Textural OCT-A Feature May Enable Early Diabetic Retinopathy Detection

A textural feature for image classification in the nasal and inferior sectors of the superficial vascular plexus (SVP) could differentiate patients with early nonproliferative diabetic retinopathy (NPDR) from healthy individuals, according to a quantitative assessment of optical coherence tomography-angiography (OCT-A) images  published in Photodiagnosis and Photodynamic Therapy.

Researchers included patients with type 2 diabetes who had mild or moderate NPDR (n=10; mean age, 60.90 years) and healthy control group participants (n=41; mean age, 46.41 years) in the single center investigation between 2019 and 2021. All study participants underwent a full ophthalmic examination that included OCT-A imaging. Textural features, which include contrast, correlation, entropy, energy, and homogeneity, and fractal dimensions (FDs), were evaluated for NPDR diagnostic utility.

Control group participants were stratified according to age into 1 of 5 subgroups. Group 1 was aged 24.60 years, group 2 was aged 35.10 years, group 3 was aged 45.00 years, group 4 was aged 54.90 years, and group 5  was aged 64.14 years. For each subgroup, normal superior, temporal, nasal, and inferior sector FDs were defined for the nerve fiber layer vascular plexus (NFLVP), SVP, intermediate capillary plexus (ICP), and deep capillary plexus (DCP). Compared between subgroups, some FDs differed significantly, including the superior and temporal sectors of the FNLVP, inferior sector of the SVP, inferior and nasal sectors of the ICP, and inferior sector of the DCP. Most differences were observed when comparing FD values between participants in groups 2 to 5 and the youngest participants in group 1. 

Using these data as reference, FD features had diagnostic accuracies with sensitivities ranging from 0.732 to 0.951 and specificities from 0.700 to 1.000, the report shows.

The best diagnostic predictor for early DR was the nasal sector of the SVP with an area under the receiving operator characteristic curve (AUROC) of 0.932, sensitivity of 0.951, and specificity of 0.800. The next best predictor was the inferior SVP sector with an AUROC of 0.910, sensitivity of 0.756, and specificity of 1.00.

“The high AUROC values of the comparisons between diabetic patients in early stages of DR and normal healthy subjects further demonstrated that the contrast and correlation in particular sectors had the diagnostic power to differentiate diabetic patients with the early stages of DR from normal healthy subjects,” according to the researchers. “And the highest AUROC of the correlation in inferior and nasal sectors in SVP not only suggested the correlation was a promising feature in the early detection of DR with [OCT-A], but also suggested the microvasculature alterations in SVP for the diabetic patients in early stages of DR.”

A small sample size is an acknowledged limitation to the research.