市数字医学研究所,重庆400038 [3]第三军医大学南医院全军泌尿外科研究所,重庆400038

出　　处：《第三军医大学学报》2011年第11期1170-1173,共4页Journal of Third Military Medical University

基　　金：国家自然科学基金(60871099,30900323);中央高校基本科研业务费资助(CDJXS10231122);重庆市科技攻关项目(CSTC2010AC2025)~~

摘　　要：目的探讨实现MRI胸腔图像中肺组织自动分割的方法。方法基于MRI胸腔图像,首先对整套图像数据集进行预处理,以增强图像中肺组织区域和周围组织的对比度。采用由粗到细的肺组织自动分割算法。该算法先采用阈值法分割出粗略的肺实质,然后通过标记连通区域统一背景,最后利用形态学的方法平滑肺边缘、填补肺区内的细小缺失并去掉黏连的气管及主支气管。结果自动分割的肺组织边缘轮廓清晰,与原图基本吻合,而且与手工分割结果的一致性较好。各时相点的平均DSI系数均超过89%,可视为有效分割。结论该算法能准确有效的自动分割出一个呼吸周期中不同时相点的MRI胸腔图像序列中的肺组织。Objective To investigate on the possibility of automatic segmentation of lung tissue in the thoracic MRI images.Methods According to the features of thoracic MRI images,an algorithm for the segmentation of lung tissue was proposed.It was characterized by the initial rough segmentation and then followed by fine segmentation.Initially,the region of lung tissue was extracted roughly in the threshold method.Then the background was unified by means of the connected region labeling.Finally,the morphological method was adopted to smooth the edge of the segmented lung tissue,fill the internal small defects and remove the attached trachea as well as main bronchus.Results The edges of the automatically segmented lung tissues in all images were distinct and almost matched the original ones.The results were also consistent with the manually segmented ones.The average dice similarity index was beyond 89%,and could be identified as effective segmentation.Conclusion The algorithm is accurate and valid for automatic segmentation of lung tissue in the thoracic MRI images collected at different instants in a respiratory cycle.

关 键 词：MRI 肺组织 分割 连通标记 形态学 

分 类 号：R445.2[医药卫生—影像医学与核医学] R816.4[医药卫生—诊断学]