眼科AI诊断:真实世界中的挑战与解决方案

李飞 ,  梁晓莹 ,  汪德明 ,  杨泽锋 ,  刘潇逸 ,  张秀兰 ,  林顺潮

科学观察 ›› 2025, Vol. 20 ›› Issue (1) :1-21

PDF (2426KB)
科学观察 ›› 2025, Vol. 20 ›› Issue (1) :1-21 DOI: 10.15978/j.cnki.1673-5668.20241203
前沿观察

眼科AI诊断:真实世界中的挑战与解决方案

作者信息 +

Ophthalmic AI Diagnostics: Challenges and Solutions in the Real World

Author information +
文章历史 +

摘要

[目的/意义] 人工智能(AI)技术的迅速发展带来了眼科诊断模式的新变革,但其在真实世界应用中仍然面临着诸多挑战。探索眼科AI在临床部署和应用中的挑战及相应的解决方案具有重要的现实意义。[方法/过程] 通过文献综述梳理眼科AI的应用现状,总结眼科AI在真实世界应用中所面临的挑战及可能的解决方案。[结果/结论] 眼科AI诊断在真实世界应用中存在数据缺乏与标注困难、准确性偏低及可靠性差等技术难关、面临医患信任度低、伦理与法规问题等诸多挑战。为了克服数据资源的短缺,我们可以采取建立数据共享平台、利用生成数据和增强学习技术等策略来丰富AI模型的训练数据。同时,为了提升医患对AI模型的信任,可以开发更高效的算法模型,整合多源数据进行训练,并加强对医疗人员的AI技术培训。此外,制定和完善相关法律法规及行业标准,建立详尽的数据隐私保护措施,对于填补AI监管的空白至关重要。

Abstract

[Objective/Significance] The rapid advancement of artificial intelligence (AI) has significantly transformed ophthalmic diagnostic approaches. However, the deployment of AI in real-world still faces numerous challenges. It is crucial to identify these challenges and explore potential solutions to ensure the effective deployment of ophthalmic AI in clinical practice. [Method/Process] A literature review was conducted to assess the state of the art of ophthalmic AI, identify challenges and explore potential solutions in real-world deployment. [Results/Conclusions] Ophthalmic diagnostic AI faces several challenges in real-world deployment, including limited data, difficulties in data annotation, issues with AI accuracy and reliability, skepticism toward AI among patients and medical professionals, as well as ethical and regulatory concerns. To address the shortage of data, strategies such as establishing data-sharing platforms, utilizing generated data, and applying reinforcement learning can be employed to improve data availability and model performance. Meanwhile, building trust in AI among medical professionals and patients requires developing more efficient algorithms, integrating multimodal data for training, and providing AI relevant courses for healthcare professionals. Furthermore, formulating and refining relevant laws, regulations, and industry standards, as well as establishing comprehensive data privacy protection, are crucial to filling gaps in AI regulation.

关键词

人工智能 / 眼科 / 真实世界 / 深度学习 / 挑战 / 解决方案

Key words

artificial intelligence / ophthalmology / real-world / deep learning / challenges / solutions

引用本文

引用格式 ▾
李飞, 梁晓莹, 汪德明, 杨泽锋, 刘潇逸, 张秀兰, 林顺潮. 眼科AI诊断:真实世界中的挑战与解决方案[J]. 科学观察, 2025, 20(1): 1-21 DOI:10.15978/j.cnki.1673-5668.20241203
Li Fei, Leung Enne Hiu Ying, Wang Deming, Yang Zefeng, Liu Xiaoyi, Zhang Xiulan, Lam Dennis Shun Chiu. Ophthalmic AI Diagnostics: Challenges and Solutions in the Real World[J]. Science Focus, 2025, 20(1): 1-21 DOI:10.15978/j.cnki.1673-5668.20241203

登录浏览全文

4963

注册一个新账户 忘记密码

0 引言

人工智能(AI),作为计算机科学的一个分支,致力于创造能够模仿智能行为的算法[1-2]。在医疗和公共卫生领域,AI已广泛应用于基于临床图像的自动化诊断[3-6]。眼球结构精细且复杂,眼病诊断通常依赖不同的影像检查设备,影像是眼病诊疗的重要前提和基础。眼科临床实践产生了丰富的影像数据,因此眼科成为医学领域与AI结合的重要前沿和窗口[7-12]。AI技术的诞生与发展,促进了眼病诊断精度及速度的显著提升,为眼科诊疗模式革新带来了巨大机遇与突破[9-10,13-14]

眼科AI诊断源于对眼科影像数据分析与模式识别技术的探索与发展[15-16]。初期的AI应用主要集中在利用机器学习算法对眼底照相和光学相干断层扫描(OCT)等影像资料进行的分析[17-22]。对于临床文本和小型影像数据集,传统的机器学习技术(如支持向量机和随机森林)展现了稳健的性能[18,23-25]。但是,随着临床数据的快速增长,特别是在眼科图像分析中,这些方法由于计算复杂度高和效率低而受限;而深度学习技术,尤其是卷积神经网络,在处理大规模、高维度数据时显示了其优越性[17,24,26-32]。近年来,大语言模型(large language models,LLM)(如ChatGPT)在多个应用场景中展示出了其不可小觑的进步和潜力,这类模型具备良好的泛化能力,能够处理和分析多模态数据(如文本、图像),也能与用户实时互动[16,33-37]

眼科AI诊断系统提供了精准、高效的辅助诊断和疾病分析,其对于眼科诊断的增益不仅体现在辅助医生提升诊断的准确性和效率上,更在于其对于眼科医疗资源配置的优化[12,15,18,38-41]。在许多发展中国家和偏远地区,专业的眼科医生和先进的医疗设备非常匮乏,而AI技术的应用能够在这些地区提供初步筛查和诊断服务,有望提升偏远地区人群的眼健康水平[19,42-46]。此外,AI系统可以提供持续的数据监测和评估,这对于慢性眼病管理尤为重要[47-50]。例如,AI能够监测糖尿病患者的视网膜状态,提醒医生及时调整治疗方案,预防病情恶化[19,47-48]

AI在眼科诊断方面呈现出极具潜力的应用前景。然而,其临床部署和应用仍面临着多方面的挑战[7-10,51]。首先,从数据层面来看,获取高质量的模型训练数据受到多种因素的限制,包括数据量不足、数据标注的高成本和复杂性,以及与隐私保护相关的问题[52-54]。其次,AI模型的准确性和可靠性、医疗设备与AI软件的兼容性,以及AI辅助下的医疗工作流的适应性,都是需要解决的关键问题[55-57]。此外,深度学习模型普遍缺乏可解释性,这导致医生对AI技术的接受度较低,进而限制了AI技术的推广和应用[58-63]。在法规与伦理方面,监管政策的缺乏和不确定性,以及患者隐私保护的伦理问题,也是AI在眼科应用中需要克服的困境[13,64-65]。这些挑战不仅限制了AI技术的普及,还影响了其在不同医疗体系中的整合性和标准化进程。因此,克服这些挑战是实现AI在眼科诊断领域广泛应用的关键。

本文旨在探讨眼科AI在真实世界应用中遇到的主要挑战,并提出创新的解决方案,以期为眼科智能医疗的改进提供有力的支持和方向。通过分析当前眼科AI技术的实际运用情况,结合最新的研究成果和案例研究,我们将深入讨论如何克服这些障碍,实现眼科AI诊断技术的有效应用,推动眼科诊疗模式的变革从理论向实践转化。

1 眼科AI诊断的应用现状

1.1 主要应用领域

随着眼科影像数据来源的丰富和成像技术的不断进步,眼科AI的应用场景已经实现了从早期的眼底彩照向OCT、超声生物显微镜、房角照相、眼底血管成像技术以及角膜地形图等全领域眼科影像数据的广泛扩展[7,10-11,66]。AI与眼科影像技术的共同发展为眼前节到眼后节等不同疾病的智能辅助诊断奠定了基础,并推动了眼科AI辅助诊断临床级产品的技术完善和标准化[60,67-72]

眼科AI诊断系统的初代产品主要使用单一模态影像数据作为输入,实现单一疾病的分类诊断或病灶特征识别[73-77]。作为第一个专用于糖尿病性视网膜病变(DR)检测的AI设备,IDx-DR在2018年获得了美国食品和药物管理局的正式批准。该系统基于眼底照片诊断DR,并提出转诊的建议[78]。该设备在后续真实的初级保健环境验证中展现了出色的性能,诊断视力威胁性DR的敏感度、特异度分别达100%、97.8%[79]。这类系统通常使用单一模态的影像数据,专注于一个具体疾病的高精度识别和诊断。

随着研发的深入,眼科AI诊断系统已能够通过单一模态的影像数据(如OCT图像、眼底照相)来诊断多种眼病[80-83]。例如:基于OCT的AI系统不仅能诊断黄斑变性,还能识别黄斑水肿、视网膜脱离等其他眼底疾病[84-89];类似地,眼底多病种AI诊断系统CARE可以基于眼底照相诊断多种眼底病,在内部验证中达到0.955的受试者曲线下面积(AUC)值,并在外部社区医院验证数据中达到了0.983的AUC值[30]。这种单模态多病种的诊断方式提高了诊断设备的效率和利用率,为医生提供了更全面的疾病诊断信息。

然而,由于单一模态数据信息的有限性,基于单一模态的眼科AI诊断系统在诊断更为复杂的眼病(如青光眼)时,性能表现有限[90]。为此,研究人员在大样本OCT-视野配对数据上验证了双模态诊断算法相比单模态算法的优越性。基于OCT和视野的互补性,双模态算法能更好地区分正常人与其他非青光眼眼病患者,完美模拟青光眼真实世界诊断流程,AUC达到0.943,优于任一单一模态[91]

眼科AI诊断系统的发展不仅在提升眼病诊断效率和准确度上取得了突破,而且正在从单纯的眼病分类诊断向疾病分级、分型、分期以及疾病转归预测等更广阔的方向发展[74,76-77,80-81,92-96]。通过采用数据融合深度学习网络对不同类型及模态的影像数据进行综合分析,可以实现疾病的精准分级分期诊断。例如,结合眼底彩照和OCT/眼底荧光素血管造影图像,已成功实现了年龄相关性黄斑变性(AMD)等眼底疾病的精准分类诊断[51,59,81,97-99]。在青光眼的分型诊断中,电子房角镜系统能通过分析三维前节OCT的影像特征,智能区分宽角和窄角以定性诊断青光眼类型,更突破了该领域的技术瓶颈,实现周边虹膜前粘连的程度和范围识别,从而辅助医生进行更精准的治疗规划[29]

在人体中,眼球是唯一能直接观测终末血管的器官[100]。基于眼底影像,AI还能实现心血管疾病、糖尿病、肾病等全身性疾病的辅助诊断[101-102]。这些应用场景为眼科智能医学的发展开辟了全新的前景,并为全科医学大规模健康管理提供了重要的临床应用价值。

1.2 优势与潜力

AI技术尤其是深度学习技术的迅速进步,极大地增强了对眼科影像数据的分析效能。当前的眼科AI诊断系统能够高效地处理大量影像数据,提供高精度的疾病诊断,显著加快了眼病诊断的速度并提高了准确性[7,10 -11],并已被证明在眼病的筛查中具备经济可行性[103]。这些系统能够迅速识别并精确分析眼底照片、OCT图像等中的微小病变,例如DR、青光眼、AMD等[29,79,91,97,102,104-107]。与传统的诊断方法相比,AI技术的应用显著降低了误诊和漏诊的风险,同时促进了疾病的早期发现和预防[61,63,69,71,108-111]。早期诊断不仅可以提高治疗的有效性,还能显著降低治疗成本,减轻患者负担。

随着智能手机、互联网、云端、大数据和AI技术的融合进步,智能手机有望成为医疗诊断成像的多功能平台,为远程眼科AI诊断提供新的突破口[112]。通过强大的图像处理和分析技术,以及网络的便捷数据存储与传输功能,基于智能手机的眼底照相适用于大多数需要眼底检查的情况[113-114]。结合智能手机软件中的自动化辅助诊断技术或云平台,基于智能手机的眼底照相能够为那些难以获得专业眼科护理设施的人群提供远程医疗服务[113],且已在青光眼、社区视网膜疾病、早产儿视网膜病变、儿童眼病等的早期诊断以及农村地区的初级眼护理方面已经取得了初步应用[94,108,115-118]

