A new study funded by the National Eye Institute (NEI) explored the use of cost-effective smartphone-based fundus imaging (SBFI) systems to screen for retinopathy of prematurity (ROP), both with and without the assistance of artificial intelligence (AI).
Tell me about the research.
Conducted between January 2021 and April 2022, a cross-sectional comparison study screened 156 premature infants (312 eyes) in a single-center ROP teleophthalmology program in India using SBFI systems and widefield digital fundus imaging (WDFI).
Two masked readers evaluated all images for the following criteria:
- Zone
- Stage
- Plus
- Vascular severity scores (VSS)
What types of SBFI systems were used?
SBFI imaging was taken using one of two devices: the Make-In-India (MII) Retcam and the Keeler Monocular Indirect Ophthalmoscope (MIO).
Gotcha. Then what did they do?
The images were then arranged into three data sets—training (70%), validation (10%), and test (20%)—and used to train a deep learning program (ResNet18) that assessed binary classification of the data (normal vs. preplus or plus disease) in comparison with human graders.
And the findings?
According to the study authors, there were no statistically significant differences between the two SBFI systems, and human graders were able to effectively use these images to detect referral-warranted ROP (RW-ROP) and treatment-requiring ROP (TR-ROP).
The AI system was similarly effective, with a specificity of 58.6% compared to the human grader’s specificity of 83.49% for TR-ROP.
So what does that mean?
The authors concluded that SBFI systems may be more cost-effective than WDFI systems; however, they have a narrower field of view and are not yet widely used in real-world telemedicine settings.
And for the future?
This study offers suggestive evidence that these systems are effective, and utilizing them in telemedicine settings could mean expanding access to eyecare to detect ocular disease, particularly in low- to middle-income countries.