肺癌是当今对人类健康与生命危害最大的恶性肿瘤之一。早期肺癌一般表现为肺结节,如能及时从肺部CT图像中检测到肺结节,便能及早发现肺癌,经治疗后可有效延长患者的生存时间,所以CT图像是肺癌诊断和疾病治疗的重要依据。但对全肺进行螺旋CT扫描产生的大量图像给人工检测肺结节带来了困难,因此,基于CT图像的肺结节计算机辅助检测(CAD)技术应运而生。由于CAD能有效辅助放射科医生提高肺结节的检测准确率与工作效率,降低漏诊与误诊率,因此,CAD成了目前生物医学工程领域的研究热点之一。尽管目前报道的CAD系统所采用的方法各有不同,但基本上都是遵循以下步骤完成:1)CT图像的预处理;2)肺结节的分割;3)特征提取及优化选择;4)肺结节的分类识别。其中对结节的精确分割与否直接影响到后续的特征选择与优化,而特征选择与优化又进而影响到分类器的分类属性,所以肺结节分割是基于CT图像的肺结节计算机辅助检测的关键步骤。
肺结节可细分为实质性结节(solid nodule)和亚实质性结节(sub-solid nodule)。其中完全屏蔽肺实质的结节称为实质性结节,否则称为亚实质性结节。实质性结节表现为边界比较规则的类圆形病灶,且密度较高、边界清晰,因此较容易分割,对实质性肺结节的分割国内外均有大量文献报道。与实质性肺结节相比,亚实质性肺结节其密度表现为磨玻璃影(GGO),且边缘不清晰(多带毛刺)、没有特定的形状。实质性结节中93%以上为良性病灶,而因为带有GGO,亚实质性肺结节的恶性化程度较实质性结节而言表现得较高。因此,亚实质性结节的精确分割对发现早期肺癌更具应用价值,也面临更大的难度和挑战。
模糊聚类算法是一种基于模糊数学的常用的灰度图像分割方法,适合解决灰度图像中存在的模糊和不确定性问题。而经典的模糊聚类算法及其数种改进算法在聚类过程中具有明显的缺点和不足,仅考虑了每个像素的灰度值分别与各聚类中心的距离,未考虑相邻像素之间的影响及利用图像的空间信息,在分割时可能会丢失图像部分信息,所以不适用于亚实质性肺结节分割。针对肺CT图像中亚实质性肺结节的特点,对模糊C均值聚类(FCM)及其改进算法核模糊聚类(KFCM)和加权核模糊聚类(WKFCM)进行实践,提出一种改进的利用血管及类别结构信息加权的适用于亚实质肺结节的核模糊聚类(IWKFCM)3维分割方法。该方法首先从肺CT序列图像的中心层中手动选取结节感兴趣区域(ROI),然后在由ROI临近层确定的3维感兴趣区域(VOI)内进行IWKFCM聚类,最后对聚类结果进行3维连通域标记及形态学处理得到最终结节的分割结果。本文分别采用36个LIDC标准数据和18个临床数据对所提出的分割方法进行评价,以放射科医生手动分割的区域作为金标准计算重合率,其均值分别为71.65%及76.18%,且假阳性率及假阴性率均低于17%。实验结果表明,相较于FCM,KFCM与WKFCM等未改进算法,IWKFCM能够获得更准确的分割结果,并且分割效果同时优于目前文献报道的其他非模糊数学方法,为基于CT图像的肺结节计算机辅助检测提供了一种分割亚实质性肺结节的工具。

Accurately and reliably automated segmentation of pulmonary tumors could play an important role in lung cancer diagnosis and radiation oncology. However, it remains a very difficult task in particular for segmenting pulmonary tumors associated with sub-solid nodules that are partially obscured in lung CT images. In this study, we propose and test an improved weighed kernel fuzzy C-means (IWKFCM) method that incorporates vessels structure information and classes’ distribution as weights to segment sub-solid pulmonary nodules. For this purpose, a region of interest (ROI) of a nodule in center CT slice is manually defined. The IWKFCM algorithm is applied to identify and cluster the potential nodule pixels located in this manually-defined center slice and its adjacent (surrounding) slices. The sub-solid nodule is then segmented and defined through 3D connected component labeling and morphological post-processing. This segmentation method is tested using two datasets including 36 nodules selected from a public dataset (LIDC) and 18 nodules depicted on CT images collected from our local hospital. The average overlap ratios between the automated and radiologists’ segmentation of nodules of two datasets are 76.18% and 71.65% respectively. In both datasets, the false-positive ratio (FPR) and false-negative ratio (FNR) are smaller than 17%. Experimental results show that the proposed method enables us to achieve more accurate result in segmenting sub-solid pulmonary nodules than the other previously reported clustering methods. The segmentation results could also provide a consultative reference for more accurately extracting image features and optimal classification of pulmonary nodules in developing computer-aided detection (CAD) schemes.