Purchase this article with an account.
P. Gunvant, Y. Zheng, P.G. Schlottmann, D. Garway–Heath, E.A. Essock; Comparison of OCT and VCC RNFL Estimates in Identifying Glaucoma Using Wavelet–Fourier Analysis . Invest. Ophthalmol. Vis. Sci. 2005;46(13):2510.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Aim: To compare the performance of a scanning laser polarimeter (GDx–VCC, Laser Diagnostic Technologies, Inc.) and the Optical Coherence Tomograph (Stratus OCT, Carl Zeiss Meditec Inc.) in differentiating glaucoma and healthy eyes using Wavelet–Fourier Analysis (WFA) of retinal nerve fiber layer (RNFL) estimates. A second aim was to compare the RNFL curves obtained using the two devices. Methods: One eye of 55 individuals (20 healthy and 35 glaucoma, age–matched classified on the basis of visual field) was randomly selected and RNFL estimates were obtained using the GDx–VCC (64 sectors) and OCT (512 sectors). The mean deviation for glaucoma group was –3.70 (SD =2.28) and mean pattern standard deviation was 4.34 (SD=2.68). The WFA method extracts features using a two–level discrete wavelet transform (DWT), a fast Fourier transform (FFT), and principal component analysis (PCA). Linear discriminant analysis was applied to the resultant features to classify eyes. A k–fold cross–validation method was used to randomly split the dataset into independent training and testing sets and the analysis was repeated 400 times. Sensitivity, specificity and ROC area were calculated to characterize classification performance by WFA, the GDx–VCC standard metric (the Nerve Fiber Indicator, NFI) and conventional OCT metrics (Inferior Average and Average RNFL Thickness). Significance was determined using the method of DeLong et al. (Biometrics, 1988). The individual and mean RNFL thickness TSNIT pattern (i.e., thickness across 360 degrees) obtained with the GDX–VCC and the OCT was compared. Results: The WFA using the RNFL estimates from OCT was best at differentiating glaucoma and healthy eyes (sensitivity/specificity/ROC area =0.80/0.98/0.94), followed by WFA using GDx–VCC estimates (0.79/0.94/0.92), OCT Inferior Average (0.88/0.85/0.92), OCT Average RNFL (0.86/0.80/0.89) and the NFI (0.66/1.00/0.90). Sensitivity at a fixed specificity of 95% was 83% for WFA–OCT, 78% for WFA–GDx–VCC, 77% for OCT Inferior Average, 69% for NFI and 53% for OCT–Average RNFL. However, with the present sample size these differences in ROC area were not significant (p>0.01). There were considerable differences in curve shape, including mean thickness, between the devices. Conclusions: There are important differences between RNFL estimates with the two devices and also a tendency for WFA–OCT to outperform the other measures.
This PDF is available to Subscribers Only