| 140 | 0 | 86 |
| 下载次数 | 被引频次 | 阅读次数 |
深度学习技术在糖尿病性黄斑水肿（diabetic macular edema,DME）的频域光学相干断层扫描（spectral domain optical coherence tomography,SD-OCT）影像分割中发挥着重要作用。针对数据隐私保护、计算成本控制和不确定性量化等关键挑战，本文提出一种基于联邦学习的DME分割算法（DME segmentation algorithm based on federated learning,DMESA-FL）。首先，通过在卷积神经网络（convolutional neural network,CNN）中增加尺度感知金字塔融合模块和全局金字塔引导模块来捕获多尺度上下文信息并融合全局上下文信息流和解码路径特征。接着，将改进的CNN作为联邦学习框架的预测模型并采用序列训练来更新全局模型以增强数据安全性。最后，通过为所有客户端增加特征离散化预处理模块以减少CNN的计算量并提升其泛化能力。在特征离散化的过程中，构建基于粗糙集的适应度函数以评估数据不确定性并采用遗传算法（genetic algorithm,GA）搜索SD-OCT影像的最佳特征离散化方案。此外，通过在网络的损失函数中引入不确定性约束项以将粗糙集的平均近似精度作为先验知识有效融入CNN。DMESA-FL与主流的SD-OCT眼底影像分割算法的对比结果表明，DMESA-FL能够在数据非共享的情况下高效地训练具有不同客户端的模型，从而实现DME的精确分割。
Abstract：Deep learning technology plays a crucial role in the segmentation of spectral domain optical coherence tomography(SD-OCT) images for diabetic macular edema(DME). A DME segmentation algorithm based on federated learning(DMESA-FL) is proposed to address key challenges such as data privacy protection, computational cost control, and uncertainty quantification. Initially, a scale-aware pyramid fusion module and global pyramid guidance modules are incorporated into the convolutional neural network(CNN) to capture multi-scale contextual information and fuse the global contextual information flow with the features of the decoding path. Subsequently, the improved CNN is employed as the prediction model within the federated learning framework, and sequential training is adopted to update the global model, thereby enhancing data security. Ultimately, a feature discretization preprocessing module is introduced for all clients to reduce the computational burden of CNN and improve its generalization capability. During the feature discretization process, a fitness function based on rough sets is constructed to assess data uncertainty, and a genetic algorithm(GA) is utilized to search for the optimal breakpoints in SD-OCT images(the optimal feature discretization scheme for SD-OCT images). Additionally, an uncertainty constraint term is introduced into the loss function of the network for effectively integrating the average approximation precision of rough sets as prior knowledge into CNN. The comparative results between DMESA-FL and the state-of-the-art SD-OCT fundus image segmentation algorithms demonstrate that DMESA-FL can efficiently train models across different clients without data sharing,thereby achieving precise segmentation of DME.
[1] ROBERTS P K , VOGL W D , GERENDAS B S , et al.Quantification of fluid resolution and visual acuity gain in patients with diabetic macular edema using deep learning:a post hoc analysis of a randomized clinical trial[J]. JAMA Ophthalmology, 2020, 138(9):945-953.
[2] SUN H, SAEEDI P, KARURANGA S, et al. IDF diabetes atlas:global , regional and country-level diabetes prevalence estimates for 2021 and projections for 2045[J]. Diabetes Research and Clinical Practice, 2022, 183:109119.
[3]贾伟平.血糖监测技术的进步与展望[J].上海交通大学学报（医学版），2022, 42(9):1171-1175.JIA W P. New progress and prospects of blood glucose monitoring technology[J]. Journal of Shanghai Jiao Tong University(Medical Science), 2022, 42(9):1171-1175.(in Chinese)
[4] TAN G S, CHEUNG N, SIMóR, et al. Diabetic macular oedema[J]. The Lancet Diabetes&Endocrinology, 2017, 5(2):143-155.
