Skip to main content

National Eye Institute researchers have developed and validated an artificial intelligence-based method to assess patients with Stargardt, an eye disease that can cause vision loss in children. The method quantifies the disease-related loss of light-sensitive retinal cells, providing information to monitor patients, understand the genetic causes of the disease, and develop therapies to treat it. The findings published today in JCI Overview.

“These results provide a framework for assessing the progression of Stargardt disease, which will help control significant patient-to-patient variability and facilitate therapeutic trials,” said Michael F. Chiang, MD, director of the NEI, which is part of the National Institutes. of health.

About 1 in 9,000 people develop the most common form of Stargardt, or ABCA4-associated retinopathy, an autosomal recessive disease caused by variants of the ABCA4 gene, which contains genetic information for a transmembrane protein in light-sensitive photoreceptor cells. People develop Stargardt when they inherit two mutated copies of ABCA4, one from each parent. People who have only one mutated copy of ABCA4 are genetic carriers, but do not develop the disease. Rarer forms of Stargardt are associated with variants of other genes.

Yet even among patients who all have ABCA4 gene variants, there can be a wide range in terms of age of onset and disease progression. One patient may have very early loss of light-sensitive photoreceptors throughout the retina, while another may be an adolescent with involvement limited to the fovea, the area of ​​the retina that provides the sharpest central vision whose one needs to read and see other fine details. Yet another patient can reach their forties without loss of vision.

“Different variants of ABCA4 are probably responsible for the different characteristics or phenotypes of the disease. However, conventional approaches to analyzing structural changes in the retina have not allowed us to correlate genetic variants with phenotype,” said study co-lead Brian P. Brooks, MD, Ph.D. ., head of the NEI Ophthalmic Genetics & Visual Branch of function. Dr. Brooks co-directed the study with Brett G. Jeffrey, Ph.D., Head of the Human Visual Function Core of NEI’s Ophthalmic Genetics and Visual Function Branch.

The researchers followed 66 Stargardt patients (132 eyes) for five years using a retinal imaging technology called spectral domain optical coherence tomography (SD-OCT). SD-OCT 3D cross-sectional retinal images were segmented and analyzed using deep learning, a type of artificial intelligence in which huge amounts of imaging data can be fed into an algorithm, which learns then to detect patterns that allow the images to be classified.

Thanks to the deep learning method, the researchers were able to quantify and compare the loss of photoreceptors and different layers of the retina according to the patient’s phenotype and ABCA4 variant.

Specifically, the researchers focused on photoreceptor health in an area known as the ellipsoid zone – a feature of the inner/outer segment boundary of the photoreceptor that is diminished or lost due to disease. The researchers also examined the outer nuclear layer in the immediate region surrounding the ellipsoid zone loss zone.

They found that loss of the ellipsoid zone (a measure of severe photoreceptor degeneration) and thinning of the outer nuclear layer beyond these zones (a measure of subtle photoreceptor degeneration) followed a temporal and predictable space. Based on this predictability, they could generate a way to classify the severity of 31 different ABCA4 variants.

Importantly, they also discovered that photoreceptor degeneration was not limited to the area of ​​ellipsoid zone loss. Instead, progressive thinning of the photoreceptor layer—subtle to the physician’s eye, but quantitatively measurable—was evident in areas away from the ellipsoid zone loss boundary. This represented the actual main front of the disease, suggesting that this would be an area to watch closely to determine if a new therapy was having an effect.

“We now have sensitive structural outcome measures for Stargardt disease that are applicable to a wide range of patients, which is essential to move forward with therapeutic trials,” Jeffrey said.

The study was funded by the NEI Intramural Research Program. The study was conducted at the NIH Clinical Center, ClinicalTrials.gov ID: NCT01736293.