Purchase this article with an account.
Jonathan D Oakley, Magí Andorra, Elena H Martinez-Lapiscina, Daniel Russakoff, Pablo Villoslada; Comparison of Automated Retinal Segmentation across OCT Devices using Independent Analysis Software. Invest. Ophthalmol. Vis. Sci. 2016;57(12):5956.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Given the same patient eye, different commercial optical coherence tomography (OCT) devices report different thicknesses for nominally the same structure. For clinical trials and patient management, it is therefore of interest to use a common analysis algorithm. We investigated the compatibility of one such analysis tool across two devices.
Within the same session, 233 patient eyes of MS patients were imaged using the Cirrus HD-OCT (Carl Zeiss Meditec) and Heidelberg Spectralis (Heidelberg Engr). The OrionTM analysis software (Voxeleron) was used to segment both the Cirrus and Spectralis scans. The two primary clinical endpoints of Total Retinal Thickness (TRT) and the Ganglion Cell layer plus the Inner Plexiform Layer (GC-IPL), were compared within the automatically centered ETDRS grid. Correlation of measures used the coefficient of determination (R2), and Bland-Altman plots were used to investigate any systematic discrepancies.
Within the ETDRS grid Cirrus and Spectralis average TRT thicknesses had an R2 of 0.94. The individual sectors temporal (T), superior (S), nasal (N) and inferior (I) had R2 values of 0.85, 0.88, 0.94 and 0.91, respectively. The GC-IPL had an R2 of 0.82 for the macular and 0.78, 0.68, 0.85 and 0.57 for the respective zones (T,S,N,I). Bland-Altman plots showed no obvious systematic discrepancies (Fig 1 and 2), although the mean differences of ~4/~5 microns informs us of larger Spectralis thicknesses, albeit within the inherent axial resolution of each device 5-7 microns.
Excellent compatibility is possible using TRT as a primary structural endpoint in cross-platform OCT analysis using the Cirrus and Spectralis devices. Good to excellent compatibility was seen using the GC-IPL endpoint and the same devices. In this study, no correction for scanner specific measurement bias was necessary, meaning that multi-center trials using different OCT devices can be facilitated using 3rd party analysis software.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
Bland Altman Plot, with mean and 95% confidence intervals, of the Macular TRT for the Cirrus and Spectralis devices, where difference is Cirrus-Spectralis. Correlation R = 0.036.
Fig 2 - Bland Altman Plot, with mean and 95% confidence intervals, of the Macular GC-IPL for the Cirrus and Spectralis devices, where difference is Cirrus-Spectralis. Correlation R = 0.029.
This PDF is available to Subscribers Only