“We use AI all the time, we just don't realize it,” Ranya Habash, MD, a corneal refractive surgeon at Bascom Palmer and the head of AI for AECOS told Dr. Singh in an episode dedicated to understanding the potential role of AI in ophthalmology.
However, she added, it would be best to understand the “A” in AI as augmented. In her view, AI has every potential to add clinical efficiencies that will permit physicians to spend more time on the reason they got into medicine in the first place—and something that is suffering in the current clinical setting: interacting with the human patients in their practices.
The Future Role of AI in Ophthalmology
Dr. Habash explained how generative AI differs from the behind-the-scenes digital automations that eye care practitioners have come to rely on (IOL calculations, for example). AI systems that produce content, such as ChatGPT (Chat Generative Pre-Trained Transformer), are not confined by their current level of knowledge. They have to be trained to function as desired, and they continuously improve over time. “Every time someone corrects its output, it learns how to improve its answers,” she said. “It generates new data.”
Where such a technology might be applicable, for example, is in listening to exam room conversations between a doctor and a patient, and then synthesizing the most important information into clinical notes. Such services would free up physicians’ time to do more patient-centric tasks. Dr. Habash added that some doctors are exploring AI’s use in writing prior authorizations and denial letters. They are using it to find the citations they need so that insurance companies “have no wiggle room.”
Learning As It Goes
ChatGPT’s learning potential was demonstrated in a study that evaluated whether the generative content functions could provide answers to two 260-question simulated exams used to study for the yearly OKAP exams.5 The study found 55.8% and 42.7% accuracy in the two trials, which is remarkable, considering the system was used “out of the box” without any training. Dr. Habash said she is aware of a more recent figure of 81% with ChatGPT-4. As well, even within the study, ChatGPT performed better on question sets in some subspecialities than others (75% accuracy in general medicine and retina vs 35% in uveitis). “The more information that you can give it, the more feedback you can give it, the quicker it will learn,” she said.
To many, AI remains an intriguing concept, but one that sometimes engenders concerns that machines will replace humans, and medicine is too important a sector to risk something going wrong. Dr. Habash pointed out that AI, if it is augmented intelligence, is meant to improve the capabilities of human actors. A second brain, if you will, that will allow the individual using it to perform more efficiently.
At Bascom Palmer, Dr. Habash and colleagues placed fundus cameras equipped with AI for automatic detection of suspicious findings in primary care offices in an experiment to boost the number of screenings for diabetic retinopathy, which hovered at around 40%. The team was hopeful for even a slight increase. After the cameras were installed, the rate improved to 82%.
“Then, we found pathology in 50% of the patients who were screened,” Dr. Habash said, adding that her team was able to expedite those patients to retina specialists for intervention instead of waiting 6 months for an appointment.
The APPRAISE Study
Ultimately, how accepting ophthalmologists are of AI—however it is defined—will hinge on its applications. That certainly seems to be the case based on the results of the Acceptance and Perception of Artificial Intelligence Usability in Eye Care (APPRAISE) for Ophthalmologists survey,6 which asked 1,176 ophthalmologists from 70 countries to “evaluate the perspectives of ophthalmologists regarding AI in four major eye conditions: diabetic retinopathy, glaucoma, age-related macular degeneration, and cataract."
Overwhelmingly, and perhaps unsurprisingly, respondents were more willing to use AI in clinical assistance capacities (office tasks, scripting, etc.) than as a diagnostic tool. Dr. Habash interpreted these findings to mean that ophthalmologists welcome AI support for clinical efficiencies, but they are not enthusiastic to let AI generate or be responsible for clinical decisions.
The Future Will Be Collaborative
Dr. Habash stressed the need for medical practitioners to collaborate with tech industry partners as well as ophthalmic device manufacturers to develop AI-assisted technologies. In fact, she brought representatives from Microsoft, Inc., to give a panel talk at this year’s ASCRS meeting in which they highlighted ways those in the medical and technology fields can work together to usher in an age of “precision medicine.”
1. Kramer BA, Berdahl J, Gu X. Merchea M. Real-world incidence of monofocal toric intraocular lens repositioning: analysis of the American Academy of Ophthalmology IRIS Registry. J Cataract Refract Surg. 2022;48(3):298-303. doi: 10.1097/j.jcrs.0000000000000748.
2. Hernandez R, Almenara C, Soriano D, et al. Toric intraocular lens implantation vs femtosecond laser-assisted arcuate keratotomy for correction of moderate astigmatism in cataract surgery. J Cataract Refract Surg. 2022;48(8):887-893.
3. Pandit RT, Devgan U, Chapman JR Jr. Twist and out intraocular lens removal. J Cataract Refract Surg. 2020;46(8):1072-1074. doi: 10.1097/j.jcrs.0000000000000161.
4. Micheletti JM, Weber N, McCauley MB, et al. Punch and rescue technique for scleral fixation of dislocated single-piece intraocular lenses. J Cataract Refract Surg 2022;48(2):247-250. doi: 10.1097/j.jcrs.0000000000000845.
5. Teebagy S, Colwell L, Wood E, Yaghy A. Improved performance of ChatGPT-4 on the OKAP Exam: a comparative study with ChatGPT-3.5. medRxiv. April 2023.
6. Gunasekeran DV, Zheng F, Lim GYS, et al. Acceptance and Perception of Artificial Intelligence Usability in Eye Care (APPRAISE) for ophthalmologists: a multinational perspective. Front Med (Lausanne). 2022;13(9):875242.
