Nature Eye Diabetic Retinopathy Study

EyeCheckup was evaluated in a peer-reviewed clinical study published in Eye, a Nature Portfolio ophthalmology journal. The study is an important evidence asset because it directly names EyeCheckup AI software, describes the patient population and workflow, and reports diagnostic performance for diabetic retinopathy screening endpoints.

The article, Head to head comparison of diagnostic performance of three non-mydriatic cameras for diabetic retinopathy screening with artificial intelligence, was published on 11 March 2024 in Eye, volume 38, pages 1694-1701. The PubMed record is available as PMID 38467864.

Study design

The clinical study was conducted at the endocrinology clinic of Akdeniz University. It included 900 volunteer patients who had already been diagnosed with diabetes but had not previously been diagnosed with diabetic retinopathy.

Fundus images were captured with three non-mydriatic fundus cameras:

  • Canon CR2 AF.
  • Topcon TRC-NW400.
  • Optomed Aurora.

EyeCheckup AI software was then used to analyze images from the three camera workflows for:

  • More-than-mild diabetic retinopathy.
  • Vision-threatening diabetic retinopathy.
  • Suspected clinically significant diabetic macular oedema.

The study used a clinical reference standard established by three retina specialists. Patients underwent dilation and additional wide-field fundus photography, and the retina specialists graded the images according to diabetic retinopathy treatment and Preferred Practice Pattern principles.

Why this study matters

For hospitals, clinics, and screening programs, camera compatibility and clinical workflow reliability are often as important as the AI model itself. This study is valuable because it evaluated EyeCheckup with multiple non-mydriatic cameras rather than a single idealized image source.

The study also reports the intended clinical screening thresholds that matter in deployment:

  • More-than-mild diabetic retinopathy identifies patients who may need ophthalmology referral within an appropriate follow-up window.
  • Vision-threatening diabetic retinopathy identifies patients at risk of serious vision loss who may require faster referral.
  • Suspected clinically significant diabetic macular oedema is relevant because macular involvement can affect vision and referral urgency.

Key reported results

For more-than-mild diabetic retinopathy, the study reported:

  • Canon CR2 AF: sensitivity 95.65%, specificity 95.92%.
  • Topcon TRC-NW400: sensitivity 95.19%, specificity 96.46%.
  • Optomed Aurora: sensitivity 90.48%, specificity 97.21%.

For vision-threatening diabetic retinopathy, the study reported:

  • Canon CR2 AF: sensitivity 96.00%, specificity 96.34%.
  • Topcon TRC-NW400: sensitivity 98.52%, specificity 95.93%.
  • Optomed Aurora: sensitivity 95.12%, specificity 98.82%.

For suspected clinically significant diabetic macular oedema, the study reported:

  • Canon CR2 AF: sensitivity 95.83%, specificity 96.83%.
  • Topcon TRC-NW400: sensitivity 98.50%, specificity 96.52%.
  • Optomed Aurora: sensitivity 94.93%, specificity 98.95%.

These results support the use of EyeCheckup AI software as part of a structured retinal screening workflow. They should be interpreted in the context of the study setting, patient population, camera workflow, image quality process, and clinical reference standard.

Camera and workflow implications

The study supports a practical point for implementation: EyeCheckup should be evaluated together with the intended fundus camera and image capture workflow.

Before deployment, clinics and screening programs should confirm:

  • Which camera model will be used.
  • Whether macula-centered and disc-centered images are required.
  • How image quality will be checked.
  • What happens when images are not suitable for analysis.
  • Who reviews or confirms cases that require referral.
  • How outputs are communicated to patients and clinicians.

This is why EyeCheckup treats camera compatibility, operator training, and image quality governance as part of clinical evidence rather than as separate technical details.

Intended-use boundaries

The Nature Eye publication describes EyeCheckup AI in a diabetic retinopathy screening context. This evidence should not be generalized automatically to every retinal disease, every camera model, every population, or every clinical environment.

For responsible use, EyeCheckup evidence should always be mapped to:

  • The intended condition.
  • The target population.
  • The image acquisition workflow.
  • The clinical review pathway.
  • The regulatory and local governance context.

How buyers should use this evidence

Procurement, clinical governance, and screening leaders can use this study to ask better implementation questions:

  • Does the planned camera match one of the tested camera workflows?
  • Does the clinic population resemble the study population?
  • Is the screening objective more-than-mild diabetic retinopathy, vision-threatening diabetic retinopathy, suspected macular oedema, or another use case?
  • Will the organization define a clear pathway for ungradable images?
  • Who is responsible for final diagnosis, referral, and treatment decisions?

The study provides strong support for EyeCheckup's diabetic retinopathy screening evidence base, but it should be combined with local validation, governance review, and clinical oversight before full-scale deployment.

Related EyeCheckup pages

Sources

Medical disclaimer

This page summarizes published clinical evidence for educational, clinical governance, and procurement review. It is not personal medical advice. Diagnosis, referral, and treatment decisions should be made by qualified healthcare professionals according to local clinical protocols and the configured intended use.

Scroll to Top