Validation of the Non-Mydriatic and Mydriatic Remidio (FOP) Camera with Artificial Intelligence in Screening Diabetic Retinopathy at Northern-Tanzania (2023-2024)
Ally Magero *
Mbeya Zonal Referral Hospital, Tanzania.
William Makupa
Kilimanjaro Christian Medical Center, Tanzania.
Maria Kissanga
Kilimanjaro Christian Medical Center, Tanzania.
*Author to whom correspondence should be addressed.
Abstract
Aims: This study aimed to evaluate the validity of non-mydriatic fundoscopy using a Remidio FOP Camera with Artificial intelligence to screen for Diabetic Retinopathy in Northern Tanzania, 2023-2024.
Study Design: This study employed a community-based cross-sectional design.
Place and Duration of Study: The study was conducted at the Diabetic and eye clinics of the health facilities of the Kilimanjaro and Arusha regions for a duration of 10 months, for data collection from October 2023 to July 2024.
Methodology: Type 2 Diabetic patients from 18 years and above were included in this study, and all patients with pupillary disorders that led to pupillary obstruction. The calculated sample size was 380 eyes in this study from 380 patients, This includes only one eye of interest from each participant, then the fundus images were taken from each consented participant, being non-mydriatic and mydriatic and then compared for quality and diagnosis through the AI (Medios). The reference group was the Mydriatic arm since it is the standard way of doing Fundoscopy.
Results: The sensitivity and specificity of non-mydriatic Remidio FOP Camera to grade fundus images was 43.9% (95 CI: 41.9 – 46.3) and 98.8 (95 CI: 94.2 – 99.7) respectively with PPV 99.3% (95 CI: 93.5 -99.7) and NPV 34.8 (95 CI: 31.6 -37.7) and Kappa Agreement score of 0.26 (95 CI: 0.000 – 0.008) with the p-value < 0.005. Also the sensitivity and specificity of mydriatic Remidio FOP has 87.5% sensitivity (95 CI: 83.2 – 89.7) in diagnosing Diabetic Retinopathy (DR) from gradable images with a specificity of 88.5% (95 CI: 82.9 – 88.7) in giving No Diabetic Retinopathy (No DR) gradable images with a Positive predictable value of 78.9% (95 CI: 73.3 – 79.1) and a Negative predictable value of 93.5% (95 CI: 93.1 – 99.7). This shows a Mydriatic FOP Camera when used has the most chance of giving a correct diagnosis of DR or No DR.
Conclusion: The Remidio FOP camera can be a useful equipment for screening Diabetic retinopathy cases on outreaches and hospitals, BUT it has to be used as a mydriatic tool to give a more precise images of the fundus, and the use of Artificial intelligence has revealed significance in grading the images as well.
Limitations: The artificial intelligence used can give only a diagnosis of Diabetic retinopathy or not which hinders further assessment and types of DR.
Recommendations: More Artificial intelligence algorithms should be developed to be able in diagnosing detailed Diabetic retinopathy.
Keywords: FOP-fundus on photo, DR- diabetic retinopathy, AI – artificial intelligence, DM – diabetic mellitus