融合多尺度信息的肺结节自动检测算法

刘秀玲; 戚帅帅; 熊鹏; 刘京; 王洪瑞; 杨建利

doi:10.7507/1001-5515.201910047

融合多尺度信息的肺结节自动检测算法

doi: 10.7507/1001-5515.201910047

刘秀玲^{1, 2},
戚帅帅¹,
熊鹏^{1, 2},
刘京³,
王洪瑞^{1, 2},
杨建利^{1, 2, ,}

详细信息

通讯作者:
杨建利，Email：yangjianli_1987@126.com

计量
- 文章访问数: 338
- HTML全文浏览量: 141
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2019-10-25
- 修回日期: 2020-04-06
- 发布日期: 2020-03-17

An automatic pulmonary nodules detection algorithm with multi-scale information fusion

Xiuling LIU^{1, 2},
Shuaishuai QI¹,
Peng XIONG^{1, 2},
Jing LIU³,
Hongrui WANG^{1, 2},
Jianli YANG^{1, 2
, ,}

More Information

Corresponding author: YANG Jianli, Email: yangjianli_1987@126.com

摘要

摘要: 肺结节是早期肺癌的主要表现形式，准确检测肺结节对肺癌的早期诊断和治疗具有重要意义。然而，由于肺部计算机断层扫描（CT）图像背景复杂、检测范围大，且肺结节大小不一、形态各异，所以快速准确检测肺结节是一项极具挑战的工作。为此，本文提出了一种融合多尺度特征的肺结节自动检测算法，实现了肺结节的准确检测。首先，在用于大规模图像识别的深度卷积网络（VGG16）上设计了具有三层模块结构的肺结节检测模型，利用网络第一层模块提取 CT 图像中肺结节特征并粗略地估计肺结节位置；然后利用网络第二层模块融合多尺度的图像特征信息进一步增强肺结节细节特征；而网络第三层模块融合分析第一层和第二层模块的特征，得到多尺度下肺结节候选框；最后利用非极大值抑制方法对多尺度下肺结节候选框进行概率分析，得到最终的肺结节位置。本文应用肺部影像数据库联盟（LIDC）公共数据集上的肺结节数据对所提算法进行了验证，平均检测精度达到 90.9%。本研究成果可应用于肺结节自动筛查系统，有助于提升肺结节筛查精度。
- 多尺度 /
- 肺结节检测 /
- 计算机断层扫描图像 /
- 特征融合
Abstract: Lung nodules are the main manifestation of early lung cancer. So accurate detection of lung nodules is of great significance for early diagnosis and treatment of lung cancer. However, the rapid and accurate detection of pulmonary nodules is a challenging task due to the complex background, large detection range of pulmonary computed tomography (CT) images and the different sizes and shapes of pulmonary nodules. Therefore, this paper proposes a multi-scale feature fusion algorithm for the automatic detection of pulmonary nodules to achieve accurate detection of pulmonary nodules. Firstly, a three-layer modular lung nodule detection model was designed on the deep convolutional network (VGG16) for large-scale image recognition. The first-tier module of the network is used to extract the features of pulmonary nodules in CT images and roughly estimate the location of pulmonary nodules. Then the second-tier module of the network is used to fuse multi-scale image features to further enhance the details of pulmonary nodules. The third-tier module of the network was fused to analyze the features of the first-tier and the second-tier module of the network, and the candidate box of pulmonary nodules in multi-scale was obtained. Finally, the candidate box of pulmonary nodules under multi-scale was analyzed with the method of non-maximum suppression, and the final location of pulmonary nodules was obtained. The algorithm is validated by the data of pulmonary nodules on LIDC-IDRI common data set. The average detection accuracy is 90.9%.
- multi scale /
- pulmonary nodule detection /
- computed tomography images /
- feature fusion

HTML全文

图 1 肺结节自动检测算法实现框图

Figure 1. Block diagram of the realization of pulmonary nodules automatic detection algorithm

下载: 全尺寸图片幻灯片

图 2 检测层网络结构

Figure 2. Network structure of detection layers

下载: 全尺寸图片幻灯片

图 3 高层特征图 CONV6 与低层特征图 Conv_fc 的融合过程

Figure 3. Fusion process of high-level feature map CONV6 and low-level feature map Conv_fc

