人工智能是对人类智能的模拟和拓展。基于深度学习的人工智能可以很好地利用图像的内在特征，如轮廓、框架等，来分析图像。研究人员通常利用图像来诊断眼底病，因此将人工智能应用于眼底检查是有意义的。在眼科领域，人工智能通过分析光学相干断层扫描图像、眼底照片和超宽视野图像，已经在检测多种眼底疾病上取得了类似医生的性能。它也已经被广泛应用于疾病进展预测。然而，人工智能在眼科的应用也存在一些潜在的挑战，黑盒问题是其中之一。研究人员致力于开发更多的可解释的深度学习系统，并确认其临床可行性。人工智能在最流行的眼底病中的最新应用、可能遇到的挑战以及未来的道路将一一阐述。
Artificial intelligence (AI) is about simulating and expanding human intelligence. AI based on deep learning (DL) can analyze images well by using their inherent features, such as outlines, frames and so on. As researchers generally diagnoses ocular fundus diseases by images, it makes sense to apply AI to fundus examination. In ophthalmology, AI has achieved doctor-like performance in detecting multiple ocular fundus diseases through optical coherence tomography (OCT) images, fundus photographs, and ultra-wide-field (UWF) images. It has also been widely used in disease progression prediction. Nonetheless, there are also some potential challenges with AI application in ophthalmology, one of which is the black-box problem. Researchers are devoted to developing more interpretable deep learning systems (DLS) and confirming their clinical feasibility. This review describes a summary of the state-of-the-art AI application in the most popular ocular fundus diseases, potential challenges and the path forward.
Background: Research innovations inoculardisease screening, diagnosis, and management have been boosted by deep learning (DL) in the last decade. To assess historical research trends and current advances, we conducted an artifcial intelligence (AI)–human hybrid analysis of publications on DL in ophthalmology.
Methods: All DL-related articles in ophthalmology, which were published between 2012 and 2022 from Web of Science, were included. 500 high-impact articles annotated with key research information were used to fne-tune alarge language models (LLM) for reviewing medical literature and extracting information. After verifying the LLM's accuracy in extracting diseases and imaging modalities, we analyzed trend of DL in ophthalmology with 2 535 articles.
Results: Researchers using LLM for literature analysis were 70% (p= 0.000 1) faster than those who did not, while achieving comparable accuracy (97% versus 98%, p = 0.768 1). The field of DL in ophthalmology has grown 116% annually, paralleling trends of the broader DL domain. The publications focused mainly on diabetic retinopathy (p = 0.000 3), glaucoma (p = 0.001 1), and age-related macular diseases (p = 0.000 1) using retinal fundus photographs (FP, p = 0.001 5) and optical coherence tomography (OCT, p = 0.000 1). DL studies utilizing multimodal images have been growing, with FP and OCT combined being the most frequent. Among the 500 high-impact articles, laboratory studies constituted the majority at 65.3%. Notably, a discernible decline in model accuracy was observed when categorizing by study design, notwithstanding its statistical insignificance. Furthermore, 43 publicly available ocular image datasets were summarized.
Conclusion: This study has characterized the landscape of publications on DL in ophthalmology, by identifying the trends and breakthroughs among research topics and the fast-growing areas. This study provides an efcient framework for combined AI–human analysis to comprehensively assess the current status and future trends in the feld.
Background: Research innovations inoculardisease screening, diagnosis, and management have been boosted by deep learning (DL) in the last decade. To assess historical research trends and current advances, we conducted an artifcial intelligence (AI)–human hybrid analysis of publications on DL in ophthalmology.
Methods: All DL-related articles in ophthalmology, which were published between 2012 and 2022 from Web of Science, were included. 500 high-impact articles annotated with key research information were used to fne-tune alarge language models (LLM) for reviewing medical literature and extracting information. After verifying the LLM's accuracy in extracting diseases and imaging modalities, we analyzed trend of DL in ophthalmology with 2 535 articles.
Results: Researchers using LLM for literature analysis were 70% (p = 0.000 1) faster than those who did not, while achieving comparable accuracy (97% versus 98%, p = 0.768 1). The field of DL in ophthalmology has grown 116% annually, paralleling trends of the broader DL domain. The publications focused mainly on diabetic retinopathy (p = 0.000 3), glaucoma (p = 0.001 1), and age-related macular diseases (p = 0.000 1) using retinal fundus photographs (FP, p = 0.001 5) and optical coherence tomography (OCT, p = 0.000 1). DL studies utilizing multimodal images have been growing, with FP and OCT combined being the most frequent. Among the 500 high-impact articles, laboratory studies constituted the majority at 65.3%. Notably, a discernible decline in model accuracy was observed when categorizing by study design, notwithstanding its statistical insignificance. Furthermore, 43 publicly available ocular image datasets were summarized.
Conclusion: This study has characterized the landscape of publications on DL in ophthalmology, by identifying the trends and breakthroughs among research topics and the fast-growing areas. This study provides an efcient framework for combined AI–human analysis to comprehensively assess the current status and future trends in the feld.
