Subtle Early Changes in Diabetic Retinas Revealed by a Novel Method that Automatically Quantifies Spectral Domain Optical Coherence Tomography (SD-OCT) Images

Abstract

Purpose : Diabetes inevitably results in diabetic retinopathy (DR) with time. SD-OCT imaging is used to assess diabetic changes in the retina, however these changes are apparent only late in the disease, when the retinal anatomy is already compromised. The purpose of this study was to develop an automated algorithm to detect occult retinal changes on SD-OCT images prior to the development of obvious anatomical or clinical changes.

Methods : Spectral domain OCT scans (Zeiss Cirrus HD-OCT 5000) were obtained from 16 diabetic subjects aged 53-76 years without clinical evidence of DR. 16 healthy subjects, matched for sex and age were used as controls. Subjects with high myopia (≤ -6.00 diopters) or tilted OCT scans were excluded. Twelve distinct layers were first segmented using a novel automated algorithm that combines shape, intensity, and spatial information. A novel normalized reflectivity scale (NRS) ranging from 0 units (vitreous) to 1000 units (retinal pigment epithelium [RPE]) was then applied to calculate the reflectivity for each segmented layer from the raw data. Tortuosity of retinal layers was quantified using the mean absolute curvature κ of the boundary between adjacent layers.

Results : Normalized reflectivity varied significantly by diagnosis (F_1,658 = 24.18; p < 0.0001) and retinal layer (F_11,658 = 457.3; p < 0.0001). Reflectivity per layer differed by side of the fovea (F_11,658 = 6.13; p < 0.0001). Reflectivity in diabetic subjects in all layers was in average 35.1 NRS greater than in their matched controls (Figure 1). Curvature varied significantly by layer (F_11,658 = 11.9; p < 0.0001) and side (F_1,658 = 6.63; p = 0.010). Post hoc testing revealed significant differences in the IPL and INL where κ averaged 1.91 mm⁻¹ and 1.12 mm⁻¹ greater, respectively, in diabetes compared to control (Figure 2).

Conclusions : An automated image analysis algorithm allows the identification of subtle changes in diabetic retinas that occur prior to the disruption of normal anatomy. Detection of such changes carries the promise of early diagnosis of DR in diabetics.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.

 

Normalized reflectivity per layer and diagnosis. No layer showed significant differences between diabetics and controls.

 

Mean and standard deviation of tortuosity per layer and diagnosis. **Control and diabetic subjects show significant differences (p < 0.05).