基于改进Snake模型的肺部图像分割

Lung segmentation based on improved Snake model

肺实质分割是基于CT图像的计算机辅助检测技术必不可少的步骤。针对现有活动轮廓模型对初始位置敏感、深度凹陷区域分割不准确和抗噪性差等缺点,提出了一种基于改进Live-Wire算法确定Snake模型初始轮廓的交互式分割方法。该方法结合并改进Live-wire算法和一般的阈值法对图像进行预分割,将得到的边缘作为Snake模型的初始轮廓,通过Snakes模型演化得到肺实质轮廓结果。实验结果表明该方法能快速地对肺部图像进行分割,与传统方法相比具有人工交互次数减少、抗噪音性好、更具鲁棒性和效率性的优点。

Lung segmentation is an indispensable step in pulmonary nodules computer-aided detection based CT images. Aimed at decreasing the Snake dependence on initial contour points, improving the accuracy of segmenting the depth sag area and resistance of noise, this paper proposes an interactive lung parenchyma segmentation algorithm, which uses improved Live-Wire algorithm to define Snake initial contour. It modifies the traditional Live-Wire algorithm by combining it with the threshold algorithm. After the pre-segmentation, the edge is obtained. Then taking the edge as the initial contour, Snake model can get lung parenchyma by evolution. Experimental results show that the method can extract the lung paren-chyma effectively and quickly. Compared with traditional method, this method can reduce the number of human-computer interactions and can well resist noise, which is more robust and efficient.