FEW-SHOT MEDICAL IMAGE CLASSIFICATION VIA META-LEARNING

Medical images are generally used to represent the interior structure of the human body and play a very essential part in illness diagnosis and therapy. Medical image classification and reconstruction are major research fields in medical image analysis. How to increase the accuracy and efficiency of medical image classification is an ongoing research area. Deep learning-based algorithms have demonstrated significant benefits in image classification in recent years. In general, the availability of an extensive dataset is a necessary condition for such state-of-the-art approaches to perform successfully. However, when it comes to classifying electroencephalogram (EEG) signals, obtaining a large amount of data is impracticable owing to issues of safety, protection, and privacy. Meta-learning algorithms can bring fresh ideas in the challenge of few-shot EEG signal classification without relying on large datasets. The concept of meta-learning is to gather previous knowledge by training on other relevant data, and then fine-tune the training model on actual samples. This is a method that is capable of rapidly adapting to new samples depending on prior experience. Meta-learning can be considered as a bi-level programming problem due to its two-stage optimization process.

In this thesis, we begin by preprocessing the raw EEG data and then classifying it using bi-level programming techniques (i.e., meta-learning algorithms). Furthermore, we propose a new three-level programming model on the basis of the bi-level programming by introducing a feasible set of the lower-level variables. Due to the presence of the third-level constraint, the previous meta-learning approaches become inapplicable. Inspired by the BDA and ADMM algorithms, we present a new algorithm and its convergence analysis. Following that, we evaluate and contrast meta-learning approaches to the more traditional ones. The experimental results demonstrate the meta-learning method's success in training EEG prediction models, which is critical for the development of non-invasive Brain-Machine Interfaces (BMIs).

Reconstruction of medical images is also a hot topic in the medical industry. The essence of the image reconstruction problem is to solve an inverse problem, that is, to rebuild the original signal from the acquired signal or observation data. Because this requires denoising, model building, and other challenges, it is not a simple or straightforward problem to solve. To reconstruct the original signal of medical images, we developed a bi-level programming model inspired by the elastic net and l_p-l_1-2 model. The top objective function is a mixture of the l₁ and l₂ norms; the lower objective function is utilized to deal with noise pollution. This bi-level programming problem is difficult to solve due to the non-smoothness of the upper layer objective. As a result, we first leverage the properties of the bottom layer objective function to convert the bi-level programming model to a single-layer problem under certain prior information, and then solve this optimization problem using the alternating direction method of multipliers (ADMM). Numerical results demonstrate that this model and the associated algorithm are both practical and successful for reconstructing medical images that have been affected by various noise sources.

[1] BERGER H. On the electroencephalogram of man[J]. Electroencephalography and Clinical Neurophysiology, 1969, 28(Suppl.): 37-74.
[2] ABOU-KHALIL B, MISULIS K E. Atlas of eeg & seizure semiology[M]. Butterworth-Heinemann, 2006.
[3] CHERNECKY C C, BERGER B J. Laboratory tests and diagnostic procedures[M]. Elsevier Health Sciences, 2012.
[4] ZHANG H, WEI C, ZHAO M, et al. A novel convolutional neural network model to remove muscle artifacts from eeg[C]//ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021: 1265-1269.
[5] SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[J]. Advances in Neural Information Processing Systems, 2017, 30: 4078-4088.
[6] GLOWINSKI R, MARROCCO A. On the solution of a class of non linear dirichlet problems by a penalty-duality method and finite elements of order one[C]//Optimization Techniques IFIP Technical Conference. Springer, 1975: 327-333.
[7] GABAY D, MERCIER B. A dual algorithm for the solution of nonlinear variational problems via finite element approximation[J]. Computers & Mathematics with Applications, 1976, 2(1):17-40.
[8] BOYD S, PARIKH N, CHU E, et al. Distributed optimization and statistical learning via the alternating direction method of multipliers[J]. Foundations and Trends® in Machine learning, 2011, 3(1): 1-122.
[9] HE B, YUAN X. On the o(1/n) convergence rate of the douglas–rachford alternating direction method[J]. SIAM Journal on Numerical Analysis, 2012, 50(2): 700-709.
[10] ANG K K, CHIN Z Y, WANG C, et al. Filter bank common spatial pattern algorithm on bci competition iv datasets 2a and 2b[J]. Frontiers in Neuroscience, 2012, 6: 39.
[11] MOUSAVI E A, MALLER J J, FITZGERALD P B, et al. Wavelet common spatial pattern in asynchronous offline brain computer interfaces[J]. Biomedical Signal Processing and Control, 2011, 6(2): 121-128.
[12] TALUKDAR M T F, SAKIB S K, PATHAN N S, et al. Motor imagery eeg signal classification scheme based on autoregressive reflection coefficients[C]//2014 International Conference on Informatics, Electronics & Vision (ICIEV). IEEE, 2014: 1-4.
