【摘要】 根据肺部计算机横断扫描(computed tomography, CT)图像准确提取肺气管对肺呼吸功能测定和疾病诊断具有重要意义。现有的肺气管分割方法需要依赖大量人机交互才能提升分割精度，而深度学习在医学图像处理领域有比较广泛的应用，尤其是在肺部结节检测和良恶性诊断方面，但深度学习用于肺部CT图像的肺气管分割由于图像噪声和部分容积效应的影响会造成肺气管分割的泄漏，难以分割出微小的气管。原始肺部CT图像中包含骨骼、病床等非感兴趣区域，处理数据量的增大会消耗更多的数据处理时间，且极易造成误差。利用肺气管树的解剖结构信息，对肺气管分割采用分步处理，提出了一种基于Attention-Unet的肺气管分割方法。实验结果表明，将基于深度学习的Attention-Unet网络应用于肺部CT图像的肺气管分割，能提高分割的速度和精度，并有效防止泄漏。更多还原

【Abstract】 Accurately extracting the pulmonary trachea from lung CT images is of great significance for the determination of pulmonary respiratory function and disease diagnosis. The existing pulmonary trachea segmentation tasks must rely on a lot of human-computer interaction to improve the segmentation accuracy. Deep learning has a wide range of applications in the field of medical image processing, especially in the detection of pulmonary nodules and diagnosis of benign and malignant pulmonary nodules. However, the influence of image noise and partial volume effect will cause leakage of pulmonary trachea segmentation, making it difficult to segment tiny trachea. Considering that the original lung CT image contains non-interesting areas such as bones and hospital beds, it not only consumes more data processing time due to the increase in the amount of processing data, but also easily causes errors. In this paper, used the anatomical structure information of the pulmonary airway tree, the pulmonary trachea segmentation is processed step by step, and proposes a method for pulmonary trachea segmentation based on Attention-Unet. The experimental results demonstrate that applying deep learning-based Attention-Unet network to segment pulmonary trachea from CT images, this method can improve the speed and accuracy of segmentation, and effectively prevent leakage.更多还原