In-vivo Imaging of Human Cone Photoreceptors without the Use of Adaptive Optics

Abstract

 

Purpose

 

Optical aberrations in the human cornea and lens limit the contrast of high spatial frequency retinal structure, for example when imaging the cone photoreceptors. A common way to overcome this is through the use of adaptive optics (AO) where the combination of a wavefront sensor (WFS) and deformable mirror (DM) measure and correct for these distortions in real time. We sought to explore the ability of post-processing techniques to recover this high spatial frequency information from uncorrected retinal images.

 

Methods

 

The AO fundus camera at Ohio State University was used to acquire retinal images on normal human subjects at 2 and 4° in the temporal retina with and without AO. At each location, 10-20 retinal images were acquired at 20 Hz (6 ms exposures, 680nm imaging light, 0.75° field of view, 6 mm diameter pupil, 2nd order refractive correction) along with the corresponding Shack-Hartmann WFS image. The images were processed and deconvolved using (i) information derived solely from the raw images and (ii) supplementing the images with WFS measurements. The image-based PSFs were estimated using the cones as partially-resolved beacons, a separate iterative PSF estimation technique that optionally includes the WFS information, and computed PSFs using only the WFS data. The results were stacked and compared to images of the same retinal locations acquired under standard AO imaging conditions.

 

Results

 

The image-based PSF algorithm showed consistent features at the diffraction limit of the camera, averaging to a resolution comparable with AO. The image-only methods worked well, suggesting that the WFS sensor data is not as critical as might be expected. Work continues on the various algorithms and further performance improvement is expected.

 

Conclusions

 

The ability to acquire cone images without the use of a DM results in simpler and less expensive retinal imaging systems. Moreover, the ability to extract PSFs from various parts of the images allows images larger than the isoplanatic patch to be constructed, provided Nyquist sampling is maintained. The coming months will see the imaging of more normal human subjects along with AMD and RP patients.

 

 

Stacked unprocessed images from (a) fovea, (b) 2° and (c) 4° temporal elongation.

 

 

Examples using image-based PSF estimates. The image eccentricities are (a) fovea, (b) 2° and (c) 4° temporal.

 

Keywords: 626 aberrations • 549 image processing • 552 imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)