切换至 "中华医学电子期刊资源库"

中华眼科医学杂志(电子版) ›› 2020, Vol. 10 ›› Issue (02) : 70 -75. doi: 10.3877/cma.j.issn.2095-2007.2020.02.002

论著

人工智能关于视盘区多任务深度学习模型在青光眼分类中的应用
张悦1, 余双2, 马锴2, 初春燕2, 张莉1, 庞睿奇1, 王宁利1, 刘含若1,()   
  1. 1. 100730 首都医科大学附属北京同仁医院 北京同仁眼科中心 北京市眼科研究所 眼科学与视觉科学北京市重点实验室
    2. 518000 深圳,腾讯天衍实验室
  • 收稿日期:2020-03-05 出版日期:2020-04-28
  • 通信作者: 刘含若
  • 基金资助:
    国家自然科学基金项目(81700813); 北京市医院管理局"青苗"计划专项经费项目(QML20180205); 北京市科技新星项目(Z191100001119072); 首都医科大学附属北京同仁医院拔尖人才培养计划医药协同科研创新研究专项(Z181100001918035)

The application of artificial intelligence multi-task deep learning model of optic disc area in the classification of glaucoma

Yue Zhang1, Shuang Yu2, Kai Ma2, Chunyan Chu2, Li Zhang1, Ruiqi Pang1, Ningli Wang1, Hanruo Liu1,()   

  1. 1. Beijing Tongren Eye Center, Beijing Key Laboratory of Ophthalmology and Visual Sciences, Beijing Tongren Hospital, Capital Medical University, Beijing 100730, China; Beijing Institute of Ophthalmology
    2. Tencent Jarvis Laboratory, Shenzhen 518000, China
  • Received:2020-03-05 Published:2020-04-28
  • Corresponding author: Hanruo Liu
引用本文:

张悦, 余双, 马锴, 初春燕, 张莉, 庞睿奇, 王宁利, 刘含若. 人工智能关于视盘区多任务深度学习模型在青光眼分类中的应用[J/OL]. 中华眼科医学杂志(电子版), 2020, 10(02): 70-75.

Yue Zhang, Shuang Yu, Kai Ma, Chunyan Chu, Li Zhang, Ruiqi Pang, Ningli Wang, Hanruo Liu. The application of artificial intelligence multi-task deep learning model of optic disc area in the classification of glaucoma[J/OL]. Chinese Journal of Ophthalmologic Medicine(Electronic Edition), 2020, 10(02): 70-75.

目的

探讨基于人工智能深度学习结合视盘改变分类多任务模型对青光眼的检测能力。

方法

收集2017年8月至2018年8月来自29个省直辖市自治区的医疗机构的共计21 969例(32 548只眼)患者的临床资料。其中,男性12 762例(18 745只眼),女性9207例(13 803只眼);年龄17~75岁,平均年龄(57.1±7.9)岁。由两级人工阅片小组根据眼底图像诊断并标注为正常眼与青光眼,并将其划分为训练集、同源测试集及非同源测试集。使用基于Python 3.6和Pytorch 0.4的腾讯觅影眼底照片辅助诊断系统,以NVIDIA Tesla P40作为运行芯片完成模型的训练,以ResNet 34作为骨干网络的多任务卷积神经网络模型,以青光眼分类任务为主任务,视盘萎缩任务为辅助任务,并选取在调优集上表现最好的模型作为最终模型。此外,利用模型类别激活图对模型做出预测的特征区域进行解释。采用敏感度、特异度及工作特性曲线下面积对其进行预测和性能评价。

结果

模型在三个测试集中诊断青光眼的灵敏度分别为95.9%、95.4%及95.7%;特异度分别为97.7%、91.6%及92.2%;工作特性曲线下面积分别为0.993、0.968及0.974。在引入视盘萎缩多任务后,敏感度和特异度得到了有效提升。

结论

基于深度学习结合视盘改变分类的多任务模型对于青光眼的检测具有较高的准确率。同时,通过可解释性实验有效探索和分析了本模型对于高度近视眼等具有与青光眼相似眼底特征的鉴别能力。

Objective

The aim of this study was to investigate the detection ability of multi-task model based on deep learning combined with optic disc change classification in glaucomatous optic neuropathy.

