Progress in the application of artificial intelligence in eye diseases

doi:10.3877/cma.j.issn.2095-2007.2026.01.009

Abstract

Abstract:

With the rapid development of artificial intelligence(AI), its application in the medical field has received widespread attention. The application of AI in ophthalmology is gradually developing towards a more comprehensive and in-depth level, demonstrating good performance in the diagnosis and treatment of eye diseases. The application of AI in common ophthalmic diseases was reviewed, looking forward to future development trends.

Key words: Artificial intelligence, Cornea, Glaucoma, Cataract, Uveitis, Retinal diseases

Aoxin Gu, Zhifeng Wu. Progress in the application of artificial intelligence in eye diseases[J]. Chinese Journal of Ophthalmologic Medicine(Electronic Edition), 2026, 16(01): 48-53.

/ / Recommend

Add to citation manager EndNote|Ris|BibTeX

URL: https://zhykyxzz.cma-cmc.com.cn/EN/10.3877/cma.j.issn.2095-2007.2026.01.009

https://zhykyxzz.cma-cmc.com.cn/EN/Y2026/V16/I01/48

Figures/Tables 1

References 56

[1]	Kline R. Cybernetics, automata studies, and the Dartmouth conference on articial intelligence[J]. IEEE Ann Hist Comput, 2010, 33(4): 5-16.
[2]	Bali J, Garg R, Bali RT. Artificial intelligence(AI)in healthcare and biomedical research: Why a strong computational/AI bioethics framework is required[J]? Indian J Ophthalmol, 2019, 67(1): 3-6.
[3]	Labib KM, Ghumman H, Jain S, et al. A review of the utility and limitations of artificial intelligence in retinal disorders and pediatric ophthalmology[J]. Cureus, 2024, 16(10): e71063.
[4]	Sonmez SC, Sevgi M, Antaki F, et al. Generative artificial intelligence in ophthalmology: current innovations, future applications and challenges[J]. Br J Ophthalmol, 2024, 108(10): 1335-1340.
[5]	Mai J, Schmidt-Erfurth U. Role of artificial intelligence in retinal diseases[J]. Klin Monbl Augenheilkd, 2024, 241(9): 1023-1031.
[6]	Serghiou S, Rough K. Deep learning for epidemiologists: An introduction to neural networks[J]. Am J Epidemiol, 2023, 192(11): 1904-1916.
[7]	Zhang Z, Wang Y, Zhang H, et al. Artificial intelligence-assisted diagnosis of ocular surface diseases[J]. Front Cell Dev Biol, 2023, 11: 1133680.
[8]	Kuo BI, Chang WY, Liao TS, et al. Keratoconus screening based on deep learning approach of corneal topography[J]. Transl Vis Sci Technol, 2020, 9(2): 53-59.
[9]	Zhao L, Yin Y, Hu T, et al. Comprehensive management of post-LASIK ectasia: From prevention to treatment[J]. Acta Ophthalmol, 2023, 101(5): 485-503.
[10]	Quanchareonsap W, Kasetsuwan N, Reinprayoon U, et al. Deep learning algorithm for keratoconus detection from tomographic maps and corneal biomechanics: A diagnostic study[J]. J Curr Ophthalmol, 2024, 36(1): 46-53.
[11]	Kundu G, D′Souza S, Modak D, et al. Using artificial intelligence for an efficient prediction of outcomes of deep anterior lamellar keratoplasty(DALK)in advanced keratoconus[J]. Transl Vis Sci Technol, 2025, 14(5): 30-37.
[12]	Shao Y, Jie Y, Liu ZG. Guidelines for the application of artificial intelligence in the diagnosis of anterior segment diseases[J]. Int J Ophthalmol, 2023, 16(9): 1373-1385.
[13]	Wang D, Sklar B, Tian J, et al. Improving artificial intelligence-based microbial keratitis screening tools constrained by limited data using synthetic generation of slit-lamp photos[J]. Ophthalmol Sci, 2024, 5(3): 100676.
[14]	Hart C, Chen X, Ahmed M, et al. AI-MK: artificial intelligence for assessing and monitoring microbial keratitis[J]. BMJ Open Ophthalmol, 2026, 11(1): e002556.
