Evaluation of Two OCT 3 Border Detection Algorithms

Abstract

Abstract: : Purpose: To evaluate the existing (A1) and a new version (A2) of Carl Zeiss Meditec software algorithm for quantification of retinal nerve fiber layer (RNFL) thickness using third generation optical coherence tomography (OCT 3, Carl Zeiss Meditec Inc., Dublin, CA) in normal and glaucomatous human eyes. Methods: OCT 3 images (fast peripapillary circular scans) were analyzed for 2 different aspects: glaucoma discriminating power and reproducibility. Glaucoma discriminating power: Images obtained from normal and glaucomatous eyes were used to assess the discriminating power of the software. Areas under ROC (receiver operator characteristic) curves were calculated for mean RNFL thickness obtained by each algorithm against the clinical evaluation (normal or glaucoma). Reproducibility: Eyes were scanned with at least 30 minutes interval in a session. Three sessions were carried out for each eye on 3 different days within a 60-day period. A variance components analysis using random effects mixed models was applied to determine intraclass correlation coefficients (ICC), and inter- and intra-session standard deviations. Results: Glaucoma discriminating power: 189 eyes were enrolled for this analysis (107 normal (92 subjects, 46.9 ± 13.4 yrs), 64 early (46 subjects, 62.2 ± 11.7 yrs, SITA MD -4.2 ± 1.9), and 18 advanced glaucoma (17 subjects, 61.5 ± 19.0 yrs, SITA MD -15.5 ± 5.5)). RNFL was thinner in normal eyes with A2 software (97.3 ± 12.3 vs. 95.9 ± 10.9 µm, p=0.01, paired t-test), while there was no difference for glaucomatous eyes (p>0.14). There was no difference in area under ROC curve for A1 versus A2 (0.82 vs. 0.82). Reproducibility: Eight eyes of 8 normal subjects were enrolled (mean age 30.5 ± 7.0 years). A2 software constantly showed smaller intra- and inter-session mean standard deviation than A1 software for RNFL (intra-session 4.13 vs. 2.75 µm (A1 vs. A2); inter-session 3.80 vs. 2.01 µm). ICC was higher in A2 than A1 (0.77 vs. 0.87 (A1 vs. A2)). Conclusions: Although A2 software showed no difference in terms of glaucoma discriminating power, intra- and inter-session reproducibility of RNFL thickness measurements were better with A2 than A1 software. Our findings suggest that A2 software shows an improvement over A1 software by reducing inter-session variability.

Keywords: imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, S • image processing