Methods

21 969 participants (32 548 eyes) were included from medical institutions of 29 Provinces, municipalities and autonomous regions from August 2017 to August 2018. There were 12 762 male and 9207 female, and the age range was 17 to 75 with the average age (57.1±7.9) years-old. The two-level group diagnosed and labeled the fundus images as normal eyes and glaucomatous eyes, and divided them into training set, homologous testing set and two non-homologous testing sets. Our research was based on Tencent Miying Fundus Photo Auxiliary Diagnosis System using Python 3.6 and Pytorch 0.4. We trained our model with NVIDIA Tesla P40 as running chip. Using the multi-task convolution neural network model based on ResNet 34, with the main task of glaucoma classification and the auxiliary task of optic disc atrophy, the model with the best performance in the optimization set was selected as the final model. Moreover, class activation map was used to explain the feature areas of the model. The sensitivity, specificity and area under the curve are used to predict and evaluate its performance.

Results

The sensitivity of the model was95.9%, 95.4% and 95.7%, respectively; specificity of that was 97.7%, 91.6% and 92.2%, respectively; area under the curve of that was 0.993, 0.968 and 0.974, respectively. The sensitivity and specificity of the model were effectively improved after the introduction of multi-task model of optic disc atrophy.

Conclusions

The multi-task model based on deep learning combined with optic disc change classification has a high accuracy for the detection of glaucomatous optic neuropathy. Moreover, the ability of this model to distinguish high myopia from glaucoma was explored and analyzed by interpretable experiment.

