The Research And Implement Of Lung Nodules Detection Based On Semi-Supervised Learning In CT Images

In recent years,due to the deterioration of human's living environment, lung diseases, such as lung cancer, have become a main course of human death and low-quality life. Early detection of lung cancer may allow for timely therapeutic intervention and thus a favorable prognosis for the patients. With the development of CT technology, CT has become the major tool for the early detection of lung cancer. Lung diseases are often displayed as nodules in the CT images. So doctors need find lung nodules in CT images, and now CT machine produces at least tens of images, even hundreds of images for the lung, which leads to heavy burden for the doctors and brings on missed diagnose. The Computer Aided Diagnosis (CAD) of lung nodules can help doctors reduce the possibility of missed diagnose, and provide the prediction of benign or malignant nodules.This paper discusses the relevant work about the CAD of lung nodules, and consists of the following three parts:1. A complete scheme of detecting lung nodules.This scheme includes the process and format conversion of DICOM Files, lung extraction based on region growing, ROI(region of interst) extraction, circle-like detection, ROIs'feature extraction, and lung nodule prediction based on semi-supervised learning method Co-Forest. The reason why this scheme uses the semi-supervised learning method is that unlabelled datas, which are CT images that are undiagnosed by doctors here, are easy to collect, while the labelled CT images are hard to retrieve, and this case is what the semi-supervised learning method exactly can solve.2. A classification scheme to distinguish between benign and malignant nodules. Identifing the patient which disease he takes is very difficult through the CT images, even if experienced experts aren't able to identify the disease exactly without pathology test. This paper adopts 13 kinds of features which are provided by hospital, and uses SVM to distinguish between benign and malignant nodules.3. A CAD System of lung nodules. It provides a nice user interface and good user experience。The System uses the Server/Browser model, which can provide users make use of it in any time and any place, meanwhile, interface and algorithm are separated, which is helpful to update algorithms.At the last part of this paper, we do the relevant experiments. The results show that the methods we proposed are effective,and they solve the diagnosis of lung nodules in some degree.