Evaluation of Retinal Image Quality, Lesion Detection and Diabetic Retinopathy AI Performance Using a New Automated Non-Mydriatic Tabletop Fundus Camera: A Prospective Validation Against a Reference Standard System

Background

High-quality fundus imaging is critical for effective screening and diagnosis of retinal diseases, particularly diabetic retinopathy. Image quality assessment IQA plays a vital role in ensuring diagnostic accuracy and can be evaluated using both objective device-based and subjective human evaluator-based methods. Subjective assessment remains the most trusted approach, especially since clinical interpretation depends on human review.In this study, we will evaluate and compare the study device against two reference standard non-mydriatic fundus imaging systems

Study device Remidio InstaZ Non-Mydriatic Fundus Camera

Reference standard Topcon NW400 fundus camera system.

Additional Reference Standard device for AI evaluation, objective 2 Remidio Fundus-On-Phone NM10 FOP NM-10

The FOP NM-10 is a compact, smartphone-based, non-mydriatic camera that offers a 40° field of view and is designed for accessibility, ease of use, and cost-effectiveness. It includes an in-built image quality assessor and is equipped with offline AI algorithms for detecting diabetic retinopathy and glaucoma. These AI models have demonstrated strong diagnostic performance, with validated sensitivity and specificity rates of up to 100 percentage, 93.4percentage, and 88.4percentage, 85.4percentage for DR and glaucoma, respectively.

The InstaZ study device, in contrast, is a high-resolution tabletop fundus camera equipped with features such as 3D auto-tracking, auto-capture, and a wider 45 degree field of view. It allows imaging without dilation or dark room conditions and supports imaging enhancing diagnostic versatility. This device does not consists of an inbuilt Artificial Intelligence AI for detecting DR.

The primary objective of the study is to assess and compare the image quality of fundus photographs captured using an automated tabletop non-mydriatic fundus camera Remidio InstaZ and reference standard fundus camera system Topcon NW400. Another primary objective is to evaluate the device agnostic performance of MediosHI DR artificial AI algorithm integrated on FOP NM10 when applied to the novel study device images in diabetic individuals by running the InstaZ images through the same AI algorithm in script form.

Methodology

Consecutive adult patients Greater than 18 years attending the general ophthalmology unit and willing to provide informed consent, will be recruited. Participants must have good fixation and be cooperative during imaging. Exclusion criteria include significant media opacities e.g., advanced cataract, corneal opacity, inability to fixate e.g., nystagmus, amblyopia, active ocular infection or post-operative status, hypersensitivity to light, or contraindications to pharmacological dilation e.g., angle-closure glaucoma.

All participants will undergo a standardized eye examination including history, vision assessment, refraction, slit-lamp evaluation, and dilated fundus examination. Following dilation, all the participants will undergo cataract grading if any using LOCS III classification system by trained ophthalmologist.