基于少标注样本的眼底彩照异常检测算法

糖尿病视网膜病变（DR）是全球导致失明的主要疾病之一，因此早期检测至关重要。眼底彩照是一种非侵入性的检测方法，广泛应用于DR的早期检测和治疗。然而，由于DR眼底彩照标注的高成本和稀缺性，基于少标注样本的眼底彩照异常检测成为一个极具研究意义且富有挑战性的任务。

鉴于现有的异常检测工作难以处理复杂数据中的异常，本文提出了一个名为Lesion2Void的无监督异常检测框架。该框架使用随机擦除眼底彩照图像块并重建完整图像的方法来识别图像的异常区域。Lesion2Void假设病变图像块是相互独立的，与它们的邻近像素无关，而正常图像块可以根据邻近像素的信息来预测。因此，在测试阶段，通过测量被擦除的图像块的重建误差，就可以将图像识别为正常或异常。实验结果表明，Lesion2Void在公开数据集上的DR眼底彩照异常检测方面表现优越。

针对无监督异常检测存在没有真实异常样本的训练以及缺乏真实异常先验知识的问题，本文中提出了一种名为AugPaste的单标注样本眼底彩照异常检测框架，用于检测眼底图像中的DR病灶。该框架利用单个标注DR样本中真实DR病灶进行增强，并合成人工DR样本用于异常检测。具体而言，该框架通过对随机选择的DR病变区域进行增强来构建DR病灶库。随后，计算显著性图以确保DR病灶不被粘贴在视盘、视杯和血管上。采用MixUp技术将病变区域粘贴在高斯分布采样得到的正常图像位置上，以合成DR样本用于训练检测网络。本文在四个公开的眼底彩照数据集上进行了广泛的实验，结果表明AugPaste方法明显优于几个最先进的无监督和小样本异常检测方法，甚至接近全监督方法的效果。

本文的工作具有重要的实际应用价值，可以帮助实现DR早期检测和治疗，同时为解决DR数据标注的高成本和稀缺性问题提供了新的解决方案。

Diabetic retinopathy (DR) is one of the leading causes of blindness globally, making early detection critical. Fundus image is a non-invasive screening method widely used for early detection of DR. However, due to the high cost and scarcity of annotated DR data, anomaly detection for fundus images with few labeled data remains a challenging task.

Given the difficulty of detecting anomalies in complex data, an unsupervised anomaly detection framework called Lesion2Void is proposed. The framework employs a method of randomly erasing fundus image patches and reconstructing the complete image to identify the abnormal regions. The method assumes that lesion patches are independent of each other and independent of their neighboring pixels, whereas normal patches can be predicted based on the information from the neighborhood. Therefore, in the testing phase, an image can be identified as normal or abnormal by measuring the reconstruction errors of the erased patches. Experiments are conducted on the public dataset, demonstrating the superiority of Lesion2Void for DR anomaly detection in fundus images.

To address the limitations of unsupervised anomaly detection, which lacks knowledge of real anomalies, a one-shot anomaly detection framework called AugPaste is proposed for detecting DR fundus images. The framework augments real DR lesions to synthesize artificial DR samples for training. Specifically, the framework constructs a DR lesion bank by augmenting randomly selected DR lesion patches. Then, a saliency map is computed to avoid pasting DR lesions on the optic disc/cup and vessels, and MixUp is used to paste lesion patches at positions sampled from a Gaussian distribution in normal images to synthesize DR samples to train the detection network. Extensive experiments are conducted on four publicly fundus image datasets indicates that AugPaste significantly outperforms several state-of-the-art unsupervised and few-shot anomaly detection methods, and is comparable to the fully-supervised method.

This work has important practical applications, can help achieve early detection and treatment of DR, and provides a new solution to the problem of high cost and scarcity of labeled DR data.

