Purchase this article with an account.
Haogang Zhu, Lei Li, Yaxing Wang; Compensate systemic variability of parapapillary retinal nerve fiber layer thickness measurement. Invest. Ophthalmol. Vis. Sci. 2019;60(9):4820.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Parapapillary retinal nerve fiber layer thickness (RNFLT) is a standard measurement for the management of glaucoma. It’s variability is however associated with the factors such as myopia, disc-fovea angle and disc-fovea distance. This study establishes transformation of RNFLT according to these factors in order to compensate systemic variability.
OCT RNFLT profile together with axial length, disc-fovea angle and distance were collected on 2521 eyes of 2521 healthy subjects. It is assumed that the RNFLT is systemically affected by these factors such that the population variability can be reduced when the variability is compensated. A neural network is trained to predict the vertical (scaling) and horizontal (shift) transformation of RNFLT at each location. The objective function is the average variance of transformed RNFLT across all locations. Two constraints were also applied so that the vertical and horizontal transformation is smooth. Diagnostic performance was tested with original and transformed RNFLT on 264 randomly reserved healthy subjects and 264 glaucoma subjects. The diagnostic criterion is the number of locations that is out of the 95% confidence interval at the corresponding location in the training data.
The mean (standard deviation [SD]) axial length is 23.26 ± 1.05 mm, the mean (SD) disc-fovea distance is 4.94 ± 0.42 mm and the mean (SD) disc-fovea angle is 7.66 ± 3.53 degree. The variance of the RNFLT on the test healthy data is plotted in Figure 1. The reduction of variance mainly focuses on the superior and inferior regions where early glaucoma commonly presents. The diagnostic ROC curve before and after the transformation is plotted in Figure 2. The area under the ROC curve is improved from 0.81 to 0.88 by applying the transformation.
The analytical model compensates the systemic variability in RNFLT measurement and lead to better diagnostic performance. The method can be particularly useful with the population with extreme factors such as high myopia and tilted disc.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
This PDF is available to Subscribers Only