Telecentric model eye for correction of image distortion in adaptive optics ophthalmoscopes

Abstract

Here, we demonstrate a telecentric model eye for measuring scanning, sampling, and optical distortion in AO ophthalmoscopes across a 10-diopter focus range, seeking to improve the reproducibility of adaptive optics (AO) ophthalmoscopy biomarkers. The model eye lens provides diffraction-limited performance when imaging with 800 nm light over circular 2.0, 3.0, 4.6, and 9.2° fields of view through 8, 6, 4, and 2 mm diameter aperture stops, respectively. Measurements of double-pass wavefront aberrations using both model and real retinas show that the use of opal glass model retinas, rough model retina surfaces, and wavefront sensing beacon scanning mitigate first-pass aberrations. This is particularly important, as first-pass aberrations are often assumed but not always achieved in AO ophthalmoscopes. Using the model eye with custom distortion estimation algorithms, we recorded 0.06% non-isotropic scaling repeatability, 0.02° shear repeatability, 0.5% reproducibility for both metrics and a root-mean-square residual distortion of 0.1 pixels across the field of view and focus range.

Fig. 1.

Traditional and telecentric model eyes for estimating ophthalmoscope image distortion, with S being the ophthalmoscope’s exit pupil plane and f the lens focal length eye.

Fig. 2.

Depiction of experimental setup for estimating the first pass wavefront (green) preservation after a second pass (red) through an adaptive optics ophthalmoscope in which a deformable mirror induces an off-center wavefront bump. When using a reflective retina a symmetric double-pass wavefront is seen, while with a scattering retina only a single-pass (outgoing) wavefront is seem.

Fig. 3.

Pictures of tested substrates, captured with oblique illumination with a white LED, the reflection of which can be seen on the polished surfaces, but not on the rough ones.

Fig. 4.

SHWS centroid displacement quiver (magnified ×10) plots and wavefronts captured with an AO ophthalmoscope and a model eye (see Fig. 2) with various model retinas when poking an off-center deformable mirror actuator. The numbers show the single pass (SP) and double pass (DP) wavefront peak amplitude in relation to that of the second pass. In the human data, tip and tilt (nominally zero), as well as defocus and astigmatism (eye refractive error) were removed from the wavefront reconstruction to emphasize the deformable mirror actuator wavefront.

Fig. 5.

SHWS images and lenslet photon count maps for various model eyes and one real eye with 5 and 8 mm entrance pupils, captured with a scanned wavefront sensing beacon. The diagonal cross pattern seen in the reticle on glass and human retina, due to birefringence, suggests some polarization preservation in the double pass process. Note how the stronger scattering surfaces broaden the SHWS spots, almost completely filling the pupil with a tan/pink color.

Fig. 6.

Sequence of wavefronts captured by a SHWS in an AO ophthalmoscope while iteratively correcting the pushing of a deformable mirror actuator when using a model eye with paper and silver mirror substrates as model retinas, when using AO control gains of 0.2, 0.6 and 1.

Fig. 7.

Sequence of steps for estimating image distortion in an adaptive optics ophthalmoscope with a horizontal resonant scanner and a vertical non-resonant scanner, using images captured with the proposed telecentric model eye. Once the image distortion transformations are known, they can be used to correct retinal images through the steps highlighted in red.

Fig. 8.

Quiver plots showing model eye lens distortion (see Section 2.3) for the grid dots in the images shown Fig. 7, for -5, 0, and 5 D over a 2° full field of view. The quiver arrow length has been magnified 50 times (see top right corner each plot), to facilitate visualization.

Fig. 9.

Telecentric model eyes for mounting on an AO ophthalmoscope optical table (left, emulating a 5 D myopic eye), and a stage for alignment of human subjects for retinal imaging (right, emulating an emmetropic eye). Both model eyes include an iris diaphragm (leftmost black component), a 100 mm focal length achromatic doublet (central black component along metal rods), and a square grid of dots on a mount with tip/tilt adjustment (rightmost black component).

Fig. 10.

Distortion maps after cumulative correction of distortion due to scanning velocity variation and sampling jitter (left); shear and non-isotropic scaling (middle); and 3^rd order optical distortion (right) for 2° full field of view in a custom adaptive optics scanning light ophthalmoscope. Each arrow in the quiver plot (see magnification on top right corner of plots) corresponds to a dot on a square distortion grid.

Fig. 11.

Fitting error between the dots of the real and perfect distortion grids after correction of scaling and shear (blue dots) and additional corrections of distortion described by a 3^rd order polynomial (green dots). Ellipse semi-axes represent two times the standard deviation of the fitting errors for all the dots, providing an idea of the uncertainty in the centroid corrections or optical distortions not captured by the proposed geometrical transformations.

Fig. 12.

AO ophthalmoscope image distortion scaling (left plot) and shear (right plot) coefficients for -5, 0 and 5D of refractive error. Note that the short horizontal black segments are vertical error bars (standard deviation of 5 measurements), showing high measurement repeatability, while the small differences between measurements from two different model eyes show ∼0.5% reproducibility. The black scaling coefficient curves, calculated as the average of the corresponding green and blue curves, show that the eye was not perfectly telecentric, with an average magnification change of ∼2.5% over the 10D range.

Fig. 13.

Fitted 3^rd order distortion polynomial coefficients as defined in Eqs. (15) and (16).