LLM为眼科AI诊断进一步拓宽了应用领域。目前,这些模型已初步应用于青光眼、神经眼科疾病、葡萄膜炎、角膜病等的诊断,并展现出了卓越的性能[119-123]。LLM凭借其大规模参数和复杂的计算结构,经过大量文本数据的训练,在理解和生成自然语言方面表现出色[34,37]。这些模型具有强大的数据拟合能力,能够从海量数据中提取有用信息,为各种复杂临床眼病提供更精准、高效的诊断[33-34,37]。此外,LLM能够整合多模态、多类型的眼科数据,实现人机交互、智能问答等应用,从而进一步提升眼科AI诊断的泛化性和可解释性[16,33-34]

2 眼科AI诊断系统在实际应用中的主要挑战

2.1 数据挑战

眼科AI诊断系统在实际应用中面临的主要挑战之一是数据难题,包括数据获取困难、数据标注的高成本和复杂性。此外,高质量、精标注的公共眼科数据集的匮乏也是一大挑战。

2.1.1 高质量数据的获取难题

AI模型性能高度依赖丰富、精确的高质量数据输入。然而,在眼科AI诊断系统的实际应用中,获取高质量数据存在诸多困难[124-126]。在眼科AI领域,高质量的训练数据集应该能够反映疾病表型的所有可能变化,包括不同分期(严重程度)、种族、伪影、不同设备类型及临床表现的混杂变化等[127]。不同品牌和型号的眼科影像设备(如OCT和眼底照相机)在技术规格上存在显著差异,导致从不同设备获得的数据质量和格式不一致,影响数据的一致性和可比性[43,128]。此外,图像质量常受多种因素(如伪影、患者配合程度、眼动、设备操作等)影响,导致数据的准确性和可靠性欠佳[129-133]。另外,眼科数据通常聚焦于特定种族或人群,缺乏广泛的人口覆盖,限制了研究成果的普适性和泛化性[134-135]。此外,不同阶段的疾病数据也往往不均衡,使得疾病诊断模型无法全面反映各阶段病情的真实需求[136]。患者隐私保护也是数据获取困难的一个重要因素,医疗数据的收集和处理必须遵守相关法律规定,确保患者信息的保密性和安全性[64,137-138]

2.1.2 数据标注的高成本性和复杂性

精细标注眼科影像数据是一项繁重且耗时的工作,涉及诊断标签的标注以及疾病特征的标注。诊断标签的标注涉及区分影像数据是否可以定义为某一类型疾病,例如青光眼或非青光眼,DR变或AMD[72,80,82-83,88-89,131,139-147]。疾病特征的标注则更为复杂,例如:对于眼底照相,针对不同疾病的诊断需求,常需要标注视盘、视杯和微血管瘤、出血点等多处细节[77,85,93,147-152];对于OCT图像,则可能需要分割各层视网膜分层以计算各层厚度等[153-154]。由于这些标注本身具有一定的主观性和技术要求,标注通常需要专业的医生,尤其是经验丰富的眼科医生,由于专业的眼科医生本就缺乏,导致这些标注工作尤为耗时且成本高。数据标注的复杂性不仅仅是成本问题,还涉及医疗影像本身的多样性和复杂性。例如:不同的成像设备(如OCT、眼底照相机)可能产生的具备较大差异的图像,即使是经验丰富的医生在进行标注时也可能存在解读差异[25,155];病变的发展阶段、疾病的严重程度及其在图像上的表现形式也增加了标注的难度。标注的不一致性和主观性可能进一步影响训练数据的质量,从而影响AI模型的性能[16,97]

2.1.3 高质量、精标注公共数据集的缺乏

创建公共数据集是克服数据获取难题和缓解数据短缺的一种有效途径。然而,目前仍然非常缺乏高质量、精标注、标准化的公共数据集[61,156]。构建高质量标准化的数据集是眼科AI辅助诊断系统开发的关键环节。已有的公开眼科数据集普遍存在图像质量参差不齐、数据量小及无统一标注标准等问题[141,157-159]。例如:印度Eyepacs数据集存在失焦、曝光等图像质量问题[160];法国Messidor-1及Messidor-2数据集虽有1200张图像,但2个数据集的DR分级标准却不同[161];大多数青光眼公共数据集数据量只有数十到数百,且普遍缺乏详细的视盘、视杯标注,来源设备也普遍单一,目前仅有青光眼领域有REFUGE数据集同时保障了数据量充足、视盘、视杯详细标注、青光眼分类基于临床诊断、纳入不同影像设备等特征[159,162]。不同医疗机构的设备、规范和数据存储模式的差异,导致了眼科AI辅助诊断系统的性能受限。同时,机构之间缺乏完善的数据共享机制,使得数据难以匹配和共享,进而限制了大规模数据集的构建。目前,不少眼科诊断模型尤其是DR领域的大多数AI系统基于的是公共数据集,这些数据集训练得到的算法,可能会不适用于疾病背景和严重程度明显不同的临床环境,这可能为眼科AI诊断系统的真实应用带来局限性[9,163]

2.2 技术挑战

眼科AI诊断系统在真实世界应用中面临的技术挑战主要包括AI诊断模型的准确性和可靠性限制、设备与软件的兼容性问题、图像质量的影响,以及医患对AI诊断技术的信任问题。

2.2.1 AI诊断模型的准确性和可靠性问题

由于眼病的复杂性,同个眼病的多种AI诊断模型可能采用不同的诊断标准,导致其通用性和可比性受限[8]。不同研究机构和团队可能采用不同的诊断标准和数据集,这导致了AI诊断模型的通用性和可比性受限[9,55]。例如,在青光眼的AI诊断中,不同的研究可能侧重不同的视盘特征,从神经纤维层损伤到视盘边缘的形态变化,这些差异使得模型在多样的临床情景中的应用复杂化[130,164-165]。因此,建立统一的诊断标准和数据集是提高AI诊断模型准确性和可靠性的关键。

尽管诸多眼科AI诊断模型展现了出色的性能表现,但大多模型仅基于小样本数据训练,缺乏真实世界的验证,导致在实际应用中的准确性较低[25]。此外,现有模型普遍缺乏外部验证,其泛化性和可靠性有待提高[16,166]。目前眼科AI诊断模型大都缺乏前瞻性数据进行训练,而在真实世界应用中,疾病患病率、疾病的人群分布及图像质量的差异等多种因素都有可能造成模型性能降低。因此,在模型正式投入真实世界进行应用前,都有必要接受严格的前瞻性研究以检验其效能。

2.2.2 设备与图像质量的限制问题

眼科AI诊断模型在临床应用中面临跨设备与软件的兼容性问题[167]。不同眼科检查设备的仪器标准各异,导致AI模型往往仅针对特定机型数据进行训练,限制了其通用性和适用性[107,168]。眼科成像设备在扫描范围、分辨率、型号等存在诸多差异,例如OCT检测存在30°、40°、45°多种扫描范围及不同设备型号(Topcon、Zeiss、Heidelberg等)[169]。在面对五花八门的设备时,需要针对新设备重新训练AI模型而无法直接应用,间接导致后续开发成本大幅度提高[6,9,15]。同时,眼科AI诊断尚缺乏可实现数据实时、全面共享,以及可协调大数据存储库的通用眼科信息标准[52,170]。这进一步加剧了跨设备与软件的兼容性问题,导致眼科AI诊断模型在不同设备和软件平台上的应用变得复杂和困难。

另一方面,现有绝大部分AI模型基于上级医院中高质量影像数据研发,低质量图像多在数据预处理阶段就被清洗排除[129-133,171],而实际应用场景中,受限于拍摄人员技术和居民配合度,采集的图像数据质量相对较低,导致AI模型性能下降[59,72,98,171-172]

2.2.3 医患对AI诊断技术的信任问题

AI的黑盒子本质是一项核心技术挑战[58]。目前为了实现输入图像的有效可视化,研究者广泛应用梯度加权类激活映射来揭示病灶区域与影像特征,或其他可视化方法(如遮挡测试)以佐证诊断依据[10,31,55,58]。以上方法虽然确实能在一定程度上反映AI诊断模型的病灶定位依据,但缺乏足够清晰的临床决策逻辑内涵,也因此导致模型的可解释性欠佳。

当前眼科AI辅助诊断系统主要依靠单模态影像数据进行诊断,缺乏对多模态数据的综合应用,这使得其诊断结果的临床解释性不足,难以获得医生和患者的广泛认可[7,9]。眼病的发生和发展是一个复杂的过程,涉及多层次的机制,在不同的阶段表现出不同的病理损伤、临床症状和体征[11,15,66]。单纯依靠影像数据难以实现疾病的精准诊断,因为它无法包含所需的全面信息。例如,干眼作为一种累及广泛人群的常见眼表疾病,患者多表现出眼干、眼涩、眼红等主观不适症状,通过眼科常规检查可以发现患者的泪膜的稳定性降低,还涉及角膜上皮、角膜神经、眼表血管等多种组织的变化,仅基于单一的影像数据无法实现干眼的精准分型和分级[73,173]。真实世界的临床诊断需要综合分析患者病史、症状、体征及相关的多模态影像数据、临床指标、组学信息等多模态数据。然而,现有眼科智能诊断模型大都仅仅聚焦于单模态影像数据,缺乏对多模态数据的有机整合与分析,难以建立符合临床诊断逻辑的决策模型[7,25]

2.2.4 AI模型对医疗工作流程的适应性问题

眼科AI诊断模型在临床应用中还需适应、融入医疗工作流程。这一难题被称为“人工智能鸿沟”,也是机器学习流程中最难实现的一环[174]。目前,眼科AI诊断模型主要是作为独立的评估工具来使用的,难以被有效地整合到医疗工作流程中。例如,AI诊断模型可能会输出一个诊断结果,该结果可用于独立诊断还是辅助医生诊断、如何被医生接受和验证、如何整合到电子病历中、如何被用于指导治疗决策等问题都亟待探索。AI模型如何嵌入临床系统、辅助简化临床流程、如何监督和调适AI辅助的诊断流程等问题,都需要进一步明确规范[7,11,13,170]。此外,AI模型需要被设计为可以与医生和其他医疗人员进行有效交互,以在实际应用中得到广泛接受和使用[33-34]。例如,AI诊断模型需要能够提供清晰和直观的结果,需要能够接受医生的反馈和修改,需要能够与其他医疗系统进行集成等[6,33-34,174]

2.3 法规与伦理问题

法规与伦理问题也是眼科AI诊断系统在真实世界应用中的重要障碍,涉及责任归属、隐私保护等一系列复杂伦理问题。目前,针对AI模型的相关应用及监管政策仍非常缺乏。

2.3.1 AI应用的伦理困境

眼科AI诊断模型通常涉及临床来源的大数据。若医疗AI的应用过程中缺乏适当的专业指导或合理的知识性引导,可能会带来不良结果,例如误诊、漏诊或引发医学AI应用过程中的相关伦理问题[13,64]。另外,目前的AI系统无法保障其在临床诊疗以及患者护理中的安全性、有效性及公平公正性,且算法本身缺乏透明度;相关监管部门也缺乏对医学AI系统进行审查监督的统一标准[13,96,175]。这些潜在风险导致一系列伦理担忧,例如医疗事故责任由谁承担、医疗纠纷由谁解决等。

AI系统的训练过程中,大量医疗数据的需求构成了对隐私和数据保护的严峻挑战。在眼科领域,这一问题尤为敏感和复杂,因为虹膜和视网膜信息可以作为个人的身份识别标志,而且年龄、性别等生物学特征也能通过深度学习技术从眼底彩照中推断出来。这些因素使得对个人隐私的保护变得更为重要和复杂[43,176]。尽管训练AI算法时使用的患者数据并非全部保存在系统内部,但不当的数据处理和共享仍有可能导致患者隐私泄露[64-65,175]。目前,关于个人数据该如何在深度学习系统训练过程中被储存及使用的问题仍有待解决。因此,为了促进眼科AI技术在真实临床环境中的应用,亟需开发更先进的数据保护和安全措施[177]

