Automated Method for the Long-term Quantiative Analysis of Macular Neovascularization using Optical Coherence Tomography Angiography (OCTA)

Abstract

Purpose : The present study aimed to evaluate an algorithm of automatic segmentation and quantification of macular neovascularization (MNV) and to evaluate quantitatively, throughout a long-term follow-up of the OCTA, the predictive value of these variables, in patients presenting age-related macular degeneration.

Methods : Consecutive patients with wet AMD underwent multimodal imaging, including optical coherence angiography tomography (OCTA) at baseline and last follow up. The Sorensen-Dice coefficient (DSC) was calculated to evaluate the performance of automatic segmentation versus manual segmentation. Changes in total area, vascular area, vascular density, and fractal dimension were correlated with the final exudative signs on Spectral Domain Optical Coherence Tomography (SD-OCT). Receiver Operator Characteristic (ROC) curves were generated to evaluate the predictive value of these variations.

Results : Forty-two eyes were included. The average follow up time was 14 months. The average DSC was 0.72 +/- 0.15 baseline and 0.72 +/- 0.13 follow up. The correlation coefficient between the automatic and manual measurements was in the range 0.464-0.949 (p = 0.0019, p <0.0001). The inter-class correlation was in the range 0.423-0.843. The ROC curves and the area under the curve show a specificity of 69% for the presence of exudative recurrences for the variation of the whole area, of 63% for the vascular density and 62% for the vascular area. The fractal dimension was not specific for the presence of SD-OCT recurrences.

Conclusions : The algorithm designed in this study seems reliable and allows a faster and more consistent evaluation of OCTA images. Of the quantiative variables used in the recent literature, the variation of the total surface seems the most specific for the presence of the exudative recurrences on conventional multimodal imagery.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.