簡易檢索 / 詳目顯示

回結果列表
研究生： 黃雅琳
Huang, Ya-Lin
論文名稱： 基於條件式相互學習之聯邦學習: 阿茲海默症在T1 加權磁振造影之分類辨識
Federated Learning via Conditioned Mutual Learning for Alzheimer’s Disease Classification on T1-weighted MRI
指導教授： 李祈均
Lee, Chi-Chun
口試委員: 吳尚鴻
Wu, Shan-Hung
郭柏志
Kuo, Po-Chih
郭立威
Kuo, Li-Wei
學位類別： 碩士
Master
系所名稱： 電機資訊學院 - 電機工程學系
Department of Electrical Engineering
論文出版年： 2021
畢業學年度： 109
語文別： 英文
論文頁數： 59
中文關鍵詞： 聯邦學習 、知識蒸餾 、相互學習 、個性化學習 、阿茲海默症 、磁振造影
外文關鍵詞： Federated Learning, Knowledge Distillation, Mutual Learning, Personalized Learning, Alzheimer's Disease, Magnetic Resonance Image
相關次數： 點閱：2  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 數據驅動的深度學習被認為是一種應用在建構強大醫學數據模型上具有發展潛力的方法，而這類模型需要大量且不同的數據去做訓練才能足夠有效。然而，醫療數據的收集成本高昂並且需要資金維護，除此之外，醫療數據對於隱私的高度敏感性使得現有醫療數據庫多呈現小規模且分散於各研究單位。聯邦學習是一種能同時達到數據私有與協作學習的訓練架構，讓模型在訓練上能夠獲得所有可用數據的資訊卻不需要直接共享數據本身。在眾多相關研究中，聯邦知識蒸餾是一種透過平均多個客戶站點對單一公共數據集的預測分數，並透過知識蒸餾完成資訊共享的訓練方式。然而，由於不同磁振造影數據集的採集協議、招募標準...... 等會有所歧異，故在不同收集站點的磁振造影數據之間會有數據異質性，而這對使用平均共識來達到資訊共享的個人客戶來說並不理想。
    在本論文中，我們提出了一個基於條件式相互學習的聯邦學習架構(FedCM)，通過考慮到客戶端的本地識別能力以及客戶端之間的數據統計分佈相似度來進行資訊選擇，以提高協作訓練效果。這項研究是第一個利用3D 卷積神經網路在多個阿茲海默症數據集的T1 加權磁振造影之分類辨識的聯合學習，而我們的方法在現實世界的數據與情景上，實現了具有前景的辨識率。除此之外，我們進一步將人類大腦重點腦區中(region of interests, ROIs)對模型預測結果有影嚮力的區域進行了可視化以及相關性排名，發現我們的方法可以使模型關注在與阿茲海默症相關的區域，並且列出了幾個具有研究潛力的區域以供未來進行更深入的研究。

    Data-driven deep learning has been considered a promising method for building robust models for medical data, which often requires a huge amount of diverse data to be sufficiently effective. However, expensive cost of collecting and privacy constraints have resulted in small and distributed medical data sets. Federated
    learning via model distillation is data private collaborative learning where the model can leverage all available data without direct sharing. The data knowledge is shared by distillation through the multisite average prediction scores on the public dataset. However, the average consensus is suboptimal to an individual client due to data domain shift in MRI data caused by acquisition protocols, recruitment criteria, etc.
    In this thesis, we propose a federated conditioned mutual learning (FedCM) to improve the performance by knowledge selection considering the clients’ local recognition ability and the data statistical distribution similarity between clients. This work is the first federated learning on multi-dataset Alzheimer’s disease classification by 3DCNN using T1-weighted structural MRI. Our method achieves promising recognition rates on real-world data and scenarios. Further visualization and relevance ranking on the region of interests (ROI) in human brains implies that our method can make the model mainly focus on the AD-specific brain region. Several potential regions are listed for future investigation.

    Acknowledgements 摘要 i Abstract ii 1 Introduction 1 1.1 Background 1 1.2 Motivation 4 1.3 Organization of the Thesis 4 2 Database and features 7 2.1 Database 8 2.1.1 ADNI 8 2.1.2 AIBL 8 2.1.3 OASIS 9 2.1.4 Re-­Labelling 9 2.1.5 Clients Split 10 2.2 Data Preprocessing 11 3 Preliminaries of Federated Learning 13 3.1 Federated Averaging 13 3.2 Inference Attack 15 3.3 Federated Distillation 15 3.3.1 Knowledge Distillation 16 3.3.2 Deep Mutual Learning 17 3.3.3 Federated Learning via Model Distillation 17 4 Federated Learning via Conditioned Mutual Learning 19 4.1 Research Methodology 19 4.1.1 Task Definition 19 4.1.2 Framework 20 4.1.3 Entropy Ratio Conditioning 22 4.1.4 Jensen­Shannon Divergence Conditioning 22 4.2 Experimental Setup and Results 24 4.2.1 Experimental Setup 24 4.2.2 Comparison Models 25 4.2.3 Results 26 5 Experiments with public dataset in different scenarios 31 5.1 Generative Adversarial Networks (GANs) 33 5.1.1 Info GAN 33 5.1.2 α­WGAN 35 5.1.3 Generated Images 36 5.2 Experimental Setup and Results 38 5.2.1 Experimental Setup 38 5.2.2 Comparison Models 38 5.2.3 Results 39 6 Analysis 41 6.1 Client­wise analysis 42 6.2 Relevance analysis 50 7 Conclusions 53 References 55