[15]	Preston FG, Meng Y, Burgess J, et al. Artificial intelligence utilising corneal confocal microscopy for the diagnosis of peripheral neuropathy in diabetes mellitus and prediabetes[J]. Diabetologia, 2022, 65(3): 457-466.
[16]	Meng Y, Preston FG, Ferdousi M, et al. Artificial intelligence based analysis of corneal confocal microscopy images for diagnosing peripheral neuropathy: A binary classification model[J]. J Clin Med, 2023, 12(4): 1284.
[17]	Rabah CB, Petropoulos IN, Stettner M, et al. Deep learning-assisted differentiation of four peripheral neuropathies using corneal confocal microscopy[J]. Ann Clin Transl Neurol, 2026, 13(4): 747-754.
[18]	王佳浩，李王婷，杨雅涵，等. 人工智能在眼前段疾病的应用[J]. 眼科学报，2022，37(3)：171-177.
[19]	Shimizu E, Tanji M, Nakayama S, et al. AI-based diagnosis of nuclear cataract from slit-lamp videos[J]. Sci Rep, 2023, 13(1): 22046.
[20]	Labib KM, Ghumman H, Jain S, et al. Applications of artificial intelligence in ophthalmology: Glaucoma, cornea, and oculoplastics[J]. Cureus, 2024, 16(11): e73522.
[21]	张晓玲，杨多瑾，王慧琴，等. 儿童先天性白内障影响因素分析及人工智能筛查诊断系统的构建[J]. 中国医药导报，2025，22(12)：119-124，130.
[22]	娄炜，陈子盎，章尧，等. 人工智能在人工晶状体屈光力计算的应用[J]. 眼科学报，2023，38(12)：790-799.
[23]	Raimundo M, Findl O. Update on intraocular lens formulas[J]. Curr Opin Ophthalmol, 2025, 36(1): 4-9.
[24]	Gopi-Kannan P, Balasubramanian C, Jarin T. Explainable dual-stage context-aware SwinUNet for accurate optic disc and optic cup segmentation in glaucoma assessment[J]. Int Ophthalmol, 2026, 46(1): 227-232.
[25]	Boverhof BJ, Corro Ramos I, Vermeer KA, et al. The cost-effectiveness of an artificial intelligence-based population-wide screening program for primary open-angle glaucoma in the netherlands[J]. Value Health, 2025, 28(9): 1317-1326.
[26]	Jan CL, Vingrys A, Henwood J, et al. Performance of a deep learning system and performance of optometrists for the detection of glaucomatous optic neuropathy using colour retinal photographs[J]. Bioengineering(Basel), 2024, 11(11): 1139-1145.
[27]	蓝子俊，谢珺，郭燕，等. 基于线性化注意力和双重注意力的视杯盘分割模型[J]. 生物医学工程学杂志，2023，40(5): 920-927.
[28]	Cheng L, Hong J, Chen H, et al. Deep learning-based automatic differentiation of acute angle closure with or without zonulopathy using ultrasound biomicroscopy: a comparison of diagnostic performance with ophthalmologists[J]. BMJ Open Ophthalmol, 2025, 10(1): e002114.
[29]	Ohn K, Jun H, Kim YS, et al. Deep learning-based generation of retinal nerve fibre layer thickness maps from fundus photographs: A comparative analysis of u-net architectures for accessible glaucoma assessment[J]. Life(Basel), 2026, 16(4): 559.
[30]	Gobira M, Nakayama LF, Regatieri CVS, et al. Comparing no-code platforms and deep learning models for glaucoma detection from fundus images[J]. Cureus, 2025, 17(3): e81064.
[31]	Huang AS, Fam A, Zhao H, et al. ChatGPT-assisted glaucoma diagnosis: a health-equitable multi-ancestry analysis using visual field and optical coherence tomography data[J]. Am J Ophthalmol, 2026, 283：129-137.
[32]	Jacquot R, Sève P, Jackson TL, et al. Diagnosis, classification, and assessment of the underlying etiology of uveitis by artificial intelligence: A systematic review[J]. J Clin Med, 2023, 12(11): 3746-3752.
[33]	Sorkhabi MA, Potapenko IO, Ilginis T, et al. Assessment of anterior uveitis through anterior-segment optical coherence tomography and artificial intelligence-based image analyses[J]. Transl Vis Sci Technol, 2022, 11(4): 7-13.
[34]	Mellak Y, Achim A, Ward A, et al. A machine learning framework for the quantification of experimental uveitis in murine OCT[J]. Biomed Opt Express, 2023, 14(7): 3413-3432.
