CT图像中肺结节检测一直是肺癌CAD系统的关键和难点。提出了一种孤立性肺结节自动检测算法，首先对原始CT图像进行有效、准确的肺实质分割；采用寻找局部灰度最大值方法对ROI进行初始分割；再对分割出的各ROI进行特征提取，利用SVM方法对每个特征进行定量描述，根据SVM单特征分类准确率对Mahalanobis距离进行加权改进，最后采用基于改进的Mahalanobis距离进行肺结节分类。实验结果表明，该算法可以较好地提取出CT图像中的孤立性肺结节，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帮助信息。

It is hard to detect lung nodules automatically from CT images for lung CAD system. A detection algorithm is proposed for solitary pulmonary nodules(SPNs) in thoracic CT images in this paper. Firstly, lung field is segmented from original CT image effectively and accurately. Secondly, areas of local maximum gray are found, to segment regions of interest(ROIs) roughly. Then, features of each ROI are extracted, each feature is described quantitatively by the accuracy of SVM classification with each single feature separately, and Mahalanobis distance is weighted by the quantitative parameters. Finally, ROIs are classified to nodule or non nodule with the improved Mahalanobis distance. Experiment results indicated that the algorithm can detect SPNs effectively, it is with relatively high sensitivity and low false neglected rate, and it can provide doctors helpful information to diagnose lesions in early stage of lung cancer.