The use of adaptive optics (AO) for imaging the retina of living human eyes has flourished. But the conventional AO setup, which uses a dynamic wavefront sensor in a closed feedback loop with a deformable mirror, has disadvantages for retinal imaging: The wavefront-sensor beacon is sometimes bright enough to interfere with examinations; the resolution limit of the wavefront sensor itself can limit system resolution; and an undilated pupil (for example, 3 mm in diameter) can even prevent the use of a wavefront sensor. To avoid these problems, a group at the University of Houston (Houston, TX) has created a wavefront-sensorless AO system that uses a stochastic parallel-gradient-descent algorithm to image a retina in real time (25 Hz).
In the dual-deformable-mirror scanning imaging system, a “woofer” and “tweeter” corrected low- and high-order aberrations, respectively. The parameter to be optimized was the reflected light from the retina that passed through a confocal pinhole averaged over the 1.5º field of view during the 35 ms frame-exposure time. Prior to gathering image data, a Shack-Hartmann wavefront sensor was used to check and correct for lower-order aberrations (mostly defocus) and was held static during the “sensorless” dynamic operation. Because the technique automatically focused on the most reflective retinal layer, a rapid alternating defocus may be needed to image other layers. Contact Heidi Hofer at [email protected].