Automatic segmentation of pulmonary fissures is a nontrivial task in CT(Computer Tomography) chest images,due to incomplete,disrupted,deformed and accessory fissures. In this paper,we present a approach to fuse pulmonary structure characteristics for fissure segmentation. Firstly,we fuse the prior knowledge of trachea and pulmonary arteries to i-dentify fissure region of interest. Then fissures directional field is exploited to enhance fissures and a multi-plane filter is pro-posed to remove noise for fissure pre-segmentation. Finally fissure region of interest and fissure pre-segmentation are com-bined for fissure segmentation. Compared with manual fissure references,our method obtained a high segmentation accuracy with median F1-score of 0.881 and 0.878 for the left and right lung images respectively.%CT(Computer Tomography)图像中自动分割肺裂是很困难的,肺裂往往存在不完整、形变、断裂和附裂等现象.本文提出一种融合肺部解剖结构特征来实现自动分割肺裂的方法.首先结合肺部气管和动脉血管信息定位肺裂感兴趣区域.然后利用肺裂方向信息增强肺裂,并利用多剖面滤波器滤除噪声从而对肺裂进行预分割.最后融合已定位的肺裂感兴趣区域和肺裂预分割结果来自动分割肺裂.与人工参考对比,提出的算法在人体左肺和右肺中分割的肺裂的F1-score中值分别为0.881和0.878.
展开▼