2.3.2 法律规定与监管机制的缺乏

目前全球范围内尚缺乏针对AI应用及监管的详细规范和政策[13,174,178-179]。国家药品监督管理局(CDE)对AI算法的应用有严格的评估和审批要求,它们被归类为Ⅲ类医疗器械,只有通过评估和审批才能应用于临床[13]。相关部门应完善AI相关医疗器械的审批制度和流程,并将AI系统的前瞻性真实世界验证试验作为监管审批的核心指标[179-180]。同时,需要制定AI领域的规范和政策,确保有效的数据管理,并通过透明度和适当的安全标准使人们对AI产生信任,以确保AI在眼科临床实践中安全、可靠的发展[181]

在数据保护方面,使用患者健康数据时需要严格遵守相关法律。在欧洲,《通用数据保护条例》(GDPR)、《网络安全指令》和《医疗器械条例》等条例的发布,都旨在加强患者数据的保护[181-183]。尽管GDPR的框架并非专为AI设计,但其许多组件已在实践中应用于AI领域[182-183]。在美国,《健康保险流通与责任法案》(HIPAA)为医疗数据的保密性提供了广泛的解决方案,与欧洲的立法相呼应[184]。我国也出台了《人工智能辅助诊断技术管理规范(试行)》等一系列政策,以规范医疗AI技术的研发与应用[185]。然而,目前尚缺乏全面涵盖AI诊断体系全流程的法律法规,未来的法律法规制定仍需进一步完善。因此,迫切需要快速制定AI领域的规范和政策,确保有效的数据管理,并通过提升透明度和安全标准,建立公众对AI的信任,从而确保AI技术在眼科临床实践中得以平稳、安全的发展和应用。

3 眼科AI临床应用困境的潜在解决方案

3.1 数据管理

3.1.1 建立数据共享模式

为推进眼科智能诊断系统的发展,建立标准化信息共享的眼科数据平台至关重要。数据的数量、质量和多样性直接影响模型性能。已有研究表明,数据共享可以显著提高AI模型的性能和泛化能力,通过整合多中心数据构建共享平台,不仅能改善诊断模型在不同人群中的表现,还能有效减少训练过程中的偏差[186]。尽管数据共享面临监管和隐私挑战,但区块链技术、分布式数据存储和加密算法等手段可以确保数据的安全性和隐私性。例如,“数字面具”技术可在保护隐私的同时提取数据特征[177]

数据共享方法可以分为传统方法和创新方法2大类[187]。在传统的数据共享方法中,开放共享数据集和基于数据使用协议的共享方法被广泛应用,以目前国际最大眼科多病种多模态精标数据集iChallenge为例(https://ichallenge.grand-challenges.org),该数据集包含多种眼科疾病和影像模态,促进了研究的透明度和创新性,为眼科AI领域贡献了众多算法开发思路,极大地促进了国际眼科影像与智能诊断领域的发展[162,188-191]。然而,传统共享方法存在数据碎片化和无法充分保护隐私等问题。创新方法(如可信研究环境(TRE)和联邦学习)通过减少或消除数据传输过程中的隐私风险,提高了数据安全性和隐私保护。联邦学习已应用于DR、青光眼以及AMD、早产儿视网膜病变的诊断等领域[186]。该方法通过在各数据持有者的本地系统内训练模型,有效避免了数据集中存储的风险,从而更好地保护了数据隐私[9,187]

3.1.2 生成数据与强化学习技术

面对数据获取的挑战,生成数据和强化学习技术提供了创新解决方案。生成数据技术通过生成模拟的眼科影像或文本数据,补充真实数据的不足。生成对抗网络(GANs)在创建高质量生成眼科图像方面表现优异。GAN包括生成器和鉴别器。通过反复迭代训练,生成器不断提高生成类似真实数据的能力,最终生成的图像在AI模型训练中表现出与真实数据相近的效果。例如,StyleGAN和StyleGAN2已经成功生成高分辨率彩色眼底图,且眼科医生在视觉图灵测试中难以区分这些生成图像与真实图像[192]。生成数据的多样性对于AI算法在不同场景和患者群体中的表现至关重要,且成本效益显著,避免了真实数据的隐私和道德法规。然而,生成数据在细节上的准确性可能不足,尤其在反映微小病变时,这可能会影响模型的诊断精度。

基于扩散模型的生成式AI在眼科中的应用取得了显著进展,表现出优于传统GAN的潜力[193-194]。扩散模型通过逐步去噪生成高质量的合成图像,其鲁棒性和灵活性使其在处理眼科复杂图像时表现突出[193,195-196]。研究显示,利用扩散模型生成的眼底图像在真实感和细节处理上优于部分GAN方法,尤其是在视网膜血管和视盘结构的表现上更为精准[197]。这种生成能力对于小样本数据集尤为重要。通过逐步去噪和迭代优化,扩散模型能够在有限数据条件下生成分辨率较高的图像,为眼科图像的研究和临床应用提供了新的可能[197-198]。此外,与传统GAN相比,扩散模型在避免模式坍缩方面表现更好,能够生成更具多样性的合成图像,满足眼科教育、疾病筛查和研究模拟的需求[195,197-198]。然而,由于扩散模型的计算需求较高,其在高分辨率图像生成中的应用仍然受到限制,未来需要通过优化算法和提高计算资源利用效率来解决这些问题。

强化学习技术在眼科AI中显示出巨大潜力,尤其在数据获取和标签成本高昂的环境下。强化学习通过模拟与环境的互动,以较少的标记数据逐步优化模型性能。在眼科图像分割中,强化学习比传统监督学习需要更少的像素级标签,显著减少了标签成本。例如,强化学习模型在自动驾驶中需要的标签数据量比传统模型少30%,同样适用于基于图像的眼科AI诊断。这种技术进步有助于推进AI在临床实践中的应用,优化诊断效率并降低成本[199]

3.2 技术创新

3.2.1 开发更高效的算法与模型

在眼科AI的真实世界诊断中,开发更高效的算法和模型是解决技术挑战和实现创新的关键。首先,高效的算法能显著提高诊断准确性和处理速度,尤其在眼科图像分析中,深度学习模型(如卷积神经网络)已成功应用于DR、青光眼、AMD的自动诊断中。为了增强模型在不同数据分布下的泛化能力,创新的模型架构(如Transformer)比传统CNN更擅长处理多模态数据,捕捉全局信息和长距离依赖关系,提高复杂临床环境中的诊断精度[200]。基于Transformer的基础模型,通过大数据和自监督学习,无需重新训练即可处理多种任务和复杂医学数据(如影像和电子健康记录)。这些高效模型在眼科AI中通过少量数据微调,提高了诊断的准确性和鲁棒性,支持多种类型眼科疾病的精准分析,推动了AI在真实临床环境中的广泛应用和创新解决方案。此外,新兴的模型还应该体现在优化计算效率和实时性方面,例如,MobileNet和EfficientNet等轻量级模型在减少计算资源需求的同时保持了高精度,适用于计算资源受限的场景,如移动设备等应用环境[201]

提高AI模型的可解释性是增强其临床应用的重要方向。Grad-CAM等技术[202]是当前实现AI模型可解释性的主流方式,医护人员可以直观地理解AI的诊断过程,从而增加对AI决策的信任度。分层深度学习方法是提高可解释性的另一途径。它采用多个决策模块共同确定最终诊断,包括预诊断模块、图像分割模块、特征提取模块和最终决策模块,中间预测过程可以可视化,可以从眼底图像中提取解剖特征,从而理解AI诊断决策的全过程[203]。进一步地,检查决策边界为理解AI模型如何做出决策提供了宝贵的见解。决策边界在输入空间中分隔不同类的边界,理解这一边界可以揭示模型如何区分不同类别,以及哪些特征对于决策至关重要[204]。LLM的兴起全面革新了AI领域,进一步提高了AI在临床应用中的可解释性和可及性。AI能够通过自然语言与医生和患者进行交互,提供精准诊断的同时解释其诊断依据及推理思路,从而增加医患对AI的信任[205]。总之,开发高效的算法与模型不仅要在性能上提升眼科AI的诊断水平,还需要在临床实用性、实时性和可解释性上实现创新,为未来眼科AI的应用奠定基础。

3.2.2 提升设备的智能化水平

眼科AI的推广不仅依赖于软件创新,还需要硬件设备的智能化提升。AI系统可以嵌入眼科成像设备中,无需过多的专业知识即可进行图像高质量采集与分析(例如便携式眼底照相机和智能手机),结合云、物联网和智能手机技术的远程医疗工具,实时采集和分析生理数据,提供精准的诊断支持[27]。现实中,全自动免散瞳视网膜照相机在门诊的使用证明了自动化筛查工具的可行性,操作员只需接受简单培训,这些设备无需高技能人员操作复杂设备即可实现大规模筛查,简化了眼科图像采集与分级流程[206]

同时,可穿戴设备在医疗AI应用中扮演着至关重要的角色。这些设备能够实时监控患者的生理数据,为AI算法提供丰富的数据输入[207]。通过分析这些数据,AI系统能更精准地预测疾病风险,监测疾病进展,并提出个性化治疗建议。可穿戴设备的远程监测和智能诊断功能也有助于减轻医疗资源压力,提升服务效率[208]。眼科AI的未来将依赖于设备智能化与临床需求的深度融合。通过不断优化设备性能和整合新兴技术,眼科AI诊断设备的诊断效率和准确性将得到提升,从而推动医疗技术的进步并实现医疗公平。因此,推进设备的智能化升级是实现AI赋能眼科诊疗创新的关键。

3.3 教育与培训

3.3.1 加强医务人员对AI技术的理解

目前,医务人员群体中AI人才的储备明显不足,特别是在眼科AI领域,专业培训资源的分布不均衡,且相对匮乏。因此,培养跨学科的“眼科+AI”复合型医疗人才已成为医学教育领域的核心议题[209-211]。为了提升医务人员在眼科诊断中应用AI的技能,需采取多维度策略,包括基础理论与技术培训、提供模拟环境和实际操作机会,以及确保持续的学习和知识更新。医疗机构需要定期组织医务人员参加线上或线下课程,加强对AI核心概念的认识,同时应提供模拟环境和实际操作机会,使医务人员在实践中熟练掌握AI技术。结合临床应用和案例分析,邀请专家进行现场讲解,详细阐述AI技术在眼科诊断中的具体应用场景。通过对成功与失败的经典案例进行分析,帮助医务人员更加深刻地理解AI技术的优势和局限性。此外,持续的学习和知识更新是关键。各地区应加快构建高水平、多层次、宽领域的智能眼科学术交流平台,鼓励国内外学术交流与合作,使更多医务人员能够站在行业前沿,充分吸收和借鉴全球最先进的眼科技术和治疗经验,掌握最新的AI技术进展和转化[212]

3.3.2 提高患者对AI技术的认知与接受度

患者对AI技术的接受度对于其成功应用至关重要。为了提升患者对AI技术的接受度,医疗机构和技术提供商需要采取多元化的策略。首先,有效的科普宣传和教育是关键。通过向患者清晰地介绍AI技术的能力、优势和潜在风险,可以显著减少他们对新技术的疑虑,帮助他们更好地理解和接受这些技术。其次,保持透明沟通同样重要。向患者解释AI技术是如何处理他们的数据的,以及采取了哪些措施保护他们的隐私和数据安全,可以缓解他们的担忧,增强对AI技术辅助诊断的信任[213]。此外,提供个性化的体验也很关键。通过展示成功案例和临床验证数据,证明AI技术在实际应用中的有效性和安全性,可以进一步增加患者的信任。最后,尊重患者的选择至关重要。对于那些不愿使用AI技术的患者,应提供替代方案,确保他们仍能获得高质量的医疗服务。通过这些措施,可以有效地提高患者对AI技术的接受度,推动其在医疗领域的广泛应用。

3.4 政策与标准

3.4.1 制定相关法律法规与行业标准

为确保AI在医疗领域的持续可控发展,有必要制定和完善针对AI技术应用及监管的法律法规,以对科研人员、临床医生和医疗保健消费者形成正确的引导。欧盟率先于2024年5月通过《人工智能法案》,该法案根据是否独立于产品将AI系统分成2大类,并针对AI应用的风险进行了统一分级[214-215]。2024年7月,美国、英国及欧盟的监管机构发表了联合声明,就加强对AI监管达成共识,承诺共同维护一个公正开放的AI市场环境[216]。我国现行法律已经涵盖AI领域的部分内容,例如《生成式人工智能服务管理暂行办法》就把监管重点投放在生成式AI。在立法规划上,我国政府已将AI立法列入议程,将逐步建立和完善AI相关的法律法规、伦理规范以及政策体系[214]。从长远来看,AI立法需要注意硬实力与软实力相协调,国内与国际相接轨,从已有法律法规出发不断完善AI的风险规制,同时也为AI技术的发展存留足够的探索空间。