[5] CHEN Z C, FU S S, WU Z R, et al. Relationship between plasma angiogenic growth factors and diabetic foot ulcers[J]. Clinica Chimica Acta, 2018, 482:95-100.
[6] MAZLIN V , XIAO P , SCHOLLER J , et al. Real-time non-contact cellular imaging and angiography of human cornea and limbus with common-path full-field/SD OCT[J].Nature Communications, 2020, 11:1868.
[7] VUJOSEVIC S , ALDINGTON S J , SILVA P , et al.Screening for diabetic retinopathy:new perspectives and challenges[J]. The Lancet Diabetes&Endocrinology ,2020, 8(4):337-347.
[8] CHEN Q, ZENG L R, DING W P. FRCNN:a combination of fuzzy-rough-set-based feature discretization and convolutional neural network for segmenting subretinal fluid lesions[J]. IEEE Transactions on Fuzzy Systems, 2025, 33(1):350-364.
[9] VIEDMA I A, ALONSO-CANEIRO D, READ S A, et al.Deep learning in retinal optical coherence tomography(OCT):a comprehensive survey[J]. Neurocomputing, 2022,507:247-264.
[10] HARISHA M S, BHOSALE A A, NARENDER M. Deep learning-powered segmentation and classification of diabetic retinopathy for enhanced diagnostic precision[J]. Artificial Intelligence in Health, 2024, 1(4):30.
[11] ROY A G, CONJETI S, KARRI S P K, et al. ReLayNet:retinal layer and fluid segmentation of macular optical coherence tomography using fully convolutional networks[J].Biomedical Optics Express, 2017, 8(8):3627-3642.
[12] SCHLEGL T, WALDSTEIN S M, BOGUNOVIC H, et al.Fully automated detection and quantification of macular fluid in OCT using deep learning[J]. Ophthalmology, 2018,125(4):549-558.
[13] HE X X, FANG L Y, TAN M K, et al. Intra-and interslice contrastive learning for point supervised OCT fluid segmentation[J]. IEEE Transactions on Image Processing,2022, 31:1870-1881.
[14] XING G, CHEN L, WANG H L, et al. Multi-scale pathological fluid segmentation in OCT with a novel curvature loss in convolutional neural network[J]. IEEE Transactions on Medical Imaging, 2022, 41(6):1547-1559.
[15] LIN J L, CHEN Q. An intelligent sensor data preprocessing method for OCT fundus image watermarking using an RCNN[J]. Computer Modeling in Engineering&Sciences,2024, 138(2):1549-1561.
[16] MINAEE S, BOYKOV Y, PORIKLI F, et al. Image segmentation using deep learning:a survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022,44(7):3523-3542.
[17] ZHU T Y, LI K Z, HERRERO P, et al. Deep learning for diabetes:a systematic review[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(7):2744-2757.
[18] CHEN Q, WANG J H, LIN J L. Generative AI exacerbates the climate crisis[J]. Science , 2025 , 387(6734):587.
[19] ZENG L R, HUANG M X, LI Y C, et al. Progressive feature fusion attention dense network for speckle noise removal in OCT images[J]. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2024, 21(4):748-756.
[20] IDOGEN O S. The perspectives of eye care professionals on the integration of artificial intelligence in eye care practices:a systematic review[J]. Artificial Intelligence in Health,2024, 1(2):66.
[21] LI H, LI C C, WANG J, et al. Review on security of federated learning and its application in healthcare[J]. Future Generation Computer Systems, 2023, 144:271-290.
[22] RIEKE N, HANCOX J, LI W Q, et al. The future of digital health with federated learning[J]. NPJ Digital Medicine,2020, 3:119.
[23] CHEN Q, DING W P, HUANG X M, et al. Generalized interval type-II fuzzy rough model-based feature discretization for mixed pixels[J]. IEEE Transactions on Fuzzy Systems, 2023, 31(3):845-859.
[24] ZENG L R, CHEN Q, HUANG M X. RSFD:a rough setbased feature discretization method for meteorological data[J]. Frontiers in Environmental Science, 2022, 10:1013811.
