Smartphone-Based Telescreening May Benefit Detection of Retinopathy of Prematurity

J. Peter Campbell, MD, MPH

Credit: OHSU

According to new research, low field-of-view, cost-effective smartphone-based fundus imaging (SBFI) systems may be both effective and safe in telemedicine screening for treatment-requiring retinopathy of prematurity (ROP).¹

The analysis of 2 different SBFI systems suggests neither approach was superior to the other, but both human graders and an autonomous artificial intelligence (AI)-based system were able to appropriately refer treatment-requiring ROP using SBFI photos.

“These results suggest that these SBFI or other lower field-of-view camera systems may be safely and effectively used for ROP telescreening in regions where widefield digital fundus imaging (WDFI)-based ROP telemedicine or in-person ophthalmoscopic screening are not available,” wrote the investigative team, led by J. Peter Campbell, MD, MPH, Department of Ophthalmology, Casey Eye Institute, Oregon Health & Science University.¹

Limited screening ability in low- and middle-income countries has contributed to the significant disease burden of ROP. The use of telemedicine with remote grading of images obtained by WDFI cameras could be an effective way to improve screening in these regions by expanding the geographic reach of limited specialists. However, a significant barrier to the implementation and subsequent scaling of telemedicine programs is the high costs required to provide regions with WDFI cameras.²

Lower cost options, including either handheld cameras or SBFI systems, have been piloted, but provide a narrower field of view than WDFI cameras and have not been tested in a real-world, operational telemedicine setting. SBFI systems work by combining modern smartphone cameras with standard indirect ophthalmoscopy lenses to produce an affordable, portable, noncontact fundus imaging system. For the current analysis, investigators compared a WDFI device currently deployed in an operational telemedicine program (Retcam Shuttle) with 2 similar SBFI devices (Make-In-India [MII] Retcam adapter and head-mounted Keeler Monocular Indirect Ophthalmoscope [MIO] adapter).

In comparing the clinical diagnosis obtained using SBFI with WDFI in a masked fashion, investigators evaluated the efficacy of SBFI for the detection of referral-warranted (RW)-ROP and treatment-requiring (TR)-ROP, as well as the diagnostic accuracy of grader-assigned vascular severity score. Examinations occurred during routine image-based ROP screening at a single-center ROP teleophthalmology program in India between January 2021 - April 2022.

Two masker readers were presented with patient demographic characteristics, including birth weight and gestational age, and determined ROP diagnoses, including zone, stage, plus, and presence of aggressive ROP, and VSS (from 1 - 9) for individual eyes. Investigators calculated the sensitivity and specificity of any ROP, RW-ROP, and TR-ROP for each SBFI device relative to WDFI using Pearson χ² testing. In addition, they calculated the area under the receiver operative characteristic curve for the average grader-assigned VSS compared to RW-ROP and TR-ROP.

The data set included 312 eyes from 156 infants (mean gestational age, 33.0 weeks; 75 [48%] female) were included for paired analysis. The analysis indicated sensitivity and specificity were not found to be statistically different between the 2 SBFI systems, so results were combined. Investigators found the SBFI devices had only moderate sensitivity for the diagnosis of more than mild ROP (59%).

As a screening tool, however, using the cutoff of at least RW-ROP identified on the device, the sensitivity for detection of TR-ROP was 100% and the specificity was 83.49% with human graders. Moreover, data showed the AUCs with grader-assigned VSS were 0.95 (95% CI, 0.91 - 0.99) for RW-ROP and 0.96 (95% CI, 0.93 - 0.99) for TR-ROP.

Images were preprocessed and separated into training (70%), validation (10%), and test (20%) data sets. The algorithm then assigned a diagnosis of no plus, preplus, or plus to each image. After a cutoff of a median diagnosis of preplus or plus, the sensitivity of detection for TR-ROP was 100% with a specificity of 53.9% and the sensitivity for RW-ROP was 83.3%, with 56.5% specificity, for the AI-based screening.

Based on these results, SBFI systems had similar performance to one another and could potentially be utilized in regions where WDFI-based ROP telemedicine or in-person screenings are not available. Campbell and colleagues suggested the main advantage of SBFI systems is a significantly lower cost, noting the Retcam Shuttle costs approximately $125,000 while the MIO system costs approximately $1608 and the MII Retcam system costs approximately $585.

“In other words, for the cost of a single Retcam Shuttle, approximately 75 to 200 of these systems could be deployed instead for telemedicine screening, potentially dramatically increasing screening coverage,” investigators wrote.

References

1. Young BK, Cole ED, Shah PK, et al. Efficacy of Smartphone-Based Telescreening for Retinopathy of Prematurity With and Without Artificial Intelligence in India. JAMA Ophthalmol. Published online May 11, 2023. doi:10.1001/jamaophthalmol.2023.1466
2. Patil J, Patil L, Parachuri N, et al. Smartphone based ROP (S-ROP) screening-opportunities and challenges. Eye (Lond). 2020;34(9):1512-1514. doi:10.1038/s41433-020-0913-1