基于增强AlexNet的深度学习的阿尔茨海默病的早期诊断

doi:10.11896/j.issn.1002-137X.2017.6A.011

摘要/Abstract

摘要： 在中国,越来越多的老人正在忍受着阿尔茨海默病(Alzheimer’s Disease,AD)的痛苦。阿尔茨海默病俗称老年痴呆症,临床上表现为失忆、丧失语言能力等。目前,中国的阿尔茨海默病患者人数已居世界第一。因此,早期诊断阿尔茨海默病变得十分急迫。研究表明,轻度认知障碍(Mild Cognitive Impairment,MCI)转化为阿尔茨海默病的概率很高,它是介于阿尔茨海默病和正常(Healthy Control,HC)之间的一种状态。随着大数据时代的来临,机器学习方法在疾病诊断方面受到热捧。所以,研究提出使用深度学习方法实现对阿尔茨海默病、轻度认知障碍和健康人群的诊断。数据库来自网络公开数据库ADNI。原始的核磁共振图像(Magnetic Resonance Imaging,MRI)的预处理得到首都医科大学附属北京天坛医院的指导。使用卷积神经网络对降维后的实验数据进行训练。因为目前的网络模型不是针对医学图像的,所以实验的重点在于改进现有网络模型,使之达到良好的诊断效果。改进的网络模型是在图像分类方面十分出色的AlexNet网络模型。实验根据阿尔茨海默病的特点提出改进原始模型的4种算法,采用并行方式计算,使用曙光W780-G20服务器,利用8块NVIDIA Tesla K80进行 GPU计算,获得4个分类器:AD vs.HC,AD vs.MCI,MCI vs.HC和AD vs.MCI vs.HC。数据集中图像总数量超过7万张,耗时不超过30分钟。最终,通过绘制ROC曲线,计算敏感度、特异度、精确度,对测试结果进行评估,得到了较好的测试结果。

关键词: 阿尔茨海默病,轻度认知障碍,深度学习,卷积神经网络,增强的AlexNet网络模型,脑图像,核磁共振图像

Abstract: More and more people suffer from Alzheimer’s disease(AD) in China.AD is characterized by loss of memory and language ability,associated with aging.Currently,the number of Chinese patients has ranked first in the world.So,early diagnosis of AD is particularly urgent.Studies have shown that mild cognitive impairment (MCI) has a high pro-bability converted to AD.MCI may be a transition between healthy control (HC) and AD.With the advent of the era of big data,the machine learning algorithm is more and more popular in the diagnosis of disease.The method of deep lear-ning helps us classify AD,MCI and HC.The data set of magnetic resonance imaging (MRI) is from Alzheimer disease neuroimaging initiative (ADNI) as the data set.The pre-treatment of the raw brain MRI is directed by Beijing Tiantan Hospital affiliated to Capital Medical University.Images after dimensionality reduction are learned by deep convolutionalneural network (CNN) automatically.The current architecture of network is not for medical images.So,experiments focuss on improving existing networks,so as to achieve good diagnostic results.AlexNet is an excellent architecture for images classification which the experiments choose to improve.In this paper,we proposed 4 algorithms to improve the original model according to the characteristic of AD.Data ran in parallel with 8 GPUs of NVIDIA Tesla K80 of W780-G20 by Sugon.Then,we obtained 4 classifiers,AD vs.HC,AD vs.MCI,MCI vs.HC and AD vs.MCI vs.HC.Models were trained no more than 30 minutes with more than 70000 images.Finally,algorithms were evaluated by drawing ROC curve and computing sensitivity and specificity,and the better results were showed. 〖BHDWG1,WK42,WK43,WK42W〗第6A期吕鸿蒙 ,等:基于增强AlexNet的深度学习的阿尔茨海默病的早期诊断

Key words: Alzheimer’s disease,Mild cognitive impairment,Deep learning,Convolutional neural network,Intensive Alex Net,Brain imaging,MRI

吕鸿蒙,赵地,迟学斌. 基于增强AlexNet的深度学习的阿尔茨海默病的早期诊断[J]. 计算机科学, 2017, 44(Z6): 50-60. https://doi.org/10.11896/j.issn.1002-137X.2017.6A.011

LV Hong-meng, ZHAO Di and CHI Xue-bin. Deep Learning for Early Diagnosis of Alzheimer’s Disease Based on Intensive AlexNet[J]. Computer Science, 2017, 44(Z6): 50-60. https://doi.org/10.11896/j.issn.1002-137X.2017.6A.011

