Abstract
Purpose :
Macular fluid has been established as a principal biomarker in retinal diseases with significant prognostic implications. It has thus become an important treatment criterion when planning personalized treatment intervals. The aim of this study is to fully automatically quantify and monitor sub- and intraretinal fluid (SRF, IRF) volume in patients with nAMD, DME and RVO.
Methods :
In this pos-hoc analysis, we included 2311 patients from five clinical, multicenter trials (BRVO/CRVO: 310/311, nAMD: 1109, DME: 610), who received a standard anti-VEGF therapy over a 12-month period. SD-OCT scans were evaluated at baseline, month 1, 2, 3, 12, using a fully automated and validated algorithm based on deep learning. Measurements were acquired for three concentric circles with diameters of 1, 3 and 6mm, termed the fovea, para- and perifoveal area, and were compared over time. Kruskal-Wallis/Mann-Whitney U test and Wilcoxon signed-rank test were used to compare distributions.
Results :
Our analysis comprised 11.151 SD-OCT volumes. For each disease, there was a statistically significant difference in how fluid was distributed across regions: there was at every time-point more IRF and SRF/mm2 found foveally, followed by the para-, and perifoveal area (p<0.001), except for baseline SRF in nAMD, where the highest amount per mm2 was seen parafoveally (p<0.001).
IRF reduction seen after the first month of treatment differed significantly between diseases, with the highest reduction seen in RVO (95.9% of initial IRF volume), followed by nAMD (91.3%) and DME (37.3%). At month 2, 3 and 12, both RVO and nAMD showed no further significant reductions, whereas in DME, there was still seen a decrease of 7%, 4% and 6% of the initial IRF volume, respectively (total reduction at month 12: 70%).
In all diseases, SRF decrease was most pronounced within the first month: 94.7% in RVO, 98.4% in nAMD and 86.3% in DME. Subsequently, RVO and nAMD patients showed a decrease of less than 1%, whereas in DME, there was still a decrease of 6% at m2, and 2% at m3/m12.
Conclusions :
With a growing and daunting amount of OCT imaging data, our aspiration is an automated way of evaluating structural biomarkers that are of clinical significance. Our study proves the feasibility of fully automatically detecting and tracking fluid over time and highlights the role of such a decision-support software in providing patients with an individualized treatment plan.
This is a 2020 ARVO Annual Meeting abstract.