[35]	Keino H, Aman T, Furuya R, et al. Automated quantitative analysis of anterior segment inflammation using swept-source anterior segment optical coherence tomography: A pilot study[J]. Diagnostics(Basel), 2022, 12(11): 2703-2708.
[36]	娄岩，邵毅，郜原，等. 人工智能在葡萄膜炎诊断中的应用指南(2025)[J]. 眼科新进展，2025，45(3)：169-176.
[37]	杨卫华，邵毅，许言午. 眼科人工智能临床研究评价指南(2023)[J]. 国际眼科杂志，2023，23(7)：1064-1071.
[38]	Mhibik B, Kouadio D, Jung C, et al. Automated detection of vitritis using ultrawide-field fundus photographs and deep learning[J]. Retina, 2024, 44(6): 1034-1044.
[39]	Amiot V, Jimenez-Del-Toro O, Guex-Crosier Y, et al. Automatic transformer-based grading of multiple retinal inflammatory signs in uveitis on fluorescein angiography[J]. Comput Biol Med, 2025, 193: 110327.
[40]	Kim J, Nguyen NV, Soifer MA, et al. Deep ensemble learning to detect retinal vascular leakage on ultrawide-field fundus photographs of patients with uveitis[J]. Transl Vis Sci Technol, 2026, 15(4): 15-22.
[41]	Leingang O, Riedl S, Mai J, et al. Automated deep learning-based AMD detection and staging in real-world OCT datasets(PINNACLE study report 5)[J]. Sci Rep, 2023, 13(1): 19545-19551.
[42]	Holland R, Taylor TRP, Holmes C, et al. Specialized curricula for training vision language models in retinal image analysis[J]. NPJ Digit Med, 2025, 8(1): 532-539.
[43]	Neri G, Rebecchi C, Oakley JD, et al. Deep learning model for automated classification of macularneovascularization subtypes in AMD[J]. Invest Ophthalmol Vis Sci, 2025, 66(9): 55-61.
[44]	Ansari G, Schärer N, Pfau K, et al. Evaluating the progression of retinal sensitivity loss in geographic atrophy using machine-learning-based structure-function correlation(OMEGA 2)[J]. Invest Ophthalmol Vis Sci, 2025, 66(11): 34-39.
[45]	Cattaneo J, Forte P, Forte G, et al. Faricimab efficacy in type 1 macular neovascularization: AI-assisted quantification of pigment epithelium detachment(PED)volume reduction over 12 months in Naïve and switch eyes[J]. Int J Retina Vitreous, 2025, 11(1): 3-9.
[46]	Cattaneo J, De-Oliveira-Figueiredo EC, Montesel A, et al. Early real-world outcomes of intravitreal aflibercept 8 mg in treatment-Naïve neovascular AMD: AI-assisted fluid volume analysis[J]. Int J Retina Vitreous, 2025, 11(1): 42-47.
[47]	董力，李赫妍，魏文斌. 人工智能在糖尿病视网膜病变中的应用进展[J/OL]. 中华眼科医学杂志(电子版)，2024，14(1)：57-61.
[48]	Labib KM, Ghumman H, Jain S, et al. A review of the utility and limitations of artificial intelligence in retinal disorders and pediatric ophthalmology[J]. Cureus, 2024, 16(10): e71063.
[49]	Madi MMM, Farr PC, Bester D. Diabetic retinopathy detection: ai models and approaches[J]. J Ophthalmol, 2026, PMID: 8857887.
[50]	白君华，巩迪，陈宜，等. 眼底荧光造影无灌注区智能分割[J]. 生物医学工程学进展，2025，46(1)：9-16.
[51]	李博，邹北骥，肖晓霞，等. 基于多尺度双编码器和加强跳跃连接的硬性渗出物分割算法[J]. 智能计算机与应用，2025，15(5)：28-36.
[52]	Kaliki S, Vempuluru VS, Ghose N, et al. Artificial intelligence and machine learning in ocular oncology: Retinoblastoma[J]. Indian J Ophthalmol, 2023, 71(2): 424-430.
[53]	Antonio-Aguirre B, Gadiraju A, Ownagh V, et al. Novel artificial intelligence applications for pediatric retina[J/OL]. Curr Opin Ophthalmol, 2025, 36(3): 456-461.
[54]	Shoeibi N, Abrishami M, Hosseini SM, et al. Development and validation of machine learning classifiers for predicting treatment-needed retinopathy of prematurity[J]. BMC Med Inform Decis Mak, 2025, 25(1): 221-225.
[55]	He D, Luo X, Ying B, et al. Machine learning models for predicting treatment-requiring retinopathy of prematurity in the e-ROP study[J]. Transl Vis Sci Technol, 2025, 14(8): 14-19.
[56]	Zhang X, Liang H, Wang R, et al. Research on the construction of an AI diagnostic model for plus disease of retinopathy of prematurity based on cross-center fusion datasets[J]. Front Pediatr, 2026, 14: 1765353.