参考文献

[1] World Alzheimer Report 2009.http://www.alz.co.uk/research/files/WorldAlzheimerReport.pdf.
[2] PETERSEN R C,SMITH G E,WARING S C,et al.Mild cognitive impairment:clinical characterization and outcome[J].Archives of Neurology,1999,56(3):303-308.
[3] LIU S,CAI W,et al.Early diagnosis of Alzheimer's disease with deep learning[C]∥Proceedings of the 2014 IEEE 11th International Symposium on Biomedical Imaging (ISBI).Beijing,China:2014:1015-1018.
[4] GUPTA A,AYHAN M,MAIDA A.Natural image bases to represent neuroimaging data[C]∥Proceedings of the 30th International Conference on Machine Learning (ICML-13).Atlanta,USA,2013:987-994.
[5] KLOPPEL S,STONNINGTON C M,BARNES J,et al.Accuracy of dementia diagnosis-a direct comparison between radiologists and a computerized method[J].Brain,2008,131(11):2969-2974.
[6] VAN LEEMPUT K,MAES F,VANDERMEULEN D,et al.Automated model-based tissue classification of MR images of the brain[J].IEEE Transactions on Medical Imaging,1999,18(10):897-908.
[7] TRIPOLITI E E.A supervised method to assist the diagnosis and monitor progression of Alzheimer’s disease using data from an fMRI experiment[J].Artificial Intelligence in Medicine,2011,53(1):35-45.
[8] CHRIS H,VIKAS S,LOPAMUDRA M,et al.Spatially aug-mented LP boosting for AD classification with evaluations on the ADNI dataset[J].Neuroimage,2009,48(1):138-149.
[9] BADER B W,KOLDA T G.Tensor decompositions and applications.Preprint of article[J].Siam Review,2009,51(3):455-500.
[10] TUCKER L R.Some mathematical notes on three-mode factor analysis[J].Psychometrika,1966,31(3):279-311.
[11] TUCKER L R.The extension of factor analysis to three dimensional matrices[C]∥Proceedings of the Contributions to Mathematical Psychology.New York,USA,1964:110-127.
[12] PHAN A H,CICHOCKI A S,TICHAVSKY P.On Fast algorithms for orthogonal Tucker decomposition[C]∥Proceedings of the IEEE International Conference on Acoustics,Speech and Signal Processing.Florence,Italy,2014:6766-6770.
[13] MUTI D,BOURENNANE S.Survey on tensor signal algebraic filtering[J].Signal Processing,2007,87(2):237-249.
[14] LETEXIER D,BOURENNANE S.Adaptive Flattening for Mu-ltidimensional Image Restoration[J].Signal Processing Letters IEEE,2008,15:229-232.
[15] KANDEL E R.An introduction to the work of David Hubel and TorstenWiesel[J].Journal of Physiology,2009,587(12):2733-2741.
[16] 王青海.BP神经网络算法的一种改进[J].青海大学学报(自然版),2004,22(3):82-84.
[17] HINTON G E,OSINDERO S,YW T.A Fast Learning Algorithm for Deep Belief Nets[J].Neural Computation,2006,18(7):1527-1554.
[18] BENGIO Y S.Practical Recommendations for Gradient-BasedTraining of Deep Architectures[M]∥ Neural Networks:Tricks of the Trade.Springer Berlin Heidelberg,2012:437-478.
[19] LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[20] SHUIWANG J,MING Y,KAI Y.3D Convolutional NeuralNetworks for Human Action Recognition[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2013,35(1):221-231.
[21] HUBEL D H,WIESEL T N.Receptive fields,binocular interaction and functional architecture in the cat’s visual cortex[J].The Journal of physiology,1962,160(1):106-154.
[22] HARTLINE H K.The receptive fields of optic nerve fibers[J].American Journal of Physiology--Legacy Content,1940,130(4):690-699.
[23] SIMARD P Y,DAVE,STEINKRAUS,PLATT J C.Best practices for convolutional neural networks applied to visual document analysis[C]∥Proceedings of the Seventh International Conference on Document Analysis and Recognition.Edinburgh,Scotland,2003:958-963.
[24] LECUN Y,BENGIO Y.Convolutional networks for images,speech,and time series[M].The Handbook of Brain Theory & Neural,1995.
[25] MORRIS J C.The Clinical Dementia Rating (CDR):currentversion and scoring rules[J].Neurology,1993,43(11):2412-2414.
[26] TOMBAUGH T N,MCLNTYRE N J.The mini-mental stateexamination:a comprehensive review[J].Journal of the American Geriatrics Society,1992,40(9):922-935.
[27] PAYAN A,MONTANA G.Predicting Alzheimer’s disease:a neuroimaging study with 3D convolutional neural networks.http://arxiv.org/pdf/1502.02506.pdf.
[28] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageNet classification with deep convolutional neural networks[C]∥Proceedings of the Advances in Neural Information Processing Systems.South Lake Tahoe,US:2012:1097-1105.
[29] HINTON G,DENG L,YU D,et al.Deep Neural Networks for Acoustic Modeling in Speech Recognition[J].IEEE Signal Processing Magazine,2012,29(6):82-97.
[30] LIU F,SHEN C.Learning Deep Convolutional Features forMRI Based Alzheimer’s Disease Classification.http://arxiv-web.arxiv.org/pdf/1404.3366v1.
[31] JIA Y,SHELHAMER E,DONAHUE J,et al.Caffe:Convolutional Architecture for Fast Feature Embedding[J].EprintArxiv,2014:675-678.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed