Investigation of pulmonary nodule classification using multi-scale residual network enhanced with 3DGAN-synthesized volumes

摘要: It is often difficult to distinguish between benign and malignant pulmonary nodules using only image diagnosis. A biopsy performed when malignancy suspected based on CT examination. However, biopsies are highly invasive, patients with may undergo unnecessary procedures. In this study, we automated classification of a three-dimensional convolutional neural network (3DCNN). addition, increase the number training data, utilized generative adversarial networks (GANs), deep learning technique used as data augmentation method. approach, regions different sizes centered were extracted from images, large pseudo-pulmonary synthesized 3DGAN. The 3DCNN has multi-scale structure in which multiple each region inputted integrated into final layer. During 3DCNN, pre-training was first 3DGAN-synthesized nodules, then comprehensively classified by fine-tuning pre-trained model real nodules. Using an evaluation process that involved 60 confirmed cases pathological diagnosis biopsies, sensitivity determined be 90.9% specificity 74.1%. accuracy improved compared case without pre-training. 2DCNN results our previous study slightly better than results. it shown even though train limited such medical can GAN.