北京生物医学工程

基于深度学习的前房角开闭状态自动识别

Automatic recognition of the open angle and angle closure of the anterior chamber angle based on deep learning

作者：王文赛邢恩铭秦鲁宁周盛杨军林松

单位：中国医学科学院北京协和医学院，生物医学工程研究所(天津 300192) 天津医科大学眼科医院 (天津 300384) 通信作者：林松，主管技师。E-mail：linsong123123@sina.com]

关键词：深度学习；前房角；超声生物显微镜； VGG16；自动识别

分类号：R318.04

出版年·卷·期（页码）：2021·40·3（221-226）

摘要：

目的基于深度学习 ( deep learning, DL ) 和前房角超声生物显微镜 ( ultrasoundbiomicroscopy, UBM)图像进行前房角开闭状态的自动识别,为原发性闭角型青光眼的临床自动诊断提供辅助分析。方法数据集为天津医科大学眼科医院采集的眼科疾病患者的前房角 UBM 图像,由眼科专家将 UBM 图像分为房角开放和房角关闭两类,按照 6∶2∶2 的比例随机设置训练集、验证集和测试集。为提高深度学习模型的鲁棒性和识别精度,对训练集图像随机进行了旋转、平移和反转等不影响房角形态的数据增强操作。比较 VGG16、VGG19、DenseNet121、Xception 和 InceptionV3 网络模型在本文数据集上的迁移学习结果,根据迁移学习结果对 VGG16 进行卷积层和全连接层的微调,用微调后的VGG16 模型实现前房角开闭状态的自动识别。用接收者操作特征曲线下面积和准确率等评价指标对模型识别结果进行定量评价,用类激活热力图可视化模型识别前房角开闭状态时的主要关注区域。结果类激活热力图表明微调后的 VGG16 模型识别前房角开闭状态的主要关注区域为房角中心区域,与眼科专家的识别依据一致。该模型的识别准确率为 96接收者操作特征曲线下面积为 0。结论基于深度学习和前房角 UBM 图像能够以较高的准确率实现前房角开闭状态的自动识别,有利于原发性闭角型青光眼自动诊断技术的发展。

Objective Based on the deep learning ( DL ) algorithm and UBM ( ultrasound biomicroscope) images of the anterior chamber angle,the automatic recognition of the open angle and angle closure of the anterior chamber angle is performed,which provides auxiliary analysis for the clinical automatic diagnosis of primary angle?closure glaucomaMethods The data set is the UBM image of the anterior chamber angle of patients with eye diseases collected by the Tianjin medical university eye hospital images into two types:open angle and closed angle,and randomly set the training set,validation set,and test set according to the ratio of 6∶2∶2 DenseNet121,Xception, and InceptionV3 network models on the data set in this article transfer learning results,fine-tune the convolutional layer and fully connected layer of VGG16,and use the fine?tuned VGG16 model to automatically realize the anterior chamber Recognition of angle opening and closing status evaluate the model recognition results,and the class activation map is used to visualize the main areas of interest when the model recognizes the opening and closing state of the anterior chamber angleResults The class activation map shows that the focus area of the fine?tuned VGG16 model is the central area of the corner 96Conclusions Based on deep learning and UBM images of the anterior chamber angle,the automatic recognition of the opening and closing state of the anterior chamber angle can be realized with high accuracy,which is beneficial to the development of automatic diagnosis technology for primary angle?closure glaucoma.

参考文献：

[1] Tham YC, Li X, Wong TY, et al. Global prevalence of glaucoma and projections of glaucoma burden through 2040: a systematic review and meta-analysis[J].Ophthalmology,2014,121:2081–2090.

[2] Mansoori T, Balakrishna N. Anterior segment morphology in primary angle closure glaucoma using ultrasound biomicroscopy[J].?Journal of Current Glaucoma Practice. 2017,11(3):86-91.

[3]?刘丰伟,李汉军,张逸鹤,等.人工智能在医学影像诊断中的应用[J].北京生物医学工程,2019,38(2):206-211.

Liu FW，Li HJ，Zhang YH，et al.?Application of artificial intelligence in medical imaging diagnosis[J].?Beijing Biomedical Engineering,2019,38(2):206-211.

[4] Ting DSW,Cheung CY, 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 ,2017,318:2211–2223.

[5]?Gargeya R, Leng T. Automated identification of diabetic retinopathy using deep learning[J]. Ophthalmology,2017,124:962–969.

[6]?Chen X, Xu Y, Wong DWK, et al. Glaucoma detection based on deep convolutional neural network[C]//2015 37th annual international conference of the IEEE engineering in medicine and biology society (EMBC). Milan:IEEE, 2015: 715-718.

[7] Xu BY, Chiang M, Chaudhary S,et al. Deep learning classifiers for automated detection of gonioscopic angle closure based on anterior segment OCT images[J]. American Journal of Ophthalmology，2019,208:273-280.

[8] Fu H, Baskaran M, Xu Y, et al. A deep learning system for automated angle-closure detection in anterior segment optical coherence tomography images[J]. American Journal of?Ophthalmology，?2019,203:37-45.

[9]?Shorten C , Khoshgoftaar TM . A survey on image data augmentation for deep learning[J]. Journal of Big Data, 2019, 6(1):1-48.

[10]Huang G, Liu Z, van Der Maaten L, et al. Densely connected convolutional networks [C] //Proceedings of the IEEE conference on computer vision and pattern recognition. Hawaii :IEEE，2017: 4700-4708.

[11]?Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Las Vegas :IEEE，2016: 2818-2826.

[12]?Chollet F. Xception: Deep learning with depthwise separable convolutions[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Hawaii :IEEE，2017: 1251-1258.

[13]?Donahue J, Hendricks LA, Rohrbach MA, et al. Long-term recurrent convolutional networks for visual recognition and description[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 677-691.

[14]?Pan SJ, Yang Q. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10): 1345-1359.

[15]?Selvaraju RR, Cogswell?M,Das A，et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[J]. International Journal of Computer Vision,2020, 128:336–359.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录