Visual percepts evoked with an intracortical 96-channel microelectrode array inserted in human occipital cortex

Implant stability.

Following UEA implantation, the mean electrode impedance increased (particularly during the first 15 days) and then decreased back to baseline values (Figure 1D). For each experimental session, there were only minor variations in the impedances of all 96 electrodes, with most of the impedances being in the range of 37 to 56 kΩ (47 ± 4.8 kΩ; mean ± SD).

Each experimental session was preceded and concluded by a 2-minute period of spontaneous neural recordings. The neural activity appeared to be reliable and repeatable on a day-to-day basis. The number of electrodes that yielded multiunit activity varied from day to day, but some electrodes recorded reliable spikes approximately 65% of the time over the 175 days of the study. Figure 1E shows 3 representative sets of recordings obtained on days 3, 77, and 154, and the summary recording statistics for each electrode over the 175 experimental days. The colors in this figure represent the number of days on which more than 50 reliable spikes were recorded on a given electrode over the 2-minute recording period (out of 50 recording sessions). Most viable electrodes had signal amplitudes of 60 μV and noise amplitudes of less than 50 μV. Signal-to-noise ratios (SNRs) greater than 2 were required for the spikes to be included in the analysis and spike amplitudes over 900 μV were frequently observed. The SNR of the recordings did not change significantly over time (P = 0.294).

Spontaneous and electrically induced phosphenes.

The subject experienced spontaneous positive visual phenomena (phosphenes) prior to implantation, as is common in many blind persons (17–19), although these are often under reported or under recognized. After UEA implantation, the subject experienced a transient increase in spontaneous phosphenes that could appear at any time during the day and that did not correlate with an increase in the multiunit activity recorded by the implanted microelectrodes. In some experiments, we performed simultaneous EEG recordings and we did not observe any epileptiform or pathological activity associated with the perception of spontaneous phosphenes.

The number of these spontaneous phosphenes decreased gradually over time. In the first days after implantation, the average number of phosphenes was approximately 1 every 5 to 10 seconds. Six weeks after the implantation, the frequency was reduced to approximately 1 phosphene every 30 to 60 seconds, and after 12 weeks she reported episodes of spontaneous phosphenes only very occasionally.

Before electrically induced visual perceptions could be studied, we had to ensure that the subject was able to reliably discriminate electrically induced phosphenes from the spontaneous phosphenes described above. Since the features of both spontaneous and electrically induced phosphenes were very similar, early in the experimental period it was very difficult for the subject to discriminate them. After training, the subject began to correctly discriminate electrically induced from spontaneous phosphenes. Two key issues facilitated her ability to discriminate phosphenes. First, during the training phase, the subject began to understand that all evoked phosphenes were small spots of light, and that they always appeared in conjunction with a low-frequency “stimulating-now” tone that was used to indicate the start of electrical stimulation. Further, she learned that the evoked phosphenes were always localized to the same general region of her visual space; thus, the subject could recognize them more easily.

To verify that the subject was correctly discriminating stimulus-evoked and spontaneous phosphenes, a sequence of superthreshold stimuli interspersed randomly with sham trials was delivered to the subject, with a “stimulating-now” tone indicating the delivery of each stimulus (real or sham). The subject indicated phosphene perception with a button press. Over the course of 4 days, 1000 superthreshold stimuli interspersed with 500 sham trials were delivered to different microelectrodes. The subject correctly identified 947 of the superthreshold stimuli, and only infrequently (23 out of 500) indicated perception of an evoked phosphene when a sham trial was presented. We also used a 2-alternative forced-choice (2-AFC) paradigm where the subject had to indicate if the first or the second stimulus was associated with a phosphene. The percentage of correct responses in these experiments was greater than 95%. Her reliable performance in these experiments allowed us to begin the optimization of the various electrical stimulation parameters.

Stimulation parameter optimization.

