结合线性形状特征与模糊连接度的三维CT图像肺裂检测

DETECTION OF PULMONARY FISSURE IN 3D CT IMAGES BY COMBINING LINEAR SHAPE FEATURES AND FUZZY CONNECTEDNESS

针对三维CT图像中噪声、密度分布不均匀等因素造成的肺裂检测难题,提出一种线性形状特征与模糊连接度相结合的检测方法。首先,模仿放射科医生的读片方式,利用肺裂在互相垂直的2D切片中呈高亮度线形结构并被低密度肺实质包裹的特点,定义窄条微分(DoS)滤波器用于图像增强;然后,指定3D空间中感兴趣区域内一点作为种子点,根据增强后肺裂图像的灰度均匀性与灰度差值特征来构造亲和力函数以计算种子点与其他体素之间的模糊连接度,再通过选取合适的阈值对模糊连接度进行阈值分割;最后,通过基于形态学的后处理来移除分支点,并由连接元分析以去除粘黏在肺裂周围的无关组织以得到最终完整的肺裂检测结果。临床数据实验和人工定义金标准验证的结果表明,该方法可对肺裂进行较准确、有效的检测。

Considering the difficulty in pulmonary fissures detection caused by the factors such as noise and intensity inhomogeneity in 3D CT images, we propose a detection method which combines linear shape features with fuzzy connectedness. First, it imitates the way of the radiologist reading slices, and defines a derivative of stick （DoS） filter for images enhancement by utilising the feature of pulmonary fissures that in mutually perpendicular 2D fissure slices they show the high bright linear structures and are wrapped by low-density lung parenchyma; Then, it assigns a specific point within the region of interest in 3D space as the seed, constructs the affinity function according to intensity homogeneity and intensity difference features of the enhanced pulmonary fissures image to calculate the fuzzy connectedness between the seed and other voxel. Besides, it segments the fuzzy connectedness by choosing an appropriate threshold; Finally, it removes the branch points through morphology-based post-processing operation and by connected component analysis it eliminates non-fissure tissues adhering around the pulmonary fissure so as to obtain an eventually complete pulmonary fissure detection result. The clinical data experiment and the artificially defined gold standard verification result show that the method can detect pulmonary fissures more accurately and effectively.