Fundus photography technique, clinical site experience, and the ability to acquire interpretable diabetic retinopathy fundus photos in a telemedicine program

Abstract

Purpose : In preparation for an artificial intelligence diabetic retinopathy initiative at Temple University Hospital, we changed our fundus photography screening protocol from a single fundus image per eye to two fundus images per eye. We also expanded the screening program from one primary care office to eight. We wished to determine whether these changes resulted in a significant difference in acquiring interpretable photographs and if there was a learning curve for new fundus photographers when expanding clinical sites.

Methods : The one-image protocol required a single macula-centered fundus photograph while the two-image protocol required both a macula-centered and an optic-nerve centered fundus photograph. The one-image protocol was in use at one site from 2016-2017 and the two-image protocol was in use at eight sites (including the original site and seven new sites) from 2018-2020. The percentage of interpretable photographs for each protocol and for each clinical site was measured and compared.

Results : 1377 photographs were taken using the one-image protocol and 928 (67.4%) were deemed interpretable. 3804 photographs were taken using the two-image protocol and 3009 (79.1%) were deemed interpretable (Figure 1). This was a statistically significant improvement (Χ² = 237.29, p-value < 0.001). When the 2018-2020 retinal screening exams were compared between the original site and new sites, there was no statistical difference in percentage of interpretable photos (Χ² = 2.46, p-value = 0.11), (Table 1).

Conclusions : Adding a second field of view yielded a greater percentage of interpretable photographs, which suggests that the two-image protocol may have helped minimize the impacts resulting from multiple additional clinical sites and varying photographer experience.

This is a 2021 ARVO Annual Meeting abstract.

 

Figure 1. Comparing percentage of interpretable photos between one-image and two-image protocols

 

Table 1. Comparing the number of interpretable photos betwen the original site and the new sites in 2018-2020