Purchase this article with an account.
Ivania Pereira, Hemma Resch, Stephan Holzer, Florian Schwarzhans, Jing Wu, Georg Fischer, Clemens Vass; Normalization of Retinal Vessel Density profile in healthy subjects. Invest. Ophthalmol. Vis. Sci. 2016;57(12):2963.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Having demonstrated the strong correlation between the retinal nerve fiber layer (RNFL) and retinal vessel density (RVD) in a mostly young to middle-aged healthy population, one should not forget that, in clinical routine, ocular or systemic diseases may affect RVD. This in turn may jeopardize the benefits of RVD, which facilitates individualized normal limits. It is, therefore, crucial to analyze different normalization methods of the individual RVD profile to cancel out the potential influence of generalized constriction or dilatation of retinal vessels.
A sample of 202 healthy subjects who all underwent FD-OCT of the macula and the circumpapillary area was used to test three different linear normalization methods: Average RVD subtraction, min-max and Z-scores normalization. First, the RVD was calculated according to automatic measurement of the circumpapillary retinal vessels using Gaussian convolution. Then for each normalization method, the normalized RVD (n-RVD) was obtained according to each subject’s statistical parameters of the RVD (e.g. minimum, maximum, average). Finally, for each sector and each method a multivariate linear regression model was calculated, across all subjects, with RNFL as dependent and n-RVD as well as the respective statistical parameters as independent parameters. Partial and total correlation coefficients (CoCo) were calculated for each method and for all 256 sectors of the RNFL profile and were averaged across the 256 sectors.
All three methods performed similarly, in terms of number of significant sectors and R-values. Min-max normalization presented a mean R-value of 0.47 for the multivariate model, significant in 247 (out of 256) sectors. From these, n-RVD was included in 242 sectors and 211 sectors had either maximum or minimum values of RVD included. Mean partial CoCo of n-RVD was 0.45.
Although most sectors of the RNFL profile presented a significant correlation with statistical parameters (in addition to n-RVD), the close similarity between mean values of partial CoCo of n-RVD and the R-value of the multivariate model shows that it was the normalized RVD profile that contained most of the information. Therefore, we expect that systemic or generalized ophthalmic influences in vessel thickness will not reduce the utility of normalized RVD.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only