Ophthalmologists should be proud that their field “is at the forefront of artificial intelligence–based innovations in biomedical research that may lead to improvement in early detection and surveillance of ocular disease, prediction of progression, and improved quality of life.”¹ With that in mind, it’s important to take stock of some artificial intelligence technologies that are in use and explore how they serve as augmented intelligence systems for the modern ophthalmologist — that is, how they enhance clinical practice and patient care rather than replacing the expertise of ophthalmologists.

The umbrella term “artificial intelligence” is used to describe computer-based or machine-based algorithms designed to mimic neural intelligence. Artificial intelligence systems may be as simple as the spell-check tools in word processors or as complex as facial recognition interfaces on smartphones. A more specific term, “augmented intelligence,” frames artificial intelligence as adjunctive and complementary to human intelligence — that is, seeing artificial intelligence as a means of data gathering that, in turn, allows humans to make more informed decisions.

AUGMENTED INTELLIGENCE IN MEDICAL SETTINGS

Some clinicians feared that artificial intelligence systems would replace ophthalmologists, swapping the nuanced experienced of trained providers for the cold, technical precision of software systems. These fears have not been (and will likely never be) realized. Rather, artificial intelligence systems have served as an adjunct to medical practitioners, providing more precise data about patients which, in turn, allow them to improve the quality of care.

The American Medical Association House of Delegates considers all artificial intelligence in medical settings to be augmented intelligence. The House of Delegates focuses on the assistive role of this technology, “emphasizing that [augmented intelligence’s] design enhances human intelligence rather than replaces it.”²

Medical tools that leverage the power of artificial intelligence to evaluate data and provide reports to providers are similar to the other monitoring, screening, and detection tools already in use. In this view, artificial intelligence platforms will never replace human doctors. Instead, these platforms will strengthen ophthalmologists’ ability to provide precision medicine, detect disease early, and treat vision-threatening conditions. Framing artificial intelligence as augmented intelligence properly represents this technology as the opportunity that it presents.

ARTIFICIAL INTELLIGENCE FOR DIAGNOSTICS AND DISEASE MONITORING IN OPHTHALMOLOGY

Artificial intelligence systems in the clinical setting have primarily concentrated on disease monitoring, screening, and detection. Such tools will be especially useful for patients with neovascular AMD (nAMD) and diabetic retinopathy (DR), 2 of the leading causes of blindness worldwide. Forecasters have estimated that prevalence of these conditions will only increase as the population ages, and providers will have to rely on artificial intelligence to assist in the screening and detection of these diseases.

Although some artificial intelligence platforms for DR screening have been restricted to in-office use, home-based monitoring for conversion from intermediate AMD to neovascular AMD has served as a proof of concept that outside-clinic use of technology leveraging artificial intelligence transforms these platforms into augmented intelligence systems. Both in-office and home-based artificial intelligence tools have potential to improve the quality of care that ophthalmologists can provide to patients with DR and nAMD.

IN-OFFICE AUGMENTED INTELLIGENCE

Many retina specialists are familiar with imaging platforms that screen patients for DR via artificial intelligence–based analyses of fundus photographs. The real-world validity and reliability of one such platform, IDx-DR (Digital Diagnostics), was analyzed in a 2022 study by Mehra et al.³ In that study, nonmydriatic fundus images were acquired and reviewed by IDx-DR from approximately 1,050 adult patients in a primary care practice setting. Among those screened, 92% had fundus photographs that were gradable by IDx-DR. IDx-DR determined that 14% of patients had greater than mild nonproliferative DR; among these patients, IDx-DR demonstrated a 100% sensitivity, 89% specificity, 28% positive predictive value, and 100% negative predictive value compared with manual overread assessment.

The real-world data from Mehra et al were similar to the data by Abramoff et al, who explored the diagnostic performance of IDx-DR in a pivotal study.⁴ Those researchers found that IDx-DR exceeded prespecified superiority endpoints for sensitivity, specificity, and imageability rate when patients were imaged in a primary care practice setting.

