多专家注释的视杯和视盘不确定性量化

doi:10.19678/j.issn.1000-3428.0063897

摘要/Abstract

摘要： 现有基于深度学习的视杯和视盘分割方法在模型训练时，仅使用图像的单个注释或从多个注释中获取唯一的注释信息，忽略原始多专家标注中嵌入的一致性或差异性信息，从而导致模型和预测结果过度自信等问题。提出一种基于多解码器不确定性感知体系的模型MUA-Net。通过引入专业知识推断模块，将各个专家注释的专业知识水平作为先验知识嵌入编码器和解码器的瓶颈中，以形成包含专家线索的高级语义特征。利用可同时学习多个注释的多解码器结构调节多专家之间的分歧，重构多专家注释过程，并对不确定或分歧区域进行量化。提出一种双分支软注意机制，增强多解码器分割预测的模糊区域，得到最终校准的分割结果。实验结果表明，该模型在RIGA数据集上能以较高的不确定性预测合理的区域，与MRNet模型相比，该模型在视杯分割中的平均精度、Dice系数、交并比分别提升了0.75、0.39、0.41个百分点。

关键词: 不确定性估计, 多解码器, 多专家注释, 视杯视盘分割, 软注意机制

Abstract: Existing methods for segmenting images of optic cups and optic discs based on deep learning have some limitations in terms of training. For example, they may use only a single annotation of an image or obtain unique information from multiple annotations and, thus, ignore consistency or difference information embedded in the original annotations by multiple experts; this typically causes overconfidence in the model and prediction results.In this study, a multi-decoder uncertainty-aware model called MUA-Net is proposed.First, an Expertise-aware Inferring Module (EIM) is introduced to embed information on the level of professional knowledge of each expert annotation into the bottleneck of an encoder and a decoder as a priori knowledge to construct high-level semantic features containing information provided by experts.This enables the proposed multi-decoder structure to learn multiple annotations simultaneously to adjust the divergence between multiple experts, reconstruct the multi-expert annotation process, and quantify uncertain or divergent regions.Finally, a dual-branch soft attention mechanism is proposed to further enhance the fuzzy region predicted by using multi-decoder segmentation to obtain calibrated segmentation results and improve segmentation performance.The experimental results show that the model can predict reasonable regions with high uncertainty on the RIGA dataset.Compared with the MRNet model, the mean accuracy, Dice coefficient, and Intersection over Union(IoU) of the model in the cup segmentation are improved by 0.75, 0.39, and 0.41 percentage points respectively.

Key words: uncertainty estimation, multi-decoder, multi-expert annotation, optic cup and optic disc segmentation, soft attention mechanism

中图分类号:

TP391.41

刘丽霞, 宣士斌, 刘畅, 李嘉祥. 多专家注释的视杯和视盘不确定性量化[J]. 计算机工程, 2023, 49(1): 250-257,269.

LIU Lixia, XUAN Shibin, LIU Chang, LI Jiaxiang. Quantification of Optic Cup and Optic Disc Uncertainty with Multi-Expert Annotations[J]. Computer Engineering, 2023, 49(1): 250-257,269.

