Towards photosensor movement--adaptive image analysis in an electronic retinal prosthesis

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Ophthalmology

Background:

• Electronic vision prostheses aim to restore rudimentary vision for individuals with profound blindness.
• These devices generate artificial light spots, known as phosphenes, in the visual field via stimulating electrodes.
• Current systems require effective image processing to translate visual input into useful phosphene patterns.

Purpose of the Study:

• To evaluate different image analysis schemes for generating phosphene images in electronic vision prostheses.
• To quantify the efficacy of these schemes using a mutual-information function.
• To propose an adaptive image analysis strategy for improved visual prosthesis performance.

Main Methods:

• A numerical experiment was conducted to observe phosphene image generation under various image analysis schemes.
• The mutual-information function was employed to assess the efficiency of each scheme, penalizing redundancy.
• Laplacian of Gaussian (∇²G) kernels, geometrically transformed based on phosphene layout, were investigated.

Main Results:

• An effective image analysis scheme utilizing geometrically transformed ∇²G kernels was identified.
• The proposed scheme demonstrated efficacy in generating a meaningful phosphene image from visual stimuli.
• The study suggests adapting analysis kernels based on photosensor movement for enhanced performance.

Conclusions:

• The developed image analysis scheme offers a promising approach for improving visual perception in electronic vision prostheses.
• Adaptive kernel strategies, responsive to camera motion, represent a significant advancement.
• This research contributes to the development of more effective visual restoration technologies for blindness.