图1 多任务卷积神经网络框架图 该模型主要由多个相连的卷积层、残差模块、全局池化层和全连接层构成。Conv,代表全卷积层;GAP,代表全局池化层;FC,代表全连接层;Sigmoid,代表激活函数;⊕,代表按位相加;7×7和3×3,代表所在卷积层的卷积核大小;×3、×4、×6和×3,代表所在的残差模块被重复的次数;各模块下方的数字,代表特征通道数
表1 数据集划分及构成信息[例数(%)]
表2 单任务青光眼分类模型与多任务模型多中心性能的比较
图2 单任务模型与多任务模型在测试集上的工作特性曲线图 图A示在同源测试集上单任务与多任务模型得到的工作特性曲线下面积分别为0.991和0.993;图B示在非同源测试集1上单任务与多任务模型得到的工作特性曲线下面积分别为0.962和0.968;图C示在非同源测试集2上单任务与多任务模型得到的工作特性曲线下面积分别为0.962和0.974
图3 模型预测类别激活图 图A1示青光眼但无视盘萎缩的原始图像,图A2示图A1的青光眼任务模型激活图,图A3示图A1的视盘萎缩任务模型激活图;图B1示青光眼且存在视盘萎缩的原始图像,图B2示图B1的青光眼任务模型激活图,图B3示图B1的视盘萎缩任务模型激活图;图C1示非青光眼且无视盘萎缩的原始图像,图C2示图C1的青光眼任务模型激活图,图C3示图C1的视盘萎缩任务模型激活图;图D1示非青光眼但存在视盘萎缩的原始图像,图D2示图D1的青光眼任务模型激活图,图D3示图D1的视盘萎缩任务模型激活图
[1]
Quigley HA, Broman AT. The number of people with glaucoma worldwide in 2010 and 2020[J]. Br J Ophthalmol, 2006, 90(3): 262-267.
[2]
Weinreb RN, Aung T, Medeiros FA. The pathophysiology and treatment of glaucoma: a review[J]. JAMA, 2014, 311(18): 1901-1911.
[3]
Ting DSW, Peng L, Varadarajan AV, et al. Deep learning in ophthalmology: The technical and clinical considerations[J]. Prog Retin Eye Res, 2019, 72: 100759.
[4]
He KZX, Ren S, Sun J. Deep residual learning for image recognition[J]. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, 29(5): 770-778.
[5]
Zhou BKA, Lapedriza A, Oliva A, et al. Learning deep features for discriminative localization[J]. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, 38(10): 2921-2929.
[6]
Lecun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[7]
Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks[J]. Nature, 2017, 546(7660): 686-686.
[8]
Ehteshami BB, Veta M, Johannes van Diest P, et al. Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer[J]. JAMA, 2017, 318(22): 2199-2210.
[9]
Raumviboonsuk P, Krause J, Chotcomwongse P, et al. Deep learning versus human graders for classifying diabetic retinopathy severity in a nationwide screening program[J]. NPJ Digit Med, 2019, 2: 25.
[10]
Gulshan V, Peng L, Coram M, et al. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs[J]. JAMA, 2016, 316(22): 2402-2410.
[11]
Ting DSW, Cheung CYL, 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-J Am Med Assoc, 2017, 318(22): 2211-2223.
[12]
Chakrabarty L, Joshi GD, Chakravarty A, et al. Automated Detection of Glaucoma From Topographic Features of the Optic Nerve Head in Color Fundus Photographs[J]. J Glaucoma, 2016, 25(7): 590-597.
[13]
Annan L, Jun C, Wong DWK, et al. Integrating holistic and local deep features for glaucoma classification[J]. Conf Proc IEEE Eng Med Biol Soc, 2016, 43(11): 1328-1331.
[14]
Li Z, He Y, Keel S, et al. Efficacy of a Deep Learning System for Detecting Glaucomatous Optic Neuropathy Based on Color Fundus Photographs[J]. Ophthalmology, 2018, 125(8): 1199-1206.
[15]
Liu H, Li L, Wormstone IM, et al. Development and Validation of a Deep Learning System to Detect Glaucomatous Optic Neuropathy Using Fundus Photographs[J]. Jama Ophthalmol, 2019, 137(12): 1353-1360.
[16]
Kermany DS, Goldbaum M, Cai W, et al. Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning[J]. Cell, 2018, 172(5): 1122-1131.
[17]
Singh A, Dutta MK, Partha SM, et al. Image processing based automatic diagnosis of glaucoma using wavelet features of segmented optic disc from fundus image[J]. Comput Meth Prog Bio, 2016, 124: 108-120.
[18]
Waisbourd M, Pruzan NL, Johnson D, et al. The Philadelphia Glaucoma Detection and Treatment Project: Detection Rates and Initial Management[J]. Ophthalmology, 2016, 123(8): 1667-1674.
[19]
Stevens GA, White RA, Flaxman SR, et al. Global prevalence of vision impairment and blindness: magnitude and temporal trends, 1990—2010[J]. Ophthalmology, 2013, 120(12): 2377-2384.
[20]
Bourne RR, Stevens GA, White RA, et al. Causes of vision loss worldwide, 1990—2010: a systematic analysis[J]. Lancet Glob Health, 2013, 1(6): 339-349.
[21]
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(11): 2081-2090.
[22]
Liang YB, Friedman DS, Zhou Q, et al. Prevalence of primary open angle glaucoma in a rural adult Chinese population: the Handan eye study[J]. Invest Ophthalmol Vis Sci, 2011, 52(11): 8250-8257.
[23]