http://www.ecice06.com/CN/Y2023/V49/I1/250

图/表 7

20230701181319

20230701181322

20230701181325

20230701181328

20230701181331

20230701181334

20230701181337

参考文献

[1] FU H Z, CHENG J, XU Y W, et al.Joint optic disc and cup segmentation based on multi-label deep network and polar transformation[J].IEEE Transactions on Medical Imaging, 2018, 37(7):1597-1605.
[2] ZHANG Q M, LIU L Y, MA K, et al.Cross-denoising network against corrupted labels in medical image segmentation with domain shift[EB/OL].[2022-01-10].https://www.semanticscholar.org/paper/Cross-denoising-Network-against-Corrupted-Labels-in-Zhang-Liu/d14cbc0b64e75330a85fe4b9fb038536abbb3695.
[3] LI G, LI C S, ZENG C, et al.Region focus network for joint optic disc and cup segmentation[C]//Proceedings of the AAAI Conference on Artificial Intelligence.Menlo Park, USA:AAAI Press, 2020:751-758.
[4] WARFIELD S K, ZOU K H, WELLS W M.Simultaneous truth and performance level estimation (STAPLE):an algorithm for the validation of image segmentation[J].IEEE Transactions on Medical Imaging, 2004, 23(7):903-921.
[5] ZHANG S H, FU H Z, YAN Y G, et al.Attention guided network for retinal image segmentation attention guided network for retinal image segmentation[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin, Germany:Springer, 2019:797-805.
[6] JENSEN M H, JØRGENSEN D R, JALABOI R, et al.Improving uncertainty estimation in convolutional neural networks using inter-rater agreement[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin, Germany:Springer, 2019:540-548.
[7] GUAN M, GULSHAN V, DAI A, et al.Who said what:modeling individual labelers improves classification[EB/OL].[2022-01-10].https://www.xueshufan.com/publication/2963772355.
[8] BLUNDELL C, CORNEBISE J, KAVUKCUOGLU K, et al.Weight uncertainty in neural networks[EB/OL].[2022-01-10].https://arxiv.org/abs/1505.05424.
[9] KENDALL A, GAL Y.What uncertainties do we need in Bayesian deep learning for computer vision?[EB/OL].[2022-01-10].https://arxiv.org/abs/1703.04977.
[10] HÜLLERMEIER E, WAEGEMAN W.Aleatoric and epistemic uncertainty in machine learning:an introduction to concepts and methods[J].Machine Learning, 2021, 110(3):457-469.
[11] ABDAR M, POURPANAH F, HUSSAIN S, et al.A review of uncertainty quantification in deep learning:techniques, applications and challenges[J].Information Fusion, 2021, 76:243-297.
[12] GAWLIKOWSKI J, TASSI C R N, ALI M, et al.A survey of uncertainty in deep neural networks[EB/OL].[2022-01-10].https://arxiv.org/abs/2107.03342.
[13] GAL Y, GHAHRAMANI Z.Dropout as a Bayesian approximation:representing model uncertainty in deep learning[EB/OL].[2022-01-10].https://arxiv.org/abs/1506.02142.
[14] LAVES M H, IHLER S, KORTMANN K P, et al.Well-calibrated model uncertainty with temperature scaling for dropout variational inference[EB/OL].[2022-01-10].https://arxiv.org/abs/1909.13550.
[15] HUMT M, LEE J, TRIEBEL R.Bayesian optimization meets Laplace approximation for robotic introspection[EB/OL].[2022-01-10].https://arxiv.org/abs/2010.16141.
[16] LAKSHMINARAYANAN B, PRITZEL A, BLUNDELL C.Simple and scalable predictive uncertainty estimation using deep ensembles[EB/OL].[2022-01-10].https://arxiv.org/abs/1612.01474.
[17] BELUCH W H, GENEWEIN T, NURNBERGER A, et al.The power of ensembles for active learning in image classification[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:9368-9377.
[18] GUSTAFSSON F K, DANELLJAN M, SCHON T B.Evaluating scalable Bayesian deep learning methods for robust computer vision[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.Washington D.C., USA:IEEE Press, 2020:1289-1298.
[19] STAHL N, FALKMAN G, KARLSSON A, et al.Evaluation of uncertainty quantification in deep learning[C]//Proceedings of International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems.Berlin, Germany:Springer, 2020:556-568.
[20] KOHL S A A, ROMERA-PAREDES B, MEYER C, et al.A probabilistic U-net for segmentation of ambiguous images[EB/OL].[2022-01-10].https://arxiv.org/abs/1806.05034.
[21] BAUMGARTNER C F, TEZCAN K C, CHAITANYA K, et al.Phiseg:capturing uncertainty in medical image segmentation[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin, Germany:Springer, 2019:119-127.