Since threshold currents for the detection of phosphenes are a function of multiple parameters (amplitude of the biphasic stimulus, pulse frequency, pulse duration, inter-phase interval, train length, and polarity of the stimulus), we initially performed several experiments to optimize these parameters. Optimum stimulation was defined as the combination of stimulation parameters that provided reliable perceptions with the minimum amount of current in a given stimulation period. Current-induced tissue damage during electrical stimulation was avoided by keeping stimulus charge density below the damage threshold boundary (20, 21). In order to reduce response variability, we performed these optimization experiments by simultaneously passing the same currents via 4 neighboring (abutting) electrodes.

Representative psychometric curves of these optimization experiments are shown in Figure 2, which shows the probability of phosphene perception as a function of the charge per phase for different pulse durations (Figure 2A), frequencies (Figure 2B), and train durations (Figure 2C). The lower charge per phase was reached using biphasic stimuli with phase durations of 170 μs/phase, although the behavioral results were similar for phase durations of 100 and 170 μs (P = 0.26, ANOVA test). Increasing the phase duration to 400 μs or 800 μs resulted in a significant increase in the charge per phase. A relatively small amount of charge per phase was required when we used a frequency of 300 Hz and a train duration of 166 ms (Figure 2, B and C). The polarity of the stimulation (negative/positive phase first) did not have a significant influence on stimulation thresholds (P = 0.108, t test).

Open in a separate window

Figure 2

Relationship between phosphene thresholds and charge per phase.

(A) Probability of phosphene perception for different pulse durations (pulse width, PW). (B) Effects of the stimulation frequency. (C) Effects of the duration of the train (TD) of stimulation pulses. One hundred twenty responses were used for each psychometric curve. The dashed horizontal blue lines represent the thresholds (50% probability of detection).

Based on these results, we selected the following parameters for inducing phosphene percepts: stimulus trains composed of 50 charge-balanced cathodic first biphasic stimuli, a phase duration of 170 μs/phase, an interphase interval of 60 μs, and a frequency of 300 Hz (train duration of 166 ms; see Figure 3A). We used these stimulus parameters in most experiments (unless otherwise specified).

Open in a separate window

Figure 3

Phosphene thresholds.

(A) Diagram illustrating the stimulation parameters of the biphasic pulse waveforms that were used for the quantification of the thresholds. (B) Distribution of phosphene thresholds for stimulation via single microelectrodes. Electrode tips are pointing away from the page. (C) Distribution of the thresholds for all the electrodes on week 21. (D) Distribution of phosphene thresholds evoked by simultaneous stimulation via sets of 4 contiguous (abutting) electrodes. Electrode tips are pointing away from the page. (E) Histogram of averaged thresholds for stimulation via the sets of 4 electrodes. (F) Evolution of the threshold for stimulation of single electrodes and for groups of 4 electrodes. Data presented as mean ± SEM. (G) Representative psychometric curves generated in response to stimulation (biphasic pulses, with a phase duration of 170 μs/phase and a frequency of 300 Hz) via 1 and 4 abutting electrodes.

Stimulation thresholds.

To measure phosphene thresholds, we used a binary search procedure (see Methods). We were able to reliably evoke phosphenes with 88 of the 96 electrodes. Mean stimulation thresholds for evoking phosphenes using single electrodes was 66.8 ± 36.5 μA (range 7–128 μA). Thresholds measured on different single electrodes ranged from 1.2 nC/phase up to 20.4 nC/phase (average threshold of 11.2 nC/phase), and threshold charge densities to evoke a visual perception at single electrodes (calculated from the geometric surface areas of the electrodes) ranged from 40 to 680 μC/cm²/phase (average threshold charge density was 374 μC/cm²/phase). Nonetheless, since we used sputtered iridium oxide film electrodes, in which the effective surface area is much larger than the nominal geometric surface area, it is highly probable that the charge densities in the tissue were lower.

Figure 3B shows the distribution of average thresholds and Figure 3C shows a representative histogram of the thresholds for all the individual electrodes in week 21, whereas Figure 3, D and E show the distribution of phosphene thresholds for groups of 4 electrodes distributed across the UEA. The threshold levels appeared to be distributed quite uniformly across the entire UEA. A longitudinal study of mean current threshold levels from the first reliable measurements showed that mean thresholds of individual electrodes increased by 19.4% over the last 4 months of the study, from 64.8 ± 25.7.5 μA up to 80.4 ± 28.8 μA (Figure 3F). Therefore, thresholds were relatively stable over time (changes were less than 20%), akin to the electrode impedances and the quality of multiunit recordings.