第一作者	疾病名称	模型基础	信息获取途径	性能
眼前段疾病
Kuo等^[8]	圆锥角膜	卷积神经网络	角膜地形图	敏感性和特异性均超过0.90，AUC为0.995
Quanchareonsap等^[10]	圆锥角膜	卷积神经网络	角膜地形图和角膜生物力学分析仪(Corvis ST)	从Pentacam获得屈光图的AI模型1的AUC为0.938，准确度为0.947。模型2将动态角膜反应和Corvis ST报告添加到模型1中，AUC为0.985，准确率为0.956。模型3将角膜生物力学指数添加到模型2中，AUC为0.991，准确度为0.956
Kundu等^[11]	圆锥角膜	随机森林	眼前节分析仪＋问卷评估	AUC为0.957，灵敏度为98%，特异性为85.6%
Wang等^[13]	角膜炎	生成对抗网络	裂隙灯显微镜	AUC为0.93，优于仅用真实数据训练的模型，提高了0.17
Hart等^[14]	角膜炎	卷积神经网络	前段光学相干断层扫描	人工智能模型准确捕捉了93%的病变。该模型的敏感性为93%，特异性为100%，阳性预测值为100%，阴性预测值为73%
Meng等^[16]	糖尿病周围神经病变	卷积神经网络	角膜共聚焦显微镜	灵敏度为0.91，特异性为0.93，AUC为0.95
Rabah等^[17]	糖尿病周围神经病变	卷积神经网络	角膜共聚焦显微镜	人工智能取得了优异的结果：AUC＝96.75%，敏感性83.87%，特异性95.07%
Shimizu等^[19]	白内障	机器学习	裂隙灯显微镜	0级白内障AUC＝0.967；1级白内障AUC＝0.928；2级白内障：AUC＝0.923；3级白内障：AUC＝0.949
张晓玲等^[21]	儿童先天性白内障	随机森林及自适应增强	裂隙灯显微镜	基于随机森林的人工智能诊断筛查系统优于基于自适应增强的模型。其双侧外部验证AUC为0.95，四重交叉为0.91；单侧外部验证为0.85，四重交叉为0.82
青光眼
Gopi-Kannan等^[24]	青光眼	卷积神经网络	眼底照相	使用Drishti GS和REFUGE数据集进行性能评估，该模型的准确率分别为99.3%和99.1%
Boverhof等^[25]	青光眼	决策树-马尔可夫混合模型	眼底照相	人工智能筛查可以更早地发现青光眼，检测到的病例数是机会性筛查的1.60倍
Jan等^[26]	青光眼	卷积神经网络	眼底照相	人工智能的敏感性为33.3%，明显低于医师的65.1%，但是人工智能具有明显更高的特异性97.4%，眼科医师为85.5%
蓝子俊等^[27]	青光眼	注意力残差网络	眼底照相	视神经分割的视网膜图像数据集中，视盘和视杯交并比分别为0.9623和0.8564；用于视神经评估的开放式视网膜图像数据集中，视盘和视杯交并比分别为0.9563和0.7844
Cheng等^[28]	青光眼	逻辑回归模型	超声生物显微镜	二元分类逻辑回归模型具有最佳性能，AUC为0.9922，准确率为96.0%。手动计算的多项式逻辑回归模型的AUC为0.9373，准确率为79.45%
Ohn等^[29]	青光眼	卷积神经网络	眼底照相	ResU-Net表现最佳，使用均方误差＝0.00061，平均绝对误差＝0.01877，结构相似性指数度量＝0.9163，峰值信噪比＝32.19 dB，弗雷歇起始距离＝30.08，可以从眼底照片中生成高保真的视网膜神经纤维厚度图
