刘莹芳, 柏正尧, 李琼. 一种基于CT图像的肺实质分割方法[J]. 云南大学学报(自然科学版), 2019, 41(3): 456-463. doi: 10.7540/j.ynu.20180442
引用本文: 刘莹芳, 柏正尧, 李琼. 一种基于CT图像的肺实质分割方法[J]. 云南大学学报(自然科学版), 2019, 41(3): 456-463. doi: 10.7540/j.ynu.20180442
LIU Ying-fang, BAI Zheng-yao, LI Qiong. A method of pulmonary parenchyma segmentation based on CT image[J]. Journal of Yunnan University: Natural Sciences Edition, 2019, 41(3): 456-463. DOI: 10.7540/j.ynu.20180442
Citation: LIU Ying-fang, BAI Zheng-yao, LI Qiong. A method of pulmonary parenchyma segmentation based on CT image[J]. Journal of Yunnan University: Natural Sciences Edition, 2019, 41(3): 456-463. DOI: 10.7540/j.ynu.20180442

一种基于CT图像的肺实质分割方法

A method of pulmonary parenchyma segmentation based on CT image

  • 摘要: 在肺部疾病的计算机辅助诊断(CAD)时,肺实质的正确分割尤为重要. 为了减少检测区域,节省运算时间,提高准确率,需要预先提取肺组织. 提出了一种改进的最大类间方差法(OTSU)结合形态学运算的肺实质分割方法. 首先对原始CT图像做滤波去噪、图像增强的预处理,自适应阈值二值化图像;然后连通区域标记获取轮廓,利用基于OTSU的改进算法去除气管、肺液等干扰,分离肺实质与背景,对左右肺叶粘连的情况采用行列扫描、区域彩色标记并将其有效分离;最后采用一系列的形态学运算对提取出来的肺实质弥合修补. 从公开数据库LIDC中选取830张CT图像,用该方法可完整分割肺实质,平均准确率为97.56%,平均召回率可达99.29%,Dice相似系数为98.42%.

     

    Abstract: The correct segmentation of lung parenchyma is very important in the computer-aided diagnosis (CAD) of lung diseases. In order to reduce detection area, save operation time and improve accuracy, it is necessary to extract lung tissue in advance. In this paper, an improved method of lung parenchyma segmentation based on the maximum inter-class variance method (OTSU) and morphological operation is proposed. First, the original CT image is filtered and denoised, image enhancement preprocessing, adaptive threshold binarization image, and then connected area label to obtain contour, then the improved algorithm based on OTSU is used to remove the trachea, lung fluid and other interference. The lung parenchyma and background were separated, The left and right pulmonary leaf adhesion are marked with color and separated effectively by row and column scan. Finally, a series of morphological operations are used to repair the extracted lung parenchyma. To select 830 CT images from the open database LIDC, and the lung parenchyma could be segmented completely by this method. The average accuracy is 97.56%, and the average recall rate is as high as 99.29%, Dice similarity coefficient is 98.42%.

     

/

返回文章
返回