基于卷积神经网络的肺结节分类算法
|
摘要
肺部的检查是每年体检的重要一部分。体检中有成百上千的病例,而每个病例中含有许多的肺部横切面CT图像。这些都需要专业医生去逐个筛查出存在肺结节的病例,不仅工作量大而且存在误筛的可能。针对上述问题,把卷积神经网络(CNN)引入筛查存在肺结节的CT图像诊断,提出一种基于CNN的分类算法。在LIDC数据库的实验结果表明,对比应用广泛的lenet-5网络和传统方法等,使用自定义的卷积神经网络将分类的正确率提升了4到10个百分点不等。AUC值为0.821 6,也是几个分类器中最大的。相比于其他方法,该方法能较为准确地识别肺部CT图像,可以为临床诊断提供较为客观的参考。
The examination of the lung is a requisite part of the medical examination. There are quantities of cases in the medical examination, and each case contains plenty of cross-sectional CT images of the lung. While, all of these requires that professional doctors screen out the cases of lung nodules one by one, which not only cause tons of work, but also a possibility of inaccurate screening. Aiming at above problems, the Convolutional Neural Network(CNN)is introduced to screen out the CT images of lung nodules, and a classification algorithm based on CNN is proposed. The experimental results on the LIDC database show that, comparing with the widely used lenet-5 networks and traditional methods, the accuracy of classification is increased by 4 to 10 points by using self-defined convolutional neural network. AUC is 0.821 6, which is the largest of several classifiers. In brief, this method can distinguish CT images of the lung more accurately than other methods, and provides more objective reference for clinical diagnosis.
The examination of the lung is a requisite part of the medical examination. There are quantities of cases in the medical examination, and each case contains plenty of cross-sectional CT images of the lung. While, all of these requires that professional doctors screen out the cases of lung nodules one by one, which not only cause tons of work, but also a possibility of inaccurate screening. Aiming at above problems, the Convolutional Neural Network(CNN)is introduced to screen out the CT images of lung nodules, and a classification algorithm based on CNN is proposed. The experimental results on the LIDC database show that, comparing with the widely used lenet-5 networks and traditional methods, the accuracy of classification is increased by 4 to 10 points by using self-defined convolutional neural network. AUC is 0.821 6, which is the largest of several classifiers. In brief, this method can distinguish CT images of the lung more accurately than other methods, and provides more objective reference for clinical diagnosis.
引文
[1]胡向丽,范立南,孙申申,等.基于PCA和改进的BP网络的肺结节识别研究[J].仪器仪表用户,2011,18(6):3-7.
[2]Zhu Y J,Tan Y Q,Hua Y Q,et al.Feature selection and performance evaluation of Support Vector Machine(SVM):Based classifier for differentiating benign and malignant pulmonary nodules by computed tomography[J].Journal of Digital Imaging,2010,23(1):51-65.
[3]Krewer H,Geiger B,Hall L O,et al.2013 IEEE International Conference on Systems,Man,and Cybernetics,Manchester,Oct 13-16,2013[C].Piscataway:IEEE Press,2011.
[4]Chen H,Zhang J,Xu Y,et al.Performance comparison of artificial neural network and logistic regression model for differentiating lung nodules on CT scans[J].Expert Systems with Applications,2012,39(13):11503-11509.
[5]Lumsden A B,Heyman E R.Prospective randomized study evaluating an absorbable cyanoacrylate for use in vascular reconstructions[J].J Vascular Surg,2006,44(5):1002-1009.
[6]Sarikaya R,Hinton C E,Deoras A.Application of deep belief networks for natural language understanding[J].IEEE Transactions on Audio Speech and Language Processing,2014,22(4):778-784.
[7]Landecker W,Chartrand R,Dedeo S.Robust sparse coding and compressed sensing with the difference map[C]//Proceedings of European Conference on Computer Vision,2014:315-329.
[8]Graves A,Mohamed A,Hinton G E.Speech recognition with deep recurrent neural networks[C]//Proceedings of IEEE International Conference on Acoustic Speech and Signal Processing,2013:6645-6649.
