Purchase this article with an account.
N. D. Leroux, H. Hofer; Improving L and M Cone Classification with a Simple Post-processing Technique to Reduce the Impact of Optical Blur. Invest. Ophthalmol. Vis. Sci. 2009;50(13):2728.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous in vivo retinal imaging studies indicate some degree of L and M cone clumping in foveal and mid-peripheral primate retina. However, these studies are difficult to interpret because optical blur may bias cone identification and lead to the mistaken assumption that the mosaic is clumped. Deconvolution has been shown to reduce the impact of optical blur in one set of macaque image data1, but this technique is computationally unwieldy and has not proven broadly applicable.
We developed and tested a simple iterative post-processing method to partially correct for the effects of optical blur on cone classification. The method was validated with realistic simulated retinal mosaics then applied to real data from 10 previously published human and macaque mosaics2,3. The improvement afforded by the method was determined and implications for interpreting mosaic organization were assessed.
The iterative method yielded valid improvement for simulated mosaics and was successfully applied to all but one of the real mosaics. L and M cone assignment error was reduced in 5 out of 9 real retinal mosaics, with reduction up to 30%. More importantly, apparent clumping was markedly reduced or eliminated in all mosaics that initially exhibited clumping of like-type cones, suggesting a selective removal of blur-induced bias. This work also indicated that current in vivo cone classification methods generally underestimate assignment error by ~ a factor of 2, likely due to inappropriate assumptions about the distribution of cone absorptance data.
The iterative method was shown to be valid and broadly useful in improving L and M cone classification for real image data. More importantly, by removing bias due to optical blur, the technique serves as a promising tool in the analysis of cone arrangement. That current methods generally underestimate classification error suggests increased caution when interpreting cone arrangement data. In light of this, potential improvements to the cone classification method as well as conditions that allow confident conclusions about arrangement to be drawn are also discussed.1. Christou,J & Roorda,A,(2004) JOSA A, 21,1393.2. Hofer et al.(2005). J neurosci, 25,9669.3. Roorda et al.(2001) Vis Res, 41,1291.
This PDF is available to Subscribers Only