Segmentation Errors in Macular Ganglion Cell Analysis as Determined by Optical Coherence Tomography.

PURPOSE

To investigate the prevalence, features, associated factors, and reproducibility of segmentation errors in macular ganglion cell inner plexiform layer (GCIPL) thickness measurement as determined by optical coherence tomography (OCT).

DESIGN

Cross-sectional study.

PARTICIPANTS

Five hundred thirty-eight glaucomatous and healthy eyes from 290 subjects with OCT-measured macular GCIPL thickness were enrolled. Eyes with macular disorders, including epiretinal membrane, macular degeneration, macular hole, and myopic maculopathy, were excluded.

METHODS

By inspecting 128 cross-sectional OCT B-scan images per eye, the presence (yes vs. no), layer (anterior vs. posterior border), location (quadrants), and area (diffuse vs. focal) of macular GCIPL segmentation error were investigated. The effects of age, refractive error, mean deviation of visual field test, circumpapillary retinal nerve fiber layer thickness obtained by OCT, and signal strength of OCT scan on the presence of macular GCIPL segmentation errors were evaluated. In eyes with segmentation errors, repeated OCT examinations were performed to investigate the reproducibility of the segmentation errors.

MAIN OUTCOME MEASURES

The prevalence, features, associated factors, and reproducibility of macular GCIPL segmentation errors were assessed.

RESULTS

Among the 538 eyes, 52 eyes (9.7%) showed segmentation errors in macular GCIPL thickness measurement. The most common features of segmentation errors were that they affected both the anterior and posterior borders, were located at the nasal quadrant (centered to the fovea), and were diffuse. In univariate analysis, the presence of segmentation error was associated significantly with younger age (P < 0.001), higher degree of myopia (P < 0.001), and lower signal strength of OCT scan (P = 0.038). In multivariate analysis, only higher degree of myopia was associated significantly with the presence of segmentation error (P < 0.001). In repeated examinations, segmentation errors were reproducible in 24 eyes (46.2%). In other cases, the features of segmentation errors changed or disappeared.

CONCLUSIONS

Although the OCT segmentation algorithm accurately detected macular GCIPL thickness in most eyes without macular disorders, in some cases, segmentation errors were found, especially in myopic eyes. In repeated examinations, approximately half of the errors were nonreproducible. These findings should be considered when assessing macular GCIPL thickness using OCT.