AOSLO Imaging and Image Montaging

The central ±300 µm in both eyes of each participant were imaged using near-infrared light for imaging and wavefront sensing, filtered dichroically (788 ± 12 nm; FF01-788/12-25, Semrock, Rochester, NY) from the output of a supercontinuum laser light source (SuperK EXTREME, NKT Photonics, Birkerød, Denmark). Adaptive optics correction, run in a closed loop at approximately 25 Hz, consisted of a Shack–Hartmann wavefront sensor (SHSCam AR-S-150-GE; Optocraft GmbH, Erlangen, Germany) and a 97-actuator deformable mirror (DM97-08; ALPAO, Montbonnot-Saint-Martin, France) placed at a pupil conjugate. The imaging raster spanned a square field of 0.85° × 0.85° of visual angle. The light reflected from the retina was detected in a photomultiplier tube (H4711-50, Hamamatsu, Japan), located behind a confocal pinhole (0.5 Airy disk diameter). Photomultiplier tube signals were sampled by a field programmable gate array board (ML506; Xilinx, San Jose, CA), producing video frames with 512 × 512 pixels (spatial resolution, 0.1 arcmin of visual angle per pixel) at approximately 27 or 30 Hz. To ensure optimal image quality during recording, the pupil's position relative to the AOSLO beam was carefully maintained.²²

To guide the participant's gaze to image selected retinal locations, a 6-arcmin square fixation target was produced by modulating the imaging beam intensity, flashing at 3 Hz. Videos were recorded at the preferred retinal locus of fixation and eight surrounding points, evenly spaced and centered on the perimeter of the central imaging field, covering a square area of approximately ±300 µm of the fovea centered on the preferred retinal locus of fixation. Optimal image quality was found by selecting the best video from 5 to 10 preferred retinal locus of fixation–centered videos recorded using different defocus settings of the deformable mirror. At surrounding locations, two or more videos were recorded as deemed necessary by image quality. All videos were 10 seconds long. Acquired AOSLO video frames were spatially stabilized by offline, stripwise image registration using a modified version of previously published software in Matlab (MathWorks Inc., Natick, MA).²³ Frames and strips displaying incomplete stabilization (e.g., owing to poor image quality, eye blinks, or drying tear film) were automatically identified and deleted. The remaining strips of each video frame were averaged to obtain a single high-quality image. Such images were automatically aligned using a previously described registration software.²⁴ Regionally aligned images were imported in Corel Photo-Paint (Cascade Parent Limited, Ottawa, Ontario, Canada), subjectively selected for best retinal structural quality, and manually blended to arrive at a single continuous foveal image montage.