    [1] A. Bakkour, J. C. Morris, D. A. Wolk, and B. C. Dickerson, “The effects of aging and alzheimer’s disease on cerebral cortical anatomy: specificity and differential relationships with cognition,” Neuroimage, vol. 76, pp. 332–344, 2013.

    [2] A. Association et al., “2018 alzheimer’s disease facts and figures,” Alzheimer’s & Demen­tia, vol. 14, no. 3, pp. 367–429, 2018.

    [3] B. J. Kelley and R. C. Petersen, “Alzheimer’s disease and mild cognitive impairment,”
    Neurologic clinics, vol. 25, no. 3, pp. 577–609, 2007.

    [4] C. Andrade and R. Radhakrishnan, “The prevention and treatment of cognitive decline and dementia: An overview of recent research on experimental treatments,” Indian journal of psychiatry, vol. 51, no. 1, p. 12, 2009.

    [5] J. R. Cockrell and M. F. Folstein, “Mini­-mental state examination,” Principles and prac­tice of geriatric psychiatry, pp. 140–141, 2002.

    [6] J. C. Morris, “Clinical dementia rating: a reliable and valid diagnostic and staging mea­sure for dementia of the alzheimer type,” International psychogeriatrics, vol. 9, no. S1, pp. 173–176, 1997.

    [7] I. Arevalo­Rodriguez, N. Smailagic, M. R. i Figuls, A. Ciapponi, E. Sanchez­Perez, A. Gi­ annakou, O. L. Pedraza, X. B. Cosp, and S. Cullum, “Mini-­mental state examination (mmse) for the detection of alzheimer’s disease and other dementias in people with mild cognitive impairment (mci),” Cochrane Database of Systematic Reviews, no. 3, 2015.

    [8] M. N. Samtani, N. Raghavan, G. Novak, P. Nandy, and V. A. Narayan, “Disease progres­sion model for clinical dementia rating–sum of boxes in mild cognitive impairment and alzheimer＇s subjects from the alzheimer＇s disease neuroimaging initiative,” Neuropsy­ chiatric disease and treatment, vol. 10, p. 929, 2014.

    [9] G. Frisoni, C. Testa, A. Zorzan, F. Sabattoli, A. Beltramello, H. Soininen, and M. Laakso, “Detection of grey matter loss in mild alzheimer’s disease with voxel based morphometry,” Journal of Neurology, Neurosurgery & Psychiatry, vol. 73, no. 6, pp. 657–664, 2002.

    [10] P. Vemuri and C. R. Jack, “Role of structural mri in alzheimer’s disease,” Alzheimer’s research & therapy, vol. 2, no. 4, pp. 1–10, 2010.

    [11] K. A. Johnson, N. C. Fox, R. A. Sperling, and W. E. Klunk, “Brain imaging in alzheimer disease,” Cold Spring Harbor perspectives in medicine, vol. 2, no. 4, p. a006213, 2012.

    [12] T. Jo, K. Nho, and A. J. Saykin, “Deep learning in alzheimer’s disease: diagnostic clas­sification and prognostic prediction using neuroimaging data,” Frontiers in aging neuro­ science, vol. 11, p. 220, 2019.
    [13] N. Rieke, J. Hancox, W. Li, F. Milletari, H. R. Roth, S. Albarqouni, S. Bakas, M. N. Galtier, B. A. Landman, K. Maier­Hein, et al., “The future of digital health with federated learning,” NPJ digital medicine, vol. 3, no. 1, pp. 1–7, 2020.
    [14] G. Folego, M. Weiler, R. F. Casseb, R. Pires, and A. Rocha, “Alzheimer’s disease detection through whole-­brain 3d­cnn mri,” Frontiers in bioengineering and biotechnology, vol. 8, 2020.
    [15] J. Xu, B. S. Glicksberg, C. Su, P. Walker, J. Bian, and F. Wang, “Federated learning for healthcare informatics,” Journal of Healthcare Informatics Research, vol. 5, no. 1, pp. 1– 19, 2021.
    [16] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication­ efficient learning of deep networks from decentralized data,” in Artificial Intelligence and Statistics, pp. 1273–1282, PMLR, 2017.
    [17] M. J. Sheller, G. A. Reina, B. Edwards, J. Martin, and S. Bakas, “Multi­-institutional deep learning modeling without sharing patient data: A feasibility study on brain tumor seg­ mentation,” in International MICCAI Brain-lesion Workshop, pp. 92–104, Springer, 2018.
    [18] X. Li, Y. Gu, N. Dvornek, L. H. Staib, P. Ventola, and J. S. Duncan, “Multi­site fmri anal­ysis using privacy­-preserving federated learning and domain adaptation: Abide results,” Medical Image Analysis, vol. 65, p. 101765, 2020.
    [19] E. Kondrateva, M. Pominova, E. Popova, M. Sharaev, A. Bernstein, and E. Burnaev, “Do­ main shift in computer vision models for mri data analysis: an overview,” in Thirteenth International Conference on Machine Vision, vol. 11605, p. 116050H, International Soci­ety for Optics and Photonics, 2021.
    [20] X. Peng, Z. Huang, Y. Zhu, and K. Saenko, “Federated adversarial domain adaptation,”
    arXiv preprint arXiv:1911.02054, 2019.