智能眼科领域行业标准化不足,政府和相关标准化组织亟需加快确立技术标准,以提高AI技术在影像分析、疾病诊断和治疗建议等方面的准确性和可靠性。针对涉及AI干预的临床随机试验,Liu等[217]和Rivera等[218]依据循证医学证据提出包含待解决最少项目集的实践指南,有利于改善AI临床试验报告的透明度和完整性,帮助评审者和广大读者更好地理解和评估试验结果。我国也在积极推动相关行业标准的制定。针对眼底彩照数据标注缺乏规范的问题,中山大学中山眼科中心牵头国内多家单位制定了《眼底彩照标注与质量控制规范》[219];针对眼科公开图像数据库可见性和可用性不足的问题,我国整理出台了《全球眼科图像公开数据库使用指南》[220]。此外,我国还发布了《人工智能在眼前节疾病诊断中的应用指南》《眼科人工智能临床研究评价指南》等多篇AI临床研究相关指南[221-222]。这些指南的制定和完善,将为眼科AI技术的应用提供宝贵的指导意见,进而推动智能眼科的临床转化向更加标准化和规范化的方向发展。

3.4.2 加强跨学科合作,推动政策制定

在学科深度交叉融合的语境下,智能医学的战略地位日益趋升。对于眼科AI诊断,跨学科合作的意义在于,充分发挥不同学科领域的优势,共同攻克现有技术难题,优化眼科疾病诊疗流程,为患者提供更高质量的医疗服务。因此,只有进一步加强跨学科合作,推动智能医学政策体系构建,才能为眼科AI发展持续赋能。

我国智能医学得以快速发展,与政策的持续演进密不可分。应AI技术人才需求,教育部于2018年推出《高等学校人工智能创新行动计划》,支持高校设立AI学科,培养具备“AI+X”复合能力的专业人才。2020年9月,国务院再次发文,加快医学教育创新步伐,推动医学与其他学科的融合[211,223]。面对AI产生的社会伦理问题,国家新一代AI治理专业委员提出发展“负责任的人工智能”理念,梳理了AI治理框架和实践指南,以筑牢AI可控发展的坚强后盾[224]。在眼科AI层面,“十四五”眼健康规划强调了信息化平台建设的重要性,提出深化AI技术与眼科服务的融合,并建立病例数据库支撑眼病研究[225]

欧盟国家在AI政策制定上侧重于立法和监管层面,以保障个人数据安全;美国则强调在安全治理的同时保持AI的技术领先优势[226]。在逐步融入AI全球治理体系的过程中,我们需要思考如何因地制宜地推进具有本土特色的治理方案与政策,加强发展政策与监管政策的目标协同,适应AI技术的快速发展和市场环境的不断变化,推动眼科AI技术的繁荣与发展[227]

4 未来展望

综上,数据管理、技术创新、教育与培训,以及政策与标准的完善,均是眼科AI实现突破的关键路径,这些潜在的解决方案将塑造智能医疗的全新格局。在数据管理方面,跨机构的数据共享模式、联邦学习和生成数据与强化学习技术将有助于在保护隐私的同时实现数据整合,为模型的泛化能力奠定坚实基础;在技术创新方面,高效的算法开发及可解释性技术的进步,将进一步提高AI的实用性。此外,加强医务人员与患者对AI技术的理解与接受度,以及推进政策与标准的完善,也将在推动眼科AI的规范化应用中扮演重要角色。这些努力不仅是对当前研究工作的深化,也是为未来探索奠定坚实基础的必要举措。

近年来,以ChatGPT为代表的LLM在全球范围内迅速发展,为眼科领域带来了新机遇[205,228]。LLM在医疗文书自动化处理中的应用尤为突出,可大幅缓解眼科医生的文书负担。传统医学文书的书写费时费力[229],而LLM具备理解自然语言并高效处理文本的能力。通过简单的语音或关键字输入,LLM可生成格式规范、内容精准的医疗文书,包括病历、术后记录和出入院小结。例如,Nuance Communications推出的Dragon Ambient Experience Express系统能够理解患者与医生之间的对话,提取关键信息,生成医疗记录并整合至信息系统中,从而提升临床工作效率。

同时,LLM有望成为辅助诊断的重要工具。结合最新医疗指南和研究,LLM可提供解释性诊断建议,提高AI诊断疾病的可解释性。世界卫生组织(WHO)的一份关于生成式AI在医学应用主题报告[230]显示,诊断是LLM具体应用的有前途的领域。现实世界的诊断涉及交互的过程,需要对疾病进行不断推理。临床医生根据从动态临床环境中获得的各种临床信息来权衡不同的诊断可能性。而LLM正是通过促进互动,以对话为导向,动态收集信息,不断汇总诊断依据,获得准确的鉴别诊断,这种能力远超一次性输入和静态分析的传统AI模式。在初级保健中心,LLM可补充医生常规诊断,确保常见病不被忽视。同样,在三级医院和专科中心,其互动性有助于探索复杂病例或罕见情况的可能性。未来,LLM将能够整合多模态数据,包括眼科图像、电子健康记录和患者报告等,其跨模态信息融合能力将助力复杂眼科疾病的精准诊断,提供更全面的判断支持。

以患者为中心的沟通在眼科保健中具有重要意义,而LLM的出现正在改变公共医疗保健的格局。传统的眼科护理服务往往昂贵且耗时,导致医疗资源分配紧张。近年来,医疗聊天机器人作为一种新兴的患者自我评估和健康教育工具,受到了广泛的关注和应用。例如,ChatGPT等LLM工具通过与患者的互动对话以及精准的后续询问,能够高效收集病史并提供分诊建议,类似于分诊护士的作用[166,231 -233]。研究表明,眼病相关的急诊中约有50%为非紧急病例,这些患者大多可在初级保健机构得到有效处理[234-235]。通过寻求LLM的帮助,患者可以获得量身定制的建议,例如,是否需要立即就医,是否可以选择初级保健解决,或是否适合虚拟咨询。这不仅能够有效减少不必要的急诊就诊和长时间的等待,还能显著缓解医疗资源压力,提高患者的满意度和医疗服务效率[33]

除了分诊功能,LLM在患者自我教育方面同样展现了巨大的潜力。对于许多患者来说,医学术语往往晦涩难懂,而LLM可以用简单易懂的语言解释复杂的诊断和建议,帮助患者更好地理解自己的病情[236-237]。在医生因时间有限无法详细解释所有细节时,LLM能够补充说明,例如疾病的常见症状、可能的治疗方案及其优劣点,支持患者做出更知情的决策。此外,LLM还可为患者提供健康教育和康复指导,包括个性化的生活方式调整建议和康复管理方案,从而帮助患者更好地掌控康复过程,加速康复进程。这种支持对于增强患者的自主性、提高医患共同决策质量以及提升治疗效果尤为关键。

尤其是在慢性眼病管理中,LLM展现了更多应用场景。以糖尿病视网膜病变(DR)为例,患者通常需要持续教育和长期管理以控制病情。DeepDR-LLM融合了LLM与深度学习技术,将其应用于糖尿病诊疗流程后,显著改善了新发糖尿病患者的自我管理行为,并提高了DR患者的转诊依从性[238]。这种智能化辅助工具可以帮助患者更直观地了解病情管理的核心要点,例如,定期检查的重要性、饮食与生活方式的调整建议,以及坚持药物治疗的必要性。未来,LLM还能够通过动态监测患者的行为,及时识别早期疾病进展风险并提供个性化指导,促使患者及时转诊,从而进一步优化慢性病管理。

总的来说,未来眼科AI研究应在技术深度和广度上持续探索,尤其是在LLM引领下的应用创新,这不仅对推动智能医学发展具有重要意义,也将为全球医疗技术的革新带来深远影响。

5 结论

经历了数十年的技术沉淀与突破,眼科AI诊断模型已得到了长足发展,随着眼科疾病的日益增多和医疗资源的紧张,眼科AI可以帮助提高诊断准确率、减轻医务人员的负担、改善患者的治疗效果。尤其是在基层医疗机构,眼科AI可以帮助缓解压力,提高偏远、落后地区的医疗水平,做到疾病的早筛查、早防治。对于大型医院和眼科专科医院,AI的辅助诊断亦可极大提升临床医生的工作效率,降低误诊率,提高患者的满意度。因此,眼科AI诊断模型从展现出色的理论性能到实际临床应用的跨越具备重大的现实意义。然而,眼科AI在实际应用中仍然面临着诸多挑战,包括数据问题、技术问题、法规与伦理问题等。这些挑战不仅影响了眼科AI的部署和应用,也制约了其在医疗领域的发展潜力。

因此,解决这些挑战是将眼科AI带向实际临床运用的必要条件。未来,眼科AI的发展需要在数据管理、技术创新、人才教育及培训、政策与标准制定等多个层面上寻求突破。展望未来,LLM的兴起为眼科AI注入了新的动力,推动AI从辅助医生诊疗全流程拓展到患者教育与慢性病管理的全面应用,眼科AI将逐步迈向实际临床情景的落地应用,真正发挥其潜在价值,全面革新眼科诊疗模式。

参考文献

[1]

McCarthy J, Minsky M L, Rochester N, et al. A proposal for the dartmouth summer research project on artificial intelligence, August 31, 1955[J]. AI Magazine, 2006, 27(4): 12-14.

[2]

Xu Y J, Liu X, Cao X, et al. Artificial intelligence: a powerful paradigm for scientific research[J]. The Innovation, 2021, 2(4): 100179.

[3]

Davenport T, Kalakota R. The potential for artificial intelligence in healthcare[J]. Future Healthcare Journal, 2019, 6(2): 94-98.

[4]

Hussain S, Mubeen I, Ullah N, et al. Modern diagnostic imaging technique applications and risk factors in the medical field: a review[J]. BioMed Research International, 2022, 2022(1): 5164970.

[5]

Oren O, Gersh B J, Bhatt D L. Artificial intelligence in medical imaging: switching from radiographic pathological data to clinically meaningful endpoints[J]. The Lancet Digital Health, 2020, 2(9): e486-e488.

[6]

Nicholson Price W 2nd, Gerke S, Glenn Cohen I. Potential liability for physicians using artificial intelligence[J]. JAMA, 2019, 322(18): 1765-1766.

[7]

Franzco S K P, van Wijngaarden PhD Franzco P. The eye in AI: artificial intelligence in ophthalmology[J]. Clinical & Experimental Ophthalmology, 2019, 47(1): 5-6.

[8]

Li J O, Liu H R, Ting D S J, et al. Digital technology, tele-medicine and artificial intelligence in ophthalmology: a global perspective[J]. Progress in Retinal and Eye Research, 2021, 82: 100900.

[9]

Li Z W, Wang L, Wu X F, et al. Artificial intelligence in ophthalmology: the path to the real-world clinic[J]. Cell Reports Medicine, 2023, 4(7): 101095.

[10]

Ting D S W, Pasquale L R, Peng L, et al. Artificial intelligence and deep learning in ophthalmology[J]. British Journal of Ophthalmology, 2019, 103(2): 167-175.

[11]

Ting D S W, Peng L, Varadarajan A V, et al. Deep learning in ophthalmology: the technical and clinical considerations[J]. Progress in Retinal and Eye Research, 2019, 72: 100759.

[12]

Teo Z L, Ting D S W. AI telemedicine screening in ophthalmology: health economic considerations[J]. The Lancet Global Health, 2023, 11(3): e318-e320.

[13]

Abdullah Y I, Schuman J S, Shabsigh R, et al. Ethics of artificial intelligence in medicine and ophthalmology[J]. Asia-Pacific Journal of Ophthalmology, 2021, 10(3): 289-298.

[14]

Sanil J, Jerrome S, Iswarya M, et al. Diagnostic accuracy of artificial intelligence-based automated diabetic retinopathy screening in real-world settings: a systematic review and meta-analysis[J]. American Journal of Ophthalmology, 2024, 263: 214-230.

