光学 精密工程, 2018, 26 (5): 1211, 网络出版: 2018-08-14   

基于三维卷积神经网络的低剂量CT肺结节检测

Detection of low dose CT pulmonary nodules based on 3D convolution neural network
作者单位
1 内蒙古科技大学 信息工程学院, 内蒙古 包头 014010
2 上海大学 计算机工程与科学学院, 上海 200444
摘要
为提高早期肺癌筛查过程中肺结节的检出率, 提出利用三维卷积神经网络进行低剂量CT肺结节检测。首先采用多方向形态学滤波算法对低剂量序列CT图像进行预处理; 接着, 利用改进三维区域生长与凸包算法相结合进行肺实质分割; 然后提取三维候选结节, 为了解决卷积神经网络对样本不平衡的敏感问题, 对三维候选结节正样本进行旋转和光照处理; 最后在不同的网络参数下, 对ELCAP数据库中50个序列低剂量肺癌筛查数据进行4组实验。实验结果表明, 通过对网络参数的不断优化, 准确度、灵敏度、特异度以及ROC曲线的AUC值分别达到了84.6%、88.89%、8032%及0.924 4。该方法能够正确地对低剂量CT肺结节进行检测, 与文献所提出肺结节检测算法相比, 准确度、灵敏度和特异度分别平均提高了5.37%、5.6%和10.42%, 综合性能较强, 可以为肺癌筛查提供有效的帮助。
Abstract
To improve the detection rate of pulmonary nodules in early lung cancer screening, a low-dose CT pulmonary nodule detection algorithm based on 3D convolution neural network was presented. First, the multi-directional morphological filtering algorithm was used to preprocess low-dose sequence CT image. The improved 3D region growth algorithm combined with the convex hull algorithm was used for lung parenchymal segmentation. Then the 3D candidate nodules were routed and illuminated in order to solve the convolution neural network on the sample imbalance sensitive issues. Finally, in situations of different network parameters, four groups of experiments were performed on the 50 sequences of low-dose lung cancer screening data in ELCAP database. The results showed that accuracy, sensitivity, specificity and ROC curve of the AUC values were 84.6%, 88.89%, 80.32% and 0.924 4 respectively by the constant optimization of network parameters. The proposed algorithm can correctly detect low-dose lung nodules, with the the accuracy, sensitivity, and specificity increased by 5.37%, 5.6% and 10.42%, respectively, which is more comprehensive and can provide effective help for lung cancer screening compared with conventional lung nodule detection algorithm.

吕晓琪, 吴凉, 谷宇, 张文莉, 李菁. 基于三维卷积神经网络的低剂量CT肺结节检测[J]. 光学 精密工程, 2018, 26(5): 1211. L Xiao-qi, WU Liang, GU Yu, ZHANG Wen-li, LI Jing. Detection of low dose CT pulmonary nodules based on 3D convolution neural network[J]. Optics and Precision Engineering, 2018, 26(5): 1211.

本文已被 13 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!