Abstract
Purpose :
Artificial intelligence has advanced diverse avenues of ophthalmic research, including ocular disease diagnosis and progression. Open-source research holds particular importance, owing to its transparency and resource sharing. However, the progress in this domain lacks a systematic overview. We conducted a literature search to understand the progress and launched EyeHorizon, a platform that documents open-source projects and facilitates the use of released models to a wide community.
Methods :
We conducted a literature search using a combination of queries on databases, including IEEE Xplore, PubMed, Web of Science, and ScienceDirect (Figure 1a). We then randomly selected 100 papers published between 2017 and 2020, as well as another 100 papers between 2021 and 2023, to assess the open-source ratio. We built EyeHorizon on Github and Google Colab, which provides a user-friendly interface and free computational resources, catering to a broad user base.
Results :
We collected 11,108 publications from IEEE Xplore (1,646), PubMed (4,537), Web of Science (3,818), and ScienceDirect (1,107). After removing duplicates, we retained 8,352 papers, revealing a consistent upward trend from 2017 to 2023 (Figure 1b). The period from 2019 to 2022 saw rapid growth, averaging 375 additional papers annually, while the increase from 2022 to 2023 was minor. Initial findings from random samples (Figure 1c) highlight two key observations: 1) a low overall ratio of open-source research (6% and 7% for two periods of time respectively), and 2) progress falling below expectations, despite encouragement for open science in recent years. To integrate and standardise the open-source research, we started EyeHorizon and incorporated the foundation model RETFound (https://www.nature.com/articles/s41586-023-06555-x) inside as the first step (Figure 2). More open-source models will be standardised and included.
Conclusions :
This research revealed that limited progress has been made in open-source research for AI in ophthalmology, which hopefully triggers the consciousness to share the resources and knowledge via releasing AI models, to alleviate the resource imbalance and poverty. Subsequent work will curate the collected literature, filter the open-source papers, and include more powerful AI models in EyeHorizon.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.