Texture-based analysis of retinal OCT images can discriminate the early impact of type 1 and type 2 diabetes on the retina: evidence from animal models

The onset and progression of diabetic retinopathy (DR) are highly variable, highlighting the need for strategies to distinguish different degrees of diabetes-induced retinal dysfunction/damage, particularly during the early, "silent" disease stages. Using texture analysis on computed retinal images obtained from optical coherence tomography (OCT) data, we have recently reported early texture changes in the retina of type 1 diabetic animals exhibiting subtle cellular and molecular retinal alterations, as well as in type 2 diabetic retinas, exhibiting even subtler and more gradually evolving changes. In this work, we sought to compare those changes in retinal texture across type 1 and type 2 diabetes models. Type 1 diabetes was experimentally induced through an intraperitoneal (i.p.) administration of streptozotocin (65 mg/kg) and type 2 diabetes by maintaining animals on high-fat diet combined with an i.p. injection of STZ (35 mg/kg). OCT volume scans were acquired, followed by automated OCT data segmentation and texture analysis. Retinal texture differed between type 1 and type 2 diabetic animals, showing slope (rate of change over time) differences across all retinal layers in the following metrics: autocorrelation, cluster prominence, correlation, homogeneity, information measure of correlation II, and sum average. Moreover, type 1 diabetic animals showed negative slopes across all retinal layers, while type 2 diabetic animals exhibited near-zero slopes. Our findings suggest that retinal texture captures distinct grades of diabetes-induced retinal impairment, supporting its potential to distinguish between different early subclinical stages of DR, although this requires validation in clinical studies.