[9]Bruna J,Szlam A,Lecun Y.Signal recovery from pooling representations[J].Proceedings of Machine Learning Research,2014,32(2):307-315.
[10]Li H X,Lin Z,Shen X H,et al.A convolutional neural network cascade for face detection[C]//Prceedings of IEEE Conference on Computer Vision and Pattern Recognition,2015:5325-5334.
[11]王华利,周俊忠,张见,等.基于深度卷积神经网络的图像分类算法[J].计算机工程与应用,2017,53(13):181-188.
[12]周俊宇,赵艳明.卷积神经网络在图像分类和目标检测应用综述[J].计算机工程与应用,2017,53(13):34-41.
[13]Qiang Y,Zhang X,Ji G,et al.Automated lung nodule segmentation using an active contour model based on PET/CT images[J].Journal of Computational and Theoretical Nanoscience,2015,12(8):1972-1976.
[14]郭克友,贾海晶,郭晓丽.卷积神经网络在车牌分类器中的应用[J].计算机工程与应用,2017,53(14):209-213.
[15]徐珊珊,刘应安,徐昇.基于卷积神经网络的木材缺陷识别[J].山东大学学报(工学版),2013,43(2):23-28.
[2]Zhu Y J,Tan Y Q,Hua Y Q,et al.Feature selection and performance evaluation of Support Vector Machine(SVM):Based classifier for differentiating benign and malignant pulmonary nodules by computed tomography[J].Journal of Digital Imaging,2010,23(1):51-65.
[3]Krewer H,Geiger B,Hall L O,et al.2013 IEEE International Conference on Systems,Man,and Cybernetics,Manchester,Oct 13-16,2013[C].Piscataway:IEEE Press,2011.
[4]Chen H,Zhang J,Xu Y,et al.Performance comparison of artificial neural network and logistic regression model for differentiating lung nodules on CT scans[J].Expert Systems with Applications,2012,39(13):11503-11509.
[5]Lumsden A B,Heyman E R.Prospective randomized study evaluating an absorbable cyanoacrylate for use in vascular reconstructions[J].J Vascular Surg,2006,44(5):1002-1009.
[6]Sarikaya R,Hinton C E,Deoras A.Application of deep belief networks for natural language understanding[J].IEEE Transactions on Audio Speech and Language Processing,2014,22(4):778-784.
[7]Landecker W,Chartrand R,Dedeo S.Robust sparse coding and compressed sensing with the difference map[C]//Proceedings of European Conference on Computer Vision,2014:315-329.
[8]Graves A,Mohamed A,Hinton G E.Speech recognition with deep recurrent neural networks[C]//Proceedings of IEEE International Conference on Acoustic Speech and Signal Processing,2013:6645-6649.
[9]Bruna J,Szlam A,Lecun Y.Signal recovery from pooling representations[J].Proceedings of Machine Learning Research,2014,32(2):307-315.
[10]Li H X,Lin Z,Shen X H,et al.A convolutional neural network cascade for face detection[C]//Prceedings of IEEE Conference on Computer Vision and Pattern Recognition,2015:5325-5334.
[11]王华利,周俊忠,张见,等.基于深度卷积神经网络的图像分类算法[J].计算机工程与应用,2017,53(13):181-188.
[12]周俊宇,赵艳明.卷积神经网络在图像分类和目标检测应用综述[J].计算机工程与应用,2017,53(13):34-41.
[13]Qiang Y,Zhang X,Ji G,et al.Automated lung nodule segmentation using an active contour model based on PET/CT images[J].Journal of Computational and Theoretical Nanoscience,2015,12(8):1972-1976.
[14]郭克友,贾海晶,郭晓丽.卷积神经网络在车牌分类器中的应用[J].计算机工程与应用,2017,53(14):209-213.
[15]徐珊珊,刘应安,徐昇.基于卷积神经网络的木材缺陷识别[J].山东大学学报(工学版),2013,43(2):23-28.