Optical coherence tomography angiography (OCTA) quantification using a new multilayer segmentation (MLS) vs fast segmentation (FS) in healthy and patients with diabetic retinopathy (DR)

Abstract

Purpose : Vessel density in the retina is a helpful metric for DR diagnosis in OCTA scans. For OCTA with CIRRUS AngioPlex, the superficial retinal layer (SRL) has been found to be the most useful in distinguishing early DR from healthy eyes. The goal of this study is to determine if this was still true with a newly developed MLS algorithm that more accurately identifies the locations of the layer boundaries.

Methods : OCTA scans were obtained by CIRRUS^TM HD-OCT 5000 with AngioPlex^® (ZEISS, Dublin, CA) using an Angio 3x3mm scan pattern. A total of 164 subjects (55 healthy and 109 DR patients) had one eye tested. The layers were calculated with both segmentation algorithms (MLS and CIRRUS algorithm (FS)), and angio slabs were generated: SRL (internal limiting membrane (ILM) to inner plexiform layer (IPL) – 10µm), deep retinal layer (DRL) (IPL – 10µm to outer plexiform layer (OPL)) and retina layer (ILM to smoothed retinal pigment epithelium (RPEFit) - 70µm) slabs. Decorrelation tail removal algorithm was applied to the DRL slab. The OCTA quantification algorithm from CIRRUS was applied to these slabs to determine vessel density. Diagnostic efficacy for differentiating normal eyes from early DR was determined for each slab based on MLS and FS by calculating the area under the receiver operating characteristic (AUC). Pairwise AUC comparisons were performed with the Method of DeLong.

Results : AUC was 0.78 using SRL derived from FS and 0.77 from MLS (comparison P = 0.66). For the deeper layer, the AUC is 0.61 for FS and 0.64 for MLS (comparison P = 0.27). Considering only patients with more severe DR (categories 15 and above) both algorithms perform just as well (comparison P > 0.05 for SRL and DRL) (see Table 1 for all values).

Conclusions : Vessel density measured in the SRL shows the best ability to distinguish healthy eyes from eyes with DR. Performance is similar using a new MLS algorithm to performance when deriving the layers using a simple estimation. In previous work, it was observed that capillary closure was seen less well in the full retinal layer than the SRL, and this work confirms that (AUC_SRL > AUC_Retina, comparison P = 0.10).

This is a 2020 ARVO Annual Meeting abstract.

 

Figure 1: MLS vs FS comparing slab type for a) all DR categories and b) DR categories 15 and above.

 

Table 1: Comparing AUC for MLS and FS considering different severity categories.