Malignant-benign classification of pulmonary nodules based on random forest aided by clustering analysis

To help the radiologists better differentiate the benign from malignant pulmonary nodules on CT images, a novel classification scheme was proposed to improve the performance of benign and malignant classifier of pulmonary nodules. First, the pulmonary nodules were segmented with the references to the results from four radiologists. Then, some basic features of the segmented nodules such as the shape, gray and texture are given by calculation. Finally, malignant-benign classification of pulmonary nodules is performed by using random forest (RF) with the aid of clustering analysis. The data with a set of 952 nodules have been collected from lung image database consortium (LIDC). The effect of proposed classification scheme was verified by three experiments, in which the variant composite rank of malignancy were got from four radiologists (experiment 1: rank of malignancy '1', '2' as benign and '4', '5' as malignant; experiment 2: rank of malignancy '1', '2', '3' as benign and '4', '5' as malignant; experiment 3: rank of malignancy '1', '2' as benign and '3', '4', '5' as malignant) and the corresponding (A(z)) (area under the receiver operating characteristic curve) are 0.9702, 0.9190 and 0.8662, respectively. It can be drawn that the method in this work can greatly improve the accuracy of the classification of benign and malignant pulmonary nodules based on CT images.