Automated Quantitative Analysis of the Fovea using OCT Angiography

Abstract

Purpose : To present an automated pipeline for quantitative analysis of the foveal vasculature and Foveal Avascular Zone (FAZ) acquired with Optical Coherence Tomography Angiography (OCTA).

Methods : Twelve eyes from six normal human subjects were imaged with a 1060-nm, 100-kHz custom-built OCTA system. Automated techniques were used to quantify the FAZ metrics (area, greatest diameter, and lowest diameter) and capillary density surrounding the FAZ. Deep convolutional neural networks were used for automated segmentation of the retinal microvasculature in order to calculate the capillary density.

Results : The morphometry of the FAZ and perifoveal capillaries determined by the automated tools were compared with the results from a human rater. The minimum diameter (manual: 501μm ± 72μm, automated: 482μm ± 75μm), maximum diameter (manual: 734μm ± 103μm, automated: 733μm ± 116μm) and area (manual: 0.308mm2 ± 0.074mm2, automated: 0.294mm2 ± 0.071mm2) were calculated. The accuracy of the automated blood vessel segmentation was evaluated by pixel-wise comparison of the manually segmented image with the thresholded output of the neural network. Using this method, blood vessel segmentation reached a mean accuracy of ~81%.

Conclusions : The methods used here for automated quantitative analysis of OCT Angiography were shown to be accurate when compared to a manual rater. Further work is required is validate the utility of these methods in creating an automated retinal vascular disease screening system.

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

 

Left: Automated results of the minimum (red) and maximum (green) FAZ diameter and perimeter (yellow). Middle: Automated vessel segmentation results. Right: Capillary perfusion map.