Accuracy of Automated Subretinal Fluid Volume Quantification Using the Macuject Algorithm

Abstract

Purpose : Quantification of retinal exudation, and in particular the amount of subretinal fluid (SRF), may be useful in the management of eyes with macular neovascularization (MNV). In this study, we evaluate the accuracy of the MACUJECT algorithm for quantifying SRF volume by comparing automated measurements with ground truth based on manual segmentation.

Methods : In this retrospective, IRB-approved study, 50 eyes of 50 subjects with MNV secondary AMD were identified from retinal clinics (Doheny-UCLA Eye Centers, Retina Consultants of Houston). All subjects underwent imaging with spectral domain OCT (Cirrus and Spectralis) for identification of exudation. SRF is defined as the hyporeflective space between outer photoreceptor border & inner surface of subretinal hyper-reflective material (if present) or retinal pigment epithelium (in absence of SHRM). The SRF in each B-scan was automatically segmented by Macuject algorithm (Macuject Pty Ltd), and volume (nanoliters) was automatically calculated. Separately, in masked fashion, certified OCT graders at Doheny Image Reading and Research Lab (DIRRL) computed a SRF volume after manually segmenting on all B-scans of the volume using 3D-OCTOR grading tool. Bivariate Pearson correlations & intraclass correlation (ICC) were performed between manual and automated measurements using SPSS. A P<0.05 considered to be statistically significant.

Results : Of 50 eyes, 25 were from Cirrus SD-OCT & 25 eyes were from Spectralis SD-OCT. The mean (± SD (range)) SRF volume was statistically (p = 0.84) similar between automated (2303 ± 283 nl (0.02 – 1083 nl) and manual methods (239 ± 332 nl (10 – 1660 nl) for entire study cohort. There was also a significant correlation between the automated & manual SRF volume (r = 0.88, p <0.001). When considering individual OCT devices, the mean SRF volume measurements between automated and manual were also statistically similar in the Cirrus SD-OCT subset (245 nl vs. 262 nl; automated vs manual) and within the Spectralis SD-OCT subset (223 nl vs. 218 nl). The correlations were stronger for Cirrus subset (r=0.95, p<0.001) compared to the Spectralis subset (r=0.85, p<0.001).

Conclusions : The SRF volume measurements were similar between the automated software and human expert manual segmentation. Automated quantification of exudation in eyes with MNV, may be useful for monitoring the response and progression of these eyes following treatment.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.