针对肺结节检测的肺实质CT图像分割

Automated Segmentation of Lung Parenchyma Designed for Detection of Lung Nodules in CT Scans

目的:针对CT图像上肺结节的自动检测,开发并评价对全肺螺旋CT扫描中的肺实质进行自动分割的一种综合方法。方法:首先利用全局阈值对CT图像进行二值化,然后消除由于支气管、细支气管等低密度影和由于结节、血管等高密度影以及由检查床造成的条状伪影等噪声,最后对包含胸膜连接结节的图像利用数学形态学运算和图像凸包运算进行完善形成肺实质掩膜。结果:利用该方法对从LIDC数据库中所有包含结节的505张CT扫描片(来自69个病例)进行肺实质分割,正确率为95.4%。其中,包含胸膜连接结节的139张CT扫描片的正确分割率为94.2%。结论:本文提出的方法较好地完成了肺实质分割任务,为利用CT图像进行计算机辅助肺结节的检测打下了基础。

Objective： To develop and evaluate an integrated method that automatically segments lung parenchyma in whole lung helical CT scans for the automated detection of pulmonary nodules. Methods： An automated lung segmentation method involved three main stages. First, the whole lung CT scan was binarized using gray-level thresholding. Second, image noises due to streaking artifact, high-intensity structures such as nodules and vessels, and low-intensity structures such as bronchi were removed. Finally, morphological operations and convex image operations were applied to smooth the edge of lung parenchyma and lung mask was then generated. Results： The segmentation method was tested on LIDC database which consisted of 505 CT scans （from 69 patients） involving lung nodules. The method achieved 95.4% correctness. For 139 CT scans which involved nodules attached to the pleural surface, that rate was 94.2%. Conclusion： The developed method achieved practical performance for automated segmentation of lung parenchyma. The automated method will be helpful for computer-aided detection of lung nodules in CT scans.