[1] TING D S W, CHEUNG G C M, WONG T Y. Diabetic retinopathy: global prevalence, major risk factors, screening practices and public health challenges: a review[J]. Clinical & experimental ophthalmology, 2016, 44(4): 260-277.
[2] LI T, BO W, HU C, et al. Applications of deep learning in fundus images: A review[J]. Medical Image Analysis, 2021, 69: 101971.
[3] EDUPUGANTI V G, CHAWLA A, KALE A. Automatic optic disk and cup segmentation of fundus images using deep learning[C]//2018 25th IEEE international conference on image processing (ICIP). IEEE, 2018: 2227-2231.
[4] ABRÀMOFF M D, GARVIN M K, SONKA M. Retinal imaging and image analysis[J]. IEEE reviews in biomedical engineering, 2010, 3: 169-208.
[5] HUANG Y, LIN L, CHENG P, et al. Lesion-based contrastive learning for diabetic retinopathy grading from fundus images[C]//Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part II 24. Springer, 2021: 113-123.
[6] YAMASHITA R, NISHIO M, DO R K G, et al. Convolutional neural networks: an overview and application in radiology[J]. Insights into imaging, 2018, 9: 611-629.
[7] TING D S W, CHEUNG C Y L, LIM G, et al. Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes[J]. Jama, 2017, 318(22): 2211-2223.
[8] PANG G, SHEN C, CAO L, et al. Deep learning for anomaly detection: A review[J]. ACM computing surveys (CSUR), 2021, 54(2): 1-38.
[9] CHALAPATHY R, CHAWLA S. Deep learning for anomaly detection: A survey[A]. 2019.
[10] CHEN B, WANG L, WANG X, et al. Abnormality detection in retinal image by individualized background learning[J]. Pattern Recognition, 2020, 102: 107209.
[11] DU Y, WANG L, MENG D, et al. Individualized Statistical Modeling of Lesions in Fundus Images for Anomaly Detection[J]. IEEE Transactions on Medical Imaging, 2022.
[12] BERGMANN P, FAUSER M, SATTLEGGER D, et al. MVTec AD–A comprehensive real-world dataset for unsupervised anomaly detection[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 9592-9600.
[13] GEORGESCU M I, BARBALAU A, IONESCU R T, et al. Anomaly detection in video via self-supervised and multi-task learning[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 12742-12752.
[14] RUDOLPH M, WANDT B, ROSENHAHN B. Same same but differnet: Semi-supervised defect detection with normalizing flows[C]//Proceedings of the IEEE/CVF winter conference on applications of computer vision. 2021: 1907-1916.
[15] TSCHUCHNIG M E, GADERMAYR M. Anomaly detection in medical imaging-a mini review[C]//Data Science–Analytics and Applications: Proceedings of the 4th International Data Science Conference–iDSC2021. Springer, 2022: 33-38.
[16] HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. science, 2006, 313(5786): 504-507.
[17] GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial networks [J]. Communications of the ACM, 2020, 63(11): 139-144.
[18] GONG D, LIU L, LE V, et al. Memorizing normality to detect anomaly: Memory-augmented deep autoencoder for unsupervised anomaly detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 1705-1714.
[19] ZHOU K, LI J, LUO W, et al. Proxy-bridged image reconstruction network for anomaly detection in medical images[J]. IEEE Transactions on Medical Imaging, 2021, 41(3): 582-594.
[20] ZHOU K, XIAO Y, YANG J, et al. Encoding structure-texture relation with p-net for anomaly detection in retinal images[C]//Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XX 16. Springer, 2020: 360-377.
[21] SCHLEGL T, SEEBÖCK P, WALDSTEIN S M, et al. Unsupervised anomaly detection with generative adversarial networks to guide marker discovery[C]//Information Processing in Medical Imaging: 25th International Conference, IPMI 2017, Boone, NC, USA, June 25-30, 2017, Proceedings. Springer, 2017: 146-157.
[22] SCHLEGL T, SEEBÖCK P, WALDSTEIN S M, et al. f-AnoGAN: Fast unsupervised anomaly detection with generative adversarial networks[J]. Medical image analysis, 2019, 54: 30-44.
