Use of an animal model to study psychophysical responses to phosphenes has been
widely debated. Since there exists little published work in this area,
specifically related to visual prostheses, it is our intention to explore, and
hopefully exploit, the limits of an animal behavioral model, combined with
neural recording from the large number of intracortical electrodes. We used the
refined map, in combination with a memory saccade behavioral task, to access the
animal’s perception of the electrical stimulation. In this task, the animal
has been reward-trained to visually fixate, then saccade to a small flash in the
visual field, using memory, after the small flash extinguishes (Figure 16a). A
magnetic eye-tracking system measures the saccades.
Figure
16a
Figure 16b
To map, and test responses to phosphenes, the animal initially performed the
visual memory saccade task, with flashes at locations that correspond to the
receptive field map. Then the visual stimuli were replaced by electrically
stimulating the corresponding electrode site. The effectiveness of stimulation,
compared to visual stimuli, can be evaluated in terms of the standard deviation
around each saccade endpoint (Figure 16b). The use of a memory saccade task
avoids the problem of phosphenes being fixed on the retinotopic map, thereby
appearing to shift with eye movements. Note that this task would not be possible
with a single, or even a few electrodes, since the animal would quickly memorize
the target location. With many electrodes, activated randomly, the animal must
continually saccade to the location he perceived an instant before. Two weeks
after the start of training, the animal was clearly looking directly to
stimulated targets for many of the stimulation channels. This was recently
confirmed by eye position data, shown in Figure 17. Black dots show
where the animal looked after electrical stimulation followed by an additional 1
sec waiting period.
We also intend to evaluate the potential for tuned-response manipulation. Based on previous visual prosthesis work,
one obvious stimulation strategy might be to maximize the resolution with which phosphenes can be induced, with
the goal of delivering visual information in the form of pixiled images (strategy 2).
However, a potentially far more effective strategy might involve the selective
activation of neurons according to their natural tuning properties (strategy 1).
For example, one might transmit information about objects by activating neurons
whose orientation preferences trace out the object’s shape, as opposed to
rendering the entire object as a simple texture map. This is a good idea in
theory, but no one knows if it is feasible.