[15]

Lee C S, Brandt J D, Lee A Y. Big data and artificial intelligence in ophthalmology, where are we now?[J]. Ophthalmology Science, 2021, 1(2): 100036.

[16]

Zhou Y K, Chia M A, Wagner S K, et al. A foundation model for generalizable disease detection from retinal images[J]. Nature, 2023, 622(7981): 156-163.

[17]

Huang X, Wang H, She C Y, et al. Artificial intelligence promotes the diagnosis and screening of diabetic retinopathy[J]. Frontiers in Endocrinology, 2022, 13: 946915.

[18]

Bs J O, Bs A S, Md J C. Artificial intelligence in ophthalmology: optimization of machine learning for ophthalmic care and research[J]. Clin Exp Ophthalmol, 2021, 49(5): 413-415.

[19]

Schmidt-Erfurth U, Sadeghipour A, Gerendas B S, et al. Artificial intelligence in retina[J]. Progress in Retinal and Eye Research, 2018, 67: 1-29.

[20]

Guymer R, Wu Z C. Age-related macular degeneration (AMD): More than meets the eye. The role of multimodal imaging in today's management of AMD:a review[J]. Clinical & Experimental Ophthalmology, 2020, 48(7): 983-995.

[21]

Liu R, Li Q C, Xu F P, et al. Application of artificial intelligence-based dual-modality analysis combining fundus photography and optical coherence tomography in diabetic retinopathy screening in a community hospital[J]. Biomedical Engineering Online, 2022, 21(1): 47.

[22]

Rajesh A E, Davidson O Q, Lee C S, et al. Artificial intelligence and diabetic retinopathy: AI framework, prospective studies, head-to-head validation, and cost-effectiveness[J]. Diabetes Care, 2023, 46(10): 1728-1739.

[23]

Ahsan M M, Luna S A, Siddique Z. Machine-learning-based disease diagnosis: a comprehensive review[J]. Healthcare (Basel), 2022, 10(3): 541.

[24]

Choi R Y, Coyner A S, Kalpathy-Cramer J, et al. Introduction to machine learning, neural networks, and deep learning[J]. Translational Vision Science & Technology, 2020, 9(2): 14.

[25]

Tong Y, Lu W, Yu Y, et al. Application of machine learning in ophthalmic imaging modalities[J]. Eye Vis (Lond), 2020, 7: 22.

[26]

Alzubaidi L, Zhang J L, Humaidi A J, et al. Review of deep learning: concepts, CNN architectures, challenges, applications, future directions[J]. Journal of Big Data, 2021, 8(1): 53.

[27]

Li F, Song D P, Chen H, et al. Development and clinical deployment of a smartphone-based visual field deep learning system for glaucoma detection[J]. NPJ Digital Medicine, 2020, 3: 123.

[28]

Li F, Su Y D, Lin F B, et al. A deep-learning system predicts glaucoma incidence and progression using retinal photographs[J]. Journal of Clinical Investigation, 2022, 132(11): e157968.

[29]

Li F, Yang Y F, Sun X, et al. Digital gonioscopy based on three-dimensional anterior-segment OCT an international multicenter study[J]. Ophthalmology, 2022, 129(1): 45-53.

[30]

Lin D R, Xiong J H, Liu C X, et al. Application of comprehensive artificial intelligence retinal expert (CARE) system: a national real-world evidence study[J]. The Lancet Digital Health, 2021, 3(8): e486-e495.

[31]

Muntean G A, Marginean A, Groza A, et al. The predictive capabilities of artificial intelligence-based OCT analysis for age-related macular degeneration progression: a systematic review[J]. Diagnostics (Basel), 2023, 13(14): 2464.

[32]

Yoon J, Han J, Ko J, et al. Developing and evaluating an AI-based computer-aided diagnosis system for retinal disease: diagnostic study for central serous chorioretinopathy[J]. Journal of Medical Internet Research, 2023, 25: e48142.

[33]

Betzler B K, Chen H C, Cheng C Y, et al. Large language models and their impact in ophthalmology[J]. The Lancet Digital Health, 2023, 5(12): e917-e924.

[34]

Thirunavukarasu A J, Ting D S J, Elangovan K, et al. Large language models in medicine[J]. Nature Medicine, 2023, 29(8): 1930-1940.

[35]

Singh S, Watson S. ChatGPT as a tool for conducting literature review for dry eye disease[J]. Clinical & Experimental Ophthalmology, 2023, 51(7): 731-732.

[36]

Korot E, Gonçalves M B, Huemer J, et al. Clinician-driven AI: code-free self-training on public data for diabetic retinopathy referral[J]. JAMA Ophthalmology, 2023, 141(11): 1029-1036.

[37]

Hu X Y, Ran A R, Nguyen T X, et al. What can GPT-4 do for diagnosing rare eye diseases? A pilot study[J]. Ophthalmology and Therapy, 2023, 12(6): 3395-3402.

[38]

Yuan A, Lee A Y. Artificial intelligence deployment in diabetic retinopathy: the last step of the translation continuum[J]. The Lancet Digital Health, 2022, 4(4): e208-e209.

[39]

Nakayama L F, Zago Ribeiro L, Novaes F, et al. Artificial intelligence for telemedicine diabetic retinopathy screening: a review[J]. Annals of Medicine, 2023, 55(2): 2258149.

[40]

Shahriari M H, Sabbaghi H, Asadi F, et al. Artificial intelligence in screening, diagnosis, and classification of diabetic macular edema: a systematic review[J]. Survey of Ophthalmology, 2023, 68(1): 42-53.

[41]

Zhao X Y, Lin Z Z, Yu S S, et al. An artificial intelligence system for the whole process from diagnosis to treatment suggestion of ischemic retinal diseases[J]. Cell Reports Medicine, 2023, 4(10): 101197.

[42]

Chou Y B, Kale A U, Lanzetta P, et al. Current status and practical considerations of artificial intelligence use in screening and diagnosing retinal diseases: Vision Academy retinal expert consensus[J]. Current Opinion in Ophthalmology, 2023, 34(5): 403-413.

[43]

Zapata M A, Royo-Fibla D, Font O, et al. Artificial intelligence to identify retinal fundus images, quality validation, laterality evaluation, macular degeneration, and suspected glaucoma[J]. Clinical Ophthalmology, 2020, 14: 419-429.

[44]

Bellemo V, Lim Z W, Lim G, et al. Artificial intelligence using deep learning to screen for referable and vision-threatening diabetic retinopathy in Africa: a clinical validation study[J]. The Lancet Digital Health, 2019, 1(1): e35-e44.

[45]

Wong T Y, Bressler N M. Artificial intelligence with deep learning technology looks into diabetic retinopathy screening[J]. JAMA, 2016, 316(22): 2366-2367.

[46]

Hwang D K, Hsu C C, Chang K J, et al. Artificial intelligence-based decision-making for age-related macular degeneration[J]. Theranostics, 2019, 9(1): 232-245.

[47]

Gu C F, Wang Y J, Jiang Y, et al. Application of artificial intelligence system for screening multiple fundus diseases in Chinese primary healthcare settings: a real-world, multicentre and cross-sectional study of 4795 cases[J]. British Journal of Ophthalmology, 2024, 108(3): 424-431.

[48]

Guan Z Y, Li H T, Liu R H, et al. Artificial intelligence in diabetes management: advancements, opportunities, and challenges[J]. Cell Reports Medicine, 2023, 4(10): 101213.

[49]

Padmanabhan S, Tran T Q B, Dominiczak A F. Artificial intelligence in hypertension: seeing through a glass darkly[J]. Circulation Research, 2021, 128(7): 1100-1118.

[50]

Ellahham S. Artificial intelligence: the future for diabetes care[J]. The American Journal of Medicine, 2020, 133(8): 895-900.

[51]

Antaki F, Chopra R, Keane P A. Vision-language models for feature detection of macular diseases on optical coherence tomography[J]. JAMA Ophthalmology, 2024, 142(6): 573-576.

[52]

Dow E R, Keenan T D L, Lad E M, et al. From data to deployment the collaborative community on ophthalmic imaging roadmap for artificial intelligence in age-related macular degeneration[J]. Ophthalmology, 2022, 129(5): e43-e59.

[53]

Halfpenny W, Baxter S L. Towards effective data sharing in ophthalmology: data standardization and data privacy[J]. Current Opinion in Ophthalmology, 2022, 33(5): 418-424.

[54]

Moss H E, Joslin C E, Rubin D S, et al. Big data research in neuro-ophthalmology: promises and pitfalls[J]. Journal of Neuro-Ophthalmology, 2019, 39(4): 480-486.

[55]

Ahuja A S, Wagner I V, Dorairaj S, et al. Artificial intelligence in ophthalmology: a multidisciplinary approach[J]. Integrative Medicine Research, 2022, 11(4): 100888.

[56]

Bajwa J, Munir U, Nori A, et al. Artificial intelligence in healthcare: transforming the practice of medicine[J]. Future Healthcare Journal, 2021, 8(2): e188-e194.

[57]

Feng X R, Xu K Z, Luo M J, et al. Latest developments of generative artificial intelligence and applications in ophthalmology[J]. Asia-Pacific Journal of Ophthalmology, 2024, 13(4): 100090.

[58]

González-Gonzalo C, Thee E F, Klaver C C W, et al. Trustworthy AI: closing the gap between development and integration of AI systems in ophthalmic practice[J]. Progress in Retinal and Eye Research, 2022, 90: 101034.

[59]

Gomez Rossi J, Rojas-Perilla N, Krois J, et al. Cost-effectiveness of artificial intelligence as a decision-support system applied to the detection and grading of melanoma, dental caries, and diabetic retinopathy[J]. JAMA Network Open, 2022, 5(3): e220269.

[60]

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.

[61]

Chen E M, Chen D, Chilakamarri P, et al. Economic challenges of artificial intelligence adoption for diabetic retinopathy[J]. Ophthalmology, 2021, 128(3): 475-477.

[62]

Quellec G, Al Hajj H, Lamard M, et al. ExplAIn: explanatory artificial intelligence for diabetic retinopathy diagnosis[J]. Medical Image Analysis, 2021, 72: 102118.

[63]

Tan W M, Wei Q L, Xing Z, et al. Fairer AI in ophthalmology via implicit fairness learning for mitigating sexism and ageism[J]. Nature Communications, 2024, 15(1): 4750.

[64]

Farhud D D, Zokaei S. Ethical issues of artificial intelligence in medicine and healthcare[J]. Iranian Journal of Public Health, 2021, 50(11): i-v.

[65]

Mennella C, Maniscalco U, De Pietro G, et al. Ethical and regulatory challenges of AI technologies in healthcare: a narrative review[J]. Heliyon, 2024, 10(4): e26297.

[66]

Obuchowicz R, Strzelecki M, Piórkowski A. Clinical applications of artificial intelligence in medical imaging and image processing:a review[J]. Cancers, 2024, 16(10): 1870.

[67]

Ting D S J, Foo V H, Yang L W Y, et al. Artificial intelligence for anterior segment diseases: emerging applications in ophthalmology[J]. British Journal of Ophthalmology, 2021, 105(2): 158-168.

[68]

Han Z K, Yang B, Deng S G, et al. Category weighted network and relation weighted label for diabetic retinopathy screening[J]. Computers in Biology and Medicine, 2023, 152: 106408.

[69]

Wang Y L, Yu M, Hu B J, et al. Deep learning-based detection and stage grading for optimising diagnosis of diabetic retinopathy[J]. Diabetes Metabolism Research and Reviews, 2021, 37(4): e3445.

[70]

Burlina P, Paul W, Alvin Liu T Y, et al. Detecting anomalies in retinal diseases using generative, discriminative, and self-supervised deep learning[J]. JAMA Ophthalmology, 2022, 140(2): 185-189.

[71]

Abdelmotaal H, Hazarbasanov R, Taneri S, et al. Detecting dry eye from ocular surface videos based on deep learning[J]. The Ocular Surface, 2023, 28: 90-98.

[72]

Ting D S W, Cheung C Y, 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(22): 2211-2223.

[73]

Storås A M, Strümke I, Riegler M A, et al. Artificial intelligence in dry eye disease[J]. The Ocular Surface, 2022, 23: 74-86.

