深度学习模型GoolgeNet-PNN对肝硬化的识别
|
摘要
针对传统机器学习人工提取特征耗时耗力,并且提取高质量特征存在一定困难等问题,将基于深度学习的方法,首次结合卷积神经网络和概率神经网络,提出了一种新的模型GoolgeNet-PNN,其自动学习特征,避免了手动提取特征的繁琐性,而且结合了PNN训练容易、收敛速度快等特点,在肝病分类的实验中取得了较好的效果;并使用了迁移学习的方法,通过在自然图像集的预训练,然后应用到医学图像,避免了因样本不足而出现的过拟合问题,实验结果最终表明识别准确率要优于其他方法,达到了98%的客观识别率。
Traditional machine learning is difficult to extract high quality features and it consumes much time and energy,Therefore, based on the deep learning method and combined with convolution neural network and probabilistic neural network, a new model called GoolgeNet-PNN is first put forward and applied. Firstly, it automatically learns features and avoids the complexity of manually extracting features. Secondly, it combines the advantages of PNN, such as easy training and fast convergence speed. It has achieved good results in the experiment of liver disease classification. What's more, combined with the migrating learning, the method firstly pre-trains in the natural image set and then is applied to the medical image, which avoids the overfitting problem caused by the shortage of samples. Finally, experimental results show recognition accuracy is better than other methods and it has reached 98% objectively.
Traditional machine learning is difficult to extract high quality features and it consumes much time and energy,Therefore, based on the deep learning method and combined with convolution neural network and probabilistic neural network, a new model called GoolgeNet-PNN is first put forward and applied. Firstly, it automatically learns features and avoids the complexity of manually extracting features. Secondly, it combines the advantages of PNN, such as easy training and fast convergence speed. It has achieved good results in the experiment of liver disease classification. What's more, combined with the migrating learning, the method firstly pre-trains in the natural image set and then is applied to the medical image, which avoids the overfitting problem caused by the shortage of samples. Finally, experimental results show recognition accuracy is better than other methods and it has reached 98% objectively.
引文
[1]周国辉,汪源源.基于超声图像特征的肝硬化分析系统[J].声学技术,2003,22(4):288-291.
[2]雷一鸣,赵希梅,王国栋,等.基于一种改进的LBP算法和超限学习机的肝硬化识别[J].计算机科学,2017,44(10):45-50.
[3]韩秀芝,赵希梅,于可歆,等.一种基于LBP特征提取和稀疏表示的肝病识别算法[J].中国生物医学工程学报,2017(6):647-653.
[4]郭丽丽,丁世飞.深度学习研究进展[J].计算机科学,2015,42(5):28-33.
[5] Hinton G E,Osindero S,Teh Y W.A fast learning algo-rithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[6] Lee H,Grosse R,Ranganath R,et al.Unsupervised learningof hierarchical representations with convolutional deepbelief networks[J].Communications of the ACM,2011,54(10):95-103.
[7] Suk H I,Lee S W,Shen D.Hierarchical feature represen-tation and multimodal fusion with deep learning for AD/MCI diagnosis[J].Neuroimage,2014,101:569-582.
[8] Prasoon A,Petersen K,Igel C,et al.Deep feature learningfor knee cartilage segmentation using a triplanar convo-lutional neural network[C]//Medical Image Computing andComputer-Assisted Intervention-MICCAI 2013.Berlin,Heidelberg:Springer,2013:246-253.
[9] Zhang Wenlu,Li Rongjian,Deng Houtao,et al.Deep con-volutional neural networks for multi-modality isointenseinfant brain image segmentation[C]//Proc IEEE Int SympBiomed Imaging,2015,108:1342-1345.
[10] Li F,Tran L,Thung K H,et al.Robust deep learningfor improved classification of AD/MCI patients[C]//International Workshop on Machine Learning in MedicalImaging,2014:240-247.
[11] Buty M,Xu Z,Gao M,et al.Characterization of lungnodule malignancy using hybrid shape and appearancefeatures[C]//International Conference on Medical ImageComputing and Computer-Assisted Intervention,2016:662-670.
[12] Li Q,Cai W,Wang X,et al.Medical image classificationwith convolutional neural network[C]//International Con-ference on Control Automation Robotics&Vision,2014:844-848.
[13] Gao M,Bagci U,Lu L,et al.Holistic classification ofCT attenuation patterns for interstitial lung diseases viadeep convolutional neural networks[J].Comput MethodsBiomech Biomed Eng Imaging Vis,2015,6(1):1-6.
[14] Anthimopoulos M,Christodoulidis S,Ebner L,et al.Lung pattern classification for interstitial lung diseasesusing a deep convolutional neural network[J].IEEETransactions on Medical Imaging,2016,35(5):1207-1216.
