Purchase this article with an account.
Travis B. Smith, Laurie M Renner, Martha Neuringer, Trevor J McGill; Computational optical sectioning of flood-illuminated adaptive optics retinal images: initial results. Invest. Ophthalmol. Vis. Sci. 2016;57(12):75.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Flood-illuminated adaptive optics (AO) systems have no confocal aperture and, consequently, produce retinal images with considerable defocused energy from reflective structures outside the focal plane. This study investigates the utility of computational optical sectioning (COS) techniques from the field of deconvolution microscopy to refocus this energy and improve image contrast.
Two healthy male subjects—a young rhesus macaque and a 25-year old human—were imaged with an Imagine Eyes rtx1 AO camera. The macaque was maintained under inhalant anesthesia and positioned with a custom headrest. We acquired depth stacks of twenty 4° en face images from focal depths ranging from the choroid to the vitreous. In place of the rtx1 onboard processing which enhances each image in isolation, all stack images underwent custom joint processing with non-uniformity and gamma corrections to preserve the signal intensity relationships across depth. Images were spatially registered with 2-D phase correlation. COS reconstruction used non-iterative 3-D deconvolution with L2 Tikhonov regularization to refocus the energy dispersed throughout the stack. The 3-D point spread function (PSF) was modeled on an Airy disk with a depth-varying quadratic phase error. For comparison with COS, images also underwent median-filtered background subtraction and depth averaging (BS-DA).
For the macaque, the spatio-temporal stability of the image stack was excellent and COS refocused cone energy well, even recovering some cones (Fig 1, orange arrow) lost with the other methods. Low stability in the human, seen in the depth stack cross-section (right column, yellow), made COS refocusing was less discernible. Both COS and BS-DA results do not have the low-frequency intensity variation and signal dropout characteristic of flood-illuminated AO systems. For both subjects, scattering tissues were not depth-resolved well enough to produce cross-sectional images comparable to an OCT b-scan.
Computational optical sectioning can remove defocused signal and improve image quality. For in vivo imaging, subject motion and reflectivity fluctuations must be minimized. Future work will reduce the number of stack images to help with these issues and expand to more subjects.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
Fig 1: En face images (0.25°) at the focal depth of the cone photoreceptors, with cross-sectional images of the depth stack below.
This PDF is available to Subscribers Only