Zhao D, Guallar E, Gajwani P, et al. Optimizing Glaucoma Screening in High-Risk Population: Design and 1-Year Findings of the Screening to Prevent (SToP) Glaucoma Study[J]. Am J Ophthalmol, 2017, 180: 18-28.
[24]
Owsley C, Rhodes LA, Mcgwin G, et al. Eye Care Quality and Accessibility Improvement in the Community (EQUALITY) for adults at risk for glaucoma: study rationale and design[J]. Int J Equity Health, 2015, 14: 135.
[25]
Tatham AJ, Medeiros FA, Zangwill LM, et al. Strategies to improve early diagnosis in glaucoma[J]. Prog Brain Res, 2015, 221: 103-133.
[26]
Samples SP. The Glaucoma Book: a Practical, Evidence-Based Approach to Patient Care[M]. New York: Springer Science & Business Media, 2010.
[27]
Pizzi LT, Waisbourd M, Hark L, et al. Costs of a community-based glaucoma detection programme: analysis of the Philadelphia Glaucoma Detection and Treatment Project[J]. Br J Ophthalmol, 2018, 102(2): 225-232.
[28]
Miller SE, Thapa S, Robin AL, et al. Glaucoma Screening in Nepal: Cup-to-Disc Estimate With Standard Mydriatic Fundus Camera Compared to Portable Nonmydriatic Camera[J]. American Journal of Ophthalmology, 2017, 182: 99-106.
[29]
Lee PP, Walt JG, Doyle JJ, et al. A multicenter, retrospective pilot study of resource use and costs associated with severity of disease in glaucoma[J]. Arch Ophthalmol-Chic, 2006, 124(1): 12-19.
[30]
Teng CC, De Moraes CG, Prata TS, et al. Beta-Zone parapapillary atrophy and the velocity of glaucoma progression[J]. Ophthalmology, 2010, 117(5): 909-915.
[31]
Masumoto H, Tabuchi H, Nakakura S, et al. Deep-learning Classifier with an Ultrawide-field Scanning Laser Ophthalmoscope Detects Glaucoma Visual Field Severity[J]. J Glaucoma, 2018, 27(7): 647-652.
[32]
Jonas JB, Nagaoka N, Fang YX, et al. Intraocular Pressure and Glaucomatous Optic Neuropathy in High Myopia[J]. Invest Ophthalmol Vis Sci, 2017, 58(13): 5897-5906.
[33]
Fu HZ, Cheng J, Xu YW, et al. Disc-Aware Ensemble Network for Glaucoma Screening From Fundus Image[J]. Ieee T Med Imaging, 2018, 37(11): 2493-2501.
[34]
Jonas JB, Fang Y, Weber P, et al. Parapapillary Gamma and Delta Zones in High Myopia[J]. Retina, 2018, 38(5): 931-938.
[35]
Zhang Z, Xu Y, Liu J, et al. Automatic diagnosis of pathological myopia from heterogeneous biomedical data[J]. Plos One, 2013, 8(6): e65736.
[36]
Tan TE, Ting DSW, Liu Y, et al. Artificial intelligence using a deep learning system with transfer learning to predict refractive error and myopic macular degeneration from color fundus photographs[J]. Invest Ophth Vis Sci, 2019, 60(9): 1013-1018.
[1] 李洋, 蔡金玉, 党晓智, 常婉英, 巨艳, 高毅, 宋宏萍. 基于深度学习的乳腺超声应变弹性图像生成模型的应用研究[J/OL]. 中华医学超声杂志(电子版), 2024, 21(06): 563-570.
[2] 罗刚, 泮思林, 孙玲玉, 李志新, 陈涛涛, 乔思波, 庞善臣. 一种新型语义网络分析模型对室间隔完整型肺动脉闭锁和危重肺动脉瓣狭窄胎儿右心发育不良程度的评价作用[J/OL]. 中华医学超声杂志(电子版), 2024, 21(04): 377-383.
[3] 孔德铭, 刘铮, 李睿, 钱文伟, 王飞, 蔡道章, 柴伟. 人工智能辅助全髋关节置换三维术前规划准确性评价[J/OL]. 中华关节外科杂志(电子版), 2024, 18(04): 431-438.
[4] 王雅楠, 刘丹, 曹正浓, 贾慧敏. 儿童迟发性先天性膈疝患儿的临床诊治特点分析[J/OL]. 中华妇幼临床医学杂志(电子版), 2024, 20(04): 410-419.
[5] 张嘉炜, 王瑞, 张克诚, 易磊, 周增丁. 烧烫伤创面深度智能检测模型P-YOLO的建立及测试效果[J/OL]. 中华损伤与修复杂志(电子版), 2024, 19(05): 379-385.
[6] 叶莉, 杜宇. 深度学习在牙髓根尖周病临床诊疗中的应用[J/OL]. 中华口腔医学研究杂志(电子版), 2024, 18(06): 351-356.
[7] 黄俊龙, 李文双, 李晓阳, 刘柏隆, 陈逸龙, 丘惠平, 周祥福. 基于盆底彩超的人工智能模型在女性压力性尿失禁分度诊断中的应用[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2024, 18(06): 597-605.
[8] 赵毅, 李昶田, 唐文博, 白雪婷, 刘荣. 腹腔镜术中超声主胰管自动识别模型的临床应用[J/OL]. 中华腔镜外科杂志(电子版), 2024, 17(05): 290-294.
[9] 郑大雯, 王健东. 胆囊癌辅助诊断研究进展[J/OL]. 中华肝脏外科手术学电子杂志, 2024, 13(06): 769-773.
[10] 陈健, 周静洁, 夏开建, 王甘红, 刘罗杰, 徐晓丹. 基于卷积神经网络实现结直肠息肉的实时检测与自动NICE分型(附视频)[J/OL]. 中华结直肠疾病电子杂志, 2024, 13(03): 217-228.
[11] 张莉, 傅楹迪, 王宁利, 左晓玲. 青光眼视路损伤及脱髓鞘在青光眼神经损伤作用中的研究进展[J/OL]. 中华眼科医学杂志(电子版), 2024, 14(04): 233-239.
[12] 刘甘甘, 荣翱. 激光虹膜周边切开术联合囊袋张力环植入治疗原发性闭角型青光眼伴白内障的远期临床研究[J/OL]. 中华眼科医学杂志(电子版), 2024, 14(04): 200-205.
[13] 潘清, 葛慧青. 基于机械通气波形大数据的人机不同步自动监测方法[J/OL]. 中华重症医学电子杂志, 2024, 10(04): 399-403.
[14] 孙铭远, 褚恒, 徐海滨, 张哲. 人工智能应用于多发性肺结节诊断的研究进展[J/OL]. 中华临床医师杂志(电子版), 2024, 18(08): 785-790.
[15] 林一鑫, 董晶, 贾建文, 黄菊梅, 武军元, 王双坤, 柳云鹏, 汪阳. 基于人工智能分析颈内动脉颅外段迂曲特征及对称性的应用性评价[J/OL]. 中华脑血管病杂志(电子版), 2024, 18(03): 202-209.
阅读次数
全文


摘要