基于深度元学习的医学图像分类研究
Research on Medical Image Classification Based on Deep Meta-Learning
DOI: 10.12677/CSA.2023.1311211, PDF,   
作者: 董大山, 张仲荣*：兰州交通大学数理学院，甘肃 兰州；张其斌*：甘肃省科技厅高新技术创业服务中心，甘肃 兰州
关键词: 医学图像分类；元学习；小样本学习；迁移学习；Medical Image Classification； Meta-Learning； Few-Shot Learning； Transfer Learning
摘要: 本文针对医学图像分类问题，提出了一种基于深度元学习的多阶段小样本学习方法，该方法结合了元学习、迁移学习的思想。通过引入预训练的视觉变压器模型(vision transformer)作为特征提取模型，提升对医学图像特征的提取能力。使用额外的数据集对原型网络进行训练，以克服医疗图像数据量不足的问题。我们对于医学图像分类任务进行微调，以提高模型的针对性，使模型更易进行适应医学图像任务。我们在两个医学图像数据集(血液、病理学)上进行了实验，并与相关工作进行了比较。实验结果表明，我们的方法在血液数据集上3way 1-shot，5-shot，10-shot准确率分别为68.06%，91.55%，95.3%，在病理数据集上3way 1-shot，5-shot，10-shot准确率分别为78.50%，91.84%，94.93%，取得了领先的性能，具有可靠的识别率。
Abstract: This paper focuses on the problem of medical image classification and proposes a multi-stage few-shot learning method based on deep meta-learning, combining the ideas of meta-learning and transfer learning. By introducing a pre-trained vision transformer model as the feature extraction model, the capability of extracting features from medical images is enhanced. Additional datasets are used to train the prototype network to overcome the problem of insufficient medical image data. We finetune the model for medical image classification tasks to improve its specificity and make it more adaptable to medical image tasks. We conducted experiments on two medical image datasets (hematology and pathology) and compared the results with related works. The experimental results demonstrate that our method achieves leading performance with reliable recognition rates. For the hematology dataset, the accuracy rates of 3way 1-shot, 5-shot, and 10-shot are 68.06%, 91.55%, and 95.3% respectively, while for the pathology dataset, the accuracy rates of 3way 1-shot, 5-shot, and 10-shot are 78.50%, 91.84%, and 94.93% respectively.
文章引用：董大山, 张仲荣, 张其斌. 基于深度元学习的医学图像分类研究[J]. 计算机科学与应用, 2023, 13(11): 2116-2124. https://doi.org/10.12677/CSA.2023.1311211

参考文献

[1] LeCun, Y., Bengio, Y. and Hinton, G. (2015) Deep Learning. Nature, 521, 436-444. [Google Scholar] [CrossRef] [PubMed]
[2] He, K., Zhang, X., Ren, S., et al. (2016) Deep Residual Learning for Im-age Recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, 27-30 June 2016, 770-778. [Google Scholar] [CrossRef] 
[3] Antonie, M.L., Zaiane, O.R. and Coman, A. (2001) Application of Data Mining Techniques for Medical Image Classification. Proceedings of the 2nd International Conference on Multime-dia Data Mining, San Francisco, 26 August 2001, 94-101.
[4] Li, Q., Cai, W., Wang, X., et al. (2014) Medical Image Classification with Convolutional Neural Network. 2014 13th IEEE International Conference on Control Automation Robotics & Vision (ICARCV), Singapore, 10-12 December 2014, 844-848. [Google Scholar] [CrossRef] 
[5] Gondara, L. (2016) Medical Image Denoising Using Convo-lutional Denoising Autoencoders. 2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW), Barcelona, 12-15 December 2016, 241-246. [Google Scholar] [CrossRef] 
[6] Hu, S., Tomczak, J. and Welling, M. (2018) Meta-Learning for Medical Image Classification.
[7] Yu, Y., Lin, H., Meng, J., et al. (2017) Deep Transfer Learning for Modality Classi-fication of Medical Images. Information, 8, Article No. 91. [Google Scholar] [CrossRef] 
[8] Gyawali, P.K., Ghimire, S., Bajracharya, P., et al. (2020) Semi-Supervised Medical Image Classification with Global Latent Mixing. Medical Image Computing and Computer Assisted Intervention-MICCAI 2020: 23rd International Conference, Lima, 4-8 October 2020, 604-613. [Google Scholar] [CrossRef] 
[9] Beluch, W.H., Genewein, T., Nürnberger, A., et al. (2018) The Power of Ensembles for Active Learning in Image Classification. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, 18-22 June 2018, 9368-9377. [Google Scholar] [CrossRef] 
[10] Dosovitskiy, A., Beyer, L., Kolesnikov, A., et al. (2020) An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale.