[22] GANTENBEIN M, ERDIL E, KONUKOGLU E.RevPHiSeg:a memory-efficient neural network for uncertainty quantification in medical image segmentation[M]//Uncertainty for Safe Utilization of Machine Learning in Medical Imaging and Graphs in Biomedical Image Analysis.Berlin, Germany:Springer, 2020:13-22.
[23] HU S, WORRALL D, KNEGT S, et al.Supervised uncertainty quantification for segmentation with multiple annotations[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin, Germany:Springer, 2019:137-145.
[24] MIRIKHARAJI Z, ABHISHEK K, IZADI S, et al.D-LEMA:deep learning ensembles from multiple annotations-application to skin lesion segmentation[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.Washington D.C., USA:IEEE Press, 2021:1837-1846.
[25] ZENG D W, LI M Q, DING Y K, et al.Segmentation with multiple acceptable annotations:a case study of myocardial segmentation in contrast echocardiography[C]//Proceedings of International Conference on Information Processing in Medical Imaging.Berlin, Germany:Springer, 2021:478-491.
[26] WANG Q L, YANG H, SHEN L, et al.Uncertainty-aware lung nodule segmentation with multiple annotations[EB/OL].[2022-01-10].https://arxiv.org/abs/2110.12372.
[27] VALIUDDIN M M A, VIVIERS C G A, VAN SLOUN R J G, et al.Improving aleatoric uncertainty quantification in multi-annotated medical image segmentation with normalizing flows[M]//Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Perinatal Imaging, Placental and Preterm Image Analysis.Berlin, Germany:Springer, 2021:75-88.
[28] ZHANG Y, KERS J, CASSOL C A, et al.U-Net-and-a-half:convolutional network for biomedical image segmentation using multiple expert-driven annotations[EB/OL].[2022-01-10].https://www.xueshufan.com/publication/3192108523.
[29] YU S, ZHOU H Y, MA K, et al.Difficulty-aware glaucoma classification with multi-rater consensus modeling[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin, Germany:Springer, 2020:741-750.
[30] YANG Y, GUO X, PAN Y, et al.Uncertainty quantification in medical image segmentation with multi-decoder U-Net[EB/OL].[2022-01-10].https://link.springer.com/chapter/10.1007/978-3-031-09002-8_50.
[31] JI W, YU S, WU J D, et al.Learning calibrated medical image segmentation via multi-rater agreement modeling[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2021:12336-12346.
[32] RONNEBERGER O, FISCHER P, BROX T.U-Net:convolutional networks for biomedical image segmentation[C]//Proceedings of International Conference on Medical Image Computing and Computer-assisted Intervention.Berlin, Germany:Springer, 2015:234-241.
[33] SHI X J, CHEN Z R, WANG H, et al.Convolutional LSTM network:a machine learning approach for precipitation nowcasting[EB/OL].[2022-01-10].https://arxiv.org/abs/1506.04214.
[34] WOO S, PARK J, LEE J Y, et al.CBAM:convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision.Berlin, Germany:Springer, 2018:3-19.
[35] ALMAZROA A, ALODHAYB S, OSMAN E, et al.Agreement among ophthalmologists in marking the optic disc and optic cup in fundus images[J].International Ophthalmology, 2017, 37(3):701-717.
[36] ZHANG S, FU H, YAN Y, et al.Attention guided network for retinal image segmentation[C]///Proceedings of International Conference on Medical Image Computing and Computer Assisted Intervention.Berlin, Germany:Springer, 2019:797-805.
[37] GU Z W, CHENG J, FU H Z, et al.CE-net:context encoder network for 2D medical image segmentation[J].IEEE Transactions on Medical Imaging, 2019, 38(10):2281-2292.
[38] YU S.Robust optic disc and cup segmentation with deep learning for glaucoma detection[J].Computerized Medical Imaging and Graphics, 2019, 74:61-71.
[39] WANG S J, YU L Q, YANG X, et al.Patch-based output space adversarial learning for joint optic disc and cup segmentation[J].IEEE Transactions on Medical Imaging, 2019, 38(11):2485-2495.
[40] WANG S J, YU L Q, LI K, et al.Boundary and entropy-driven adversarial learning for fundus image segmentation[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin, Germany:Springer, 2019:102-110.

选择文件类型/文献管理软件名称

选择包含的内容