Abstract
Purpose :
Significant advances in artificial intelligence (AI) has led to promising applications in ophthalmology. This study highlights the 100 most cited ophthalmology research papers on AI.
Methods :
Ophthalmology research papers published between 1990 and October 2021, were extracted from the Institute for Scientific Information Web of Knowledge platform, using the keywords AI, machine learning (ML), and deep learning (DL). Papers were assessed for eligibility based on title/abstract review, followed by full-text review. The primary outcome measure was the number of times cited. Secondary outcome measures were: publication year, author attributes, journal name, study design and characteristics (ophthalmology subspecialties, pathology examined, AI branches, algorithms, databases, and imaging modalities used).
Results :
The 100 publications were cited between 27 and 399 times, with a median of 46 [IQR: 32-74]. All papers were published between 1994 and 2020, with 64% of them published in 2017-2019. They were written by 1 to 30 authors, with a median of 7 [4-9] authors per paper. 482 unique authors wrote between 1 and 13 papers. 46% of corresponding authors were based in the USA and 23% in Europe. The top three journals were: Investigative Ophthalmology & Visual Science (26%), Ophthalmology (14%) and JAMA Ophthalmology (8%). 67 studies were retrospective, 21 prospective and 4 ambidirectional. Out of the 100 publications, 46 focused on retina, 35 on glaucoma and 10 on cornea and external disease. Studies mainly examined glaucoma (37%), age-related macular edema (28%), and diabetic retinopathy (27%). The most used imaging modalities were: fundus photography (42%), optical coherence tomography (OCT, 42%), and visual fields (VFs, 14%). 78 studies were geared towards the development and evaluation of a diagnostic technology. Among the 86 studies specifying their ML algorithms, 78% used supervised learning, 11% unsupervised learning and 12% used a mix of both techniques. 78 studies focused on DL. Among the 95 studies specifying their algorithm approach, convolutional NNs were the most used (51%). Among the 51 studies specifying their database, 65% used private databases and 51% used public databases.
Conclusions :
This is the first study to analyze the characteristics of the 100 most-cited ophthalmology papers on AI. The use of AI was predominantly for image recognition to develop and evaluate a new diagnostic technology.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.