A computed tomography signs quantization analysis method for pulmonary nodules malignancy grading

In order to improve the accuracy of pulmonary nodules malignancy grading, we propose a method to implement quantitative analysis for lung nodules using computed tomography (CT) signs. Firstly, we construct feature sets of CT signs by combing the radiomics features with the higher-order features extracted from a convolutional neural network. Secondly, on the basis of the mixed feature set, an evolutionary ensemble learning mechanism is used to generate a classifier to get the quantitative scores for seven lung nodule CT signs. Finally, the scores of seven CT signs are input into a multiclassifier optimized by the differential evolution algorithm to acquire the grade of malignancy. In the experimental study, 2000 lung nodule samples from the LIDC-IDRI dataset were used to train and test the evolutionary ensemble learner and malignancy classifier. The results show that the recognition accuracy of seven CT signs can reach more than 0.964. Comparison with many typical algorithms, the proposed method not only gets higher accuracy in pulmonary nodules malignancy grading but also can make the result more interpretable.