Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (13): 20-32.DOI: 10.3778/j.issn.1002-8331.2002-0051

Previous Articles     Next Articles

Review of Deep Learning Methods Applied to Lung Nodule Detection in CT Images

ZHANG Fuling, ZHANG Shaomin   

  1. School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, China
  • Online:2020-07-01 Published:2020-07-02

应用于CT图像肺结节检测的深度学习方法综述

张福玲,张少敏   

  1. 北方民族大学 计算机科学与工程学院,银川 750021

Abstract:

Lung cancer is the most deadly cancer in the world. Detection of lung nodules by chest CT images is of great significance for the early diagnosis and treatment of lung cancer. In order to reduce the workload of radiologists and at the same time reduce the rate of misdiagnosis and missed diagnosis, researchers have proposed a Computer-Aided Detection(CAD) system to assist radiologists in detecting and diagnosing pulmonary nodules. Researchers are currently experimenting with different deep learning techniques to improve the performance of computer-aided diagnostic systems in CT-based lung cancer screening. This work reviews the current typical deep learning algorithms and frameworks for CAD systems for lung cancer detection. It mainly introduces six aspects:data set introduction, 2D deep learning methods, 3D deep learning methods, data imbalance processing, model training methods and model interpretability. Finally, the main characteristics and algorithm performance of each method are comprehensively compared and analyzed, and how to improve the nodule detection performance is prospected.

Key words: deep learning, CT images, candidate nodule detection, convolutional neural network

摘要:

肺癌是世界上死亡率最高的癌症,通过胸部CT影像检测肺结节对肺癌早期诊断和治疗意义重大。为了减轻放射科医生的工作量以及同时减少误诊率和漏诊率,研究人员提出了计算机辅助检测(CAD)系统辅助放射科医生检测和诊断肺结节。目前,研究人员正在尝试不同的深度学习技术,以提高计算机辅助诊断系统在基于CT图像的肺癌筛查中的性能。这项工作回顾了作为肺癌检测的CAD系统目前典型的深度学习的算法和框架,主要从数据集介绍、2D深度学习方法、3D深度学习方法、数据不平衡问题的处理、模型训练方法以及模型可解释性这六个方面进行介绍。最后,对各个方法的主要特点和算法性能进行了综合比较分析,并对如何提高结节检测性能进行了展望。

关键词: 深度学习, CT图像, 候选结节检测, 卷积神经网络