[74]

Bo Z H, Guo Y C, Lyu J H, et al. Relay learning: a physically secure framework for clinical multi-site deep learning[J]. NPJ Digital Medicine, 2023, 6(1): 204.

[75]

Dai L, Wu L, Li H T, et al. A deep learning system for detecting diabetic retinopathy across the disease spectrum[J]. Nature Communications, 2021, 12(1): 3242.

[76]

PhD S K, PhD Z L M, PhD J S, et al. Development and validation of a deep-learning algorithm for the detection of neovascular age-related macular degeneration from colour fundus photographs[J]. Clinical & Experimental Ophthalmology, 2019, 47(8): 1009-1018.

[77]

Zago G T, Andreão R V, Dorizzi B, et al. Diabetic retinopathy detection using red lesion localization and convolutional neural networks[J]. Computers in Biology and Medicine, 2020, 116: 103537.

[78]

Abràmoff M D, Lavin P T, Birch M, et al. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices[J]. NPJ Digital Medicine, 2018, 1: 39.

[79]

Verbraak F D, Abramoff M D, Bausch G C F, et al. Diagnostic accuracy of a device for the automated detection of diabetic retinopathy in a primary care setting[J]. Diabetes Care, 2019, 42(4): 651-656.

[80]

Garifullin A, Lensu L, Uusitalo H. Deep Bayesian baseline for segmenting diabetic retinopathy lesions: advances and challenges[J]. Computers in Biology and Medicine, 2021, 136: 104725.

[81]

Alsaih K, Yusoff M Z, Tang T B, et al. Deep learning architectures analysis for age-related macular degeneration segmentation on optical coherence tomography scans[J]. Computer Methods and Programs in Biomedicine, 2020, 195: 105566.

[82]

Asiri N, Hussain M, Al Adel F, et al. Deep learning based computer-aided diagnosis systems for diabetic retinopathy: a survey[J]. Artificial Intelligence in Medicine, 2019, 99: 101701.

[83]

Tsiknakis N, Theodoropoulos D, Manikis G, et al. Deep learning for diabetic retinopathy detection and classification based on fundus images: a review[J]. Computers in Biology and Medicine, 2021, 135: 104599.

[84]

Bai J H, Wan Z Q, Li P, et al. Accuracy and feasibility with AI-assisted OCT in retinal disorder community screening[J]. Frontiers in Cell and Developmental Biology, 2022, 10: 1053483.

[85]

Virgili G, Menchini F, Murro V, et al. Optical coherence tomography (OCT) for detection of macular oedema in patients with diabetic retinopathy[J]. Cochrane Database of Systematic Reviews, 2011(7): CD008081.

[86]

Lam C, Wong Y L, Tang Z Q, et al. Performance of artificial intelligence in detecting diabetic macular edema from fundus photography and optical coherence tomography images: a systematic review and meta-analysis[J]. Diabetes Care, 2024, 47(2): 304-319.

[87]

Wang S R, Zhang Y L, Lei S B, et al. Performance of deep neural network-based artificial intelligence method in diabetic retinopathy screening: a systematic review and meta-analysis of diagnostic test accuracy[J]. European Journal of Endocrinology, 2020, 183(1): 41-49.

[88]

Yim J, Chopra R, Spitz T, et al. Predicting conversion to wet age-related macular degeneration using deep learning[J]. Nature Medicine, 2020, 26(6): 892-899.

[89]

Ajana S, Cougnard-Grégoire A, Colijn J M, et al. Predicting progression to advanced age-related macular degeneration from clinical, genetic, and lifestyle factors using machine learning[J]. Ophthalmology, 2021, 128(4): 587-597.

[90]

Ran A R, Cheung C Y, Wang X, et al. Detection of glaucomatous optic neuropathy with spectral-domain optical coherence tomography: a retrospective training and validation deep-learning analysis[J]. The Lancet Digital Health, 2019, 1(4): e172-e182.

[91]

Xiong J, Li F, Song D P, et al. Multimodal machine learning using visual fields and peripapillary circular OCT scans in detection of glaucomatous optic neuropathy[J]. Ophthalmology, 2022, 129(2): 171-180.

[92]

Yang K, Wang Q, Wu L, et al. Development and verification of a combined diagnostic model for primary Sjögren’s syndrome by integrated bioinformatics analysis and machine learning[J]. Scientific Reports, 2023, 13(1): 8641.

[93]

Oulhadj M, Riffi J, Khodriss C, et al. Diabetic retinopathy prediction based on vision transformer and modified capsule network[J]. Computers in Biology and Medicine, 2024, 175: 108523.

[94]

Young B K, Cole E D, Shah P K, et al. Efficacy of smartphone-based telescreening for retinopathy of prematurity with and without artificial intelligence in India[J]. JAMA Ophthalmology, 2023, 141(6): 582-588.

[95]

Kanagasingam Y, Xiao D, Vignarajan J, et al. Evaluation of artificial intelligence-based grading of diabetic retinopathy in primary care[J]. JAMA Network Open, 2018, 1(5): e182665.

[96]

Abràmoff M D, Cunningham B, Patel B, et al. Foundational considerations for artificial intelligence using ophthalmic images[J]. Ophthalmology, 2022, 129(2): e14-e32.

[97]

Farinha C, Cachulo M L, Coimbra R, et al. Age-related macular degeneration staging by color fundus photography vs. multimodal imaging-epidemiological implications (The Coimbra eye study-report 6)[J]. Journal of Clinical Medicine, 2020, 9(5): 1329.

[98]

Waldstein S M, Vogl W D, Bogunovic H, et al. Characterization of drusen and hyperreflective foci as biomarkers for disease progression in age-related macular degeneration using artificial intelligence in optical coherence tomography[J]. JAMA Ophthalmology, 2020, 138(7): 740-747.

[99]

Burlina P M, Joshi N, Pacheco K D, et al. Use of deep learning for detailed severity characterization and estimation of 5-year risk among patients with age-related macular degeneration[J]. JAMA Ophthalmology, 2018, 136(12): 1359-1366.

[100]

Pemp B, Polska E, Garhofer G, et al. Retinal blood flow in type 1 diabetic patients with no or mild diabetic retinopathy during euglycemic clamp[J]. Diabetes Care, 2010, 33(9): 2038-2042.

[101]

Al-Halafi A M. Applications of artificial intelligence-assisted retinal imaging in systemic diseases: a literature review[J]. Saudi Journal of Ophthalmology, 2023, 37(3): 185-192.

[102]

Li H M, Cao J, Grzybowski A, et al. Diagnosing systemic disorders with AI algorithms based on ocular images[J]. Healthcare, 2023, 11(12): 1739.

[103]

Pietris J, Lam A, Bacchi S, et al. Health economic implications of artificial intelligence implementation for ophthalmology in Australia: a systematic review[J]. Asia-Pacific Journal of Ophthalmology, 2022, 11(6): 554-562.

[104]

Wewetzer L, Held L A, Steinhäuser J. Diagnostic performance of deep-learning-based screening methods for diabetic retinopathy in primary care-a meta-analysis[J]. PLoS One, 2021, 16(8): e0255034.

[105]

Gu B, Sidhu S, Weinreb R N, et al. Review of visualization approaches in deep learning models of glaucoma[J]. Asia-Pacific Journal of Ophthalmology, 2023, 12(4): 392-401.

[106]

Chen D, Ran E A, Tan T F, et al. Applications of artificial intelligence and deep learning in glaucoma[J]. Asia-Pacific Journal of Ophthalmology, 2023, 12(1): 80-93.

[107]

Leong Y Y, Vasseneix C, Finkelstein M T, et al. Artificial intelligence meets neuro-ophthalmology[J]. Asia-Pacific Journal of Ophthalmology, 2022, 11(2): 111-125.

[108]

Hasan S U, Rehman Siddiqui M A. Diagnostic accuracy of smartphone-based artificial intelligence systems for detecting diabetic retinopathy: a systematic review and meta-analysis[J]. Diabetes Research and Clinical Practice, 2023, 205: 110943.

[109]

Wu R F, Chen Z P, Yu J L, et al. A graph-learning based model for automatic diagnosis of Sjögren’s syndrome on digital pathological images: a multicentre cohort study[J]. Journal of Translational Medicine, 2024, 22(1): 748.

[110]

Kuklinski E J, Henry R K, Shah M, et al. Screening of diabetic retinopathy using artificial intelligence and tele-ophthalmology[J]. Journal of Diabetes Science and Technology, 2023, 17(6): 1724-1725.

[111]

Morris A. Technology watch: AI can diagnose diabetic retinopathy[J]. Nature Reviews Endocrinology, 2018, 14(2): 65.

[112]

Iqbal U. Smartphone fundus photography: a narrative review[J]. International Journal of Retina and Vitreous, 2021, 7(1): 44.

[113]

Jin K, Li Y Y, Wu H K, et al. Integration of smartphone technology and artificial intelligence for advanced ophthalmic care: a systematic review[J]. Advances in Ophthalmology Practice and Research, 2024, 4(3): 120-127.

[114]

Nazari Khanamiri H, Nakatsuka A, El-Annan J. Smartphone fundus photography[J]. Journal of Visualized Experiments, 2017,125:e55958.

[115]

Natarajan S, Jain A, Krishnan R, et al. Diagnostic accuracy of community-based diabetic retinopathy screening with an offline artificial intelligence system on a smartphone[J]. JAMA Ophthalmology, 2019, 137(10): 1182-1188.

[116]

Patel T P, Aaberg M T, Paulus Y M, et al. Smartphone-based fundus photography for screening of plus-disease retinopathy of prematurity[J]. Graefe’s Archive for Clinical and Experimental Ophthalmology, 2019, 257(11): 2579-2585.

[117]

Sosale B, Sosale A R, Murthy H, et al. Medios: an offline, smartphone-based artificial intelligence algorithm for the diagnosis of diabetic retinopathy[J]. Indian Journal of Ophthalmology, 2020, 68(2): 391-395.

[118]

Wintergerst M W M, Mishra D K, Hartmann L, et al. Diabetic retinopathy screening using smartphone-based fundus imaging in India[J]. Ophthalmology, 2020, 127(11): 1529-1538.

[119]

Delsoz M, Madadi Y, Munir W M, et al. Performance of ChatGPT in diagnosis of corneal eye diseases[J]. medRxiv, 2023: 43(5): 664-670.

[120]

Delsoz M, Raja H, Madadi Y, et al. The use of ChatGPT to assist in diagnosing glaucoma based on clinical case reports[J]. Ophthalmology and Therapy, 2023, 12(6): 3121-3132.

[121]

ROJAS-CARABALI W, SEN A, AGARWAL A, et al. Chatbots vs. human experts: evaluating diagnostic performance of chatbots in uveitis and the perspectives on AI adoption in ophthalmology[J]. Ocular Immunology and Inflammation, 2024, 32(8): 1591-1598.

[122]

Madadi Y, Delsoz M, Lao P A, et al. ChatGPT assisting diagnosis of neuro-ophthalmology diseases based on case reports[J]. medRxiv, 2023: 2023.09.13.23295508.

[123]

Liu X C, Wu J G, Shao A, et al. Uncovering language disparity of ChatGPT on retinal vascular disease classification: cross-sectional study[J]. Journal of Medical Internet Research, 2024, 26: e51926.

[124]

Meskó B. Data annotators are the unsung heroes of medicine’s artificial intelligence revolution[J]. Journal of Medical Artificial Intelligence, 2020, 3: 1.

[125]

Gurnani B, Kaur K. Data annotators: the unacclaimed heroes of artificial intelligence revolution in ophthalmology[J]. Indian Journal of Ophthalmology, 2022, 70(5): 1847.

[126]

Liu P, Higashita R, Guo P Y, et al. Reproducibility of deep learning based scleral spur localisation and anterior chamber angle measurements from anterior segment optical coherence tomography images[J]. British Journal of Ophthalmology, 2023, 107(6): 802-808.

[127]

He M G, Li Z X, Liu C, et al. Deployment of artificial intelligence in real-world practice: opportunity and challenge[J]. Asia-Pacific Journal of Ophthalmology, 2020, 9(4): 299-307.

[128]

Panwar N, Huang P, Lee J Y, et al. Fundus photography in the 21st century:a review of recent technological advances and their implications for worldwide healthcare[J]. Telemedicine Journal and E-Health, 2016, 22(3): 198-208.

