【摘要】 目的:根据气管与支气管的树状特征,为了克服气管提取过程中的"阻塞"与"泄漏"问题,以适应临床教学训练需要,兼顾算法的精确性与执行效率,提出一种改进的区域生长算法。方法:首先利用经典的区域生长算法从三维螺旋CT图像中得到大气管初始区域;然后以初始区域为"大的"种子点,结合支气管在相邻CT图像中的Hessian矩阵特征向量,进行第二次区域生长,最终得到气管与支气管的全部数据。为了评价本文所提出的算法,我们对收集的13例CT气管数据进行了支气管重建,并与Kitasaka方法进行了对比。我们还请临床合作医生对算法提取的三维气管树的直观效果及临床的实用性进行了主观评价。结果:通过支气管检出率的比较,本文所提出方法明显优于Kitasaka方法。临床专家的主观评价认为本文所提出的方法对纤维支气管镜术前的气管三维结构观察和虚拟内镜开发及镜下手术模拟,具有重要的临床意义。结论:该方法自动化程度高,不仅适用于气管与支气管分割,还可推广到骨骼、人体血管系统等其他管道小目标的分割问题。更多还原

【Abstract】 Objective: In this paper,a new region growing method was proposed to robustly extract human airways that appeared as an upside-down bifurcating tree.The method could efficiently and accurately solve leaking and obstructing problems,associated with automated computerized segmentation of the airway tree.Methods: As an initial step,classical region growing was used to detect the large airways from spiral CT images.To extract more airway generations,the eigenvalues and eigenvectors of the Hessian matrix was computed to extract image features of small airway in the second-step region growing.For evaluation,thirteen sets of 3D CT data of pulmonary were reconstructed by both Kitasaka’s method and the proposed method.And clinical experts were invited to make a subjectively visual comparison.Results: Our method could detect more 3D human airway than Kitasaka’s method.Clinically subjective conclusion was that our method was important to path planning and preoperative analysis of 3D airway trees for both real and virtual bronchoscopy.Conclusions: Due to its automated scheme,the method can also be used to extract other tubular structures such as nerves and blood vessels system in 3D CT images.更多还原