The conclusions by Abramoff et al led the US Food and Drug Administration (FDA) to authorize IDx-DR as the first autonomous artificial intelligence diagnostic system in any field of medicine.^4,5 Other platforms have earned clearance from regulatory bodies, too. The EyeArt system (Eyenuk), cleared by the FDA in 2020, is the first FDA-cleared artificial intelligence platform to detect more-than-moderate DR and vision-threatening DR in 1 test. RetinAI Discovery software (RetinAI Medical), an artificial intelligence system that quantifies fluid and layer segments on OCT images in patients with nAMD, DR, DME, and retinal vein occlusion (RVO), received the CE mark and has been submitted for FDA clearance.⁶ Similarly, the RetInSight Fluid Monitor (RetInSight), an artificial intelligence software that quantifies retinal fluid volumes based on OCT images, was awarded the CE mark in May 2022.⁷

HOME-BASED AUGMENTED INTELLIGENCE

Given the rapid rate at which intermediate AMD can transition to neovascular AMD, home-based artificial intelligence systems have been shown to be especially useful for patients who are at risk of developing nAMD. Artificial intelligence–enabled preferential hyperacuity perimetry (PHP) evaluation, a component of the Notal Vision Monitoring Center’s home-based monitoring service, has been used by real-world patients for more than a decade. Patients with intermediate AMD who are at risk of conversion to nAMD may be enrolled by their clinicians in the ForeseeHome AMD Monitoring Program. Patients in the program are encouraged to submit to daily PHP testing, results of which are analyzed by an artificial intelligence software for aberrations in test results. If the algorithm triggers an alert, an in-house ophthalmologist at the Notal Vision Monitoring Center reviews the patient’s testing results and determines the best course of action for the patient. If in-person examination by their referring provider is advised, the patient’s provider is contacted so that they can initiate follow-up.

Data from the ALOFT study, a retrospective review of 3,334 eyes in 2,123 patients with bilateral dry AMD or unilateral nAMD with fellow-eye dry AMD who used the ForeseeHome monitoring service from 2010 to 2020, confirmed the efficacy of at-home monitoring via an artificial intelligence PHP evaluation. Among those using the ForeseeHome monitoring program in the real world, 52% of patients who converted to nAMD experienced a system alert before they noticed changes in vision or were diagnosed at a prescheduled office visit.⁸ It should be noted that the longitudinal data gathered during monitoring may improve the likelihood of detecting small aberrations — and could possibly help detect conversion activity that would otherwise be undetected on routine clinical examination. A recent subanalysis of the ALOFT study found that, among patients with unilateral nAMD who triggered ForeseeHome alerts but did not receive immediate diagnoses of nAMD in the fellow eye, 22% and 43% of patients reported conversion in their fellow eye at 12 and 24 months, respectively, suggesting that higher vigilance in these patients may be warranted.⁹

Given the high volume of data acquired in a program such as this one (both due to the sheer number of patients submitting to frequent testing and the nature of longitudinal data), human analysis of PHP results is impractical. However, use of artificial intelligence to analyze these data — and thereby supplement the clinical monitoring efforts of retina specialists — optimizes disease monitoring, increases the likelihood of detecting otherwise undetected disease, and allows retina specialists to treat patients before symptom onset results in irrecoverable vision loss.

THE FUTURE OF AUGMENTED INTELLIGENCE IN RETINA

As US retina specialists await FDA clearance for more artificial intelligence systems meant for clinical use (ie, RetinAI and RetInSight Fluid Monitor), they may be interested to know that home-based technology that captures and analyzes OCT images was investigated by Liu et al in a 2022 prospective observational study.¹⁰ In that trial, 15 patients with active nAMD performed daily self-imaging using the Notal Vision Home OCT system, results from which were analyzed by the Notal OCT Analyzer (NOA), an artificial intelligence algorithm. The researchers determined that the home-OCT scans analyzed using NOA and the in-office OCT scans graded by human experts agreed on the fluid status in 96% of the cases, concluding that “daily home OCT imaging is feasible among patients with nAMD.”¹⁰ Given that significant fluctuations in retinal fluid volumes are associated with worse visual acuity during the maintenance phase of anti-VEGF therapy,¹¹ capturing of OCT images at home and analyzing them with an artificial intelligence software may allow retina specialists to keep patients under more consistent monitoring and optimize retreatment schedules so as to reduce treatment burden while maximizing the effects of therapy.