葡萄膜炎
Sorkhabi等^[33]	葡萄膜炎	卷积神经网络	光学相干断层扫描	临床分级与人工智能软件检测到的颗粒计数及颗粒密度之间存在显著相关性(Spearman ρ＝0.7077，0.7035；P<0.05)。人工和人工智能检测到的颗粒之间存在显著的相关性(r＝0.9948，P<0.05)
Mellak等^[34]	葡萄膜炎	卷积神经网络	光学相干断层扫描	在有无葡萄膜炎的二元分类中实现了高达80%的准确率，并且可以捕捉到一些区分日期的特征，准确率在70%到74%之间
Mhibik等^[38]	葡萄膜炎	卷积神经网络	超宽视野眼底照片	该模型用于检测玻璃体炎症的性能良好，灵敏度为91%，特异性为89%，准确度为0.90
Amiot等^[39]	葡萄膜炎	卷积神经网络	荧光素眼底血管成像	在检测血管渗漏、毛细血管渗漏、黄斑水肿和视盘高荧光方面与专家评分者非常匹配
Kim等^[40]	葡萄膜炎	集成学习	超宽视野眼底照片	模型显示出卓越的性能，数据集wMildRVL的准确率为0.7704，灵敏度为0.7699，特异性为0.7713，AUC为0.8018，woMildRVL数据集的准确度为0.7900，灵敏度为0.7519，特异性为0.8000，AUC为0.8344
眼底疾病
Leingang等^[41]	年龄相关性黄斑变性	卷积神经网络	光学相干断层扫描	该模型在真实世界的数据集上进行了训练，并进行了广泛的评估，AUC达到了0.94
Neri等^[43]	年龄相关性黄斑变性	卷积神经网络	光学相干断层扫描	1型黄斑新生血管的最高报告敏感性和特异性分别为96.7%和84.9%；2型为100.0%和85.5%；3型为84.9%和87.9%。1、2和3型MNV的AUC分别为0.95、0.97及0.91
Ansari等^[44]	年龄相关性黄斑变性	随机森林、LASSO回归和多元自适应回归样条	微视野、光学相干断层扫描	随机森林模型在所有情况下都显示出最高的准确性
白君华等^[50]	糖尿病性视网膜病变	卷积神经网络	荧光素眼底血管成像	Enhance LadderNet模型的无灌注区域自动分割效果优于其他传统模型，准确率为85.01%，灵敏度为88.00%
李博等^[51]	糖尿病性视网膜病变	卷积神经网络	眼底照相＋公开数据集IDRiD、E-Ophtha	该模型在IDRiD数据集上召回率、准确率、戴斯相似系数及交并比分别为81.00%、78.51%、79.66%及66.29%，在E-Ophtha上分别为52.85%、63.20%、56.15%及39.96%
Kaliki等^[52]	视网膜母细胞瘤	开放式计算机视觉技术和深度学习	眼底照相	人工智能模型检测视网膜母细胞瘤的敏感性、特异性、阳性预测值和阴性预测值分别为96%、94%、97%及91%
Shoeibi等^[54]	早产儿视网膜病变	K近邻、支持向量机、随机森林、极值梯度提升、深度神经网络及变换器等六种机器学习	眼底照相	人工智能模型K近邻和深度神经网络的AUC分别为0.777和0.853，极端梯度增强和深度神经网络的灵敏度分别为0.765和0.929，深度神经网络和变压器的特异性分别为0.644和0.698