We also investigated phosphene thresholds by stimulating subsets of 2, 4, 9, and 16 contiguous electrodes at various locations across the UEA. We found only subtle differences between phosphene thresholds evoked by single electrodes and pairs of electrodes, but stimulation of larger groups of electrodes was associated with a significant reduction in thresholds (P < 0.001, ANOVA test). Figure 3G shows 2 representative psychometric curves measured with single electrodes and 4 abutting electrodes. The mean thresholds for 4-electrode stimulation were 28% lower on average (48.0 ± 22.6 μA) than those of single electrodes (66.8 ± 36.5 μA) and the threshold stimulation levels for the combinations of 4 electrodes were very stable over time (Figure 3F). Supplemental Figure 1 shows the mean thresholds for single electrodes and for groups of 2, 4, 9, and 16 contiguous electrodes.

Simultaneous stimulation and recording experiments.

In some experiments, we performed electrophysiological recordings during electrical stimulation (see Supplemental Video 1). As expected, electrical stimulation often altered the activity of the neurons surrounding the electrodes. Figure 4 shows some representative examples. In general, the electrical stimulation trains were followed by an increase in multiunit activity (85% of the cases). However, the increased firing of the neurons around the stimulated electrodes did not always correlate with the perception of phosphenes (Figure 4A). For trials in which the subject reported seeing a phosphene, we observed an increase in activity in 65.5% of the trials. The remaining 34.5% of trials showed an increase in activity trials without phosphene perception. On some occasions (15% of trials), we also found an inhibition of spiking activity on some electrodes following electrical stimulation, which usually returned to baseline after a few seconds (Figure 4B).

Open in a separate window

Figure 4

Representative simultaneous stimulation and recording experiments.

(A) Repetitive stimulation of electrode 28 with 78 μA at 0.5 Hz (red lines, 10 times, identical parameters) induced an increased firing of the neurons surrounding this electrode. Only the last 4 stimulations were associated with the perception of a phosphene (see upper enlarged panel where “–” indicates no perception and “Y” indicates perception). (B) Repetitive stimulation of electrode 9 with 64 μA at 1 Hz induced an inhibition of the neurons surrounding the electrode that recovered after a few seconds. In each of these examples, the neural recording was obtained from the stimulated electrode. Red vertical bars indicate the stimulus times.

When we stimulated several electrodes simultaneously, we observed diverse responses in different electrodes. Often there was an increase in activity in all the electrodes (84% of the trials), but sometimes we also recorded an increase in activity related to the stimulation in some electrodes and some inhibitory responses in other electrodes. In several cases (16% of trials), we also observed inhibitory responses in all of the targeted electrodes, despite the fact that the subject reported seeing a weak percept.

Overall, 74.7% of stimulation trials that evoked an increase in the activity recorded around the electrodes (either by stimulating a single electrode or several electrodes simultaneously) were associated with a reported visual perception.

Phosphene mapping.

During the 6-month study period we performed a total of 18 mapping sessions to estimate the location of the phosphenes in visual space. The subject manually indicated the location of the phosphene on a tablet with a small post in the center, which was used as the reference location (see Methods). An example map of the phosphene location of each individual microelectrode in the subject’s visual field is shown in Figure 5A. All the phosphenes were close to the central visual field and located in an ellipsoidal region in her left hemifield (shown as a red ellipse in Figure 5A) adjacent to the horizontal meridian. However, the reported location of each phosphene varied slightly between sessions, especially for some of the electrodes (Figure 5B). These variations were always less than 0.5 degrees and were influenced by gaze position. There was also a good agreement between predicted phosphene locations based on our presurgical analysis (refs. 22, 23; yellow area in Figure 5A) and the reported locations of the evoked phosphenes (blue dots with electrode numbers in Figure 5A).

