AI Quantifies Loss of Retinal Cells in Stargardt

Download PDF

NEI researchers have developed and validated an artificial intelligence (AI)–based method that surpasses conventional approaches to analyzing structural changes in the retina related to ABCA4-associated retinopathy, the most common form of Stargardt dis­ease.1 Their model provides a frame­work for evaluating disease progression and appears to help control for the significant variability among patients for age of onset and spatial pattern of photoreceptor degeneration.

“Despite the apparent heterogeneity of patients with Stargardt disease, a simple additive model of allele sever­ity, in conjunction with age, explains almost half of the between-patient vari­ability in imaging-based severity,” said lead author Maximilian Pfau, MD.

SHEDDING LIGHT. The presentations of Stargardt disease vary widely. Previous approaches to analyzing structural changes in the retina have not allowed researchers to correlate genetic variants with disparate disease characteristics. This image was originally published in the ASRS Retina Image Bank. Becca Harris and Liliya Sutherland, DO. Stargardt Disease. Retina Image Bank. 2021; Image Number 85107. © The American Society of Retina Specialists.

Homing in on the ellipsoid zone. The researchers used thickness maps of six retinal layers from OCT images of 132 eyes (66 patients) to train a deep-learn­ing (DL) algorithm to quantify retinal degeneration over time. These images enabled the algorithm to detect pat­terns for quantifying and comparing loss of photoreceptors and layers of the retina that correlate ABCA4 variants with a patient’s phenotype.

Researchers focused their interest on ellipsoid zone (EZ) loss, a measure of severe photoreceptor degeneration. They also considered the outer nuclear layer beyond the areas of EZ loss. Final­ly, they compared measures of macula-wide photoreceptor loss, genotype, and age against the genetic data.

Outcomes. Photoreceptor losses fol­lowed predictable spatial and temporal patterns. Importantly, patients with ABCA4-associated retinopathy exhibit­ed photoreceptor degeneration beyond the area of EZ loss. The researchers propose that this distant area may be the actual leading front of the disease.

The study also demonstrated dif­ferent rates of retinal change among patients, with the age of loss dependent on genotype. Thus, the DL model gen­erated a way of classifying the severity of 31 variants, including 16 variants not previously quantitatively analyzed for clinical severity.

Avenues for further research. Although the prediction model works well for macula-wide loss of photo­receptors, the genetic determinants of foveal sparing in Stargardt are less clearly understood and warrant further investigation, Dr. Pfau said.

With regard to potential therapies, a clinical trial of oral metformin for Star­gardt is now underway at the NEI, but it was initiated before this study was published. Dr. Pfau and his colleagues plan to apply the AI-based segmenta­tion to investigate potential treatment effects of metformin on photoreceptors distant to the EZ loss boundary.

Other therapeutic approaches to prevent progression of photoreceptor loss are underway, Dr. Pfau said. How­ever, these therapies will only slow the deterioration of vision, not improve it, and patients will want evidence of treatment effects. “Ultimately,” he said, “it will be necessary to translate these photoreceptor loss maps to patient-rel­evant measures, such as reading ability and mobility.”

—Miriam Karmel

___________________________

1 Pfau M et al. JCI Insight. 2022;7(2):e155373.

___________________________

Relevant financial disclosures—Dr. Pfau: Apellis: C; Novartis: L.

For full disclosures and the disclosure key, see below.

• Biomarkers Predict Anti-VEGF Response in AMD

• OCT Pins Down MOGAD and MS Diagnosis in ON

• Assessing Clinicians’ Ability to Dx Infectious Keratitis