Application of computer aided diagnosis system based on multi-stage three dimensional deep convolutional neural network in lung cancer screening
10.3760/cma.j.cn112149-20200414-00554-1
- VernacularTitle:基于多阶段三维深度卷积神经网络的计算机辅助诊断系统在肺癌筛查中的应用
- Author:
Qingcheng MENG
1
;
Pengrui GAO
;
Lanwei GUO
;
Jia DING
;
Xuejun CHEN
;
Hailiang LI
Author Information
1. 郑州大学附属肿瘤医院放射科 450000
- From:
Chinese Journal of Radiology
2020;54(6):552-556
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To evaluate the value of a novel multiphase three-dimensional deep learning neural network of computer-aided diagnosis (CAD) used in LDCT lung cancer screening.Methods:Eight thousand eight hundred and fifty volunteers with 1 111 nodules were enrolled in the lung cancer screening from November of 2013 to December of 2017, and the baseline LDCT imaging data of volunteers accompanied with clinical information were retrospectively analyzed. All volunteers in this study were designed to receive LDCT test at least once. All the imaging of volunteers were read through the methods of visual detectioin (VD), CAD, and VD Combined CAD. The criteria of the true pulmonary nodule was determinated by the consistent opinion of two specialists in chest imaging（in case of disagreement, the decision should made by the third chief physician). In terms of the numbers, types or Lung-RADS categories of nodules, the detection rate, missed diagnosis rate and false positive rate of pulmonary nodules or lung cancer among three methods were compared, and the rates between groups were compared by χ 2test. Results:Compared with VD or CAD ,the detection rate of nodules in the CAD combined VD was significantly increased (95.7% , 94.2%, vs. 80.1% P<0.05 ), and the rate of missed diagnosis was significantly reduced (5.8%, 4.3% vs. 19.9% ，χ2=101.650, 128.500 ，P<0.05); Compared with VD, the methods of CAD or VD combined CAD significantly increased the the detection rates of Lung-RADS categories (χ2 =25.083,23.449, P=0.000, 0.000) or different types of nodules (χ2=6.955,6.821, P=0.031, 0.033), but there was no statistically significant difference between CAD and VD combined CAD for Lung-RADS categories and different types of nodules (all P>0.05); Compared with VD and VD combined CAD, the positive prediction rate of CAD for lung cancer was significantly reduced, and the rate of missed diagnosis and false positive rate were significantly increased, but there was no significant difference between VD and VD combined CAD in the prediction rate, missed diagnosis rate and false positive rate of lung cancer. Conclusion:The method of CAD combined VD can reduce the detection of false positive nodules and improve the detection rate of true pulmonary nodules,which is the preferred method using in LDCT lung cancer screening for city population.