[23] AKCAY S, ATAPOUR-ABARGHOUEI A, BRECKON T P. Ganomaly: Semi-supervised anomaly detection via adversarial training[C]//Computer Vision–ACCV 2018: 14th Asian Conference on Computer Vision, Perth, Australia, December 2–6, 2018, Revised Selected Papers, Part III 14. Springer, 2019: 622-637.
[24] REN J, LIU P J, FERTIG E, et al. Likelihood ratios for out-of-distribution detection[J]. Advances in neural information processing systems, 2019, 32.
[25] LI C L, SOHN K, YOON J, et al. Cutpaste: Self-supervised learning for anomaly detection and localization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 9664-9674.
[26] ZAVRTANIK V, KRISTAN M, SKOČAJ D. Draem-a discriminatively trained reconstruction embedding for surface anomaly detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 8330-8339.
[27] SATO J, SUZUKI Y, WATAYA T, et al. Anatomy-aware Self-supervised Learning for Anomaly Detection in Chest Radiographs[A]. 2022.
[28] LI H, IWAMOTO Y, HAN X, et al. An Accurate Unsupervised Liver Lesion Detection Method Using Pseudo-lesions[C]//Medical Image Computing and Computer Assisted Intervention– MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part VIII. Springer, 2022: 214-223.
[29] WANG Y, YAO Q, KWOK J T, et al. Generalizing from a few examples: A survey on few-shot learning[J]. ACM computing surveys (csur), 2020, 53(3): 1-34.
[30] RUFF L, VANDERMEULEN R A, GÖRNITZ N, et al. Deep semi-supervised anomaly detection[A]. 2019.
[31] TABERNIK D, ŠELA S, SKVARČ J, et al. Segmentation-based deep-learning approach for surface-defect detection[J]. Journal of Intelligent Manufacturing, 2020, 31(3): 759-776.
[32] QUELLEC G, LAMARD M, CONZE P H, et al. Automatic detection of rare pathologies in fundus photographs using few-shot learning[J]. Medical image analysis, 2020, 61: 101660.
[33] PANG G, DING C, SHEN C, et al. Explainable deep few-shot anomaly detection with deviation networks[A]. 2021.
[34] TIAN Y, MAICAS G, PU L Z C T, et al. Few-shot anomaly detection for polyp frames from colonoscopy[C]//Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part VI 23. Springer, 2020: 274-284.
[35] HANSEN S, GAUTAM S, JENSSEN R, et al. Anomaly detection-inspired few-shot medical image segmentation through self-supervision with supervoxels[J]. Medical Image Analysis, 2022, 78: 102385.
[36] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Computer-Assisted Intervention– MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. Springer, 2015: 234-241.
[37] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
[38] BOTTOU L. Stochastic gradient descent tricks[J]. Neural Networks: Tricks of the Trade: Second Edition, 2012: 421-436.
[39] SUTSKEVER I, MARTENS J, DAHL G, et al. On the importance of initialization and momentum in deep learning[C]//International conference on machine learning. PMLR, 2013: 1139-1147.
[40] RUDER S. An overview of gradient descent optimization algorithms[A]. 2016.
[41] MHASKAR H N, MICCHELLI C A. How to choose an activation function[J]. Advances in Neural Information Processing Systems, 1993, 6.
[42] NAIR V, HINTON G E. Rectified linear units improve restricted boltzmann machines[C]//Proceedings of the 27th international conference on machine learning (ICML-10). 2010: 807- 814.
[43] MAAS A L, HANNUN A Y, NG A Y, et al. Rectifier nonlinearities improve neural network acoustic models[C]//Proc. icml: volume 30. Atlanta, Georgia, USA, 2013: 3.
[44] BOUREAU Y L, PONCE J, LECUN Y. A theoretical analysis of feature pooling in visual recognition[C]//Proceedings of the 27th international conference on machine learning (ICML- 10). 2010: 111-118.
[45] WANG T, WU D J, COATES A, et al. End-to-end text recognition with convolutional neural networks[C]//Proceedings of the 21st international conference on pattern recognition (ICPR2012). IEEE, 2012: 3304-3308.
