Purchase this article with an account.
Qiang Yang, Jie Zhang, Koji Nozato, Kenichi Saito, Kei Suzuki, David R Williams, Ethan A Rossi; Real-time optical stabilization and digital registration for high-resolution retinal imaging. Invest. Ophthalmol. Vis. Sci. 2014;55(13):4815.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Fixational eye movements cause the small field of view (FOV) of the adaptive optics scanning light ophthalmoscope (AOSLO) to shift as the eye moves. To obtain a high signal-to-noise ratio image, eye motion must be removed and many images averaged. This is usually done in post-processing; a reference frame is manually selected and subsequent images are registered to it. Particularly in patients with poor vision, eye motion can cause the FOV to move too far outside the reference frame to be registered. This ‘frame out’ error prevents small field of view registration from being robust for large amplitude eye motion. Here we show a new multi-scale method to optically stabilize the FOV of an AOSLO and register images in real-time.
We imaged several human eyes with an AOSLO integrated with a wide FOV SLO (WF-SLO). Cross correlation was used to measure motion from the WF-SLO and AOSLO images simultaneously. These measurements were used to drive two fast, 2-axis, tip/tilt mirrors. Large amplitude motion (up to ±4°) was stabilized using the WF-SLO stabilization mirror (in open loop), while the remaining small amplitude motion was stabilized using the AOSLO stabilization mirror (in closed loop). Real-time digital registration removed the minute residual motion left after optical stabilization. We calculated the RMS of residual eye motion at each progressively finer stage of motion compensation.
Residual motion was ~3-5 µm after stabilization from the WF-SLO driven mirror only. After stabilization from both mirrors it was ~2-3 µm. When optical stabilization was combined with digital registration, it fell to ~0.2-0.3 µm.
We demonstrate a new method that has both high dynamic range to solve the ‘frame out’ problem and sub-pixel accuracy to permit real-time registration. This capability is advantageous because: 1) it will virtually eliminate post-processing, 2) it will allow for compensation of the majority of eye movements encountered in diseased eyes with poor fixation, 3) it will permit real-time registration and image averaging across multiple channels simultaneously, 4) it will facilitate imaging modes such as autofluorescence and dark field, where many frames must be averaged, allowing image formation to be visualized in real time as images are acquired, and 5) it will allow for precise retinal stimulation and manipulation of image motion for novel studies of visual function.
This PDF is available to Subscribers Only