Joint generative adversarial network model for classification of benign and malignant pulmonary nodules

To solve the misdiagnosis and wrong diagnosis of traditional lung cancer diagnosis, a new method is proposed to classify lung nodules between benign and malignant computed tomography (CT) images. This method combines the unsupervised learning generative adversarial network and the supervised learning convolutional neural network. First, by using the characteristics of the generative adversarial network, a generative adversarial network combining DCGAN and WGAN-GP is proposed. A progressive training mode is used to generate images as the augmented data (1 000). Then, the real samples (1 400) are put into the convolutional neural network with the generated data for training. Some of real samples are used to be the test set (600). Finally, the accuracy, sensitivity, specificity and AUC values of the combined model for the benign and malignant classification of lung CT image nodules reach 96.5%, 96.67%, 96.33%, and 0.953, respectively. Related control experiments are designed to evaluate the feasibility and effectiveness of the proposed method by using the samples generated by the generative adversarial network. Results show that the ability of the lung nodule classification model is improved. © 2020, Science Press. All right reserved.