Abstract
Purpose :
To provide an automated method of characterizing the presumed vascularity of pigment epithelial detachments (PEDs) in neovascular age-related macular degeneration (nAMD) and to quantify the areas of serous, neovascular, and fibrous tissues within PEDs.
Methods :
A retrospective dataset of 28 high-resolution spectral-domain OCT (SD-OCT) B-scans taken from 22 patients with nAMD with presence of fibrovascular PED, were analyzed for this experiment. B-scans were acquired using the Heidelberg Spectralis device with a resolution of 1536x1536 in pixels. Exclusion criteria included drusenoid PEDs. For pre-processing, shadow compensation and linear standardization were applied. PEDs were segmented and then filtered using 2D kernels to create a series of generated images. A PED vascularity index score (PEDVI) was calculated for each pixel within the PED using a function of generated images. Pixels within the PED were classified as serous, neovascular, or fibrous based on PEDVI and respective areas were calculated. Accuracy of segmentation and classification within the PED were graded independently twice by two expert clinicians in masked fashion on a scale of 0-100.
Results :
Of the 28 eyes, 15 had serous fluid, 24 had neovascular tissue, and 16 had fibrous tissue within a PED. Inter-observer reproducibilities were 0.99, 0.78, 0.94, and 0.79 for grader 1; 0.95, 0.78, 0.79, and 0.74 for grader 2; intra-observer repeatabilities were 0.90, 0.73, 0.84, and 0.81 for accuracy of segmentation, and classification of serous, neovascular, and fibrous respectively. Mean inter-grader reproducibility and intra-grader repeatability were 0.85 ± 0.10 and 0.82 ± 0.07 respectively. The mean graded scores were 96.88 ± 8.81, 92.67 ± 7.84, 93.44 ± 8.12, and 92.89 ± 8.25 for segmentation, serous, neovascular, and fibrous respectively. Mean total PED area, and when present, mean serous, neovascular, and fibrous areas in mm2 were 0.26 ± 0.44, 0.218 ± 0.462, 0.108 ± 0.102, and 0.061 ± 0.077 respectively.
Conclusions :
PEDVI scores calculated from a kernel-based image processing approach demonstrates potential for quantifying PEDs and approximating relative sizes of serous, neovascular, and fibrous tissue. An automated algorithm for segmentation is underway.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.