面向肺癌CAD的CT图像疑似病灶检测算法

Detection algorithm for suspected lesions in CT images of lung cancer CAD

CT图像中疑似结节病灶区域的分割和提取是肺癌CAD系统的关键和难点。本文提出一种疑似结节病灶自动检测算法,首先对原始CT图像进行有效、准确的肺实质分割,根据肺结节、气管和血管具有不同的几何特征,构造了一组不同尺度的类圆形结构元素,采用多尺度形态学滤波方法对ROI进行初始分割,再根据各ROI的大小构造相应尺度的二维高斯模板,对各ROI区域进行自适应局部高斯模板匹配,以进一步剔除假阳性。实验结果表明,该算法可以有效地提取出CT图像中类圆形的疑似结节病灶,具有较高的灵敏度和较低的漏诊率,可以为医生诊断早期肺癌病灶提供辅助信息。

Segmenting and extracting suspected nodular lesions from CT images is the key and difficult step for lung cancer CAD system. An automatic detection algorithm is proposed for the suspected nodular lesions in thoracic CT images in this paper. First, lung parenchyma is segmented from original CT image effectively and accurately. Second, according to the different geometry shapes of lung nodules, bronchial tubes and blood-vessels, a series of circle- like structure elements with different dimensions are built, and the multi-scale morphologic filtering is adopted to do the rough segmentation for the regions of interest （ROI）. Finally, several two-dimension Gauss templates are designed based on their size of ROI areas, and local template-matching is carried out adaptively between each ROI and its template separately, and some false positive ROI are eliminated. Experiment results indicate that the algorithm can extract suspected nodular lesions effectively, it has relatively high sensitivity and low missed diagnosis rate, and it can provide doctors with auxiliary information for lesion diagnosis in early stage of lung cancer.