[25] DOLEZAL J M, SRISUWANANUKORN A, KARPEYEV D, et al. Uncertainty-informed deep learning models enable high-confidence predictions for digital histopathology[J]. Nature Communications, 2022, 13:6572.
[26] LI Y P, LüX C. Convergence analysis of sequential federated learning on heterogeneous data[C]//Proceedings of the 37th International Conference on Neural Information Processing Systems, December 10-16, 2023, New Orleans, LA, USA. New York:ACM, 2023:56700-56755.
[27] LU W J, LEI J H, QIU P, et al. UPCoL:uncertainty-informed prototype consistency learning for semi-supervised medical image segmentation[C]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2023. Cham:Springer, 2023:662-672.
[28] HUANG W K, YE M, SHI Z K, et al. Federated learning for generalization, robustness, fairness:a survey and benchmark[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(12):9387-9406.
[29] GUAN H, YAP P T, BOZOKI A, et al. Federated learning for medical image analysis:a survey[J]. Pattern Recognition, 2024, 151:110424.
[30] CHEN Q, HUANG M X, WANG H. A feature discretization method for classification of high-resolution remote sensing images in coastal areas[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(10):8584-8598.
[31] YU R G, TIAN Y, GAO J, et al. Feature discretizationbased deep clustering for thyroid ultrasound image feature extraction[J]. Computers in Biology and Medicine, 2022,146:105600.
[32] CHEN Q, HUANG M X, WANG H, et al. A feature discretization method based on fuzzy rough sets for high-resolution remote sensing big data under linear spectral model[J]. IEEE Transactions on Fuzzy Systems, 2022, 30(5):1328-1342.
[33] CHEN Q, DING W P. A genetic algorithm based on deep Q-learning in optimization of remote sensing data discretization[J/OL]. IEEE Transactions on Evolutionary Computation(Early Access), 2024[2025-05-22]. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10730790.
[34] CHEN Q, XIE L R, ZENG L R, et al. Neighborhood rough residual network-based outlier detection method in IoT-enabled maritime transportation systems[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(11):11800-11811.
[35] YU B, ZHENG Z J, CAI M J, et al. FRCM:a fuzzy rough c-means clustering method[J]. Fuzzy Sets and Systems ,2024, 480:108860.
[36] FENG S L, ZHAO H M, SHI F, et al. CPFNet:context pyramid fusion network for medical image segmentation[J].IEEE Transactions on Medical Imaging, 2020, 39(10):3008-3018.
[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] PAPPU G P, TANKALA S, TALABHAKTULA K, et al.EANet:multiscale autoencoder based edge attention network for fluid segmentation from SD-OCT images[J]. International Journal of Imaging Systems and Technology,2023, 33(3):909-927.
[39] GEETHA PAVANI P, BISWAL B, GANDHI T K, et al.Robust semantic segmentation of retinal fluids from SDOCT images using FAM-U-Net[J]. Biomedical Signal Processing and Control, 2024, 87:105481.
[40] LI S E, MA F, YAN F, et al. SMFDNet:spatial and multi-frequency domain network for OCT angiography retinal vessel segmentation[J]. The Journal of Supercomputing ,2025, 81(4):535.
基本信息:
DOI：10.12194/j.ntu.20250610002
中图分类号:R587.2;R774.5;TP181;TP391.41
引用信息:
[1]陈琼,孙靖博,李俊霖,等.基于联邦学习的糖尿病性黄斑水肿分割[J].南通大学学报(自然科学版),2025,24(03):1-11+22.DOI:10.12194/j.ntu.20250610002.
基金信息:
国家自然科学基金地区科学基金项目(12461085); 海南省自然科学基金项目(825MS091,821MS138); 海南省卫生健康科技创新联合项目(WSJK2024MS200); 海南省研究生创新科研课题(Qhyb2022-138); 海南医科大学科创项目(RZ2500002173)