[15] Szegedy C,Vanhoucke V,Ioffe S,et al.Rethinking theinception architecture for computer vision[C]//ComputerVision and Pattern Recognition,2016:2818-2826.
[16] Donahue J,Jia Y,Vinyals O,et al.DeCAF:a deep con-volutional activation feature for generic visual recogni-tion[C]//International Conference on Machine Learning,2014.
[17] Razavian A S,Azizpour H,Sullivan J,et al.CNN fea-tures off-the-shelf:an astounding baseline for recogni-tion[C]//CVPR,2014:512-519.
[18] Yosinski J,Clune J,Bengio Y,et al.How transferableare features in deep neural networks?[C]//Proceedingsof the 27th International Conference on Neural Infor-mation Processing Systems(NIPS’14),2014,27:3320-3328.
[19] Specht D F.Probabilistic neural networks☆[J].NeuralNetworks,1990,3(1):109-118.
[20] Specht D F.Probabilistic neural network and generalregression neural network[M]//Fuzzy logic and neuralnetwork handbook.Hightstown,NJ,USA:Mc Graw-HillInc,1996:301-344.
[2]雷一鸣,赵希梅,王国栋,等.基于一种改进的LBP算法和超限学习机的肝硬化识别[J].计算机科学,2017,44(10):45-50.
[3]韩秀芝,赵希梅,于可歆,等.一种基于LBP特征提取和稀疏表示的肝病识别算法[J].中国生物医学工程学报,2017(6):647-653.
[4]郭丽丽,丁世飞.深度学习研究进展[J].计算机科学,2015,42(5):28-33.
[5] Hinton G E,Osindero S,Teh Y W.A fast learning algo-rithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[6] Lee H,Grosse R,Ranganath R,et al.Unsupervised learningof hierarchical representations with convolutional deepbelief networks[J].Communications of the ACM,2011,54(10):95-103.
[7] Suk H I,Lee S W,Shen D.Hierarchical feature represen-tation and multimodal fusion with deep learning for AD/MCI diagnosis[J].Neuroimage,2014,101:569-582.
[8] Prasoon A,Petersen K,Igel C,et al.Deep feature learningfor knee cartilage segmentation using a triplanar convo-lutional neural network[C]//Medical Image Computing andComputer-Assisted Intervention-MICCAI 2013.Berlin,Heidelberg:Springer,2013:246-253.
[9] Zhang Wenlu,Li Rongjian,Deng Houtao,et al.Deep con-volutional neural networks for multi-modality isointenseinfant brain image segmentation[C]//Proc IEEE Int SympBiomed Imaging,2015,108:1342-1345.
[10] Li F,Tran L,Thung K H,et al.Robust deep learningfor improved classification of AD/MCI patients[C]//International Workshop on Machine Learning in MedicalImaging,2014:240-247.
[11] Buty M,Xu Z,Gao M,et al.Characterization of lungnodule malignancy using hybrid shape and appearancefeatures[C]//International Conference on Medical ImageComputing and Computer-Assisted Intervention,2016:662-670.
[12] Li Q,Cai W,Wang X,et al.Medical image classificationwith convolutional neural network[C]//International Con-ference on Control Automation Robotics&Vision,2014:844-848.
[13] Gao M,Bagci U,Lu L,et al.Holistic classification ofCT attenuation patterns for interstitial lung diseases viadeep convolutional neural networks[J].Comput MethodsBiomech Biomed Eng Imaging Vis,2015,6(1):1-6.
[14] Anthimopoulos M,Christodoulidis S,Ebner L,et al.Lung pattern classification for interstitial lung diseasesusing a deep convolutional neural network[J].IEEETransactions on Medical Imaging,2016,35(5):1207-1216.
[15] Szegedy C,Vanhoucke V,Ioffe S,et al.Rethinking theinception architecture for computer vision[C]//ComputerVision and Pattern Recognition,2016:2818-2826.
[16] Donahue J,Jia Y,Vinyals O,et al.DeCAF:a deep con-volutional activation feature for generic visual recogni-tion[C]//International Conference on Machine Learning,2014.
[17] Razavian A S,Azizpour H,Sullivan J,et al.CNN fea-tures off-the-shelf:an astounding baseline for recogni-tion[C]//CVPR,2014:512-519.
[18] Yosinski J,Clune J,Bengio Y,et al.How transferableare features in deep neural networks?[C]//Proceedingsof the 27th International Conference on Neural Infor-mation Processing Systems(NIPS’14),2014,27:3320-3328.
[19] Specht D F.Probabilistic neural networks☆[J].NeuralNetworks,1990,3(1):109-118.
[20] Specht D F.Probabilistic neural network and generalregression neural network[M]//Fuzzy logic and neuralnetwork handbook.Hightstown,NJ,USA:Mc Graw-HillInc,1996:301-344.