摘要
为实现肺结节自动分析与识别,研究基于模糊建模思想和迭代相对模糊连接度(IRFC)算法的自动解剖识别(AAR)方法。该方法包括5个步骤:收集图像数据,用于模型构建和测试AAR;叙述胸腔中每个器官的精确定义,根据定义提取肺部轮廓;建立分层模糊解剖模型;利用分层模型识别和定位肺部;根据层级结构提取肺部轮廓。将分割好的肺部图片作为输入送入卷积神经网络进行肺部结节检测,通过使用VGG-16网络模型,在天池医疗AI大赛的数据集上实现了92.72%的目标检测准确率。
To realize automatic analysis and recognition of pulmonary nodules,an automatic anatomy recognition(AAR)methodology based on fuzzy modeling ideas and an iterative relative fuzzy connectedness(IRFC)delineation algorithm was studied.The methodology consisted of five main steps including gathering image date for both building models and testing the AAR algorithms,formulating precise definitions of each organ in the thorax and delineating lungs following these definitions,building hierarchical fuzzy anatomy models,recognizing and locating lungs with the hierarchical models,and delineating the lungs following the hierarchy.The segmented lung images were taken as input into the convolutional neural network for pulmonary nodule detection.By the use of the VGG-16 network model,a target detection accuracy of 92.72%is achieved on the data set of the Tianchi Medical AI Contest.
引文
[1]Faizal Khan Z,Abdulaziz S Al Sayyari,Syed Usama Quadri.Automated segmentation of lung images using textural echo state neural networks[C]//International Conference on Informatics,Health&Technology,2017:1-5.
[2]Sara Soltaninejad,Irene Cheng,Anup Basu.Robust lung segmentation combining adaptive concave hulls with active contours[C]//IEEE International Conference on Systems,Man,and Cybernetics,2016.
[3]DAI Shuangfeng,LYU Ke,QU Rui,et al.The whole lung segmentation method based on 3Dregion growing method and improved convex hull algorithm[J].School of Engineering Science,2016,38(9):2358-2364(in Chinese).[代双凤,吕科,瞿锐,等.基于3D区域增长法和改进的凸包算法相结合的全肺分割方法[J].中国科学院大学工程科学学院,2016,38(9):2358-2364.]
[4]DAI Meiling,QI Jin,ZHOU Zhongxing,et al.The classification of pulmonary nodules based on texture features over Local Jet transformation space[J].Chinese Journal of Biomedical Engineering,2017,36(1):12-19(in Chinese).[代美玲,祁瑾,周仲兴,等.基于Local Jet变换空间纹理特征的肺结节分类方法[J].中国生物医学工程学报,2017,36(1):12-19.]
[5]Hongyang Jiang,He Ma,Wei Qian,et al.An automatic detection system of lung nodule based on multi-group patch-based deep learning network[C]//IEEE Journal of Biomedical and Health Informatics,IEEE Early Access Articles,2017:1-11.
[6]LE Anglin.SVM-based segmentation and recognition of lung nodule[D].Shengyang:Northeastern University,2015(in Chinese).[乐昂霖.基于SVM的肺结节分割与识别[D].沈阳:东北大学,2015.]
[7]Ahmed Soliman,Fahmi Khalifa,Ahmed Elnakib,et al.Accurate lungs segmentation on CT chest images by adaptive appearance-guided shape modeling[J].IEEE Transactions on Medical Imaging,IEEE Journals&Magazines,2017,36(1):263-276.
[8]Sunita Agarwala,Debashis Nandi,Abhishek Kumar,et al.Automated segmentation of lung field in HRCT images using active shape model[C]//IEEE Region 10 Conference,2017:2516-2520.
[9]Ratishchandra Huidrom,Yambem Jina Chanu,Khumanthem Manglem Singh.A fast automated lung segmentation method for the diagnosis of lung cancer[C]//IEEE Region 10Conference,2017:1499-1502.
[10]Awais Mansoor,Ulas Bagci,Ziyue Xu,et al.A generic approach to pathological lung segmentation[J].IEEE Transactions on Medical Imaging,2014,33(12):2293-2310.
[11]Awais Mansoor,Ulas Bagci,Ziyue Xu,et al.Correction to“ageneric approach to pathological lung segmentation”[J].IEEE Transactions on Medical Imaging,IEEE Journals&Magazines,2015,34(1):354.
[12]Awais Mansoor,Ulas Bagci,Brent Foster,et al.CIDI-lungseg:A single-click annotation tool for automatic delineation of lungs from CT scans[C]//36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society,2014:1087-1090.
[13]Syoji Kobashi,Jayaram K Udupa.Fuzzy connectedness image segmentation for newborn brain extraction in MR images[C]//35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society,2013:7136-7139.
[14]Kaya A,Can AB.A weighted rule based method for predicting malignancy of pulmonary nodules by nodule characteristics[J].Journal of Biomedical Informatics,IEEE Conferences,2015,56(8):69-79.