[11] Vaswani, A., Shazeer, N., Parmar, N., et al. (2017) Attention Is All You Need. 31st Annual Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, 4-9 December 2017, 5998-6008.
[12] Snell, J., Swersky, K. and Zemel, R. (2017) Prototypical Networks for Few-Shot Learning. 31st Annual Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, 4-9 December 2017, 4077-4087.
[13] LaValley, M.P. (2008) Logistic Regression. Circulation, 117, 2395-2399. [Google Scholar] [CrossRef] 
[14] Pei, S.C. and Lin, C.N. (1995) Image Normaliza-tion for Pattern Recognition. Image and Vision Computing, 13, 711-723. [Google Scholar] [CrossRef] 
[15] Caron, M., Touvron, H., Misra, I., et al. (2021) Emerging Properties in Self-Supervised Vision Transformers. Proceedings of the IEEE/CVF International Conference on Comput-er Vision, 11-17 October 2021, 9650-9660. [Google Scholar] [CrossRef] 
[16] Acevedo, A., et al. (2020) A Dataset of Microscopic Periph-eral Blood Cell Images for Development of Automatic Recognition Systems. Data in Brief, 30, Article ID: 105474. [Google Scholar] [CrossRef] [PubMed]
[17] Kather, J.N., Krisam, J., Charoentong, P., Luedde, T., Herpel, E., Weis, C.-A., Gaiser, T., Marx, A., Valous, N.A., Ferber, D., et al. (2019) Predicting Survival from Colorectal Cancer Histology Slides Using Deep Learning: A Retrospective Multicenter Study. PLOS Medicine, 16, e1002730. [Google Scholar] [CrossRef] [PubMed]
[18] Zheng, Q., Tian, X., Yang, M., et al. (2020) PAC-Bayesian Framework Based Drop-Path Method for 2D Discriminative Convolutional Network Pruning. Multidimensional Systems and Signal Processing, 31, 793-827. [Google Scholar] [CrossRef] 
[19] Zhuang, J., Tang, T., Ding, Y., et al. (2020) Adabelief Optimizer: Adapting Stepsizes by the Belief in Observed Gradients. 34th International Conference on Neural Information Pro-cessing Systems, Vancouver, 6-12 December 2020, 18795-18806.
[20] Kingma, D.P. and Ba, J. (2014) Adam: A Method for Stochastic Optimization.
[21] Finn, C., Abbeel, P. and Levine, S. (2017) Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks. International Conference on Machine Learning, Sydney, 6-11 August 2017, 1126-1135.
[22] Sung, F., Yang, Y., Zhang, L., Xiang, T., Torr, P.H. and Hospedales, T.M. (2018) Learning to Com-pare: Relation Network for Few-Shot Learning. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, 18-22 June 2018, 1199-1208. [Google Scholar] [CrossRef] 
[23] Gordon, J., Bronskill, J., Bauer, M., Nowozin, S. and Turner, R.E. (2018) VERSA: Versatile and Efficient Few-Shot Learning. 3rd Workshop on Bayesian Deep Learning, Montréal, 7 December 2018, 1-9.
[24] Devos, A., Chatel, S. and Grossglauser, M. (2019) [Re]meta Learning with Differentiable Closed-Form Solvers. Tech. Rep.
[25] Raghu, A., Raghu, M., Bengio, S. and Vinyals, O. (2019) Rapid Learning or Feature Reuse? Towards Understanding the Effective-ness of MAML.
[26] Sun, Q., Liu, Y., Chen, Z., Chua, T.-S. and Schiele, B. (2020) Meta-Transfer Learning through Hard Tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44, 1443-1456.
[27] Vinyals, O., Blun-dell, C., Lillicrap, T., Wierstra, D., et al. (2016) Matching Networks for One Shot Learning. NIPS’16: Proceedings of the 30th International Conference on Neural Information Processing Systems, Barcelona, 5-10 December 2016, 3630-3638.
[28] Antoniou, A., Edwards, H. and Storkey, A. (2018) How to Train Your MAML.
[29] Chen, Y., Wang, X., Liu, Z., Xu, H. and Darrell, T. (2020) A New Meta-Baseline for Few-Shot Learning.
[30] Rusu, A.A., Rao, D., Sygnowski, J., Vinyals, O., Pascanu, R., Osindero, S. and Hadsell, R. (2018) Meta-Learning with Latent Embedding Optimization.
[31] Yu, Z., Chen, L., Cheng, Z. and Luo, J. (2020) Transmatch: A Transfer-Learning Scheme for Semi-Supervised Few-Shot Learning. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, 13-19 June 2020, 12856-12864. [Google Scholar] [CrossRef]