第一作者	疾病名称	模型基础	信息获取途径	性能
眼前段疾病
Kuo等^[8]	圆锥角膜	卷积神经网络	角膜地形图	敏感性和特异性均超过0.90，AUC为0.995
Quanchareonsap等^[10]	圆锥角膜	卷积神经网络	角膜地形图和角膜生物力学分析仪(Corvis ST)	从Pentacam获得屈光图的AI模型1的AUC为0.938，准确度为0.947。模型2将动态角膜反应和Corvis ST报告添加到模型1中，AUC为0.985，准确率为0.956。模型3将角膜生物力学指数添加到模型2中，AUC为0.991，准确度为0.956
Kundu等^[11]	圆锥角膜	随机森林	眼前节分析仪＋问卷评估	AUC为0.957，灵敏度为98%，特异性为85.6%
Wang等^[13]	角膜炎	生成对抗网络	裂隙灯显微镜	AUC为0.93，优于仅用真实数据训练的模型，提高了0.17
Hart等^[14]	角膜炎	卷积神经网络	前段光学相干断层扫描	人工智能模型准确捕捉了93%的病变。该模型的敏感性为93%，特异性为100%，阳性预测值为100%，阴性预测值为73%
Meng等^[16]	糖尿病周围神经病变	卷积神经网络	角膜共聚焦显微镜	灵敏度为0.91，特异性为0.93，AUC为0.95
Rabah等^[17]	糖尿病周围神经病变	卷积神经网络	角膜共聚焦显微镜	人工智能取得了优异的结果：AUC＝96.75%，敏感性83.87%，特异性95.07%
Shimizu等^[19]	白内障	机器学习	裂隙灯显微镜	0级白内障AUC＝0.967；1级白内障AUC＝0.928；2级白内障：AUC＝0.923；3级白内障：AUC＝0.949
张晓玲等^[21]	儿童先天性白内障	随机森林及自适应增强	裂隙灯显微镜	基于随机森林的人工智能诊断筛查系统优于基于自适应增强的模型。其双侧外部验证AUC为0.95，四重交叉为0.91；单侧外部验证为0.85，四重交叉为0.82
青光眼
Gopi-Kannan等^[24]	青光眼	卷积神经网络	眼底照相	使用Drishti GS和REFUGE数据集进行性能评估，该模型的准确率分别为99.3%和99.1%
Boverhof等^[25]	青光眼	决策树-马尔可夫混合模型	眼底照相	人工智能筛查可以更早地发现青光眼，检测到的病例数是机会性筛查的1.60倍
Jan等^[26]	青光眼	卷积神经网络	眼底照相	人工智能的敏感性为33.3%，明显低于医师的65.1%，但是人工智能具有明显更高的特异性97.4%，眼科医师为85.5%
蓝子俊等^[27]	青光眼	注意力残差网络	眼底照相	视神经分割的视网膜图像数据集中，视盘和视杯交并比分别为0.9623和0.8564；用于视神经评估的开放式视网膜图像数据集中，视盘和视杯交并比分别为0.9563和0.7844
Cheng等^[28]	青光眼	逻辑回归模型	超声生物显微镜	二元分类逻辑回归模型具有最佳性能，AUC为0.9922，准确率为96.0%。手动计算的多项式逻辑回归模型的AUC为0.9373，准确率为79.45%
Ohn等^[29]	青光眼	卷积神经网络	眼底照相	ResU-Net表现最佳，使用均方误差＝0.00061，平均绝对误差＝0.01877，结构相似性指数度量＝0.9163，峰值信噪比＝32.19 dB，弗雷歇起始距离＝30.08，可以从眼底照片中生成高保真的视网膜神经纤维厚度图
