论文部分内容阅读
目的 比较改进后的大津阈值法和模糊C均值聚类(FCM)两种分割算法在CT图像肺实质分割中的应用.方法 选取40例肺部图像,分别采用改进的大津阈值法和模糊C均值聚类法,分割出肺实质区域,同时剔除肺部纵膈、气管.最后,采用主观评价和客观分析(图象一致性和信息熵指标)评价分割效果.结果 主观分析显示,改进的大津法得到的图像中虽仍存在肺实质内部间隙,但图中孤立像素点明显减少,且颗粒的边缘也更加光滑.FCM算法分割出的肺实质空洞较少,获得的图像较完整.但有些许粘连,且提取中主气管容易错误地分割到肺实质区域,造成肺实质分割不彻底.客观分析表明,在一致性准则上,两种方法分割结果相差不大,分割出的区域都具有较高的内部区域一致性;从信息熵的角度,FCM法分割效果较好.结论 肺部CT图像肺实质的分割中,对于目标与背景灰度有强对比的图像,Otsu法优于FCM法,但在阈值自适应和运算时间方面需要进一步提升;而对于存在不确定性和模糊性的图像,FCM法优于Otsu法,但在抗噪性和分割精度方面需要进一步改进.“,”Objective To compare two methods (improved OTSU and fuzzy c-means clustering) for segmentation of lung parenchyma based on CT images.Methods Two methods were analyzed and applied for the segmentation of 40 series of CT lung images.After segmentation,the lung parenchyma areas were segmented,the mediastinum and the trachea were removed.Finally,the subjective evaluation and objective analysis (consistency and information entropy) were used to evaluate the segmentation effect.Results It is demonstrated from the subjective results that the improved OTSU method decreased sharply the proportion of isolated pixel and increased smoothness of the edge of particles,while there was still internal clearance in the lung parenchyma.The lung parenchyma segmented by FCM algorithm was more complete with less holes,but with some adhesion.However in the extraction process,the main trachea was often segmented improperly into lung parenchyma,which would cause incomplete segmentation of the lung parenchyma.Objective analysis showed,in terms of consistency,the difference was not great and the areas segmented had a higher internal region consistency;for the information entropy,the segmentation result of the FCM method was better.Conclusion The experiment results show that for the CT images with a strong gray contrast between the target and background,the improved OTSU method works better in the segmentation of lung parenchyma.However,there are limitation adaptive threshold and running time that can be improved.While for the images with uncertainty and fuzziness,the FCM method is superior to the former.But as far as the segmentation accuracy and the anti-noise capability are concerned,this method needs further improvement.