Assessing the validity of a cross-platform retinal image segmentation tool in normal and diseased retina

Abstract

Purpose : Retrospective, cross-sectional study, quantitatively and qualitatively comparing automated retinal image segmentation using cross platform Orion^TM software to proprietary software on retinal images captured using Heidelberg HRA+OCT

Methods : Retinal layer segmentations of normal, intermediate dry AMD and diabetic macular edema eyes were performed using Spectralis^® HRA+OCT and automated OCT segmentation software Orion^TM. Quantitative comparisons were made between the volumes of Nerve fiber layer (NFL), Ganglion cell layer (GCL), Inner plexiform layer (IPL), Outer plexiform layer (OPL), Inner nuclear layer (INL), Outer nuclear layer (ONL), total inner retinal layer (INLY) and total outer retinal layer (OUTLY). A qualitative comparison of accuracy was made by graders who compared software segmentation to manual segmentation.

Results : In normal eyes, all retinal layer volumes calculated by the two softwares were moderate-strongly correlated (Pearson correlation, r > 0.4) except OUTLY. However, differences were statistically significant except in GCIPL and OUTLY. Qualitative analysis done using Wilcoxon test showed that Orion^TM was significantly better than Heidelberg in the segmentation of NFL and INL layers (p =< 0.01). In dry AMD eyes, GCIPL, INL, ONL, INLY, TRV layer volumes were moderate-strongly correlated (r > 0.4) between softwares and their differences were statistically significant except GCIPL. Qualitatively, Orion^TM generated significantly better segmentation only for the NFL (p =< 0.05). In eyes with DME, all layer volume values were moderate-strongly correlated (r > 0.4) between softwares and their differences were statistically significant except OPL and OUTLY. Qualitatively, Orion^TM was significantly better at segmenting INL and OPL layers (p =< 0.01).

Conclusions : Layer volumes correlated well between Orion^TM and Heidelberg softwares but were, in general, significantly different suggesting that they used different retinal landmarks for segmentation. Qualitatively, when comparing to the gold standard of manual segmentation, Orion^TM segmented normal and diseased retina more accurately. Findings suggest that the cross-platform Orion^TM retinal layer segmentation software can be used reliably to study the retinal layers in normal and diseased eyes.



Abbreviations: HB- Heidelberg, OR- Orion, TRV- Total retinal volume, GCIPL – GCL+IPL

This is a 2021 ARVO Annual Meeting abstract.

 

Qualitative Assessment

 

Retinal Segmentation