[129]

Phan S, Satoh S, Yoda Y, et al. Evaluation of deep convolutional neural networks for glaucoma detection[J]. Japanese Journal of Ophthalmology, 2019, 63(3): 276-283.

[130]

Li Z X, He Y F, 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.

[131]

Liu H R, Li L, Michael Wormstone I, et al. Development and validation of a deep learning system to detect glaucomatous optic neuropathy using fundus photographs[J]. JAMA Ophthalmology, 2019, 137(12): 1353-1360.

[132]

Chuter B, Huynh J, Bowd C, et al. Deep learning identifies high-quality fundus photographs and increases accuracy in automated primary open angle glaucoma detection[J]. Translational Vision Science & Technology, 2024, 13(1): 23.

[133]

Lee E B, Wang S Y, Chang R T. Interpreting deep learning studies in glaucoma: unresolved challenges[J]. Asia-Pacific Journal of Ophthalmology, 2021, 10(3): 261-267.

[134]

Baxter S L, Nwanyanwu K, Legault G, et al. Data sources for evaluating health disparities in ophthalmology where we are and where we need to go[J]. Ophthalmology, 2022, 129(10): e146-e149.

[135]

He Z, Tang X, Yang X, et al. Clinical trial generalizability assessment in the big data era: a review[J]. Clinical and Translational Science, 2020, 13(4): 675-684.

[136]

Jiang J W, Liu X Y, Liu L, et al. Predicting the progression of ophthalmic disease based on slit-lamp images using a deep temporal sequence network[J]. PLoS One, 2018, 13(7): e0201142.

[137]

Xiang D Y, Cai W. Privacy protection and secondary use of health data: strategies and methods[J]. BioMed Research International, 2021, 2021(1): 6967166.

[138]

Jensen P B, Jensen L J, Brunak S. Mining electronic health records: towards better research applications and clinical care[J]. Nature Reviews Genetics, 2012, 13(6): 395-405.

[139]

Alawad M, Aljouie A, Alamri S, et al. Machine learning and deep learning techniques for optic disc and cup segmentation: a review[J]. Clinical Ophthalmology, 2022, 16: 747-764.

[140]

Arias-Serrano I, Velásquez-López P A, Avila-Briones L N, et al. Artificial intelligence based glaucoma and diabetic retinopathy detection using MATLAB-retrained AlexNet convolutional neural network[J]. F1000Research, 2023, 12: 14.

[141]

Diaz-Pinto A, Morales S, Naranjo V, et al. CNNs for automatic glaucoma assessment using fundus images: an extensive validation[J]. Biomedical Engineering Online, 2019, 18(1): 29.

[142]

Loo J, Kriegel M F, Tuohy M M, et al. Open-source automatic segmentation of ocular structures and biomarkers of microbial keratitis on slit-lamp photography images using deep learning[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(1): 88-99.

[143]

Pascal L, Perdomo O J, Bost X, et al. Multi-task deep learning for glaucoma detection from color fundus images[J]. Scientific Reports, 2022, 12(1): 12361.

[144]

Wang S S, Li C, Wang R P, et al. Annotation-efficient deep learning for automatic medical image segmentation[J]. Nature Communications, 2021, 12(1): 5915.

[145]

Kaba D, Wang Y, Wang C, et al. Retina layer segmentation using kernel graph cuts and continuous max-flow[J]. Optics Express, 2015, 23(6): 7366-7384.

[146]

Lee A Y, Yanagihara R T, Lee C S, et al. Multicenter, head-to-head, real-world validation study of seven automated artificial intelligence diabetic retinopathy screening systems[J]. Diabetes Care, 2021, 44(5): 1168-1175.

[147]

Moraes G, Fu D J, Wilson M, et al. Quantitative analysis of OCT for neovascular age-related macular degeneration using deep learning[J]. Ophthalmology, 2021, 128(5): 693-705.

[148]

Asrani S, Essaid L, Alder B D, et al. Artifacts in spectral-domain optical coherence tomography measurements in glaucoma[J]. JAMA Ophthalmology, 2014, 132(4): 396-402.

[149]

Bayer A, Akman A. Artifacts and anatomic variations in optical coherence tomography[J]. Turkish Journal of Ophthalmology, 2020, 50(2): 99-106.

[150]

Lee S Y, Kwon H J, Bae H W, et al. Frequency, type and cause of artifacts in swept-source and Cirrus HD optical coherence tomography in cases of glaucoma and suspected glaucoma[J]. Current Eye Research, 2016, 41(7): 957-964.

[151]

Li A, Thompson A C, Asrani S. Impact of artifacts from optical coherence tomography retinal nerve fiber layer and macula scans on detection of glaucoma progression[J]. American Journal of Ophthalmology, 2021, 221: 235-245.

[152]

Liu Y N, Simavli H, Que C J, et al. Patient characteristics associated with artifacts in spectralis optical coherence tomography imaging of the retinal nerve fiber layer in glaucoma[J]. American Journal of Ophthalmology, 2015, 159(3): 565-576.

[153]

Bagci A M, Shahidi M, Ansari R, et al. Thickness profiles of retinal layers by optical coherence tomography image segmentation[J]. American Journal of Ophthalmology, 2008, 146(5): 679-687.

[154]

Kim J S, Ishikawa H, Gabriele M L, et al. Retinal nerve fiber layer thickness measurement comparability between time domain optical coherence tomography (OCT) and spectral domain OCT[J]. Investigative Ophthalmology & Visual Science, 2010, 51(2): 896-902.

[155]

Pinto-Coelho L. How artificial intelligence is shaping medical imaging technology: a survey of innovations and applications[J]. Bioengineering (Basel), 2023, 10(12): 1435.

[156]

Shweikh Y, Sekimitsu S, Boland M V, et al. The growing need for ophthalmic data standardization (Basel)[J]. Ophthalmology Science, 2023, 3(1): 100262.

[157]

Bajwa M N, Singh G A P, Neumeier W, et al. G1020: a benchmark retinal fundus image dataset for computer-aided glaucoma detection[C]// Proceedings of 2020 International Joint Conference on Neural Networks (IJCNN), Glasgow, UK. Piscataway, USA: IEEE, 2020.

[158]

Kumar J R H, Seelamantula C S, Gagan J H, et al. Chákṣu: a glaucoma specific fundus image database[J]. Scientific Data, 2023, 10(1): 70.

[159]

Khan S M, Liu X X, Nath S, et al. A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability[J]. The Lancet Digital Health, 2021, 3(1): e51-e66.

[160]

Jabbar M K, Yan J Z, Xu H X, et al. Transfer learning-based model for diabetic retinopathy diagnosis using retinal images[J]. Brain Sciences, 2022, 12(5): 535.

[161]

Baget-Bernaldiz M, Pedro R A, Santos-Blanco E, et al. Testing a deep learning algorithm for detection of diabetic retinopathy in a Spanish diabetic population and with MESSIDOR database[J]. Diagnostics (Basel), 2021, 11(8): 1385.

[162]

Orlando J I, Fu H Z, Breda J B, et al. REFUGE challenge: a unified framework for evaluating automated methods for glaucoma assessment from fundus photographs[J]. Medical Image Analysis, 2020, 59: 101570.

[163]

Voets M, Møllersen K, Bongo L A. Reproduction study using public data of: development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs[J]. PLoS One, 2019, 14(6): e0217541.

[164]

Phene S, Dunn R C, Hammel N, et al. Deep learning and glaucoma specialists the relative importance of optic disc features to predict glaucoma referral in fundus photographs[J]. Ophthalmology, 2019, 126(12): 1627-1639.

[165]

Kamalipour A, Moghimi S, Khosravi P, et al. Deep learning estimation of 10-2 visual field map based on circumpapillary retinal nerve fiber layer thickness measurements[J]. American Journal of Ophthalmology, 2023, 246: 163-173.

[166]

Lyons R J, Arepalli S R, Fromal O, et al. Artificial intelligence chatbot performance in triage of ophthalmic conditions[J]. Canadian Journal of Ophthalmology, 2024, 59(4): e301-e308.

[167]

Tan T F, Thirunavukarasu A J, Jin L Y, et al. Artificial intelligence and digital health in global eye health: opportunities and challenges[J]. The Lancet Global Health, 2023, 11(9): e1432-e1443.

[168]

Ruamviboonsuk P, Tiwari R, Sayres R, et al. Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study[J]. The Lancet Digital Health, 2022, 4(4): e235-e244.

[169]

Oh R, Oh J Y, Choi H J, et al. Comparison of ocular biometric measurements in patients with cataract using three swept-source optical coherence tomography devices[J]. BMC Ophthalmology, 2021, 21(1): 62.

[170]

Wang Y Y, Liu C, Hu W Y, et al. Economic evaluation for medical artificial intelligence: accuracy vs. cost-effectiveness in a diabetic retinopathy screening case[J]. NPJ Digital Medicine, 2024, 7(1): 43.

[171]

Khan B, Fatima H, Qureshi A, et al. Drawbacks of artificial intelligence and their potential solutions in the healthcare sector[J]. Biomedical Materials & Devices, 2023,1:731-738.

[172]

Dismuke C. Progress in examining cost-effectiveness of AI in diabetic retinopathy screening[J]. The Lancet Digital Health, 2020, 2(5): e212-e213.

[173]

Tsubota K, Pflugfelder S C, Liu Z G, et al. Defining dry eye from a clinical perspective[J]. International Journal of Molecular Sciences, 2020, 21(23): 9271.

[174]

Taribagil P, Hogg H J, Balaskas K, et al. Integrating artificial intelligence into an ophthalmologist’s workflow: obstacles and opportunities[J]. Expert Review of Ophthalmology, 2023, 18(1): 45-56.

[175]

Evans N G, Wenner D M, Glenn Cohen I, et al. Emerging ethical considerations for the use of artificial intelligence in ophthalmology[J]. Ophthalmology Science, 2022, 2(2): 100141.

[176]

Grzybowski A, Jin K, Zhou J X, et al. Retina fundus photograph-based artificial intelligence algorithms in medicine: a systematic review[J]. Ophthalmology and Therapy, 2024, 13(8): 2125-2149.

[177]

Yang Y H, Lyu J F, Wang R X, et al. A digital mask to safeguard patient privacy[J]. Nature Medicine, 2022, 28(9): 1883-1892.

[178]

Lee C S, Brandt J D, Lee A Y. Big data and artificial intelligence in ophthalmology: where are we now?[J]. Ophthalmology Science, 2021, 1(2): 100036.

[179]

Rumbold J M M, Pierscionek B. The effect of the general data protection regulation on medical research[J]. Journal of Medical Internet Research, 2017, 19(2): e47.

[180]

Alvin Liu T Y, Wu J H. The ethical and societal considerations for the rise of artificial intelligence and big data in ophthalmology[J]. Frontiers in Medicine, 2022, 9: 845522.

[181]

He J X, Baxter S L, Xu J, et al. The practical implementation of artificial intelligence technologies in medicine[J]. Nature Medicine, 2019, 25(1): 30-36.

[182]

Topol E J. High-performance medicine: the convergence of human and artificial intelligence[J]. Nature Medicine, 2019, 25(1): 44-56.

[183]

Park S H, Han K. Methodologic guide for evaluating clinical performance and effect of artificial intelligence technology for medical diagnosis and prediction[J]. Radiology, 2018, 286(3): 800-809.

[184]

Kish L J, Topol E J. Unpatients-why patients should own their medical data[J]. Nature Biotechnology, 2015, 33(9): 921-924.

[185]

Zhu J H. AI ethics with Chinese characteristics? Concerns and preferred solutions in Chinese academia[J]. AI & Society, 2022: 1-14.

[186]

Clark P, Oermann E K, Chen D, et al. Federated AI, current state, and future potential[J]. Asia-Pacific Journal of Ophthalmology, 2023, 12(3): 310-314.

[187]

Gim N, Wu Y, Blazes M, et al. A clinician’s guide to sharing data for AI in ophthalmology[J]. Investigative Ophthalmology & Visual Science, 2024, 65(6): 21.

[188]

Fang H H, Li F, Fu H Z, et al. ADAM challenge: detecting age-related macular degeneration from fundus images[J]. IEEE Transactions on Medical Imaging, 2022, 41(10): 2828-2847.

