Abstract
Purpose :
Global thickness of the circumpapillary retinal nerve fiber layer (cpRNFL), as obtained by optical coherence tomography (OCT) circle scans, is often used to distinguish between glaucomatous (G) and healthy (H) eyes. An issue arises, however, when the global thickness of an H eye falls in the bottom 5th percentile for its age group. To distinguish these H eyes from G eyes, we examined various automated quantitative methods and compared them to the performance of an OCT specialist.
Methods :
37 H eyes from a healthy database (>700 eyes) fell within the bottom 5th percentile in global cpRNFL thickness for their age group. 59 perimetric G eyes from a group of 188 eyes fell within the same Z-score range of global thickness as the 37 H eyes. An OCT circle scan (3.5 mm dia) was obtained (Heidelberg Eng) for each eye. The cpRNFL thickness profiles were then analyzed with various automated quantitative methods. The 3 most successful were: global and regional, root-mean-square (RMS) deviation from the matched normative thickness profile; number of exposed blood vessels; and local standard deviation (STD) of cpRNFL thickness. In addition, an OCT specialist rated all 96 cpRNFL reports (Fig. 1) on a scale of 0-100%, with 100% being definitely a G eye and 0%, definitely an H eye. [1] The area under the ROC curve (AUC) and sensitivity at 85% fixed specificity was obtained for each method. Further, the number of eyes above and below 50% was determined for the OCT specialist.
Results :
All of the quantitative methods miscategorized at least 23.5% of the eyes (Table 1). The best sensitivity was 71.2% (RMS: Temporal Inferior), for an accuracy of 76.5%. The best AUC was 0.865 (RMS: Global). For context, the OCT expert had an AUC of 0.904, a sensitivity of 78.0%, and an accuracy of 80.7%. The OCT specialist miscategorized 11 (29.7%) of the 37 H eyes (false positives; Fig. 1b) and 9 (15.3%) of the 59 G eyes (false negatives; Fig 1a). Additionally, 5 H eyes and 6 G eyes that were correctly categorized fell within an uncertain region, between 40-60%.
Conclusions :
A significant proportion of H eyes with low global thickness could not be distinguished from G eyes either by an OCT expert or our quantitative methods. Therefore, for these eyes, one must depend upon other aspects of OCT scans, visual fields, and/or other clinical information. 1. Wu et al, 2018, TVST,7:7.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.