My doctorate research at the University of Rochester (with Professors David Williams and Michael Morris), represented the first major attempt to establish the limits to which very fine retinal features, such as individual photoreceptors, can be observed through the intact optics of the living human eye. Though such features had been imaged through the optics of the eyes of several species, it had not been clear whether the optical quality was good enough to resolve these same retinal features in the living human eye. To address this issue, we constructed a high-resolution fundus camera with which cones having a spacing as small as 3.5 microns were resolved for the first time in eyes of superior optical quality. In eyes with normal optical quality, the average power spectra of our retinal images showed that it is possible to recover spatial frequencies as high as 150 cycles/deg. It also revealed Yellott's ring whose radius corresponds to the photoreceptor sampling frequency. Several cone photoreceptor images collected with this camera are shown below in Figure 1. These results provided the first demonstration that it would be possible to routinely resolve retinal structures as small as photoreceptors in most eyes if their aberrations could be corrected.

Figure 1 Several of the best images collected using our fundus camera on subjects JL and MM at 0.5 and 1.25 degrees eccentricity. The four images contain illuminated retinal patches 6.8 min of arc in diameter. Two light micrographs of the cone mosaic at roughly the same eccentricities (0.42 and 1.4 degrees) are also shown for comparison.

This led to our project on de-blurring retinal images using two statistical de-blurring methods, stellar speckle interferometry (SSI) and the bispectral imaging method (BIM). Both methods were developed in the astronomy communities during the 1970s and 1980s to greatly enhance the performance of ground-based telescopes by removing the detrimental effects of the earth's atmosphere. For our application these methods were attractive as they required no a priori knowledge about the eye's aberrations nor expensive equipment. To assess the extent to which they effectively de-blur retinal images, we simulated the imaging process of the eye using a modeled photoreceptor mosaic in conjunction with measured wave aberration data of human eyes. The model incorporated cone locations from micrographs of excised human retina (Curcio et al., 1990). A two-dimensional Gaussian function was placed at each cone and rod location, representing the expected light profile exiting the photoreceptor (MacLeod et al., 1993). Figure 2 below illustrates the improvement in retinal image quality afforded by BIM and SSI in the simulation. Unfortunately, similar results were not obtained in the living human eye.

Figure 2 Results of convolving (a) the diffraction-limited PSF and (b) JL's PSF with simulated photoreceptor mosaics at the foveal center. (c) contains the mosaic reconstruction whose Fourier phase was reconstructed using BIM and whose Fourier modulus was known a priori. (d) contains the mosaic reconstruction using BIM and SSI that required no information about the mosaic or the aberrations of the eye. Random aberrations, which produced the maximum reconstructed Strehl ratio for JL's optics, were present having a strength of s = 1.1l and r = 0.4D. The mosaics are square retinal regions 26 microns wide. The exit pupil was 6 mm and the illumination wavelength was 555 nm.

I continued on at Rochester as a postdoctorate, again working with Professors David Williams and Michael Morris, as well as the research associate, Junzhong Liang. We successfully designed and constructed the first adaptive optics retinal camera for correcting higher-order ocular aberrations in the living human eye. The system provided sufficient aberration correction to allow individual cone receptors to be resolved in normal eyes. Junzhong's vast expertise in adaptive optics provided the key to our success. Figure 3 illustrates the improvement in retinal image quality provided by our adaptive optics retinal camera.

Figure 3 Image from DM's retina at 0.8 degree eccentricity (a) without and (b) with adaptive compensation. Each image is approximately 96 microns wide. The power spectra of the images in (a) and (b) are shown in (c) and (d), respectively. In the compensated power spectrum in (d), the ring of power at ~86 cycles/deg (i.e. Yellott's ring) indicates the sampling frequency of the photoreceptors at this retinal location. (e) and (f) show bandpass-filtered images of (a) and (b), respectively. A Butterworth filter, chosen to remove high-spatial-frequency noise and to enhance contrast, passed frequencies between 0.1 and 1.2 times the sampling frequency of cones determined from the power spectrum.

Adaptive optics proved substantially superior to the de-blurring methods, SSI and BIM, that we had explored earlier. The success of our adaptive optics camera is reflected in the fact that it continues to be used by graduate students and postdoctorates in David Williams research group, including the recent in-vivo cone-classification work of Austin Roorda.

Back to Dr. Miller's homepage