from the site of stimulation. This indicates that the radius of activation is a factor of the stimulus strength.

Several research groups have found that the electric field created by one electrode can interact with the field created by a neighboring electrode. Sekirnjak et al. found that simultaneous activation of several neighboring electrodes resulted in higher thresholds than that from a single neighboring electrode. Interestingly, these interactions were not enough to interfere with the field when the closest electrode was activated (Fig. 12.7b). Similarly, Ahuja et al. [1] measured thresholds for activating salamander RGCs from two 200 mm stimulating electrodes each positioned approximately 250 mm from the cell (center to center spacing). Thresholds for single electrode activation were approximately 13.3 nC and increased to 29.4 nC when stimulation from both electrodes was applied simultaneously. A finite element model presented in the Ahuja et al. study indicates that the threshold increase arises from a reduction in the voltage gradient caused by simultaneous stimulation from the second electrode.

More work is needed to determine under which conditions electrode interactions occur and whether there are means to reduce these interactions. One possible means would be to interleave the stimulus pulses from nearby electrodes – the slight offset in time would presumably minimize the amount of interaction between neighboring electrodes.

12.2.1.5  Selective Activation

In the normal retina, the neural activity in neighboring neurons can be quite different e.g. response duration of a “sustained” cell can last several hundred milliseconds longer than that of a “transient” cell. Similarly, ON and OFF cells typically do not generate spikes at the same time. This wide array of spatially and temporally varying neural activity is transmitted from the retina and reassembled by the visual cortex into our percept of the visual world. The concern arises that if prosthetic stimulation creates identical (or similar) activity in all neighboring neurons, the signal that arrives at the cortex is quite different from the normal signal and may not be intelligible. Methods to selectively activate specific RGC types may help to more closely re-create the signaling patterns created normally by the retina and ultimately improve the quality of the resulting percept.

A formal study of selective activation methods for RGCs has not been reported. Fried et al. [10] measured thresholds in three different types of rabbit RGCs and found that alpha cells (G11) had the lowest threshold while local edge detectors (LED, G1) had the highest (Fig. 12.8). This finding suggests that low amplitude stimulus pulses may be able to selectively activate a single type of RGC (e.g. alpha). Unfortunately, this method of selective activation would at best, apply to a single RGC type only and would not distinguish between ON and OFF cells.

In contrast to the results from Fried et al., Margalit and Thoreson [28] found no difference in thresholds between ON, OFF and ON-OFF RGCs in salamander retina. However, it is not clear whether the populations reported by Margalit and Thoreson

Fig. 12.8  Different ganglion cell types have different thresholds. Each point (“X”) represents a threshold measurement in a different cell. Ganglion cell types were identified by the cell’s light response prior to measurement of threshold. Pulses were 0.1 ms duration, cathodal with a distant ground

can be correlated to those from Fried et al. For example, Fried et al. found that thresholds for ON and OFF alpha cells were similar. In the Margalit and Thoreson study, it is likely that the ON-OFF cells are a different population from either the ON or OFF types and yet their thresholds were not different. Unfortunately, the limited results from Fried et al. do not preclude the possibility that some RGC types have similar thresholds. Further studies are needed to determine the threshold differences across types and if differences exist, determine whether they can be used to selectively activate specific RGC types.

It is a daunting challenge to think in terms of replicating normal light elicited patterns with a retinal prosthesis. However, there are many incremental improvements that can be realized along the way. For example, ON and OFF varieties of midget and parasol cells are the four principal types of RGCs in the human retina. Together, it is estimated that they account for >90% of all RGCs. Methods that selectively activate only one of these types, for example, would likely lead to elicited patterns of neural activity that are more physiological and therefore result in improved percepts.

12.2.1.6  Temporal Response Properties

The rates at which RGCs generate action potentials [2, 6] as well as the precise timing with which individual action potentials [30, 31] are generated are both thought to play an important role in the neural code transmitted from the retina. The upper limit on RGC spike rates can be estimated from studies by O’Brien et al. [35] and DeVries and Baylor [6]. The maximum spike rates vary for each RGC type; alpha cells have the largest maximum spike frequency (~250 Hz) which sets an approximate upper limit for the response requirements of the prosthetic.

Fig. 12.9  Programmed sequences of short electrical pulse replicate light responses. (a) Spiking response to a 1-s light stimulus (horizontal bar). (b) Bottom: expanded time scale from (a) reveals individual spike latencies. Top: programmed sequence of short pulses derived from individual spike latencies: each cathodal pulse is arranged 0.5 ms before corresponding spike. (c) Spikes elicited by programmed sequence of short pulses (bottom) precisely match the light elicited spike pattern (top) from (b). Reprinted from [9], Fig. 7, with permission

As discussed in Sect. 12.2.1.1, short duration stimulus pulses (typically 100 ms) were shown to activate RGCs directly, without activating other elements of the presynaptic circuitry. Each short pulse elicits a single action potential [9, 49], typically within 0.5–1.0 ms of the pulse onset [1, 9, 24, 49]. At higher stimulation frequencies, short pulses continue to elicit one spike per pulse. This was tested originally up to 250 Hz in rabbit [9] and more recently up to 500 Hz in salamander [1]. These spike rates are comparable to the fastest rates of normal, light elicited spiking. Using the one spike per pulse paradigm, Fried et al. programmed sequences of pulses in order to precisely replicate typical RGC light responses (Fig. 12.9).

In contrast to the results from Fried et al., Sekirnjak et al. [49] found that repetitive stimulation at high frequencies resulted in a loss of the one spike per pulse response. At 50 Hz, a slight reduction (~20%) was found and a more significant reduction (~50%) was found at 100 Hz. It is not clear whether and/or how their findings impact the ability to precisely replicate light responses.

Even if the temporal properties of normal RGC signaling can be replicated using short pulses, several important obstacles must be surmounted before this paradigm can be implemented. For example, this method would presumably activate all RGCs close to the stimulating electrode with the same spike patterns resulting in patterns of retinal activation that are non-physiological. Methods for selective activation and avoiding the activation of passing axons are both needed.

The temporal response properties resulting from stimulation of bipolar cells were very different from the responses arising from stimulation of RGCs. Fried et al. [9] showed that bipolar cell input to RGCs decreased as stimulus pulse frequency increased; by 10 Hz the amplitude of the bipolar cell output was barely detectable. Ahuja et al. [1] similarly found that the RGC output was almost completely ­eliminated