Research On Pulmonary Detection Algorithm

Lung cancer is the one of the most deadly malignant tumor.If we can diagnose more patients,it will be a very meaningful thing.Because many researches reveal that if we can detect and cure cancers at the early stage,and the possibility is raised to the survival of the patient.Though with the development of medical technology including CT,especial the coming of more MSCT,little number of lung cancer can be diagnosed and cured.The reason is that it is difficult to find the cancer at the early stage.MSCT brings many CT images at the same time,and these images burden on doctors heavily and increase the rate of omission.The second reason is the wrong diagnosis results make the patient neglect the prevention and not to check again in hospital for a long time.When doctors diagnose the lung cancer,which is at the last stage,it is difficult to cure it.In order to solve this difficulty,and help the doctors escape from so many images, more and more researchers begin to develop the CAD(computer assistant diagnose) system to help doctors detect lung nodules,distinguish the benign or malignant pulmonary nodules and so on.Lung CAD system includes there parts,the first part is the detection of the lung nodules,the second part is the calculation of the size of the lung nodules and the third part is the judgment of the malignant or benign nodules. The first part is the most important part of all;because the detection accuracy not only directly helps doctors decrease the omission and misdiagnosis,but also is related to the next work about the malignant judgment accuracy.Based on these,this paper against pulmonary nodule detection algorithm to study and we do two aspects of work.In the following we introduce these two parts briefly.1 TMH algorithmAfter the analysis and comparison of the existed lung nodule detection algorithms,we find that these algorithms have more or less disadvantages.The 2D methods are quick and complete,but have high true positive rate.The 3D methods have acceptable true positive rate,but low speed.Based these factors,we give a new lung nodule detection algorithm in this part,which is a 2D method considering the 3D characters of nodules.We name this method TMH method that is importing the Hessian matrix to temple matching method.At first,TMH detect the lung nodules in 2D completely,and then importing the 3D characters to decrease the true positive rate and it can get good experimental results.TMH method includes there parts,the first is segmentation of the CT images in order to get lung fields.In this part rolling ball method is the key content.Then we apply the temple matching method to mark the detection point.In the end,we import the Hessian matrix and use its eigenvalues to judge whether one pixel of the image is the nodule point or not.(in 3D the nodule is ball-like and the blood vessel is cylinder).After this step,the blood vessel is decreased obviously.The advantages of the 2D temple matching method are complete and rapid,but the disadvantage is the high true positive rate.TMH method not only keeps the advantage of 2D methods, but also considering the 3D morphological characteristics.The experimental results prove the advantage of TMH method.2 Improved FCM methodThe second part of this article is importing the circularity to improvc the FCM method.At first,we use FCM to class the lung CT images.We have three classes after the first step,including lung cavity class,nodule and blood vessel class and air class.For the nodule and blood vessel class,we import the circularity and remove the blood vessel mark.With the circulation of circularity,we can transplant the FCM to solve the problem of detection nodules and get the good experimental results.