Abstract
Purpose :
To evaluate the performance of retinal specialists in detecting retinal fluid presence in spectral domain OCT macular volume scans from eyes with age-related macular degeneration (AMD), and to compare performance with the artificial intelligence (AI)-based Notal OCT Analyzer (NOA).
Methods :
In this prospective study, OCT scans were acquired from all Age-Related Eye Disease Study 2 10-year follow-on (AREDS2-10Y) participants with Cirrus or Spectralis devices. Masked investigators graded each scan for intraretinal and subretinal fluid. The same scans underwent masked grading by (i) NOA, and (ii) reading center (RC) graders, used as ground truth. The primary outcome measure was accuracy.
Results :
1,127 eyes (651 participants) were eligible (mean age 80 y). 50% required RC senior adjudication for fluid presence. Retinal fluid was present in 370 eyes. For detecting retinal fluid, the AREDS2-10Y investigators’ performance was: accuracy 0.805 (95% CI 0.780-0.828), sensitivity 0.468 (0.416-0.520), and specificity 0.970 (0.955-0.981). NOA performance was: 0.851 (0.829-0.871), 0.822 (0.779-0.859), 0.865 (0.839-0.889). For intraretinal fluid, investigator performance was 0.815 (0.792-0.837), 0.403 (0.349-0.459), 0.978 (0.966-0.987); NOA performance was 0.877 (0.857-0.896), 0.763 (0.713-0.808), 0.922 (0.902-0.940). Comparing the investigator true positive (n=173) and false negative (n=197) cases, the mean NOA-calculated fluid volume was 156 vs 33 nl (p<0.001), with fluid present in 32% vs 11% B-scans (p<0.001).
Conclusions :
In this large and challenging sample of SD-OCT scans obtained with two commonly used devices, retinal specialists had imperfect accuracy in detecting retinal fluid, with low sensitivity. This was particularly true for (i) intraretinal fluid and (ii) low fluid volume appearing on fewer B-scans (i.e., harder to identify). AI-based detection achieved a higher level of accuracy. This AI software tool could assist physicians in detecting retinal fluid, which is important for diagnostic, retreatment, and prognostic tasks in AMD.
This is a 2020 ARVO Annual Meeting abstract.