上海交通大学学报（医学版）

上海交通大学学报（医学版） ›› 2019, Vol. 39 ›› Issue (9): 1045-.doi: 10.3969/j.issn.1674-8115.2019.09.017

• 论著·临床研究 • 上一篇    下一篇

深度卷积神经网络对≤ 3 cm的亚实性肺腺癌 CT图像病理学分型预测的可视化研究

蒋蓓蓓 *，张亚平 *，张琳，刘桂雪，解学乾   

  1. 上海交通大学附属第一人民医院放射科，上海 200080
  • 出版日期:2019-09-28 发布日期:2019-11-02
  • 通讯作者: 解学乾，电子信箱：xiexueqian@hotmail.com。
  • 作者简介:蒋蓓蓓（ 1995—），女，硕士生；电子信箱： jennifer.chiang@hotmail.com。张亚平（ 1987—），女，博士生；电子信箱： zhangyaping.ok@163.com。*为共同第一作者。
  • 基金资助:
    科技部国际合作项目 （2016YFE0103000）；上海市教育委员会高峰高原学科建设计划 （20181814）；上海市科学技术委员会项目 （16411968500， 16410722300）；上海交通大学转化医学交叉研究基金 （ZH2018ZDB10）；上海市第一人民医院临床研究创新团队建设项目 （CTCCR-2018B04）

A visualization study of deep convolutional neural network to classify the pathological type of sub-soild pulmonary adenocarcinoma of ≤ 3 cm based on CT images

JIANG Bei-bei, ZHANG Ya-ping, ZHANG Lin, LIU Gui-xue, XIE Xue-qian   

  1. Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200080, China
  • Online:2019-09-28 Published:2019-11-02
  • Supported by:
    International Scientific Alliance Program of Ministry of Science and Technology of China, 2016YFE0103000; Shanghai Municipal Education Commission—Gaofeng Clinical Medicine Support, 20181814; Project of Science and Technology Commission of Shanghai Municipality, 16411968500, 16410722300; Shanghai Jiao Tong University Translational Medicine Fund, ZH2018ZDB10; Clinical Research Innovation Plan of Shanghai General Hospital, CTCCR-2018B04）。

PDF (PC)

452

摘要： 目的 ·研究卷积神经网络（ convolutional neural network，CNN）根据 CT图像对≤ 3cm的亚实性肺腺癌病理分类的可行性，并通过 CNN激活区可视化分析预测分类的医学影像基础。方法 ·随机纳入 200个经免疫组化染色证实为肺腺癌的亚实性肺结节，标注为浸润前病变（含非典型腺瘤样增生和原位腺癌）、微浸润腺癌和浸润性腺癌。 160个（80%）用于训练 CNN模型， 40个（20%）用于模型验证和激活区可视化分析。激活区的影像特征定义为 14种 CT征象。结果 · CNN对肺腺癌病理分类的准确性为 87.5%。可视化分析发现 CNN激活区主要关注浸润前病变的非实性成分（ 43.0%）和光滑边缘（ 20.2%），关注微浸润腺癌的毛刺边缘（ 18.3%），关注浸润性腺癌的实性成分（ 18.9%）和毛刺边缘（ 14.1%）。结论 · CNN能根据 CT图像对肺腺癌病理分型进行分类预测， CNN激活区的可视化能提示诊断的医学影像基础。

关键词: 卷积神经网络, CT, 肺腺癌

Abstract:

Objective · To investigate the feasibility of deep convolutional neural network (CNN) to classify the pathological type of sub-soild pulmonary adenocarcinoma of ≤ 3 cm based on CT images, and to visualize the medical imaging features derived the activation area of CNN. Methods · A total of 200 sub-solid lung nodules, which were confirmed as adenocarcinomaimmunohistochemical staining, were classified as preinvasive lesions (including atypical adenomatous hyperplasia and adenocarcinoma in situ), microinvasive adenocarcinoma and invasive adenocarcinoma, in which 160 (80%) were used to train the inception v3 CNN architecture, and the other 40 (20%) were used to test the model and visualize the activation area. The characteristics of the activated area were defined as 14 CT signs. Results · The CNN yielded an accuracy of 87.5% to classify three categories of lung nodules. The visualization study found that the CNN activation area mainly focused on the non-solid component (43.0%) and smooth margin (20.2%) of the preinvasive lesions, on the spiculated margin (18.3%) of the microinvasive adenocarcinoma, and on the solid component (18.9%) and the spiculated margin (14.1%) of the invasive adenocarcinoma. Conclusion · CNN can classify the pathological type of lung adenocarcinoma based on CT images. The visualization of activation area of CNN indicates the medical imaging characteristics of diagnosis.

Key words: convolutional neural network, computed tomography, pulmonary adenocarcinoma

中图分类号: