Pulmonary nodule segmentation is a crucial step for nodule diagnosis. But it is a tough work to segment the juxta-vessel nodule, as the attached vessel is similar with the nodule in density. This paper aims to segment juxta-vessel nodules by using geodesic distance histogram. First, we employ anisotropic diffusion to smooth nodule images and the preliminary reference point of threshold segmentation is located by pro-jecting the filtered images in x,y, andz direction. Second, iterative threshold segmentation and geodesic distance transformation are utilized to obtain initial segmentation result containing nodule and vessel. Fi-nally, the attached vessel is removed by using the characteristic of geodesic distance histogram. Two groups of data selected from lung image database consortium (LIDC) dataset (30 cases) and clinical examinations (10 cases) were employed to validate and evaluate the performance of the proposed method. With these ex-periments, we achieved an average overlap rate of 83.30% with 8.21% false positive rate over the radiolo-gists' segmentations. Comparing with other reported methods, our new method demonstrated a good per-formance in preserving nodules' boundaries robustly.%肺结节分割为其图像特征的计算和良恶性判别提供了重要依据.针对CT图像中血管和结节的灰度值接近,导致血管粘连型肺结节分割较为困难的问题,提出一种基于测地线距离直方图的血管粘连型肺结节分割算法.首先利用各项异性扩散对图像进行预处理,并沿x,y,z方向分别对图像进行灰度积分投影,寻找投影曲线的极大值作为阈值分割的初始参考点;然后采用阈值分割结合测地线距离变换的方法对结节图像进行初始分割,得到包含结节和血管的前景区域;最后在初分割的基础上,根据测地线距离变换图像的直方图分布特点去除结节粘连的血管结构,获得最终分割结果.选取LIDC数据库(30例)和临床数据(10例)中的肺结节样本进行分割实验,结果表明,文中算法分割结果和医生手动分割结果之间的平均重合度达到83.30%,假阳性率为8.21%;和其他算法相比,该算法能在去除血管的同时保留结节边缘信息,具有较好的鲁棒性.
展开▼
