The technologies

EyeArt (Eyenuk), RetinaLyze (RetinaLyze System A/S) and Retmarker (Retmarker) use artificial intelligence (AI) technology to analyse retinal images to help diagnose diabetic retinopathy. The aim is to speed up and improve diagnosis. They could be used in national screening programmes and settings with limited expertise.

EyeArt is designed to work with most types of retinal cameras. The algorithm has been trained using data from over half a million people and 2 million retinal images. Colour fundus images are uploaded to the technology's GDPR-compliant cloud, which produces a PDF report in less than a minute. The technology grades the severity of diabetic retinopathy and reports results for each patient's eye on the NHS diabetic eye screening programme (NDESP) scale. It assesses the adequacy of the image at the time of photography. It also explains the reasoning behind the grading.

RetinaLyze can be accessed online or integrated into a healthcare centre's existing software systems. This technology needs someone to check the results, and claims detection of diabetic retinopathy but not diagnosis. This may change because the company has applied for class IIa CE marking.

Retmarker sends images securely to the data centre hosting the Retmarker technology and automatically removes images that show no signs of diabetic retinopathy lesions. It can be used on the healthcare centre's premises, and patients' data can be kept securely on just the centre's server if needed. The AI technology does not need individually identifiable patient data for its analysis. The company says that the technology can be used with most digital fundus cameras. The algorithm can incorporate patient history data to assess disease activity. Images with possible lesions and images of a high enough quality to be graded are sent to a clinician for manual triage and grading. The technology allows users to select the analysis modules they need. It can be integrated into the healthcare centre's software systems.


The companies claim these technologies speed up detection of diabetic retinopathy by automating the process of examining retinal images. Currently, trained retinal screeners examine each image to determine if there is diabetic retinopathy, which requires expertise and time.

The algorithms in the technologies were trained using a database of existing human-graded images. The algorithms were then tested against another set of images to fine-tune them for real-world use.

A potential benefit is that they could free retinal screeners to grade images at a higher level.

Current care pathway

Everyone aged 12 and over with diabetes has retinal screening using a retinal camera. The camera takes an image of the person's retina, which is then checked for changes. Screening is usually every year, but may be more frequent, depending on the findings from the first eye screening.

Diabetic retinopathy develops in stages over time. As it progresses, more frequent screening appointments (every 3 to 6 months) are needed. The condition may eventually lead to vision loss.

The following publications have been identified as relevant to the care pathways:

Population, setting and intended user

The technology is intended for people with diabetes aged 12 and over who have diabetic retinopathy screening. In 2018 an estimated 3,809,119 people had diabetes in the UK.

Screening takes place at GP practices, hospitals and opticians, and the technologies would be used by retinal screeners.

Policy implications

The UK National Screening Committee (UKNSC) is responsible for making recommendations on all aspects of screening programmes. This includes changes to tests that this technology would represent.

So this briefing should not be seen as a recommendation for use in diabetic eye screening programmes.

There are formal UKNSC evaluations of AI technologies happening in the diabetic eye screening pathways and formal recommendations will be made in due course.


Technology costs


The company has not provided the cost per patient for the UK. In the US, where EyeArt replaces human reading, Medicare is reimbursing healthcare providers up to £39.93 per patient for the EyeArt test. The reimbursement amount covers staff and infrastructure costs as well as the technology.


The service is available as a subscription and an organisation can buy a number of screenings upfront. Cost per patient varies depending on the number of patients, the hours the system is in operation and the subscription level. The cost is usually EUR 0.40 to EUR 2.00 (£0.35 to £1.73). There are additional costs for data management and storage, and security and accessibility requirements. The AI algorithms rely on either a standalone web application running on a Windows PC with an internet connection or a PACS/EHR system integration. Both have monthly or yearly cost and server costs. Online training is included in the cost.


In the UK, the average cost per patient is tiered by purchased volume of episodes. These are bought upfront and must be used within 1 year unless otherwise agreed. The minimum number of episodes is 50,000 (unless otherwise agreed) at £3.02 per patient. The average cost per patient reduces with higher numbers of purchased episodes, down to a minimum cost per patient of £0.86. Server costs are not included. The cost of integrating the AI system into an existing provider depends on the scope and may partially or fully be covered by Retmarker.

Costs of standard care

The standard method of grading digital colour photographs in the UK uses trained human graders who meet specific quality standards, with multiple possible levels of grading and quality control checks.

The mean cost per patient for manual screening is estimated as £4.79, and the total cost per patient is £9.92, accounting for a mix of graders at different bands doing the test. Total costs include quality assurance, including monthly tests and training. These were based on a study by Tufail et al. (2016) and adjusted for inflation to 2020/21 prices at 3.5% per year in line with NICE recommendations.

Resource consequences

Trusts adopting these technologies should take into account that they each have different outputs, some of which may not be compatible with the grading system being used by the trust. Trusts should also consider the costs of the required level of service, system integration, secure data transfer, and data storage and management. The physical infrastructure will not need changing, apart from installing the technologies into local server infrastructures. Little to no training is needed to use them.

These technologies could be placed in different parts of the care pathway – as the first, second or arbitration grader – and reduce the workload of retinal screeners.

The technologies can analyse and triage images rapidly and speed up the screening process. Two of the technologies have a class IIa CE marking, so they have the potential to be used without human oversight. If adopted, the technologies could be cost saving for the NHS if they could replace human graders. Some of the costs saved could be used to train graders in optical coherence tomography (OCT) imaging. Many retinal scans happen in private opticians that are reimbursed by the NHS. Without further review, it's unclear how adopting these technologies would affect NHS resource use if they are primarily used in a private setting.