Abstract
Purpose :
AMD is the leading cause of irreversible blindness among the elderly in the developed world. Early detection and management can prevent blindness. We assessed the performance of a novel AI system deployed on a portable fundus camera for screening referable AMD in a South Asian population.
Methods :
This prospective study was conducted in a retina clinic of a tertiary eye hospital. Macula centred images were captured in both eyes on a standard fundus camera (Zeiss Clarus 700) and the study device (smartphone-based fundus camera). Additionally, a line scan across the macula was captured using Spectralis SD-OCT. Three fellowship-trained retina specialists provided a masked diagnosis of AMD (Gold standard) based on a pre-defined criteria using both SD-OCT and retinal images from the standard fundus camera. Consensus grading was regarded as final and was used to compare the AI. The AI indicated the presence (intermediate and advanced AMD) or absence (early or No AMD) of referable AMD using fundus images alone captured on the study device.
Results :
We analysed the results of 84 eyes of 45 patients (59% male) with a mean age of 65.7±9.8 years (range 43, 92) and 64 eyes (76%) had referable AMD. The Cohen’s kappa agreement between the graders and consensus grading was 0.84 (grader 1), 0.77 (grader 2), and 0.94 (grader 3). The AI system had a sensitivity of 96.8% (62 out of 64 cases; 95% CI 89.16% to 99.62%) for referable AMD. The two false negatives missed were intermediate AMD cases. Among 20 no referable AMD, 10 eyes were flagged referable by the AI system and among them, five were diagnosed with early AMD and five as no AMD by the specialists.
Conclusions :
Early analysis shows promising results for screening referable AMD. However, the final results would demonstrate the AI’s true diagnostic accuracy. It has the potential to make AMD screening accessible, affordable, and effective.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.