Lung cancer has the highest mortality rate among all malignant tumors. The key to reducing lung cancer mortality is the accurate diagnosis of pulmonary nodules in early-stage lung cancer. Computer-aided diagnostic techniques are considered to have potential beyond human experts for accurate diagnosis of early pulmonary nodules. The detection and classification of pulmonary nodules based on deep learning technology can continuously improve the accuracy of diagnosis through self-learning, and is an important means to achieve computer-aided diagnosis. First, we systematically introduced the application of two dimension convolutional neural network (2D-CNN), three dimension convolutional neural network (3D-CNN) and faster regions convolutional neural network (Faster R-CNN) techniques in the detection of pulmonary nodules. Then we introduced the application of 2D-CNN, 3D-CNN, multi-stream multi-scale convolutional neural network (MMCNN), deep convolutional generative adversarial networks (DCGAN) and transfer learning technology in classification of pulmonary nodules. Finally, we conducted a comprehensive comparative analysis of different deep learning methods in the detection and classification of pulmonary nodules.
肺癌是死亡率最高的恶性肿瘤，肺结节早期确诊是降低肺癌死亡率的关键。计算机辅助诊断技术在肺结节早期确诊方面被认为具有超越人类专家的潜力。而基于深度学习技术的肺结节检测和分类可通过自我学习而不断提高诊断的准确率，是实现计算机辅助诊断的重要手段。本文首先系统阐述了二维卷积神经网络（2D-CNN）、三维卷积神经网络（3D-CNN）和更快速的区域卷积神经网络（Faster R-CNN）技术在肺结节检测方面的应用，然后阐述了 2D-CNN、3D-CNN、多流多尺度卷积神经网络（MMCNN）、深度卷积生成对抗网络（DCGAN）和迁移学习技术在肺结节分类中的应用，最后针对肺结节的检测与分类中不同的深度学习方法进行了综合比较分析。.
Keywords: computer-aided diagnosis; convolutional neural network; deep learning; medical image; pulmonary nodules.