Purpose:
To investigate the feasibility, robustness and speed of real time retinal motion tracking based on a high speed wide field of view confocal imager and realtime parallel processing of partial images.Tracking retinal movements can be used for improving the image quality of various retinal imaging modalities such as OCT and adaptive optics imaging. Previously developed retina trackers were expensive, complicated and large.
Methods:
The retinal motion tracker is based on a custom confocal retina scanner. The confocal scanner provides a 45 degrees Field of View, operates at a high framerate (80Hz) and provides a resolution of 500*500 pixels.In a confocal imager light is scanned across the retina and read out sequentially by a single point or line array sensor. This is in contrast with a floodlight retinal imager such as the funduscamera, in which case the whole field of view is imaged during the full exposure time on an 2D image sensor such as CMOS or CCD. In a confocal imager the pixels are read out immediately upon a very short acquisition time. As soon as partial image or strip of a given size has been imaged, it can be compared with a corresponding part in a reference image in various positions. This is executed in real time and in a strip-by-strip fashion. By using a custom developed SIMD chip, each currently acquired strip is compared with 25 positions in the reference image. Using the 80Hz confocal retina scanner, we have recorded 20 seconds of 'videos' from 10 eyes (6 subjects). We have analyzed the performance of the tracker for various distance metrics and various sizes of the strip, in order to determine the optimum settings to achieve high robustness and maximum speed.
Results:
A strip size of 32*500 pixels provided robust tracking - no loss of tracking - for the analyzed subjects. The MAD (mean absolute distance) prooved at faster but equally robust distance metric compared to RMS.
Conclusions:
A 1 kHz retina tracker can be implemented in a compact and economical system having the size of a small funduscamera, consisting of high speed confocal retinal imager and a PC based real time parallel image processor. Further study with a larger subject group is required to validate the robustness and speed.
Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • motion-2D • retina