    [21] Z. Wang, M. Song, Z. Zhang, Y. Song, Q. Wang, and H. Qi, “Beyond inferring class representatives: User­-level privacy leakage from federated learning,” in IEEE INFOCOM 2019­IEEE Conference on Computer Communications, pp. 2512–2520, IEEE, 2019.
    [22] P. Paysan, M. Lüthi, T. Albrecht, A. Lerch, B. Amberg, F. Santini, and T. Vetter, “Face reconstruction from skull shapes and physical attributes,” in Joint Pattern Recognition Symposium, pp. 232–241, Springer, 2009.
    [23] A. Kermi, S. Marniche-­Kermi, and M. T. Laskri, “3d­computerized facial reconstructions from 3d­mri of human heads using deformable model approach,” in 2010 International Conference on Machine and Web Intelligence, pp. 276–282, IEEE, 2010.
    [24] A. Nestor, D. C. Plaut, and M. Behrmann, “Feature-­based face representations and image reconstruction from behavioral and neural data,” Proceedings of the National Academy of Sciences, vol. 113, no. 2, pp. 416–421, 2016.

    [25] C. G. Schwarz, W. K. Kremers, T. M. Therneau, R. R. Sharp, J. L. Gunter, P. Vemuri, A. Arani, A. J. Spychalla, K. Kantarci, D. S. Knopman, et al., “Identification of anony­mous mri research participants with face­re-cognition software,” New England Journal of Medicine, vol. 381, no. 17, pp. 1684–1686, 2019.
    [26] D. Li and J. Wang, “Fedmd: Heterogenous federated learning via model distillation,” arXiv preprint arXiv:1910.03581, 2019.

    [27] T. Furlanello, Z. Lipton, M. Tschannen, L. Itti, and A. Anandkumar, “Born again neural networks,” in International Conference on Machine Learning, pp. 1607–1616, PMLR, 2018.
    [28] C. R. Jack Jr, M. A. Bernstein, N. C. Fox, P. Thompson, G. Alexander, D. Harvey,
    B. Borowski, P. J. Britson, J. L. Whitwell, C. Ward, et al., “The alzheimer’s disease neu­roimaging initiative (adni): Mri methods,” Journal of Magnetic Resonance Imaging: An Official Journal of the International Society for Magnetic Resonance in Medicine, vol. 27, no. 4, pp. 685–691, 2008.
    [29] K. Ellis, A. Bush, D. Darby, D. De Fazio, J. Foster, P. Hudson, N. Lautenschlager,
    N. Lenzo, R. Martins, P. Maruff, et al., “The australian imaging, biomarkers and lifestyle (aibl) study of aging: methodology and baseline characteristics of 1112 individuals re­cruited for a longitudinal study of alzheimer’s disease,” 2009.
    [30] D. S. Marcus, T. H. Wang, J. Parker, J. G. Csernansky, J. C. Morris, and R. L. Buckner, “Open access series of imaging studies (oasis): cross­-sectional mri data in young, mid­dle aged, nondemented, and demented older adults,” Journal of cognitive neuroscience, vol. 19, no. 9, pp. 1498–1507, 2007.
    [31] J. Samper-­González, N. Burgos, S. Bottani, S. Fontanella, P. Lu, A. Marcoux, A. Routier,
    J. Guillon, M. Bacci, J. Wen, et al., “Reproducible evaluation of classification methods in alzheimer’s disease: Framework and application to mri and pet data,” NeuroImage, vol. 183, pp. 504–521, 2018.
    [32] A. Routier, N. Burgos, J. Guillon, J. Samper-­González, J. Wen, S. Bottani, A. Marcoux,
    M. Bacci, S. Fontanella, T. Jacquemont, et al., “Clinica: an open source software platform for reproducible clinical neuroscience studies,” 2019.
    [33] J. Ashburner, “A fast diffeomorphic image registration algorithm,” Neuroimage, vol. 38, no. 1, pp. 95–113, 2007.
    [34] D. Enthoven and Z. Al­Ars, “An overview of federated deep learning privacy attacks and defensive strategies,” Federated Learning Systems, pp. 173–196, 2021.
    [35] M. Nasr, R. Shokri, and A. Houmansadr, “Comprehensive privacy analysis of deep learn­ing: Passive and active white-­box inference attacks against centralized and federated learning,” in 2019 IEEE symposium on security and privacy (SP), pp. 739–753, IEEE, 2019.
    [36] B. Hitaj, G. Ateniese, and F. Perez-­Cruz, “Deep models under the gan: information leak­ age from collaborative deep learning,” in Proceedings of the 2017 ACM SIGSAC Confer­ence on Computer and Communications Security, pp. 603–618, 2017.

