Abstract
Purpose::
Scanning laser ophthalmoscopy is an attractive imaging modality because of its ability to optically section tissue, to deliver patterned stimuli to the retina, and its flexibility with light delivery and detection schemes. However, due to its sequential scanning nature, constant fixational eye movements generate distortions in each recorded image. These distortions are particularly troublesome in high-magnification systems, such as the adaptive optics scanning laser ophthalmoscope, with field sizes as small as 0.5 degrees. We demonstrate a software-based method to measure and compensate for intraframe distortions at frequencies up to 1 kHz.
Methods::
The software reads images, strip by strip, as they are made available by the frame grabber. Each strip is correlated with a reference frame using a fast and efficient cross-correlation routine, called the Map-Seeking Circuit, thereby computing the local vertical and horizontal eye displacements at up to 32 time points during the acquisition of each image. A linear interpolation between shifted strips is used to generate a corrected frame and display it in less than 7 msec after the full frame acquisition ends.
Results::
Real-time eye movement correction of 512 X 512 pixel, 1.2 degree field-of-view videos at 30 frames per second is accomplished on eyes with typical adaptive-optics-corrected image quality and normal fixational stability. Excluding tracking errors due to blinks and larger saccades, features in the eye-motion corrected video remain stable with a standard deviation of just over one pixel (equivalent to 0.14 arcmin, 0.7 microns or about 1/3 the width of a foveal cone photoreceptor). Tracking accuracy, therefore, exceeds the resolution limits of the adaptive optics system. The eye-motion-corrected videos permit direct adding of a series of frames in imaging regimes where simple shift-and-add procedures fail because of intraframe distortions.
Conclusions::
Real-time eye movement correction is demonstrated in high-magnification videos of living retina. This capability allows for rapid acquisition of high signal-to-noise frames of retinal structure. More importantly, it represents the first step toward delivery of AO-corrected, stabilized stimuli to specified locations on the retina.
Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • eye movements • image processing