Purchase this article with an account.
Luis De Sisternes, Homayoun Bagherinia, Hao Zhou, Jie Lu, Warren Lewis, Sophie Kubach, Yingying Shi, Philip J Rosenfeld, Ruikang Wang, Mary Kathryn Durbin; Analyzing automated OCT choroid thickness measurements. Invest. Ophthalmol. Vis. Sci. 2021;62(11):52.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Advances in OCT technology allow larger field of view (FOV) scans and complex processing, such as choroid segmentation. However, little is known about automated choroid quantifications in large FOVs. We evaluate the accuracy and utility of these measurements across different macula sectors.
We collected 12x12 mm swept-source OCT volumes (PLEX® Elite 9000; ZEISS, Dublin, CA) from 27 healthy and 34 disease eyes (including 13 intermediate AMD eyes). Bruch’s membrane and choroid/sclera interface were annotated manually and automatically using prototype software, and thickness maps were generated by considering the distance between these interfaces. The optic nerve head (ONH) location was automatically detected and eliminated from the analysis. We analyzed the differences and correlation between manual and automated thickness values in an extended ETDRS grid and in a custom grid (Figure 1). We also measured the ability to discriminate between healthy and AMD eyes using each sector values for both the manual and automated methods by measuring the area under the receiver operating characteristic curve (AUC).
Manual annotation resulted in smoother maps compared to the automated results due to the annotation tool employed. There were statistically significant differences (p<0.05) between manual and automated measurements in all sectors analyzed (automated values were consistently higher) due to agreement difficulty on where to draw the choroid/sclera interface. However, all sectors presented significant correlation between manual and automated measurements, with very high correlation (>0.85) in sectors not including the ONH. Lower correlation within the ONH vicinity was explained by higher difficulties in the segmentation. When comparing healthy and AMD patients, thickness was significantly higher in healthy cases (+47.6 and +44.5 µm overall in manual and automated, respectively). Discrimination was similar for both manual and automated methods excluding regions within the ONH, with higher AUC values in regions closer to the fovea. The highest AUC was observed in the superotemporal sectors in the extended ETDRS grid.
While significantly different, automated choroid thickness measurements show very high correlation and very similar discriminatory power to manual measurements excluding regions within the ONH.
This is a 2021 Imaging in the Eye Conference abstract.
Figure 1. Manual/automated choroid thickness in example AMD eye (left) and comparison analyzed data (right)
This PDF is available to Subscribers Only