Neural activity shaping utilizing a partitioned target pattern

Martin J Spencer,David B Grayden,Anthony N Burkitt,Hamish Meffin,Tatiana Kameneva

doi:10.1088/1741-2552/abecc4

Abstract

Electrical stimulation of neural tissue is used in both clinical and experimental devices to evoke a desired spatiotemporal pattern of neural activity. These devices induce a local field that drives neural activation, referred to as an activating function or generator signal. In visual prostheses, the spread of generator signal from each electrode within the neural tissue results in a spread of visual perception, referred to as a phosphene. Objective. In cases where neighbouring phosphenes overlap, it is desirable to use current steering or neural activity shaping strategies to manipulate the generator signal between the electrodes to provide greater control over the total pattern of neural activity. Applying opposite generator signal polarities in neighbouring regions of the retina forces the generator signal to pass through zero at an intermediate point, thus inducing low neural activity that may be perceived as a high-contrast line. This approach provides a form of high contrast visual perception, but it requires partitioning of the target pattern into those regions that use positive or negative generator signals. This discrete optimization is an NP-hard problem that is subject to being trapped in detrimental local minima. Approach. This investigation proposes a new partitioning method using image segmentation to determine the most beneficial positive and negative generator signal regions. Utilizing a database of 1000 natural images, the method is compared to alternative approaches based upon the mean squared error of the outcome. Main results. Under nominal conditions and with a set computation limit, partitioning provided improvement for 32% of these images. This percentage increased to 89% when utilizing image pre-processing to emphasize perceptual features of the images. The percentage of images that were dealt with most effectively with image segmentation increased as lower computation limits were imposed on the algorithms. Significance. These results provide a new method to increase the resolution of neural stimulating arrays and thus improve the experience of visual prosthesis users.

Highlights

Electrode arrays are used in clinical and experimental settings to induce responses in neural tissue [1,2,3]
This approach provides a form of high contrast visual perception, but it requires partitioning of the target pattern into those regions that use positive or negative generator signals
In order to visualize the results of a Hopfield network search process, a ‘toy’ scenario with 3 × 3 target pixels and 2 × 2 electrodes is considered for illustrative purposes

Summary

Introduction

Electrode arrays are used in clinical and experimental settings to induce responses in neural tissue [1,2,3]. An important limitation of electrical stimulation is that each stimulated electrode induces a region of visual perception, referred to a phosphene [7, 8]. This is the result of the flow of electrical current in the neural tissue [9, 10] as well as characteristics of the neural structures and connections [6]. This current spread imposes a limitation on the maximum resolution of the induced pattern of neural activation

Methods

Results

Conclusion