Abstract
Purpose: :
In commercially available spectral-domain optical coherence tomography (SD-OCT) instruments, retinal thickness values are delivered automatically via software-based detection of the retinal boundaries. However, the device-internal segmentation algorithms are prone to errors particularly in the presence of pathology. Segmentation errors may lead to an output of incorrect thickness values and thus to wrong conclusions in clinical and scientific practice. Our aim was to compare the segmentation performance of 4 prevalent SD-OCT devices with regards to technical accuracy and clinical relevance.
Methods: :
29 eyes of 19 patients (19 with macular edema, 6 with degenerative disease and 4 healthy eyes) were imaged each in the same session with Spectralis OCT (Sp, Heidelberg), Cirrus HD-OCT (Ci, Carl Zeiss), 3D-OCT 2000 (3D, Topcon) and RS-3000 (RS, Nidek) instruments using a standardized protocol. Raw data were exported and evaluated in validated custom software. We investigated a predefined subfield of 1 millimeter diameter, centered manually on the fovea, to compare identical regions for each case across the 4 devices. Errors in device-provided segmentation were corrected manually on each B-scan enclosed in the central subfield. As a measure of technical accuracy, proportions of wrongly segmented A-scans (=pixels) were noted for both the inner and outer retinal boundaries. The clinical impact of segmentation errors was assessed by comparing center point thickness (CPT) and mean central millimeter retinal thickness (CMT) between device-provided segmentation and manual correction.
Results: :
Segmentation errors occurred in 77% of cases and in these affected on average 29% of A-scans within the central subfield, resulting in a mean thickness difference of 24/13µm (CPT/CMT) between device-provided and manual segmentation. Regarding inter-device comparisons, the proportions of cases with segmentation errors ranged from 48% (Sp) to 79% (Ci), 86% (3D) and 93% (RS). In these cases, the mean proportions of A-scans with wrongly detected outer retinal boundary were 30% (Sp), 9% (Ci), 23% (3D) and 10% (RS); corresponding proportions for the inner retinal boundary were 11% (Sp), 12% (Ci), 6% (3D) and 21% (RS).In cases with segmentation errors the mean deviations in CPT/CMT were 41/28µm (Sp), 17/11µm (Ci), 30/13µm (3D) and 18/8µm (RS). Clinically relevant CMT deviations (≥8µm) occurred in 21% (Sp), 27% (Ci), 44% (3D) and 48% (RS) of cases.
Conclusions: :
Segmentation errors are frequent and mandate manual correction since they lead to substantial CMT deviations. The correlation between technical accuracy and clinical impact varies strongly between the tested SD-OCT devices, which is worth considering in clinical and scientific practice.
Keywords: imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)