北京生物医学工程

基于改进YOLO算法的肺部CT图像中结节检测研究

Study on nodule detection in lung CT images based on improved YOLO algorithm

作者：王波冯旭鹏刘利军黄青松

单位：昆明理工大学信息工程与自动化学院(昆明 650500) 云南省计算机技术应用重点实验室(昆明650500) 昆明理工大学教育技术与网络中心(昆明650500）

关键词： YOLO算法; CT图像; 肺结节检测; 多尺度预测; 目标识别

分类号：R318.04

出版年·卷·期（页码）：2020·39·6（615-621）

摘要：

目的随着机器学习的发展，如何准确高效地识别肺部CT图像中的肺结节具有重要的应用价值。方法针对肺部结构复杂、肺部结节过小、肺结节病理特征各异等特点。提出一个以YOLO算法为基础，结合Darknet-53网络和Densenet网络的思想，在多尺度间具有紧密连接的深度卷积神经网络。为保证图像有效信息和提高目标定位的精确性以及检测的召回率，首先对数据集图像尺寸大小进行固定，其次通过使用K-means算法对数据集进行聚类分析。最后将使用二元交叉熵做类别预测。实验使用美国癌症研究所公开的肺部图像数据集联盟（lung image databse consortium, LIDC）提供的数据集，对肺结节检测的准确率以及检测效率进行了实验对比。结果改进的深度卷积神经网络对肺结节检测的准确率及检测效率均有提升。在肺部CT图像中肺结节检测的平均查全率达到95.69%，对微小结节的平均查全率达到88.66%，每秒识别帧数达到32 f/s，相比当前最快的Faster R-CNN检测时间缩短了近80%。结论通过对YOLO算法的改进可以提高肺结节检测效率，为肺部CT图像肺结节实时检测提供了条件。

Objective With the development of machine learning, how to accurately and efficiently identify pulmonary nodules in CT images of the lung has important application value. Methods The pulmonary structure is complex, the pulmonary nodules are too small, and the pulmonary nodules have different pathological characteristics. Based on YOLO algorithm, a deep convolutional neural network with close connection between multiple scales is proposed, which combines the darknet-53 network and Densenet network. In order to ensure the effective information of the image and improve the accuracy of the target positioning and the recall rate of detection, the image size of the data set is fixed firstly, and then the clustering analysis of the data set is carried out by using k-means algorithm.Finally, binary cross entropy will be used to make category prediction. The study used data sets from the Lung Image database Consortium (LIDC), which is publicly available from the national cancer institute, to compare the accuracy and efficiency of Lung nodule detection. Results The improved deep convolutional neural network improves the accuracy and efficiency of lung nodule detection.In the lung CT images, the average recall rate of lung nodules detection was 95.69%, the average recall rate of tiny nodules was 88.66%, and the recognition frames per second reached 32f/s, which was nearly 80% shorter than the current fastest detection time of Faster r-cnn. Conclusions The improvement of YOLO algorithm can improve the detection efficiency of pulmonary nodules and provide conditions for real-time detection of pulmonary nodules in CT images.

参考文献：

[1] Siegel RL, Miller KD, Jemal A. Cancer Statistics, 2017[J]. CA: A Cancer Journal for Clinicians, 2017, 67(1): 7-30.
[2] Nie SD, Chen ZX, Li LH. A CI feature-based pulmonary nodule segmentation using three-domain mean shift clustering[C]// International Conference on Wavelet Analysis and Pattern Recognition, 2007. Beijing, China: IEEE Xplore, 2007.
[3] Miller KD, Siegel RL, Lin CC, et al. Cancer treatment and survivorship statistics, 2016[J]. CA: A Cancer Journal for Clinicians, 2016, 66(4): 271-289.
[4] Koning HJD, Meza R, Plevritis SK, et al. Benefits and harms of CT lung cancer screening strategies. a comparative modeling study for the U.S. Preventive Services Task Force[J]. Annals of Internal Medicine, 2014, 160(5): 311-320.
[5] Henschke CI, Naidich DP, Yankelevitz DF, et al. Early lung cancer action project: initial findings on repeat screenings[J]. Cancer, 2001, 92(1): 153-159.
[6] Puderbach M, Kauczor HU. Can lung MR replace lung CT?[J]. Pediatric Radiology, 2008, 38(Suppl 3): S439- S451.
[7] 刘淑琴. 肺癌影像医学诊断进展[J]. 河北医药, 2008, 30(6): 860-861.
[8]Ko JP, Betke M. Chest CT: automated nodule detection and assessment of change over time- preliminary experience[J]. Radiology, 2001, 218(1): 267-273.
[9] Armato SG 3rd, Giger ML, Moran CJ, et al. Computerized detection of pulmonary nodules on CT scans[J]. Radiographics, 1999, 19(5): 1303-1311.
[10] Zhao B, Gamsu G, Ginsberg MS, et al. Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm[J]. Journal of Applied Clinical Medical Physics, 2003, 4(3): 248-260.
[11] Antonelli M, Lazzerini B, Marcelloni F. Segmentation and reconstruction of the lung volume in CT images[C]// 2005 ACM Symposium on Applied Computing. Santa Fe, New Mexico,USA: ACM, 2005: 255.
[12 ]El-Baz A, Gimel'Farb G, Falk R, et al. A new CAD system for early diagnosis of detected lung nodules[C]// 2007 IEEE International Conference on Image Processing. San Antonio, TX, USA: IEEE Press, 2007, 2(II): 461-464.
[13] Li Q, Sone S, Doi K. Selective enhancement filters for nodules, vessels, and airway walls in two- and three-dimensional CT scans[J]. Medical Physics, 2003, 30(8): 2040-2051.
[14] Dou Q, Chen H, Yu L, et al. Multilevel contextual 3-D CNNs for false positive reduction in pulmonary nodule detection[J]. IEEE Transactions on Bio-medical Engineering, 2017, 64(7): 1558-1567.
[15] Teramoto A, Fujita H, Yamamuro O, et al. Automated detection of pulmonary nodules in PET/CT images: ensemble false-positive reduction using a convolutional neural network technique[J]. Medical Physics, 2016, 43(6): 2821-2827.
[16] Anirudh R, Thiagarajan JJ, Bremer T, et al. Lung nodule detection using 3D convolutional neural networks trained on weakly labeled data[C]// SPIE Proceedings 9785, Medical Imaging 2016: Computer-Aided Diagnosis. Bellingham, WA, USA: SPIE, 2016: 978532.
[17] Ding J, Li A, Hu Z, et al. Accurate pulmonary nodule detection in computed tomography images using deep convolutional neural networks[M]// Medical Image Computing and Computer Assisted Intervention ? MICCAI 2017. Berlin: Springer Nature Switzerland AG, 2017:559-567.
[18] Hu S, Hoffman EA, Reinhardt JM. Automatic lung segmentation for accurate quantitation of volumetric X-ray CT images[J]. IEEE Transactions on Medical Imaging, 2001, 20(6): 490-498.
[19] Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-Time Object Detection[M]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE Computer Society, 2016.
[20] Parham J, Stewart C. Detecting plains and Grevy's Zebras in the realworld[C]// 2016 IEEE Winter Applications of Computer Vision Workshops. Lake Placid, NY, USA: IEEE Press, 2016: 1-9.
[21] Redmon J, Farhadi A. YOLOv3: an incremental improvement[J]. 2018.
[22] Huang G, Liu Z, van der Maaten L, et al. Densely Connected Convolutional Networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, Hawaii, USA: IEEE Press, 2017.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录