Identification Of Benign And Malignant Pulmonary Nodule Based On3D Space

In recent years, with the increase of environmental pollution, lung cancer has become one of the most health hazards to human as a malignant tumor. Clinical diagnosis and results of treatment show that early detection and early treatment can effectively improve the5-year survival rate for lung cancer. It is hard to identify the pulmonary nodule benign or malignant only through the naked eye observation of lung CT images at the beginning of the lung cancer. Using the computer-aided diagnosis to process the lung CT images could identify the benign or malignant of pulmonary nodule, thus achieving the purpose of early detection and early treatment. Computer-aided diagnosis of lung cancer has become a hot research spot at world, but at present the CAD research of lung disease is usually based on two-dimension features. Lung nodule itself is a three dimensional object, and using three dimensional feature to analyze it should show more accurate and comprehensive detection and diagnosis results, so the paper proposed an identification method for benign or malignant pulmonary nodule based on the3D information of Lung CT image. The main work is as follows:1) First we segmented the pulmonary nodules from CT images with the methods of threshold segmentation, regional growth, and morphology. Then chose FCM clustering algorithm to extract lung nodules. After three dimensional reconstructions of CT pictures in sequence, we get three dimensional information of lung nodule. Then use simple feature selection to remove false positive and get more accurate lung nodules.2) Next we focus on the three-dimensional feature extraction for benign or malignant nodules, and extract31three dimensional lung nodules feature including gray level character, texture, characteristics of morphological and spatial location. Then combining with characteristics of each type of ROC curves, we optimize the feature selection, and finally choose22three dimensional feature, such as volume, fourier descriptor and so on, which make up a22-dimensional feature vector using identification of benign or malignant pulmonary nodules.3) For the identification, we use the Support vector machine (SVM) predicting method for training, which require inputting22-dimensional feature vector and output is benign or malignant nodules. Finally, this paper conducted a comprehensive assessment from sensitivity, specificity, accuracy, likelihood ratio and ROC curves and so on. From the experiment, we get the sensitive of0.7776, positive likelihood ratio of2.2410, negative likelihood ratio of0.3682, and the AUC in ROC curves of0.8340, which show the identification of this experiment is good.The results of the experiment show the new method achieved satisfactory identification effect. This study is of great significance to computer-aided diagnosis of lung cancer.