肺结节检测中特征提取方法研究

Research on the Feature Extraction Approach for SPNs Detection

计算机辅助诊断(Computer-Aided Diagnosis,CAD)系统为肺癌的早期检测和诊断提供了有力的支持.本文对孤立性肺结节特征提取问题进行研究.通过对肺结节和肺内各组织在序列CT图像上的医学征象分析和研究对比,结合专家提供的知识,提出了肺结节特征提取总体方案.该方案分别从肺部CT图像的灰度特征、肺结节形态、纹理、空间上下文特征等几个方面,对关键的医学征象进行图像分析,从而实现对ROI(Regions of Interest)区域的特征提取和量化;提出特征提取的评价方案,实验结果表明,本文提取的特征提取方案是有效的.利用本文提取的特征,肺结节检测正确率达到93.05%,敏感率为94.53%.

Image processing techniques have proved to be effective for improvement of radiologists＇ diagnosis of pubmonary nodules. In this paper, we present a strategy based on feature extraction technique aimed at Solitary Pulmonary Nodules （ SPN ） detection. In feature extraction scheme, 36 features were obtained, contained 3 grey level features, 16 morphological features, 10 texture features and 7 spatial context features. And the classifier （ SVM ） running with the extracted features achieves comparative results, with a result of 93.05% in nodule detection accuracy and 94.53% in sensitivity.