Artificial intelligence (AI) is about simulating and expanding human intelligence. AI based on deep learning (DL) can analyze images well by using their inherent features, such as outlines, frames and so on. As researchers generally diagnoses ocular fundus diseases by images, it makes sense to apply AI to fundus examination. In ophthalmology, AI has achieved doctor-like performance in detecting multiple ocular fundus diseases through optical coherence tomography (OCT) images, fundus photographs, and ultra-wide-field (UWF) images. It has also been widely used in disease progression prediction. Nonetheless, there are also some potential challenges with AI application in ophthalmology, one of which is the black-box problem. Researchers are devoted to developing more interpretable deep learning systems (DLS) and confirming their clinical feasibility. This review describes a summary of the state-of-the-art AI application in the most popular ocular fundus diseases, potential challenges and the path forward.
Artificial intelligence (AI) is about simulating and expanding human intelligence. AI based on deep learning (DL) can analyze images well by using their inherent features, such as outlines, frames and so on. As researchers generally diagnoses ocular fundus diseases by images, it makes sense to apply AI to fundus examination. In ophthalmology, AI has achieved doctor-like performance in detecting multiple ocular fundus diseases through optical coherence tomography (OCT) images, fundus photographs, and ultra-wide-field (UWF) images. It has also been widely used in disease progression prediction. Nonetheless, there are also some potential challenges with AI application in ophthalmology, one of which is the black-box problem. Researchers are devoted to developing more interpretable deep learning systems (DLS) and confirming their clinical feasibility. This review describes a summary of the state-of-the-art AI application in the most popular ocular fundus diseases, potential challenges and the path forward.
手术前常规检查在临床诊疗中被广泛应用，但在一些低风险择期手术前对患者进行常规检查，对提高医疗质量并无帮助，反而降低了医疗效率，增加了医疗费用。为提高效率，一些地区、机构和专家学者陆续通过宣传教育、发表共识、制定指南等方式控制无指征术前常规检查，但效果仍依赖于执业者的重视程度和专业水平。大数据机器学习方法以其标准化、自动化的特点为解决这一问题提供了新的思路。在回顾已有研究的基础上，我们抽取2017至2019年在中山大学中山眼科中心进行眼科手术的3.4万名患者的病史和体格检查资料大数据，涵盖年龄、性别等口学信息，诊断、既往疾病等病史信息，视功能、入院时身体质量指数(BMI)等体格检查信息。并以此为基础使用机器学习方法预测术前胸部X线检查是否存在异常，受试者操作特性曲线(receiver operating characteristic curve，ROC)曲线下面积达到0.864，预测准确率可达到81.2%，对大数据机器学习精简术前常规检查的新方式进行了先期探索。
Preoperative routine tests are widely prescribed in clinical settings. However, these tests do not help improving the quality of medical care in low-risk elective surgery. Instead, they are associated with lower efficiency and increasing fees. To improve the efficiency, many regions, institutions, and scholars have attempted to reduce preoperative routine tests without indications through propaganda, education, consensus, and guidelines. Nevertheless, the effects are still highly dependent on the expertise and emphasis of practitioners. Machine learning based on big data provide a new solution with its standardization and automation. Through literature review, we extracted the big data, including demographic features such as sex and age, histories including diagnosis and chronic diseases, and physical examination features such as visual function and body mass index. A total of 34 000 patients undergone ocular surgeries in Zhongshan Ophthalmic Center, Sun Yat-sen university from 2017 to 2019. Machine learning was adopted to predict the risk of finding abnormalities in chest X-ray examination, with an accuracy of 81.2%. Area under the Receiver Operating Characteristic curve was 0.864. The study could be an early exploration into the field of simplifying preoperative tests by machine learning.
目的：分析医学人工智能通识课程“眼科人工智能的研发与应用”的开展效果，为相关医学人工智能通识课程的开展提供参考和借鉴。方法：纵向观察性研究。观察分析2020年秋季学期眼科人工智能的研发与应用通识课程学生人群，课程考核结果以及学生对课程的整体评价。结果：共有118名本科生同学参与了课程学习。其中大部分为低年级临床医学专业本科生。期中考核得分为77.21±10.07，有56位同学(47.46%)达到80分以上。期末考核得分为82.24±6.77，有91位同学(77.12%)达到80分以上。同学对课程的评分为98.76±3.55，超过90%的同学表示课程备课认真、授课条理清晰、表达准确。结论：本课程的顺利进展证明医学人工智能联合教学模式的可行性，理论和实践穿插的教学设置帮助同学们更好地掌握知识技术，完成教学目标。
Objective: To analyze the effectiveness of medical education curriculum named “Development and Application of Ophthalmic Artificial Intelligence”, and provide reference for the development of other related curriculums. Methods: Longitudinal observational study method was adopted. During the fall semester of 2020, we conducted an education curriculum named “Development and Application of Ophthalmic Artificial Intelligence” and analyzed the results of mid-term and final examinations, and curriculum evaluation of students. Results: There were 118 undergraduate students taking the course and most of them were junior students majoring in clinical medicine. The score of the mid-term examination was in the range of 77.2±10.07, and 56 students (47.46%) got more than 80 points. The score of the final examination was in the range of 82.24±6.77, and 91 students (77.12%) got more than 80 points. The score of course evaluation of students was in the range of 98.76±3.55, and more than 90% of the students thought that teachers have made full preparations before class, together with clear teaching logic and accurate expressions in class. Conclusion: The smooth progress of our course proved the feasibility of medical artificial intelligence teaching. The teaching setting interspersed with theory and practice could help students to master knowledge and technology better, so as to achieve the teaching objectives.
