基于EM算法和曲面拟合模型的肺裂检测算法

Pulmonary Fissure Detection in CT Images Based on Expectation Maximization Algorithm and Surface Fitting Model

肺裂是识别人体肺部解剖结构的重要标记.为解决由于肺裂存在形变、不完整和断裂等现象造成的难以在CT图像中自动检测的问题,提出一种基于期望最大化算法和曲面拟合模型的肺裂检测算法.首先结合气管和动脉血管先验知识估计肺裂区域,减少肺部其他结构的干扰;然后构建肺裂方向场更好的区分肺裂和噪声,并利用期望最大化算法识别肺裂;最后采用曲面拟合模型弥补丢失的肺裂,达到更好地检测肺裂的目的.提出的算法在20个患有慢性阻塞性肺部疾病患者的肺部CT图像上进行实验,结果表明,与人工参考对比,该算法在左肺裂和右肺裂的F_1-score中值分别为0.979和0.971,能够高效率地检测肺裂.

Pulmonary fissures are important landmarks for recognition of lung anatomy.It is difficult to automatically detect pulmonary fissures due to factors such as deformation,incompleteness and disruption,a reliable and valuable approach based on expectation maximization algorithm and surface fitting model is proposed to detect pulmonary fissures.Firstly,we fuse the prior knowledge of trachea and pulmonary arteries to reduce the impact of other pulmonary structures for fissure region of interest localization.Then fissure orientation field is constructed to effectively discriminate between fissures and noise,and an expectation maximization algorithm is utilized to identify pulmonary fissures.Finally,a surface fitting model is used to make up for the undetected fissures for pulmonary fissure detection.The proposed method is validated on lung CT images of 20 patients with chronic obstructive pulmonary disease.Compared with manual fissure references,our method obtain a high segmentation accuracy with median F1-score of 0.979 and 0.971 for the left and right lung images,respectively.Experimental results show that the proposed method has a good performance in pulmonary fissure detection.