MAP AND TRACK AMD WITH MACHINE LEARNING AND AI

Prof. Schmidt-Erfurth
Prof. Schmidt-Erfurth

Severe vision loss has increased by 24 percent over the past 15 years1, and yet we now have the most advanced screening technologies. We need to ensure that doctors are maximising the potential of these solutions,” said Professor Ursula Schmidt-Erfurth, Chair of the Department of Ophthalmology and Optometry at the Medical University of Vienna.

Professor Schmidt-Erfurth, a pioneer in the use of machine learning and artificial intelligence for diagnostic image analysis, believes the development of the CE marked RetinSight GA Monitor is set to have a profound effect on patient care. Launched in October 2023, the application is accessible through Heidelberg AppWay and is being used by clinicians in the European Union and researchers globally as a gateway to further analysis –

My group has analysed all the clinical trial data from the three Apellis trials, which is more than 2,000 patients over a two-year period. Our focus is on analysing gold standard OCT images to see biomarkers reflecting disease activity. With advanced analysis of the OCT image we can examine every pixel and investigate the two layers of interest. Measuring the Retinal Pigment Epithelium (RPE) can be fully automated and it is reproducible, reliable and in real time. The photoreceptor layer is the first to be affected and we can visualise this with RetinSight’s AI-based OCT tools.

Within a minute we can have a report showing the detailed loss of RPE and photoreceptors. We do not want to wait and observe: we want to offer treatment as soon as possible. But how do we know which patients will benefit from therapy? If we only see loss of pigment we know the lesion is not currently active. Studies indicate that this is around 25 percent of patients, so we review them every three months to catch the lesions when they become active.

Our work in advanced image analysis is enabling us to diagnose disease, monitor disease progression and identify the need for personalised, precision medicine. Millions of OCT images are being used to train deep learning algorithms to identify new biomarkers. AI is an assistant to the physician, working in real time. We get complete analysis for fast, informed decision making,” she says.

Dr Rosa Dolz-Marco
Dr. Rosa Dolz-Marco

Dr Rosa Dolz-Marco, a macular specialist from Valencia, has been running GA drug trials at the Oftalvist Clinic since 2014 and is also an advocate of machine learning –

For AMD care having the highest resolution OCT is essential, along with autofluorescence imaging for every patient. I like to have it as a baseline and repeat it once a year, or every six months if necessary. If you don’t have AI tools for GA diagnosis and monitoring it is very time consuming to segment the imaging data manually. AI speeds up the process and can also provide an estimation of the progression. In the next two years we will have even more tools for early detection and intervention, so we, as clinicians, need to keep educating ourselves as the condition is very variable amongst patients. Careful patient selection is critical: in some patients the progression of GA is static, and in others it is fast progressing – it is important to spend resources wisely.

Professor Frank Holz, Chair of the Department of Ophthalmology, University of Bonn, sees the tremendous value of utilising high-quality imaging to streamline care –

AI is essential. We are taking a sequence of images and generating a huge amount of data – reading them can be time consuming and a human grader cannot extract the degree of data that machine learning can provide. Referrals can filter patients in the community and, once in the referral pathway, can be assessed faster, and decisions made on monthly or bimonthly treatment. For many patients, injections every two months suffice – as was the case for 80 percent of patients in the US trial. Personalised medicine is important as injections can induce very variable reactions amongst patients – it is important to reduce overdosing and wasting resources.

Professor Frank Holz
Prof. Frank Holz

Both Dr Dolz-Marco and Professor Holz highlight the need for the efficient use of resources due to the increasing burden of an aging population on health services globally. Access to artificial intelligence solutions in clinic has the potential to increase efficiency and support distributed patient care. In addition to the new treatments for geographic atrophy, the growing number of patients and treatment options for neovascular AMD, including biosimilars, drug delivery systems, and even gene therapy, demand early detection, customised patient care and fast, confident selection and decision making.

Therapeutic pathways are optimised for the diagnosis and monitoring of neovascular AMD. Demand for therapy for this patient cohort continues to increase in line with an aging population. Recent innovations in therapy have seen the introduction of long duration agents and surgically implantable devices proving sustained slow delivery of anti-angiogenic drugs. Even with new approaches to treatment, the burden on clinical services remains high. The addition of therapy for atrophic AMD will only confound the problem and innovation like AI-based clinical support tools and service modernisation will be the key to success.

RetinSight’s MDR-approved, CE Marked Fluid Monitor for wet AMD cases, now used by consultants in 30 clinics internationally, is enabling decision making based on rapid or slowly progressing disease, says Professor Schmidt-Erfurth.

“The AI tool, used in conjunction with HEYEX 2 and Heidelberg AppWay, produces an immediate report that visualizes where the fluid is and measures the volume: everything is precise in real time. This brings an enormous improvement in workflow and early adopters say their productivity in clinics has improved by 40 percent,” she says.

“We have moved on considerably since central retinal thickness was the single measurement. Distinct types of fluid have different impacts on visual acuity – intra-retinal fluid is the most dangerous, while sub-retinal fluid can be tolerated to a degree.

We need to know if the fluid is dynamic and the profession needs to know when to treat and how much – a personalised approach leads to faster, more efficient treatment without wasting resources,” she adds.

“The simple one-page Fluid Monitor Report indicates the treatment pathway - it is objective, reproducible and enhances overall understanding. Clinicians can feel confident that they are giving bespoke treatment at the right time - the retinal specialist is still pivotal in decision making.”

Christopher Mody
Christopher Mody

“Using automated analysis of OCT images to map and track AMD is now as easy as taking blood pressure – and just as important,” explains Christopher Mody, Clinical Director for Heidelberg Engineering.

“Artificial intelligence tools can also be a valuable aid to patient compliance, with the ability to visualise change from visit to visit and provide the full history of three key biomarkers of intra-retinal fluid, sub-retinal fluid and retinal pigment epithelium detachments,” he adds.

“High resolution imaging with AI analysis accessed through application marketplaces, like Heidelberg AppWay, has the potential to transform workflow and support the effective management of the growing number of AMD patients globally. To maximise the benefit of machine learning and AI solutions for delivering individualised treatment plans, clinicians need to educate themselves, carry out diligent case reviews, identify those patients with the highest risk of progression to sight threatening AMD, and establish robust review and monitoring,” he concludes.


Back to the overview