Research On The Method Of False Positive Removal In Computer-aided Diagnosis Of Lung Cancer

At present,lung cancer is one of the highest morbidity and mortality cancers in the world.Moreover,both of the lung cancer's morbidity and mortality are constantly increasing every year,which has become an important public health problem in China.CT imaging based Computer-Aided Diagnosis(CAD)System is helpful to improve the accuracy of lung cancer diagnostic and reduce the rate of missed diagnosis and misdiagnosis caused by the fatigue of the clinician.This paper studies the methods of computer-aided diagnosis of pulmonary nodules based on CT images,and focuses on the removal method of false positive of pulmonary nodules.The main research contents are as follows:In this paper,the distribution of false positive nodules and true positive nodules in candidate nodules are analyzed.According to the existence of a large number of false positive nodules in candidate nodules,such as vascular crossings and blood vessel terminals,a multi-scale density distribution histogram is proposed by researching the feature extraction method of candidate nodules.This method uses the center of the candidate nodules as the center to construct the spherical body windows of different scales,and statistics mean density in the normal direction,and generate histograms of density distributions in different directions.Finally,the proposed method is experimentally verified on data sets consist of 45303 candidate nodules.The results show the proposed method performs very well in identifying the true positive nodules and false positive nodules,such as vascular cross and vascular end.In addition,The detection sensitivity of nodules with false positive removed achieves 97.2%and the false positive rate is 6.45FPs/Scan.In order to improve the sensitivity of pulmonary nodule detection,the candidate nodules extracted from the nodule detection algorithm contain a large number of false positive nodules,a small amount of true positive nodules and,thus,is considered as a typical imbalance data set.Considering the characteristic of the unbalanced distribution of data set in false positive removal problem,after studying a false positive removal method based on unbalanced learning,a novel method named MBSMOTEBoost is proposed.This method first uses MB SMOTE method to sampling the data set,and then classify results by using adaboost method.Finally,false positive nodules are removed.The experimental results show that the method can effectively reduce the rate of the false positive.Moreover,the false positive rate can reach 2.9FPs/Scan without reducing sensitivity.