    [37] G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” arXiv preprint arXiv:1503.02531, 2015.

    [38] Y. Tian, D. Krishnan, and P. Isola, “Contrastive representation distillation,” arXiv preprint arXiv:1910.10699, 2019.

    [39] T. Kim, J. Oh, N. Kim, S. Cho, and S.­Y. Yun, “Comparing kullback­leibler divergence and mean squared error loss in knowledge distillation,” arXiv preprint arXiv:2105.08919, 2021.

    [40] L. Gao, H. Mi, B. Zhu, D. Feng, Y. Li, and Y. Peng, “An adversarial feature distillation method for audio classification,” IEEE Access, vol. 7, pp. 105319–105330, 2019.

    [41] Y. Zhang, T. Xiang, T. M. Hospedales, and H. Lu, “Deep mutual learning,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4320–4328, 2018.

    [42] A. Krizhevsky, G. Hinton, et al., “Learning multiple layers of features from tiny images,” 2009.

    [43] X. Chen, Y. Duan, R. Houthooft, J. Schulman, I. Sutskever, and P. Abbeel, “Infogan: Interpretable representation learning by information maximizing generative adversarial nets,” in Proceedings of the 30th International Conference on Neural Information Pro­cessing Systems, pp. 2180–2188, 2 016.

    [44] G. Kwon, C. Han, and D.­s. Kim, “Generation of 3d brain mri using auto­encoding gen­erative adversarial networks,” in International Conference on Medical Image Computing and Computer­Assisted Intervention, pp. 118–126, Springer, 2019.

    [45] K. Simonyan, A. Vedaldi, and A. Zisserman, “Deep inside convolutional networks: Visu­alising image classification models and saliency maps,” arXiv preprint arXiv:1312.6034, 2013.

    [46] J. Rieke, F. Eitel, M. Weygandt, J.­D. Haynes, and K. Ritter, “Visualizing convolutional networks for mri-­based diagnosis of alzheimer＇s disease,” in Understanding and Interpreting Machine Learning in Medical Image Computing Applications, pp. 24–31, Springer, 2018.

    [47] N. Nagaratnam, T. A. Phan, C. Barnett, and N. Ibrahim, “Angular gyrus syndrome mim­icking depressive pseudodementia,” Journal of Psychiatry and Neuroscience, vol. 27, no. 5, p. 364, 2002.

    [48] E. Rodríguez-­Cano, S. Sarró, G. Monté, T. Maristany, R. Salvador, P. McKenna, and
    E. Pomarol­-Clotet, “Evidence for structural and functional abnormality in the subgenual anterior cingulate cortex in major depressive disorder,” Psychological Medicine, vol. 44, no. 15, pp. 3263–3273, 2014.

    [49] C. Hamani, H. Mayberg, S. Stone, A. Laxton, S. Haber, and A. M. Lozano, “The subcal­losal cingulate gyrus in the context of major depression,” Biological psychiatry, vol. 69, no. 4, pp. 301–308, 2011.

    [50] R. Rodrigues, R. B. Petersen, and G. Perry, “Parallels between major depressive disorder and alzheimer＇s disease: role of oxidative stress and genetic vulnerability,” Cellular and molecular neurobiology, vol. 34, no. 7, pp. 925–949, 2014.
    [51] Y. Xian, T. Lorenz, B. Schiele, and Z. Akata, “Feature generating networks for zero-­shot learning,” in Proceedings of the IEEE conference on computer vision and pattern recog­nition, pp. 5542–5551, 2018.
    [52] G. K. Nayak, K. R. Mopuri, V. Shaj, V. B. Radhakrishnan, and A. Chakraborty, “Zero­ shot knowledge distillation in deep networks,” in International Conference on Machine Learning, pp. 4743–4751, PMLR, 2019.

    QR CODE