Open in a separate window

Figure 5

Phosphene map and relative brightness and size of perceived phosphenes.

(A) Location of the perceived phosphenes (blue dots) and electrode numbering view from the pad side. The cross indicates the center of the subject’s visual field (the intersection of her horizontal and vertical meridians). Yellow region: Expected location of the phosphenes based on a standard retinotopic map superimposed on the anatomy of the visual cortex of the patient using the procedures described by Benson et al. (22, 23) and the selected implantation site. Calibration bar = 1 degree. (B) Changes in the location of the perceived phosphenes for the same 4 electrodes in 5 different trials. Calibration bar = 1 degree.

Phosphene color.

When individual microelectrodes were stimulated at or above the subject’s 50% threshold level (the stimulus that evoked a phosphene 50% of the time), the subject usually reported that the evoked perceptions were colorless (white). For current intensities below the 50% threshold, phosphenes were usually colorless, but sometimes the reported color was yellowish or sepia toned. When we used very low intensity currents (<5 μA), the subject often perceived something resembling a very dim light, which she described as a “change of intensity.” She was able to clearly differentiate these low-intensity perceptions from spontaneous and normally evoked phosphenes, but it was difficult for her to localize the exact position and size of these blurry perceptions in her visual field. Nonetheless, on many occasions she said that they were located “at the center of her left eye.”

Modulation of phosphene brightness and size.

We asked the subject to subjectively calibrate the brightness and size of her phosphenes on a scale from 0 to 5. Phosphene brightness increased with higher currents (P < 0.001, ANOVA test). Figure 6A shows the average brightness ratings for 10 individual electrodes. The highest brightness ratings were reached with currents of approximately 90 μA, close to the largest currents that were tested. In another set of experiments, we investigated how the phosphene brightness depended on the phase duration of the biphasic pulses. We used 3 different phase durations: 100, 400, and 800 μs. While the perceived brightness increased slightly by increasing the duration of each phase of the stimulus, the changes were not significant (P = 0.727, ANOVA test).

Open in a separate window

Figure 6

Relative brightness and size of perceived phosphenes on a subjective scale ranging from 0 to 5.

(A) Averaged values of perceived brightness across 10 single electrodes. Maximum was reached for currents of approximately 90 μA. (B) Subjective brightness and (C) subjective size of perceived phosphenes as a function of the number of simultaneously stimulated electrodes. Error bars denote SEM.

The estimated phosphene size evoked by single electrode stimulation was usually very small (0.8 ± 0.8; mean ± SD of subject’s size estimates) and resembled “pin points” of light at arm’s length, although there was considerable variability in perceived phosphene size evoked by different electrodes (size estimates ranged from 0.5 to 3.5; subjective units). The size of the evoked perceptions did not change significantly as a function of current amplitude on a given electrode (P = 0.212, ANOVA test).

Increasing the number of stimulating electrodes significantly increased the subjective brightness and perceived size of the phosphenes (Figure 6, B and C). This relationship was observed when stimulation of groups of electrodes (2, 4, 9, and 16) was compared to that of single electrodes (P < 0.001), and when stimulation of larger groups of 4, 9, and 16 electrodes was compared to that of smaller groups of 2 electrodes (P < 0.001, ANOVA test).

We also investigated the possible accommodation of phosphenes to repeated suprathreshold stimulation (see Methods). Our results showed that the apparent brightness and size of the phosphenes did not change significantly with 30 successive stimulations (P = 0.727, Pearson’s correlation coefficient = 0.010).

Two-point discrimination.

A number of stimulation experiments were conducted with 50 pairs of electrodes (10 stimulations/trial). In 23.6% of the stimulations, the simultaneous stimulation of 2 electrodes evoked 2 discrete phosphenes, especially for widely separated electrode pairs. This occasionally also happened for electrodes separated by 400 μm (Figure 7A). However, very often (76.4% of the stimulations), simultaneous stimulation of electrode pairs evoked a single phosphene, albeit with a subjective estimated size that was 1.9 ± 0.1 times greater than the estimated size of the phosphenes evoked by the stimulation of either electrode in isolation (0.8 ± 0.1; P < 0.001, ANOVA test). The results were generally consistent for the same pair and testing session.

