Automated Ischemia Segmentation using OCT Angiography in Diabetic Retinopathy

Abstract

Purpose : To evaluate whether an automated ischemia segmentation algorithm for optical coherence tomography angiography (OCTA) can identify disease severity of retinopathy in patients with diabetes.

Methods : Seventy-eight eyes from 40 patients with diabetes imaged on the RTVue XR Avanti AngioVue® platform were selected for image analysis. The dataset was chosen from consecutively imaged diabetic patients at New England Eye Center. No images were excluded for poor quality scans or motion artifact. We developed an algorithm which quantified geographic patches of non-perfusion on OCTA. This algorithm was able to identify poor quality scans and scale internal parameters to optimize signal to noise, thus no scans were excluded due to poor scan quality. The patients were stratified into five groups: diabetes without clinical diabetic retinopathy, mild/moderate/severe nonproliferative diabetic retinopathy (NPDR), and proliferative diabetic diabetic retinopathy (PDR). For each eye, the algorithm provided an OCTA ischemia index (OII) at the level of the superficial plexus, deep plexus, and choriocapillaris. An automated measurement of the FAZ was made at the level of the superficial and deep plexus. A chart review was performed to determine previous treatments (focal laser, PRP, anti-VEGF) as well as presence or absence of diabetic macular edema.

Results : The OII was validated with repeated measurements over 4 days from 10 normal eyes. The OII of the superficial plexus increases with worsening diabetic retinopathy severity, but this trend was not seen in the deep plexus or choriocapillaris. Figure 1 shows pairwise comparisons between the OII of the superficial plexus in each group, even when controlling for treatment. In the superficial plexus, there was a significant difference in the OII (p<.05) between diabetics without retinopathy and mild NPDR compared to all other groups. In the deep plexus, there was a significant difference in the OII between diabetics without retinopathy compared to those with retinopathy. Automated measurement of the FAZ showed a significantly larger FAZ in PDR compared to the other groups.

Conclusions : Automated algorithms can be used to assess the degree of ischemia in OCTA images from patients with diabetic retinopathy and may discriminate between different levels of disease

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