[46] CHANDOLA V, BANERJEE A, KUMAR V. Anomaly detection: A survey[J]. ACM computing surveys (CSUR), 2009, 41(3): 1-58.
[47] LIU F T, TING K M, ZHOU Z H. Isolation-based anomaly detection[J]. ACM Transactions on Knowledge Discovery from Data (TKDD), 2012, 6(1): 1-39.
[48] CAMPOS G O, ZIMEK A, SANDER J, et al. On the evaluation of unsupervised outlier detection: measures, datasets, and an empirical study[J]. Data mining and knowledge discovery, 2016, 30: 891-927.
[49] PANG G, SHEN C, VAN DEN HENGEL A. Deep anomaly detection with deviation networks[C]//Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 2019: 353-362.
[50] ZIMEK A, SCHUBERT E, KRIEGEL H P. A survey on unsupervised outlier detection in high-dimensional numerical data[J]. Statistical Analysis and Data Mining: The ASA Data Science Journal, 2012, 5(5): 363-387.
[51] PEVNỲ T. Loda: Lightweight on-line detector of anomalies[J]. Machine Learning, 2016, 102: 275-304.
[52] PANG G, CAO L, CHEN L, et al. Learning homophily couplings from non-iid data for joint feature selection and noise-resilient outlier detection[C]//Proceedings of the 26th International Joint Conference on Artificial Intelligence. IJCAI, 2017: 2585-2591.
[53] PANG G, CAO L, CHEN L, et al. Sparse modeling-based sequential ensemble learning for effective outlier detection in high-dimensional numeric data[C]//Proceedings of the AAAI Conference on Artificial Intelligence: volume 32. 2018.
[54] CAO L. Coupling learning of complex interactions[J]. Information Processing & Management, 2015, 51(2): 167-186.
[55] AGGARWAL C C, AGGARWAL C C. An introduction to outlier analysis[M]. Springer, 2017.
[56] YONEKAWA Y, MODI Y S, KIM L A, et al. American Society of Retina Specialists clinical practice guidelines: management of nonproliferative and proliferative diabetic retinopathy without diabetic macular edema[J]. Journal of vitreoretinal diseases, 2020, 4(2): 125-135.
[57] LECHNER J, O’LEARY O E, STITT A W. The pathology associated with diabetic retinopathy [J]. Vision research, 2017, 139: 7-14.
[58] GRAHAM B. Kaggle diabetic retinopathy detection competition report[J]. University of Warwick, 2015: 24-26.
[59] LIN L, LI M, HUANG Y, et al. The SUSTech-SYSU dataset for automated exudate detection and diabetic retinopathy grading[J]. Scientific Data, 2020, 7(1): 409.
[60] CHENG P, LIN L, HUANG Y, et al. I-secret: Importance-guided fundus image enhancement via semi-supervised contrastive constraining[C]//Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part VIII 24. Springer, 2021: 87-96.
[61] FU H, WANG B, SHEN J, et al. Evaluation of retinal image quality assessment networks in different color-spaces[C]//Medical Image Computing and Computer Assisted Intervention– MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part I 22. Springer, 2019: 48-56.
[62] PORWAL P, PACHADE S, KAMBLE R, et al. Indian diabetic retinopathy image dataset (IDRiD): a database for diabetic retinopathy screening research[J]. Data, 2018, 3(3): 25.
[63] LI T, GAO Y, WANG K, et al. Diagnostic assessment of deep learning algorithms for diabetic retinopathy screening[J]. Information Sciences, 2019, 501: 511-522.
[64] ABRÀMOFF M D, FOLK J C, HAN D P, et al. Automated analysis of retinal images for detection of referable diabetic retinopathy[J]. JAMA ophthalmology, 2013, 131(3): 351-357.
[65] DECENCIÈRE E, ZHANG X, CAZUGUEL G, et al. Feedback on a publicly distributed image database: the Messidor database[J]. Image Analysis & Stereology, 2014, 33(3): 231-234.
[66] SALEHI M, SADJADI N, BASELIZADEH S, et al. Multiresolution knowledge distillation for anomaly detection[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 14902-14912.
[67] ZAVRTANIK V, KRISTAN M, SKOČAJ D. Reconstruction by inpainting for visual anomaly detection[J]. Pattern Recognition, 2021, 112: 107706.
