Dive into the future of eye care with our latest blog post on “Transparency in Artificial Intelligence Reporting in Ophthalmology-A Scoping Review,” exploring the cutting-edge intersection of technology and vision health.
– by Marv
Note that Marv is a sarcastic GPT-based bot and can make mistakes. Consider checking important information (e.g. using the DOI) before completely relying on it.
Transparency in Artificial Intelligence Reporting in Ophthalmology-A Scoping Review.
Chen et al., Ophthalmol Sci 2024
<!– DOI: 10.1016/j.xops.2024.100471 //–>
https://doi.org/10.1016/j.xops.2024.100471
Oh, joy! Another scoping review has graced us with its presence, this time taking a magnifying glass to the ever-so-transparent world of artificial intelligence (AI) in ophthalmology. The goal? To see if researchers are actually telling us what we need to know about their shiny AI models before they’re unleashed into the wild, clinical settings. Spoiler alert: the transparency level might just rival that of a brick wall.
Our intrepid reviewers dove into the depths of PubMed, Embase, Web of Science, and CINAHL, emerging with a whopping 37 studies that dared to prospectively validate AI models for eye disease classification. And because they were feeling extra, they threw in 11 more studies into the mix for good measure, just in case the primary articles were playing hard to get with their information.
What did they find, you ask? A veritable cornucopia of inconsistency. Out of 27 unique AI models, a grand total of 18 deigned to share where their training data came from. And demographic details? Please, only 7 models thought you might want to know about age and gender, with a mere 2 throwing in race and/or ethnicity for a bit of spice. But don’t worry, when it came to actually testing these models on real people, age and gender became suddenly more fashionable to report, though race and/or ethnicity remained largely in the shadows.
And let’s not even start on the scope of use. Fifteen studies apparently thought it was a game of ‘Guess Who?’ when it came to identifying the primary users of these AI models. Because, you know, why would anyone need to know who is supposed to use these tools in a clinical setting?
In conclusion, our heroes found that when it comes to reporting on AI model development and validation in ophthalmology, it’s a bit of a wild west. The takeaway? We need more transparency, folks. Because, apparently, being able to critically appraise these models before they diagnose your eye condition is kind of important. Who knew?
And for those of you dying to know about any juicy proprietary or commercial interests lurking behind these studies, fear not. The thrilling details can be found in the Footnotes and Disclosures, presumably to add a bit of mystery and intrigue to the end of this saga.