| Lung cancer is the most common cause of death from malignancy in the world. Pulmonary nodule can turn to be lung cancer in several years. Especially in the last half century, incidence and mortality of lung cancer are raised quickly in many countries. If lung cancer can be diagnosed and cured at the early time, the survival rate in 5 years would be raised form 14% to 49% [1]. So diagnosis of pulmonary nodule in early time will be the key to cure lung cancer. There are two ways of pulmonary nodule detection, X-ray sternum and CT(Computer romography). Using CT detect pulmonary nodules is better than using X-ray sternum. Small nodules which have diameter less than icm can not be detected if you use X-ray sternum, and the detected rate CT has is 8 times as X-ray sternum has [2]. Lung CT has too much information, and pulmonary nodules are easily omitted by eyes, that computer aided diagnose for pulmonary nodule gets progress. Since the information in the CT images can not be identified entirely, some symptom may be missed by an expert. We intend to use Computer Aided Diagnose (CAD) to help experts before they read CT images, by listing the areas likely to be pulmonary nodules.Detection of pulmonary nodules falls to three steps. First, segment every image of CT, extract the in-lung region and the lung wall. Second, do 2D analyze, wipe off parts that can not be nodules, and remainders are the 2D-ROIs(Region Of Interesting). Last, do 3D analyze, get 3D-ROIs, which are the regions may be the nodules. Normal way of segmentation takes a long time and can not adapt noise. This paper uses related c-means segmentation, extracts in-lung regions. Our method overcomes these defects above, and gets good result according to experiment. Normal 2D analyze and 3D analyze are only adapted to solitary pulmonary nodule (SPN), but can not be adapted to conglutinate pulmonary nodule (CPN). 95% of pulmonary nodules detection literatures are only deal with SPN, because CPN detection is a hard problem and there is not a good algorithm for it. Kanazawa [3] uses the curvature of lung wall to separate the CPN. But his method is too fussy, and with high false positive rate. This paper gives a simple method to deal with CPN, minimizes the process time, and get low false positive rate.We processes 15 CT image serials of format DICOM, which include 24 SPNs and 30 CPNs. Result of experiment shows that: extraction algorithm is good; 95.8% detected rate and 20.8% false positive rate for SPN; 90.0% detected rate and 33.3% false positive rate for CPN. This reaches a high level and are recognized by experts in lung cancer. |
