Reproducibility of Total Retinal Thickness in 5 SD-OCT Scanners using Iowa Reference Algorithm

Abstract

 

Purpose

 

A comparison of total retinal thickness (TRT) measurements is presented using the Iowa Reference Algorithm from 5 commercially available Spectral-Domain Optical Coherence tomography (SD-OCT) scanners.

 

Methods

 

Fovea-centered SD-OCT volumes from 11 subjects were obtained at the Vienna Reading Center, Austria: 4 normal, 1 atrophic AMD, 1 macular hole, 2 Stargardt’s, 3 atrophic CNV. Serial-imaging OCT scans (all 6×6×2 mm3) were obtained using: CirrusTM HD-OCT (Carl Zeiss Meditec, 512×128×1024 voxels), RTVue (Optovue, 513×101×640), RS 3000 (Nidek, 512×128×512), Heidelberg Spectralis® (Heidelberg Engineering, 512×49×496), Topcon (Topcon, 512×128×885). The Iowa Reference Algorithm, which routinely segments 10 retinal layers, measured TRT as the average of all the A-scan voxels between inner limiting membrane and Bruch’s membrane surfaces. The model of TRT measurement comparisons across the 5 SD-OCT devices is Χ_mi=α_m+β_m μ_i+ε_mi. X_mi is a measurement from a device expressing the true value μ_i with intercept α_m and slope β_m, with ε_mi having mean=0 and variance=1. This model is more general than limit of agreement analysis, as β_m are not required to be the same and be equal to 1. Mendel test was used to investigate whether the 5 β_m values are the same. Bland Altman plots are provided and 5 SD-OCT pairwise predictive equations expressed via method comparison, using “additivityTests” and “MethComp” from the R package.

 

Results

 

Mendel test on the proposed model reveals that the five β_m expressing that the differences among the TRT measurements are constant (p=0.992). Having established this, Table 1 details the ten predictive equations among 5 SD-OCT devices in a pair-wise manner. The scatter plots with predictive lines and upper/lower 95% confidence limits and Bland Altman plots are shown in Figure 1.

 

Conclusions

 

The achieved p-value of 0.992 with the proposed model for 5 SD-OCTs establishes the cross-scanner reproducibility of the Iowa Reference algorithm. TRT measurements are highly reproducible, thus facilitating multi-center studies with heterogeneous device utilization.

 

 

Table 1 Details ten predictive equations with predictive error.

 

 

Figure 1 Pair-wise comparisons of 5 SD-OCTs via Bland Altman plots (upper triangle) and scatter plots (lower triangle). The predictive line (solid), identity line (bold solid), and 95% upper/lower limits (dashed) are given with predictive equations and ±SD of predictive errors.

 

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