下载: 全尺寸图片幻灯片

图 4 不同候选框尺寸检测结果

Figure 4. Detection results with different candidate box size

下载: 全尺寸图片幻灯片

图 5 肺结节检测结果

Figure 5. Images of pulmonary nodules detection

下载: 全尺寸图片幻灯片

图 6 肺结节检测效果图

Figure 6. Effect images of pulmonary nodules detection

下载: 全尺寸图片幻灯片

图 7 肺结节测试结果精确率-召回率曲线

Figure 7. Precision-recall curves of pulmonary nodules detection

下载: 全尺寸图片幻灯片

表 1 检测层网络结构参数设置

Table 1. Setting of network structure parameters in detection layers

检测层	特征图大小/像素	步长/像素	候选框尺寸/像素	通道数	纵横比
Conv3	128 × 128	4	8 × 8	256	1，0.5，2
Conv4	64 × 64	8	16 × 16	512	1，0.5，2
Conv5	32 × 32	16	32 × 32	512	1，0.5，2
Conv_fc	16 × 16	32	64 × 64	1 024	1，0.5，2
Conv6	8 × 8	64	128 × 128	512	1，0.5，2

下载: 导出CSV

表 2 不同纵横比组合下的肺结节平均检测精度

Table 2. Average detection accuracy of pulmonary nodules with different aspect ratios

纵横比	平均检测精度（%）
1	90.6
0.5，1	90.7
0.5，1，2	90.9

下载: 导出CSV

表 3 不同算法的平均检测精度

Table 3. Average detection accuracy of different algorithms

算法	平均检测精度（%）
Faster R-CNN	86.9
FPN	89.7
RefineDet	90.7
本文方法	90.9

下载: 导出CSV

参考文献(14)

[1]	陈金东. 中国各类癌症的发病率和死亡率现状及发展趋势. 遵义医学院学报, 2018, 41(6): 5-14.
[2]	陈首英, 刘福林, 庞志刚, 等. 肺癌病人5年生存率及生存因素分析. 预防医学论坛, 2004, 10(1): 1-3.
[3]	Jacobs C, van Rikxoort E M, Twellmann T, et al. Automatic detection of subsolid pulmonary nodules in thoracic computed tomography images. Med Image Analysis, 2014, 18(2): 374-384. doi: 10.1016/j.media.2013.12.001
[4]	Santos A M, Filho A O D C, Silva A C, et al. Automatic detection of small lung nodules in 3D CT data using Gaussian mixture models, Tsallis entropy and SVM. Engineering Applications of Artificial Intelligence, 2014, 36(3): 27-39.
[5]	Fan Weikang, Jiang Huiqin, Ma Ling, et al. A modified faster R-CNN method to improve the performance of the pulmonary nodule detection//Tenth International Conference on Digital Image Processing (ICDIP 2018), Shanghai, 2018. DOI: 10.1117/12.2502893.
[6]	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition//International Conference on Learning Representations 2015 (ICLR 2015). San Diego, 2015. arXiv: 1409.1556.
[7]	葛治文. 基于深度学习的CT图像肺结节检测. 南京: 东南大学, 2018.
[8]	Pezeshk A, Hamidian S, Petrick N, et al. 3D convolutional neural networks for automatic detection of pulmonary nodules in chest CT. IEEE Journal of Biomedical & Health Informatics, 2019, 23(5): 2080-2090.
[9]	Xie hongtao, Yang Dongbao, Sun Nannan, et al. Automated pulmonary nodule detection in CT images using deep convolutional neural networks. Pattern Recognition, 2019, 85: 109-119. doi: 10.1016/j.patcog.2018.07.031
[10]	Neubeck A, Gool L V. Efficient non-maximum suppression//18th International Conference on Pattern Recognition (ICPR'06), Hong Kong: IEEE, 2006: 850-855.
[11]	Armato S G, Roberts R Y, McNitt-Gray M F, et al. The Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI)[DB]. (2000-04-10)[2019-10-25]. https:// wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI.
[12]	Zang S F, Wen L Y, Bian X, et al. Single-shot refinement neural network for object detection//Conference on Computer Vision and Pattern Recognition, Salt Lake City: IEEE, 2018: 4203-4212.
[13]	Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 39(6): 1137-1149.
[14]	Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection. Computer Vision and Pattern Recognition, 2017, 45(3): 2117-2125.