[13] AN X, KUANG D, GUO X, et al. A deep learning method for classification of eeg data based on motor imagery[C]//International Conference on Intelligent Computing. Springer, 2014: 203-210.
[14] TANG Z, SUN S, ZHANG S, et al. A brain-machine interface based on erd/ers for an upper-limb exoskeleton control[J]. Sensors, 2016, 16(12): 2050.
[15] REN Z, LI R, CHEN B, et al. Eeg-based driving fatigue detection using a two-level learning hierarchy radial basis function[J]. Frontiers in Neurorobotics, 2021, 15: 618408.
[16] AGGARWAL K, KHADANGA S, JOTY S, et al. A structured learning approach with neural conditional random fields for sleep staging[C]//2018 IEEE International Conference on Big Data (Big Data). IEEE, 2018: 1318-1327.
[17] BHOSALE S, CHAKRABORTY R, KOPPARAPU S K. Calibration free meta learning based approach for subject independent eeg emotion recognition[J]. Biomedical Signal Processing and Control, 2022, 72: 103289.
[18] MIYAMOTO K, TANAKA H, NAKAMURA S. Meta-learning for emotion prediction from eeg while listening to music[C]//Companion Publication of the 2021 International Conference on Multimodal Interaction. 2021: 324-328.
[19] BANLUESOMBATKUL N, OUPPAPHAN P, LEELAARPORN P, et al. Metasleeplearner: A pilot study on fast adaptation of bio-signals-based sleep stage classifier to new individual subject using meta-learning[J]. IEEE Journal of Biomedical and Health Informatics, 2020, 25(6): 1949-1963.
[20] CHOI S. Meta-learning: Towards fast adaptation in multi-subject eeg classification[C]//2021 9th International Winter Conference on Brain-Computer Interface (BCI). IEEE, 2021: 1-1.
[21] LI D, ORTEGA P, WEI X, et al. Model-agnostic meta-learning for eeg motor imagery decoding in brain-computer-interfacing[C]//2021 10th International IEEE/EMBS Conference on Neural Engineering (NER). IEEE, 2021: 527-530.
[22] VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[J]. Advances in Neural Information Processing Systems, 2016: 3630–3638.
[23] SUNG F, YANG Y, ZHANG L, et al. Learning to compare: Relation network for few-shot learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 1199-1208.
[24] XIAN Y, LAMPERT C H, SCHIELE B, et al. Zero-shot learning—a comprehensive evaluation of the good, the bad and the ugly[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 41(9): 2251-2265.
[25] WEISS K, KHOSHGOFTAAR T M, WANG D. A survey of transfer learning[J]. Journal of Big Data, 2016, 3(1): 1-40.
[26] BEN-DAVID S, BLITZER J, CRAMMER K, et al. Analysis of representations for domain adaptation[J]. Advances in neural information processing systems, 2006: 137–144.
[27] LIU B, WANG X, DIXIT M, et al. Feature space transfer for data augmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 9090-9098.
[28] ZHANG H, CISSE M, DAUPHIN Y N, et al. mixup: Beyond empirical risk minimization[C]//International Conference on Learning Representations. 2018.
[29] SHORTEN C, KHOSHGOFTAAR T M. A survey on image data augmentation for deep learning[J]. Journal of Big Data, 2019, 6(1): 1-48.
[30] LAKE B M, SALAKHUTDINOV R, TENENBAUM J B. Human-level concept learningthrough probabilistic program induction[J]. Science, 2015, 350(6266): 1332-1338.
[31] CUBUK E D, ZOPH B, MANE D, et al. Autoaugment: Learning augmentation policies from data[C]//In Proceedings of the Conference on Computer Vision and Pattern Recognition. 2019:113–123.
[32] DING Y, ZHANG H, LI P, et al. An efficient semismooth newton method for adaptive sparse signal recovery problems[J]. arXiv:2111.12252, 2021.
[33] LIN M, SUN D, TOH K C. Efficient algorithms for multivariate shape-constrained convex regression problems[J]. arXiv:2002.11410, 2020.
[34] DENG L. The mnist database of handwritten digit images for machine learning research[J]. IEEE Signal Processing Magazine, 2012, 29(6): 141-142.
[35] NEWSON J J, THIAGARAJAN T C. Eeg frequency bands in psychiatric disorders: a review of resting state studies[J]. Frontiers in Human Neuroscience, 2019, 12: 521.
[36] MACLAURIN D, DUVENAUD D, ADAMS R. Gradient-based hyperparameter optimization through reversible learning[C]//International conference on machine learning. PMLR, 2015: 2113-2122.
[37] DEMPE S, ZEMKOHO A. Bilevel optimization[M]. Springer, 2020.
[38] FINN C, ABBEEL P, LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]//International Conference on Machine Learning. PMLR, 2017: 1126-1135.
