【作者】 张玮； 谢凤英； 秦世引； 孙应实； 张晓鹏；

【Author】 Zhang Wei Xie Fengying Qin Shiyin Sun Yingshi Zhang Xiaopeng ( School of Automation, Beihang University, Beijing 100083; Image Processing Center,Beihang University, Beijing 100083 Radiology, Beijing Cancer Hospital .Beijing 100036)

【机构】 北京航空航天大学自动化科学与电气工程学院； 北京航空航天大学图像中心； 北京肿瘤医院放射科；

【摘要】 为成功实现直肠癌DWI(Diffusion Weighted Ima- ging)图像的序列分割,根据同一序列相邻图像内肿瘤位置变化不会太大的特点,本文提出了一种结合大津(Otsu)阈值和区域生长的分割算法。算法首先应用Otsu阈值算法来对 DWI序列中的单张切片进行感兴趣区(ROI)内准确分割;对相邻切片也进行相同ROI的Otsu阈值分割,并依此阈值判断相邻切片是否存在肿瘤,如存在则进行形态学腐蚀和连通域标记;根据这两张切片来确定种子点,利用区域生长算法来自动提取相邻切片中的目标,最后更新ROI为分割下张切片做准备。本算法仅需一次用户交互就能快速准确的从 DWI序列中分割出目标——肿瘤,具有算法简单,效率高的优点。更多还原

【Abstract】 In this paper, a segmentation algorithm for the rectal cancer DWI series is proposed which is based on the technique of one-dimension maximal variance between-class and the region growing method, with some prior knowledge that the tumor location would not change too much in the nearby slices of the same DWI series. Firstly, the ROI ( region of interest) in a slice of DWI series is segmented by the Otsu threshold algorithm ; and then the same region in the nearby slice is segmented by the Otsu threshold algorithm too, and whether there are tumors or not is judged by the threshold, if there are tumors, morphological erosion and connected domain labeling are operated; finally, the marker it obtained according to the two slices, and the nearby slice is segmented by the region growing method. After the segmentation, the ROI is renewed to prepare the segmentation of the next slice. Experiment shows that the algorithm described in this paper can obtain the boundary of the desired object from a series of medical images quickly and reliably with only little user intervention.更多还原