葡萄膜炎
Sorkhabi等^[33]	葡萄膜炎	卷积神经网络	光学相干断层扫描	临床分级与人工智能软件检测到的颗粒计数及颗粒密度之间存在显著相关性(Spearman ρ＝0.7077，0.7035；P<0.05)。人工和人工智能检测到的颗粒之间存在显著的相关性(r＝0.9948，P<0.05)
Mellak等^[34]	葡萄膜炎	卷积神经网络	光学相干断层扫描	在有无葡萄膜炎的二元分类中实现了高达80%的准确率，并且可以捕捉到一些区分日期的特征，准确率在70%到74%之间
Mhibik等^[38]	葡萄膜炎	卷积神经网络	超宽视野眼底照片	该模型用于检测玻璃体炎症的性能良好，灵敏度为91%，特异性为89%，准确度为0.90
Amiot等^[39]	葡萄膜炎	卷积神经网络	荧光素眼底血管成像	在检测血管渗漏、毛细血管渗漏、黄斑水肿和视盘高荧光方面与专家评分者非常匹配
Kim等^[40]	葡萄膜炎	集成学习	超宽视野眼底照片	模型显示出卓越的性能，数据集wMildRVL的准确率为0.7704，灵敏度为0.7699，特异性为0.7713，AUC为0.8018，woMildRVL数据集的准确度为0.7900，灵敏度为0.7519，特异性为0.8000，AUC为0.8344
眼底疾病
Leingang等^[41]	年龄相关性黄斑变性	卷积神经网络	光学相干断层扫描	该模型在真实世界的数据集上进行了训练，并进行了广泛的评估，AUC达到了0.94
Neri等^[43]	年龄相关性黄斑变性	卷积神经网络	光学相干断层扫描	1型黄斑新生血管的最高报告敏感性和特异性分别为96.7%和84.9%；2型为100.0%和85.5%；3型为84.9%和87.9%。1、2和3型MNV的AUC分别为0.95、0.97及0.91
Ansari等^[44]	年龄相关性黄斑变性	随机森林、LASSO回归和多元自适应回归样条	微视野、光学相干断层扫描	随机森林模型在所有情况下都显示出最高的准确性
白君华等^[50]	糖尿病性视网膜病变	卷积神经网络	荧光素眼底血管成像	Enhance LadderNet模型的无灌注区域自动分割效果优于其他传统模型，准确率为85.01%，灵敏度为88.00%
李博等^[51]	糖尿病性视网膜病变	卷积神经网络	眼底照相＋公开数据集IDRiD、E-Ophtha	该模型在IDRiD数据集上召回率、准确率、戴斯相似系数及交并比分别为81.00%、78.51%、79.66%及66.29%，在E-Ophtha上分别为52.85%、63.20%、56.15%及39.96%
Kaliki等^[52]	视网膜母细胞瘤	开放式计算机视觉技术和深度学习	眼底照相	人工智能模型检测视网膜母细胞瘤的敏感性、特异性、阳性预测值和阴性预测值分别为96%、94%、97%及91%
Shoeibi等^[54]	早产儿视网膜病变	K近邻、支持向量机、随机森林、极值梯度提升、深度神经网络及变换器等六种机器学习	眼底照相	人工智能模型K近邻和深度神经网络的AUC分别为0.777和0.853，极端梯度增强和深度神经网络的灵敏度分别为0.765和0.929，深度神经网络和变压器的特异性分别为0.644和0.698

Please choose a citation manager

Content to export