[39] NICHOL A, ACHIAM J, SCHULMAN J. On first-order meta-learning algorithms[J].arXiv:1803.02999, 2018.
[40] FRANCESCHI L, DONINI M, FRASCONI P, et al. Forward and reverse gradient-based hyperparameter optimization[C]//International Conference on Machine Learning. PMLR, 2017: 1165-1173.
[41] FRANCESCHI L, FRASCONI P, SALZO S, et al. Bilevel programming for hyperparameter optimization and meta-learning[C]//International Conference on Machine Learning. PMLR, 2018: 1568-1577.
[42] LIU R, MU P, YUAN X, et al. A general descent aggregation framework for gradient-based bilevel optimization[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022.
[43] LIU R, MU P, YUAN X, et al. A generic first-order algorithmic framework for bi-level programming beyond lower-level singleton[C]//International Conference on Machine Learning. PMLR, 2020: 6305-6315.
[44] LIU R, MA L, YUAN X, et al. Task-oriented convex bilevel optimization with latent feasibility[J]. IEEE Transactions on Image Processing, 2022.
[45] FAZEL M, PONG T K, SUN D, et al. Hankel matrix rank minimization with applications to system identification and realization[J]. SIAM Journal on Matrix Analysis and Applications, 2013, 34(3): 946-977.
[46] XU M, WU T. A class of linearized proximal alternating direction methods[J]. Journal of Optimization Theory and Applications, 2011, 151(2): 321-337.
[47] LIU R, LIU X, ZENG S, et al. Optimization-derived learning with essential convergence analysis of training and hyper-training[J]. preprint, 2022.
[48] BAUSCHKE H H, COMBETTES P L, et al. Convex analysis and monotone operator theory in hilbert spaces[M]. Springer, 2011.
[49] CABOT A. Proximal point algorithm controlled by a slowly vanishing term: applications to hierarchical minimization[J]. SIAM Journal on Optimization, 2005, 15(2): 555-572.
[50] ZOU H, HASTIE T. Regularization and variable selection via the elastic net[J]. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 2005, 67(2): 301-320.
[51] SABACH S, SHTERN S. A first order method for solving convex bilevel optimization problems[J]. SIAM Journal on Optimization, 2017, 27(2): 640-660.
[52] SOLODOV M. An explicit descent method for bilevel convex optimization[J]. Journal of Convex Analysis, 2007, 14(2): 227.
[53] TIBSHIRANI R. Regression shrinkage and selection via the lasso[J]. Journal of the Royal Statistical Society: Series B (Methodological), 1996, 58(1): 267-288.
[54] CANDÈS E J, WAKIN M B. An introduction to compressive sampling[J]. IEEE Signal Processing Magazine, 2008, 25(2): 21-30.
[55] YIN W. Analysis and generalizations of the linearized bregman method[J]. SIAM Journal on Imaging Sciences, 2010, 3(4): 856-877.
[56] AMINI F, HU G. A two-layer feature selection method using genetic algorithm and elastic net[J]. Expert Systems with Applications, 2021, 166: 114072.
[57] TEIPEL S J, GROTHE M J, METZGER C D, et al. Robust detection of impaired resting state functional connectivity networks in alzheimer’s disease using elastic net regularized regression[J]. Frontiers in Aging Neuroscience, 2017, 8: 318.
[58] DEMIRER M, DIEBOLD F X, LIU L, et al. Estimating global bank network connectedness[J]. Journal of Applied Econometrics, 2018, 33(1): 1-15.
[59] BUNEA F, SHE Y, OMBAO H, et al. Penalized least squares regression methods and applications to neuroimaging[J]. NeuroImage, 2011, 55(4): 1519-1527.
[60] CHO S, KIM H, OH S, et al. Elastic-net regularization approaches for genome-wide association studies of rheumatoid arthritis[C]//BMC Proceedings. BioMed Central, 2009: 1-6.
[61] HE B. Modified alternating directions method of multipliers for convex optimization with three separable functions[J]. Operations Research Transactions, 2015, 19: 57-70.
[62] CHEN C, HE B, YE Y, et al. The direct extension of admm for multi-block convex minimization problems is not necessarily convergent[J]. Mathematical Programming, 2016, 155(1): 57-79.
[63] BIRD J J, FARIA D R, MANSO L J, et al. A deep evolutionary approach to bioinspired classifier optimisation for brain-machine interaction[J]. Complexity, 2019: 1-14.
[64] JOLLY B L K, AGGRAWAL P, NATH S S, et al. Universal eeg encoder for learning diverse intelligent tasks[C]//2019 IEEE Fifth International Conference on Multimedia Big Data (BigMM). 2019: 213-218.
[65] 刘浩洋, 户将, 李勇锋, 等. 最优化：建模、算法与理论[M]. 高等教育出版社, 2020.