Total Retinal Thickness Using Iowa Reference Algorithm: Measurement Reproducibility in 5 SD-OCT Scanners

Abstract

 

Purpose

 

A comparison of retinal thickness assessment obtained from the Iowa Reference Algorithm on commercially available spectral-domain (SD) optical coherence tomography (OCT) scanners from five different manufacturers.

 

Methods

 

Fovea-centered SD-OCT volumes from 11 eyes were obtained at the Vienna Reading Center: 4 normal eyes and 4 dry age-related macular degeneration, 1 lamellar macular hole and 2 Stargardt’s disease. Serial-imaging OCT scans (all 6×6×2 mm³) were obtained using: Cirrus HD-OCT (Carl Zeiss Meditec)(512×128×1024 voxels), RTVue (Optovue)(513×101×640), RS-3000 (Nidek)(512×128×512), Heidelberg Spectralis (Heidelberg Engr) (512×49×496) and Topcon-2000 (Topcon)(512×128×885). The Iowa Reference Algorithm, which routinely provides segmentation of 10 retinal layers, measured total retinal thickness (TRT), defined as averaged distances between the Inner Limiting Membrane and Bruch’s Membrane surfaces. The average of TRT obtained from all 5 scanners for each subject represented the patient-specific reference value of TRT. Scanner-specific bias was calculated as the mean signed difference between the reference standard and patient/scanner-specific TRT measurement and expressed in µm as mean ± stdv. TRT values from the 5 scanner segmentations were used to determine cross-scanner reproducibility of the Iowa Reference Algorithm.

 

Results

 

In Fig. 1, scanner bias µ_bias(in µm) for Cirrus, Topcon, RTVue, RS3000 and Spectralis are -1.33, -1.50, 2.88, (high) 8.14 and (max) -8.19, with σ_bias of 1.64, 1.92, (min) 1.11, 1.81 and (max) 2.14, respectively. Fig. 2 shows that excellent cross-scanner reproducibility can be obtained by scanner-based bias correction - TRT measurements are tightly clustered around the same value for each subject.

 

Conclusions

 

Iowa Reference Algorithm allows comparisons of TRT across clinically-relevant SD-OCT scanners, i.e. Cirrus, Spectralis, Topcon, RTVue, and RS3000. The achieved variance of around 2 µm suggests - after correcting for the systematic scanner-specific measurement bias - that retinal thickness values obtained with the Iowa Reference Algorithm are highly comparable across scanners thus facilitating multi-center studies with heterogeneous device utilization.

 

 

Fig. 1: Scanner-specific biases are expressed as mean and stdv. - given in µm and voxels.

 

 

Fig. 2: TRT after scanner-specific bias correction shows excellent cross-scanner reproducibility of RT measurement.

 

Keywords: 688 retina • 552 imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • 551 imaging/image analysis: non-clinical