[68] BERGMANN P, FAUSER M, SATTLEGGER D, et al. Uninformed students: Student-teacher anomaly detection with discriminative latent embeddings[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 4183-4192.
[69] PERERA P, NALLAPATI R, XIANG B. Ocgan: One-class novelty detection using gans with constrained latent representations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 2898-2906.
[70] LYU J, CHENG P, TANG X. Fundus image based retinal vessel segmentation utilizing a fast and accurate fully convolutional network[C]//Ophthalmic Medical Image Analysis: 6th International Workshop, OMIA 2019, Held in Conjunction with MICCAI 2019, Shenzhen, China, October 17, Proceedings 6. Springer, 2019: 112-120.
[71] KRULL A, BUCHHOLZ T O, JUG F. Noise2void-learning denoising from single noisy images[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 2129-2137.
[72] ISOLA P, ZHU J Y, ZHOU T, et al. Image-to-image translation with conditional adversarial networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 1125-1134.
[73] OTSU N. A threshold selection method from gray-level histograms[J]. IEEE transactions on systems, man, and cybernetics, 1979, 9(1): 62-66.
[74] WAGNER T, LIPINSKI H G. IJBlob: an ImageJ library for connected component analysis and shape analysis[J]. Journal of Open Research Software, 2013, 1(1).
[75] HE L, REN X, GAO Q, et al. The connected-component labeling problem: A review of state-of-the-art algorithms[J]. Pattern Recognition, 2017, 70: 25-43.
[76] DEVRIES T, TAYLOR G W. Improved regularization of convolutional neural networks with cutout[A]. 2017.
[77] ZHANG H, CISSE M, DAUPHIN Y N, et al. mixup: Beyond empirical risk minimization[A]. 2017.
[78] LYU J, ZHANG Y, HUANG Y, et al. AADG: Automatic Augmentation for Domain Generalization on Retinal Image Segmentation[J]. IEEE Transactions on Medical Imaging, 2022, 41(12): 3699-3711.
[79] HUANG Y, LYU J, CHENG P, et al. SSiT: Saliency-guided Self-supervised Image Transformer for Diabetic Retinopathy Grading[A]. 2022.
[80] MONTABONE S, SOTO A. Human detection using a mobile platform and novel features derived from a visual saliency mechanism[J]. Image and Vision Computing, 2010, 28(3): 391-402.
[81] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-cam: Visual explanations from deep networks via gradient-based localization[C]//Proceedings of the IEEE international conference on computer vision. 2017: 618-626.
[82] NESTEROV Y E. A method of solving a convex programming problem with convergence rate 𝑂(1/k^2)[C]//Doklady Akademii Nauk: volume 269. Russian Academy of Sciences, 1983:
[83] HE K, ZHANG X, REN S, et al. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification[C]//Proceedings of the IEEE international conference on computer vision. 2015: 1026-1034.
[84] ROTH K, PEMULA L, ZEPEDA J, et al. Towards total recall in industrial anomaly detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022: 14318-14328.
[85] RUDOLPH M, WEHRBEIN T, ROSENHAHN B, et al. Fully convolutional cross-scale-flows for image-based defect detection[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 2022: 1088-1097.
[86] YU J, ZHENG Y, WANG X, et al. Fastflow: Unsupervised anomaly detection and localization via 2d normalizing flows[A]. 2021.
[87] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2980-2988.
[88] AGARWAL P K, HAR-PELED S, VARADARAJAN K R, et al. Geometric approximation via coresets[J]. Combinatorial and computational geometry, 2005, 52(1): 1-30.
[89] KOBYZEV I, PRINCE S J, BRUBAKER M A. Normalizing flows: An introduction and review of current methods[J]. IEEE transactions on pattern analysis and machine intelligence, 2020, 43(11): 3964-3979.
[90] HAN K, WANG Y, CHEN H, et al. A survey on vision transformer[J]. IEEE transactions on pattern analysis and machine intelligence, 2022, 45(1): 87-110.
[91] DELONG E R, DELONG D M, CLARKE-PEARSON D L. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach[J]. Biometrics, 1988: 837-845.