论文部分内容阅读
Automatic lung tumor localization plays a significant supporting role on diagnosing and treatment of lung cancer.It could be able to avoid clinicians making mistakes of misdiagnosing tumor region or missing any tumors and could also check whether clinicians' diagnosis of tumor region was right or wrong.Currently,there are two types of localization methods which are methods based on texture and methods based on model.In this paper,we proposed a localization method based on Gaussian Mixture Model(GMM):firstly,after image preprocessing,we constructed GMM with training samples consisting of normal lung CT images from similar anatomical position of different persons; the model's parameters were figured out with EM algorithm; secondly,the GMM would discriminate the foreground and background of CT image using a probability density range determined experimentally; thirdly,the seed points of tumor region would be achieved by processing the foreground.Compared with tumor region marked by clinicians manually,the precision of our method is 92.9%.Meanwhile,we also compared the performances of GMM and GLCM(Gray Level Co-occurrence Matrix)on lung tumor localization.