[189]

Fang H H, Li F, Wu J D, et al. Open fundus photograph dataset with pathologic myopia recognition and anatomical structure annotation[J]. Scientific Data, 2024, 11(1): 99.

[190]

Fu H Z, Li F, Sun X, et al. AGE challenge: angle closure glaucoma evaluation in anterior segment optical coherence tomography[J]. Medical Image Analysis, 2020, 66: 101798.

[191]

Wu J D, Fang H H, Li F, et al. GAMMA challenge: glaucoma grading from multi-modality images[J]. Medical Image Analysis, 2023, 90: 102938.

[192]

Kim M, Kim Y N, Jang M, et al. Synthesizing realistic high-resolution retina image by style-based generative adversarial network and its utilization[J]. Scientific Reports, 2022, 12(1): 17307.

[193]

Heinke A, Radgoudarzi N, Huang BB, et al. A review of ophthalmology education in the era of generative artificial intelligence[J]. Asia-Pacific Journal of Ophthalmology, 2024, 13(4): 100089.

[194]

Kim H K, Ryu I H, Choi J Y, et al. A feasibility study on the adoption of a generative denoising diffusion model for the synthesis of fundus photographs using a small dataset[J]. Discover Applied Sciences, 2024, 6(4): 188.

[195]

Ilanchezian I, Boreiko V, Kühlewein L, et al. Generating realistic counterfactuals for retinal fundus and OCT images using diffusion models[EB/OL]. [2023-12-04]. https://arxiv.org/abs/2311.11629v2.

[196]

Kazerouni A, Aghdam E K, Heidari M, et al. Diffusion models in medical imaging: a comprehensive survey[J]. Medical Image Analysis, 2023, 88: 102846.

[197]

Müller-Franzes G, Niehues J M, Khader F, et al. A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial networks for medical image synthesis[J]. Scientific Reports, 2023, 13(1): 12098.

[198]

Nderitu P, do Rio J M N, Webster L, et al. Conditional diffusion models and retinal image synthesis in diabetic retinopathy[J]. Investigative Ophthalmology & Visual Science, 2023, 64(8): 2389.

[199]

Nath S, Korot E, Fu D J, et al. Reinforcement learning in ophthalmology: potential applications and challenges to implementation[J]. The Lancet Digital Health, 2022, 4(9): e692-e697.

[200]

Wu J H, Koseoglu N D, Jones C, et al. Vision transformers: The next frontier for deep learning-based ophthalmic image analysis[J]. Saudi Journal of Ophthalmology, 2023, 37(3): 173-178.

[201]

Abbas Q, Daadaa Y, Rashid U, et al. HDR-Efficientnet: a classification of hypertensive and diabetic retinopathy using optimize EfficientNet architecture[J]. Diagnostics, 2023, 13(20): 3236.

[202]

Selvaraju R R, 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(2): 336-359.

[203]

Xu Y L, Hu M, Liu H R, et al. A hierarchical deep learning approach with transparency and interpretability based on small samples for glaucoma diagnosis[J]. NPJ Digital Medicine, 2021, 4(1): 48.

[204]

Chang J, Lee J, Ha A, et al. Explaining the rationale of deep learning glaucoma decisions with adversarial examples[J]. Ophthalmology, 2021, 128(1): 78-88.

[205]

Yang Z F, Wang D M, Zhou F Q, et al. Understanding natural language: potential application of large language models to ophthalmology[J]. Asia-Pacific Journal of Ophthalmology, 2024, 13(4): 100085.

[206]

Berman G, Pendley A M, Wright D W, et al. Breaking the barriers: methodology of implementation of a non-mydriatic ocular fundus camera in an emergency department[J]. Survey of Ophthalmology, 2025, 70(1): 153-161.

[207]

Seo H, Chung W G, Kwon Y W, et al. Smart contact lenses as wearable ophthalmic devices for disease monitoring and health management[J]. Chemical Reviews, 2023, 123(19): 11488-11558.

[208]

RaviChandran N, Teo Z L, Ting D S W. Artificial intelligence enabled smart digital eye wearables[J]. Current Opinion in Ophthalmology, 2023, 34(5): 414-421.

[209]

杨艳, 邱佳宁, 罗静, . 人工智能赋能眼科课程设计教学改革[J]. 科学咨询(教育科研), 2024(5): 101-105.

[210]

(Yang Y, Qiu J N, Luo J, et al. Exploration of artificial intelligence empowering teaching reform of ophthalmology curriculum[J]. Policy & Scientific Consult, 2024(5): 101-105.)

[211]

王国胤, 瞿中, 赵显莲. 交叉融合的“人工智能+” 学科建设探索与实践[J]. 计算机科学, 2020, 47(4): 1-5.

[212]

(Wang G Y, Qu Z, Zhao X L. Practical exploration of discipline construction of artificial intelligence+[J]. Computer Science, 2020, 47(4): 1-5.)

[213]

项毅帆, 梁凌毅, 周毅, . 医学人工智能通识课程的效果评估[J]. 眼科学报, 2022, 37(3): 165-170.

[214]

(Xiang Y F, Liang L Y, Zhou Y, et al. Effect evaluation of general education curriculum of medical artificial intelligence[J]. Eye Science, 2022, 37(3): 165-170.)

[215]

龚迪, 李王婷, 李小萌, . 中国智能眼科发展和研究现状之我见[J]. 国际眼科杂志, 2024, 24(3): 448-452.

[216]

(Gong D, Li W T, Li X M, et al. Opinion on the development and research status of intelligent ophthalmology in China[J]. International Eye Science, 2024, 24(3): 448-452.)

[217]

徐汉辉. 基于“医患信任”理论的医疗AI可信度问题的探讨[J]. 工程研究-跨学科视野中的工程, 2020, 12(3): 252-259.

[218]

(Xu H H. On credibility of medical artificial intelligence technologies based on physician-patient trust[J]. Journal of Engineering Studies, 2020, 12(3): 252-259.)

[219]

丁晓东. 全球比较下的我国人工智能立法[J]. 比较法研究, 2024(4): 51-66.

[220]

(Ding X D. China’s artificial intelligence legislation: a global comparative perspective[J]. Journal of Comparative Law, 2024(4): 51-66.)

[221]

丁晓东. 人工智能风险的法律规制: 以欧盟《人工智能法》为例[J]. 法律科学(西北政法大学学报), 2024, 42(5): 3-18.

[222]

Ding X D. Legal regulation of artificial intelligence risks: an example from the EU artificial intelligence act[J]. Science of Law (Journal of Northwest University of Political Science and Law), 2024, 42(5): 3-18.)

[223]

秦天弘. 欧盟《人工智能法案》生效将完善监管规则[N]. 2024-08-06( 1).

[224]

(Qin T H. The entry into force of the EU's Artificial Intelligence Act will improve regulatory rules[N]. 2024-08-06( 1).)

[225]

Liu X X, Cruz Rivera S, Moher D, et al. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension[J]. The Lancet Digital Health, 2020, 2(10): e537-e548.

[226]

Rivera S C, Liu X X, Chan A W, et al. Guidelines for clinical trial protocols for interventions involving artificial intelligence: the SPIRIT-AI extension[J]. The Lancet Digital Health, 2020, 2(10): e549-e560.

[227]

China Association for Quality Inspection. Annotation and quality control specifications for fundus color photograph[J]. Intelligent Medicine, 2021, 1(2): 80-87.

[228]

邵毅, 杨卫华, 陈蔚, . 全球眼科图像公开数据库使用指南(2022)[J]. 眼科新进展, 2022, 42(12): 925-932.

[229]

Shao Y, Yang W H, Chen W, et al. Guidelines for the use of global public databases on ophthalmic images(2022)[J]. Recent Advances in Ophthalmology, 2022, 42(12): 925-932.)

[230]

邵毅, 接英, 刘祖国. 人工智能在眼前节疾病诊断中的应用指南(2023)[J]. 国际眼科杂志, 2023, 23(9): 1421-1430.

[231]

Shao Y, Jie Y, Liu Z G. Guidelines for the application of artificial intelligence in the diagnosis of anterior segment diseases(2023)[J]. International Eye Science, 2023, 23(9): 1421-1430.)

[232]

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.

[233]

Yang W H, Shao Y, Xu Y W. Guidelines on clinical research evaluation of artificial intelligence in ophthalmology(2023)[J]. International Eye Science, 2023, 23(7): 1064-1071.)

[234]

赵志君, 庄馨予. 中国人工智能高质量发展: 现状、问题与方略[J]. 改革, 2023(9): 11-20.

[235]

(Zhao Z J, Zhuang X Y. High-quality development of artificial intelligence in China: current situation, problems and strategies[J]. Reform, 2023(9): 11-20.)

[236]

王妍茜, 王成虎, 张竞月, . 人工智能应用于眼科的积极作用及其伦理问题[J]. 国际眼科杂志, 2022, 22(6): 1020-1024.

[237]

(Wang I, Wang C H, Zhang J Y, et al. Positive role and ethical problems of artificial intelligence in ophthalmology[J]. International Eye Science, 2022, 22(6): 1020-1024.)

[238]

国家卫生健康委. “十四五”全国眼健康规划(2021-2025年)[J]. 中国眼镜科技杂志, 2022(2): 3-9.

[239]

National Health Commission of the People's Republic of China. The 14th Five-Year National Eye Health Plan (2021-2025)[J]. China Glasses Science-Technology Magazine, 2022(2): 3-9.)

[240]

赵申洪. 全球人工智能治理的困境与出路[J]. 现代国际关系, 2024(4): 116-137.

[241]

(Zhao S H. The dilemma and solution of global artificial intelligence governance[J]. Contemporary International Relations, 2024(4): 116-137.)

[242]

朱旭峰, 楼闻佳. 发展还是监管? 人工智能政策的国际比较研究[J]. 学海, 2024(4): 76-85.

[243]

(Zhu X F, Lou J W. To develop or to regulate? an international comparison on artificial intelligence policies[J]. Academia Bimestris, 2024(4): 76-85.)

[244]

Zhang K, Yang X, Wang Y, et al. Artificial intelligence in drug development[J]. Nature Medicine, 2025, 31(1): 45-59.

[245]

Read-Brown S, Hribar M R, Reznick L G, et al. Time requirements for electronic health record use in an academic ophthalmology center[J]. JAMA Ophthalmology, 2017, 135(11): 1250-1257.

[246]

Guidance W H O. Ethics and governance of artificial intelligence for health[EB/OL]. World Health Organization, 2021.(2021-06-28) [2024-09-01]. https://www.who.int/publications/i/item/9789240029200.

[247]

Tsui J C, Wong M B, Kim B J, et al. Appropriateness of ophthalmic symptoms triage by a popular online artificial intelligence chatbot[J]. Eye (Lond), 2023, 37(17): 3692-3693.

[248]

Waisberg E, Ong J, Zaman N, et al. GPT-4 for triaging ophthalmic symptoms[J]. Eye (Lond), 2023, 37: 3874-3875.

[249]

Zandi R, Fahey J D, Drakopoulos M, et al. Exploring diagnostic precision and triage proficiency: a comparative study of GPT-4 and bard in addressing common ophthalmic complaints[J]. Bioengineering, 2024, 11(2): 120.

[250]

Channa R, Zafar S N, Canner J K, et al. Epidemiology of eye-related emergency department visits[J]. JAMA Ophthalmology, 2016, 134(3): 312-319.

[251]

Shah R, Edgar D F, Khatoon A, et al. Referrals from community optometrists to the hospital eye service in Scotland and England[J]. Eye (Lond), 2022, 36(9): 1754-1760.

[252]

Jeblick K, Schachtner B, Dexl J, et al. ChatGPT makes medicine easy to swallow: an exploratory case study on simplified radiology reports[J]. European Radiology, 2024, 34(5): 2817-2825.

[253]

Lyu Q, Tan J, Zapadka M E, et al. Translating radiology reports into plain language using ChatGPT and GPT-4 with prompt learning: results, limitations, and potential[J]. Visual Computing for Industry, Biomedicine, and Art, 2023, 6(1): 9.

[254]

Li J J, Guan Z Y, Wang J, et al. Integrated image-based deep learning and language models for primary diabetes care[J]. Nature Medicine, 2024, 30(10): 2886-2896.

PDF (2426KB)

1705

访问

0

被引

详细

导航
相关文章

/