Purpose : In single-pixel camera ophthalmoscope (SPCO) imaging, the reconstructed image resolution depends on the number of illumination patterns. This limits either the resolution or the duration of the image capture. Here, we examined a method to reduce the imaging time by pre-analyzing the spatial features of the retinal fundus, minimizing the impact of eye movements on the final image quality.
Methods : Images of the optical nerve head (ONH) area of six normal subjects were obtained taken with a commercially available retinal camera (EasyScan, iOptics, The Netherlands). Afterward, the images were analyzed and categorized regarding their spatial features. With the obtained information, we were able to select illumination patterns, which presumably give us spatial information to reconstruct an image of the ONH. At the same time, patterns with non-relevant or identical information were rejected and therefore not displayed during the retinal illumination. This method allowed us to limit the amount of patterns, which need to be displayed while maintaining similar final reconstruction image quality. Image quality can be improved when the overall imaging time is kept constant independent of the amount of patterns. The ONH is the preferred area of interest as its high contrast area makes it easy to compare with images taken in the experimental prototype of SPCO.
Results : Initial experimental results obtained using a model eye and a maximum reduction in the number of patterns of 80% confirmed the viability of this approach. Furthermore, images of the ONH in real eyes show good results in terms of image quality albeit a reduction no better than 50% could be achieved. Lowering the amount of patterns and hence decreasing illumination time lowers the severe impact of eye movement without losing significant object information. Final image quality is similar to previous images reconstructed while displaying all structured patterns.
Conclusions : When using pre-selected spatial patterns, the illumination time for a SPCO could be decreased while obtaining constant image quality. Pre-analyzing spatial features of a retinal area of interest could lead to a generic fundus dictionary, which could benefit future single-pixel imaging and compressive sensing ophthalmic devices.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.