Classification of Lung Nodules Based on GAN and 3D CNN.

An efficient model based on generative adversarial networks (GAN) and the three-dimensional convolutional neural network (3D CNN) is presented and implemented to classify lung nodules, useful for false positives removal. The model is trained on LIDC-IDRI dataset. In the first stage, the transfer learning based on deep convolutional generative adversarial networks (DCGAN) is proposed to preliminarily classify pulmonary nodules. In the second stage, the three-dimensional convolutional neural network (3D CNN) is introduced to finish the task of removing false positives. In the first stage, the transfer learning network converges more rapidly than the direct training network. Besides, the accuracy of the test set in the first stage is 94.50%. In the second stage, the highest accuracy rate of false positives removal is 95.30%. The research and analysis show that the proposed algorithm has better performance in the classification of lung nodules.