基于深度学习的人工智能在肺结节诊断领域的进展  

Research progress of artificial intelligence based on deep learning in the field of pulmonary nodules

作　　者：张俊[1] 侯聪 刘新疆[1,2] ZHANG Jun;HOU Cong;LIU Xinjiang(Department of Radiology,Affiliated Hospital of Binzhou Medical University,Binzhou 256603,China;Department of Radiology,Shanghai Pudong Hospital(Pudong Hospital Affiliated to Fudan University),Shanghai 201399,China)

机构地区：[1]滨州医学院附属医院放射科,山东滨州256603 [2]上海市浦东医院(复旦大学附属浦东医院)放射科,上海201399

出　　处：《分子影像学杂志》2020年第3期365-368,共4页Journal of Molecular Imaging

基　　金：山东省自然科学基金计划(ZR2016HL43);山东省医药卫生科技发展计划(2015WS0479)。

摘　　要：肺癌是死亡率最高的恶性肿瘤,肺结节的早期检测是降低肺癌死亡率的关键。基于深度学习的人工智能技术可通过自我学习,不断提高肺结节检测和诊断的准确率,是实现计算机辅助诊断的重要手段。本文介绍了人工智能、机器学习、深度学习的概念及三者间的关系,阐述了4种常见的深度学习模型:卷积神经网络、海量训练人工神经网络、自编码器和深度信念网络。卷积神经网络是最常用的深度学习模型,主要包括二维卷积神经网络、三维卷积神经网络和多流、多尺度的卷积神经网络,其中的多流、多尺度的卷积神经网络更有利于肺结节的分类;海量训练人工神经网络在有限的肺结节训练样本中具有优势;自编码器可以在较低维空间下对肺结节进行检测;深度信念网络是一种生成模式,与极限学习机结合可提高肺结节的诊断率。另外,本研究分析了目前人工智能存在的问题:标记图像过少、可解释性和可控制性不足、存在伦理和法律问题。总之,基于深度学习的人工智能不仅改变了影像学,也改变了所有其他的医学领域,具有广阔的应用前景。Lung cancer is the malignant tumor with the highest mortality rate. Early diagnosis of lung nodules is the key to reducing the mortality of lung cancer. Artificial intelligence technology based on deep learning can continuously improve the accuracy of lung nodule detection and diagnosis through self-learning, which is an important means to achieve computer-aided diagnosis. This article first briefly introduces the concepts of artificial intelligence, machine learning, deep learning, and the relationship between the three. The paper describes four common deep learning models: convolutional neural network, massive-training artificial neural network, auto-encoder, and deep belief network. The convolutional neural network is the most commonly used deep learning model, mainly including two dimensional convolutional neural network,three dimensional convolutional neural network and multi-stream multi-scale convolutional neural network, of which multi-stream multi-scale convolutional neural network it is more conducive to the classification of lung nodules. The massive-training artificial neural network has advantages in limited lung nodule training samples. The auto-encoder can detect lung nodules in a lower dimensional space.The deep belief network is a generation mode. Combining with extreme learning machine could improve the diagnosis rate of pulmonary nodules. Finally, it analyzes the current problems of artificial intelligence: too few labeled images;insufficient interpretability and controllability. There are ethical and legal issues. In short,artificial intelligence based on deep learning has changed not only imaging, but also all other medical fields, and has broad application prospects.

关 键 词：深度学习 机器学习 人工智能 肺结节 神经网络 

分 类 号：R734.2[医药卫生—肿瘤] TP18[医药卫生—临床医学] TP391.41[自动化与计算机技术—控制理论与控制工程]