The Classification Of Pulmonary Nodule Based On Semi-supervised Clustering Algorithm

According to the World Health Organization in 2016,the latest data show that lung cancer in the global incidence and mortality are the highest.Lung cancer has almost no symptoms at an early stage,and it was too late that most patients found it.Then it is very important that detection the lung cancer more earlier to improve survival rate.Lung cancer is generally present in the form of pulmonary nodules.Then detection of lung cancer more earlier becomes detection of pulmonary nodules more earlier.For doctors,due to the influence of subjectivity and other factors,it is so easily lead to misdiagnosis and missed diagnosis that selection of pulmonary nodules from a large number of CT images.Therefore,with the aid of computer aided diagnosis(computer-aided diagnosis,CAD)is particularly important.Semi-supervised learning is the hot spot of many scholars in recent years.Semi-supervised FCM algorithm is one of the more classic algorithms in semi-supervised clustering algorithm,with a low complexity and a good application effect in practical problems,which is favored by the majority of researchers.The classification of benign and malignant lung nodules is of great significance for early detection and diagnosis of lung cancer.However,in practical applications,the number of labeled images is so little,and acquisition the labeled samples will consume a lot of manpower.In this case,the use of semi-supervised learning algorithm is an idea to improve the classification performance effectively.As a classic semi-supervised learning algorithm,the traditional semi-supervised FCM can not make full use of the marker information in the case of unbalanced distribution of unlabeled samples and labeled samples.In view of this problem,this paper presents two improved algorithms,the main work is as follows:1.The classification of pulmonary nodules based on a priori semi-supervised FCM,it is mainly to solve the problem that the number of labeled samples is far less than the number of unlabeled samples,resulting the traditional semi-supervised FCM degenerates into the classic FCM,and resulting a low classification.Based on the phenomenon,this paper presents a semi-supervised FCM algorithm based on a priori distribution.Firstly,the algorithm calculates probability of the prior distribution of samples,and then assigns different weights to the labeled and unlabeled samples and incorporates them into the clustering process of the semi-supervised FCM,thereby strengthening the guidance ofsmall number of labeled samples during the clustering process.The proposed algorithm proves that compared with the traditional semi-supervised FCM algorithm,the proposed algorithm can achieve better classification of pulmonary nodules by experiments on the American LIDC database.2.The improved semi supervised FCM classification of lung nodules based on Particle Swarm Optimization.It does not improve the classification accuracy of the data set when the data is unbalanced,and it is not clear whether the data set can balance the problem in the case of the semi-supervised FCM algorithm when the data is balanced when the data is balanced.For the classification of pulmonary nodules based on a priori semi-supervised FCM,the classification accuracy is improved obviously when the data set is unbalanced,but the classification accuracy is not obvious when balancing the data set,and the semi-supervised FCM algorithm can not predict whether the data set is balanced in advance.Therefore,this paper proposes an improved semi-supervised FCM algorithm based on particle swarm optimization.By introducing the sample capacity information,the algorithm is applied to the balanced or unbalanced data set.Experiments show that the proposed semi-supervised FCM method based on the particle swarm optimization has good applicability.