Abstract
Purpose: :
(1) To evaluate the efficacy of Wavelet–Fourier Analysis (WFA) and Fast Fourier Analysis (FFA) (Essock et al. IOVS 2005) of retinal nerve fiber layer (RNFL) thickness estimates obtained from Optical Coherence Tomograph (OCT) for differentiating healthy and early glaucomatous eyes. (2) To compare the classification performance of WFA and FFA to the performance of standard OCT output measures (Inferior Average and RNFL Average). (3) To evaluate whether including measures of RNFL asymmetry improved the analysis.
Methods: :
RNFL estimates were obtained using OCT (OCT–3 Stratus software 4.04, Carl Zeiss Meditec Inc.) from 152 eyes of 152 individuals (83 healthy and 69 early glaucoma, as defined by standard automated perimetry (average MD –1.75 (SD=1.44) and –.3.67 (SD=1.65) respectively). Eyes with glaucoma were further staged with Hodapp, Parish and Anderson criteria (1993) and only eyes with early glaucomatous defect were included in this study. Independent training and test samples were obtained using 10–fold cross–validation. WFA and FFA were performed on the RNFL estimates and linear discriminant functions for both were obtained using Fisher’s linear discriminant analysis. Performance of the shape–based analysis methods (WFA and FFA), and the standard OCT output metrics (Inferior Average and RNFL Average) were evaluated by calculating sensitivity, specificity and area under the ROC curve (AUC).
Results: :
Performance was generally better for the shape–based methods (AUC of 0.94 [WFA] and 0.88 [FFA]) than for the standard metrics (0.81 [Inferior Average] and 0.74 [RNFL Average]). Specifically, WFA performance was significantly better (DeLong, alpha = 0.05, Bonferroni adjusted) than the others and Inferior Average performance was significantly better than RNFL Average. The sensitivity specificity for WFA, FFA, Inferior Average, and RNFL Average, respectively, was 0.80/0.82, 0.71/0.88, 0.61/0.90, and 0.54/0.85. Adding asymmetry measures had little effect on performance.
Conclusions: :
The performance of shape–based analysis methods in differentiating glaucomatous from healthy eyes was greater than that of the standard machine output. The performance of WFA was significantly better than the others.
Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • imaging/image analysis: clinical • visual fields