These examples of using artificial intelligence–powered tools to perform high-frequency monitoring at home are truly a demonstration of augmented intelligence. These extend the capabilities of the physician beyond the office, and in turn enhance the outcomes for patients.

MORE TO COME

As the big data era dawns in medicine, ophthalmology has the unique opportunity to dictate the terms of how to make artificial intelligence become augmented intelligence. Retina specialists should continue to embrace their role in leading patient care into the future and leverage their influence and organizational strength to direct the integration of new technology in a way that most benefits their patients. RP

REFERENCES

1. Sherif NA, Chew EY, Chiang MF, et al. Artificial intelligence at the national eye institute. Curr Opin Ophthalmol. 2022;33(6):579-584. doi:10.1097/ICU.0000000000000889
2. American Medical Association. Augmented intelligence in medicine. Accessed October 19, 2022. https://www.ama-assn.org/practice-management/digital/augmented-intelligence-medicine
3. Mehra AA, Softing A, Guner MK, Hodge DO, Barkmeier AJ. Diabetic retinopathy telemedicine outcomes with artificial intelligence–based image analysis, reflex dilation, and image overread. Am J Ophthalmol. 2022;244:125-132. doi:10.1016/j.ajo.2022.08.008
4. Abràmoff MD, Lavin PT, Birch M, Shah N, Folk JC. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digit Med. 2018;1(1):1-8. doi:10.1038/s41746-018-0040-6
5. US Food and Drug Administration. FDA permits marketing of artificial intelligence–based device to detect certain diabetes-related eye problems. Press release. April 11, 2018. https://www.fda.gov/news-events/press-announcements/fda-permits-marketing-artificial-intelligence-based-device-detect-certain-diabetes-related-eye
6. RetinAI Medical AG. RetinAI receives FDA clearance for Discovery, our medical image and data management platform for ophthalmology. Press release. May 1, 2022. https://www.retinai.com/press-releases/retinai-receives-fda-clearance-for-discovery
7. RetInSight. RetInSight Fluid Monitor provides next evolutionary step in monitoring of neovascular age-related macular degeneration. Press release. May 18, 2022. https://retinsight.com/wp-content/uploads/2022/05/retinsight-fluid-monitor-provides-next-evolutionary-step-in-monitoring-of-neovascular-age-related-macular-degeneration/
8. Mathai M, Reddy S, Elman MJ, et al; ALOFT Study Group. Analysis of the long-term visual outcomes of ForeseeHome remote telemonitoring. The ALOFT study. Ophthalmol Retina. 2022;6(10):922-929.
9. Ho AC, Schechet SA, Mathai M, et al; ALOFT Study Group. The predictive value of false positive ForeseeHome alerts in the ALOFT study. Ophthalmol Retina. 2022 [epub ahead of print]: S2468-6530(22)00511-5. doi:10.1016/j.oret.2022.10.009
10. Liu Y, Holekamp NM, Heier JS. Prospective, longitudinal study: daily self-imaging with home OCT for neovascular age-related macular degeneration. Ophthalmol Retina. 2022;6(7):575-585. doi:10.1016/j.oret.2022.02.011
11. Chakravarthy U, Havilio M, Syntosi A, et al. Impact of macular fluid volume fluctuations on visual acuity during anti-VEGF therapy in eyes with nAMD. Eye (Lond). 2021;35(11):2983-2990. doi:10.1038/s41433-020-01354-4