AI could soon tell patients how often to see the ophthalmologist

Age-related macular degeneration (AMD), retinal vein occlusion (RVO), and diabetic retinopathy (DR) are all commonly treated by injections of an anti-vascular endothelial growth factor (anti-VEGF) into the eye at intervals to slow down disease progress and prevent blindness.

The number of people affected by diabetes is expected to exceed 640 million by 2040.

That fact fosters an environment in which the barriers of the current telescreening programs, i.e., cost and delays in image grading, can be surmounted and DR detected earlier followed by rapid treatment. 

Once an AI system gains the ability to recognize patterns or markers of a disease, it can become a tool for automated diagnosis.

With eyesight a central human sense in jeopardy from these diseases, patients are eager to know that they are being treated often enough to avoid rapid worsening of their condition.

Ophthalmologists want to make sure they see each patient frequently enough not to miss important developments.

The challenge
In an aging population, cases of AMD, RVO or DME are on the rise globally, making it hard for specialized eye clinics to keep up with the growing demand for regular treatments.

"As doctors, we want to give each patient the necessary attention and treatment frequency that they need,” according to Sebastian Wolf, head of the Ophthalmology Department of the Inselspital that currently sees 6000 patients with AMD, RVO and DR. "But it is also an organizational challenge to meet all patients' needs and be able to study all relevant eye imaging data to assess individual disease progression and take treatment decisions in the short time given."

To monitor progression of the chronic eye conditions, Optical Coherence Tomography (OCT), an imaging tool that generates 3D images of the eye at extremely high resolution, is usually applied.

In collaboration with the ARTORG Center for Biomedical Engineering Research, the Inselspital has developed automated OCT analysis tools based on artificial intelligence, which can assist eye doctors in the assessment of a whole patient OCT-set in just a few seconds.

Together with RetinAI, a startup specialized in AI-based eye care technologies, they now have conducted a retrospective study of patients to assess how well AI can predict anti-VEGF treatment demand from the start.

The setup
According to researchers, the study looked at OCT-data from 340 patients with AMD and 285 patients with RVO or DME, treated with anti-VEGF at the Inselspital between 2014 and 2018.

Based on morphological features automatically extracted from the OCT volumes at baseline and after two consecutive visits, as well as patient demographic information, two machine learning models were trained to predict the probability of the long-term treatment frequency demand of a new patient (one for AMD and one for RVO and DME).

Based on the first three visits, it was possible to predict if a patient had a low or a high treatment demand for both the AMD and the RVO and DME groups with similar accuracy.

More importantly, the study revealed that it is possible to predict reasonably well at the initial visit and even before the first injection if a patient will less often require injections.

Three advantages
"We have shown that machine learning classifiers can predict treatment demand when a patient is first diagnosed with a chronic eye disease," Mathias Gallardo, postdoctoral researcher the ARTORG AI in Medical Imaging (AIMI) lab and member of the new Center for Artificial Intelligence in Medicine (CAIM), said in a statement. "Hence, artificial intelligence may assist in establishing patient-specific treatment plans for the most common chronic eye conditions in the near future."

Planning the ideal treatment frequency for each patient has multiple benefits.

Firstly, patients can be sure their disease is being treated in the best way possible without subjecting them to too frequent visits and unpleasant injections into the eye.

Secondly, individualized planning can help clinics cope with ever growing patient numbers allowing for the highest possible capacity utilization of specialized medical skills and infrastructure.

Thirdly, objectivized on-demand planning helps avoid overprovision and can lead to improved cost-efficiency and less healthcare expenditures.

High-yield confluence of clinical, data science, and industrial research
This study illustrates once more the proven eye-level collaboration between clinicians and data scientists of the Inselspital and the ARTORG Center, which produces technology solutions suitable for everyday use because they were designed directly as a response to clinical needs.

A further important aspect to provide a roadmap for the clinical implementation of such technology was the startup RetinAI.

"We are extremely happy to apply the EU funding we received to build patient-focused solutions in ophthalmology, making sure that technology can be transformed into products that can really benefit patients and improve treatment at scale," RetinAI CEO Carlos Ciller said in a statement.

With its headquarters at sitem-insel the startup also is spatially located exactly at the interface between clinic and science.

This unique Bernese environment for clinically driven AI technologies will be further capitalized by the new Center for Artificial Intelligence in Medicine (CAIM), combining the best of the three worlds of healthcare, science, and industry for the benefit of patients.