Open in a separate window

Figure 7

Examples of perceptions evoked by simultaneous stimulation of multiple electrodes.

(A) The stimulation of the 2 electrodes shown in the inset electrode map induced the perception of 2 closely spaced phosphenes. (B) Box-and-whisker plot of subjective phosphene size for electrode pairs separated from 400 to 3600 μm. Inset: Relative location of electrodes 1 and 10 in the UEA, which are separated by 400 μm (light blue color), and electrodes 1 and 89, which are separated by 3600 μm (orange color). In the box-and-whisker plot, the boundary of the box closest to zero indicates the first quartile, and the boundary of the box farthest from zero indicates the third quartile. The horizontal line is the median. The whiskers show the maximum and minimum values, with the exception of outliers (small circles) and extremes (stars). (C) Simultaneous stimulation of 4 contiguous electrodes was perceived as 3 small dots. (D) Stimulation of 12 electrodes induced the perception of a horizontal line. (E) Simultaneous stimulation of the 2 groups of 4 blue electrodes evoked the percept of a line with a horizontal orientation, whereas the simultaneous stimulation of the 2 groups of 4 red electrodes evoked the percept of a line with a vertical orientation. (F) Stimulation of these electrodes elicited the perception of a lowercase letter i. (G) Stimulation of these 2 groups of electrodes unexpectedly induced the perception of an uppercase letter L. (H) Stimulation of these electrodes elicited the perception of an uppercase letter O.

To further quantify the subjective perception of phosphene size when stimulating pairs of electrodes, we examined 4 combinations of pairs of electrodes that had interelectrode distances of 400, 800, 1200, 1600, 2000, 2400, 2800, 3200, and 3600 μm. These experiments were performed with 36 pairs of electrodes (10 stimulations/trial). Our results showed that the subjective estimated size correlated well with the degree of separation between electrodes (Figure 7B). Furthermore, the subject reported that the percepts had more elongated shapes when we increased the distance between the stimulating electrodes. This suggests that the phosphene’s size and appearance is not only a function of the number of electrodes being stimulated, but also of their spatial distribution in the UEA.

We also stimulated 40 electrode pairs with a delay between stimuli to identify the minimum time interval that facilitates the perception of discrete phosphenes (10 stimulations/trial). For these experiments, we used a fixed pulse amplitude of 90 μA for both electrodes and delays of 50, 100, 200, 250, 300, 400, 500, 700, 1000, 2000, and 4000 ms between the 2 trains of stimulating pulses. For electrodes separated by 400 μm, when stimulation of the 2 electrodes was temporally separated by more than 250 ms, the subject perceived 2 different, distinct phosphenes 90% of the time. This discrimination reached 100% for electrodes separated by 565 μm (along the diagonal) or further away. Further increasing the delay between both stimulus trains (from 300 to 4000 ms) facilitated the identification of 2 distinct phosphenes.

Discrimination and recognition of complex 2D stimulus patterns.

We next investigated if the simultaneous stimulation of more than 2 electrodes induced more complex perceptions. For these experiments, we simultaneously stimulated different combinations of 3 to 16 electrodes distributed over the UEA. Increases in the number of simultaneously stimulated electrodes increased the probability of complex visual percepts. Furthermore, the subject reported that the perceptions elicited by groups of simultaneously stimulated electrodes were easier to perceive and clearer than when we stimulated single electrodes.

First, we simultaneously stimulated 30 different groups of 4 contiguous electrodes distributed across the UEA, plus 4 groups of 3 electrodes (at the corners of the array). Figure 7, C–E illustrate several representative examples of patterned stimuli that were used, and the subject’s sketches of the 2D percepts that the stimuli evoked (drawn on a digitizer tablet). Some stimulation patterns evoked percepts of closely spaced dots (Figure 7C), whereas other patterns evoked horizontal or almost horizontal lines (Figure 7D). It was more difficult to induce vertical lines, but Figure 7E shows a representative example of 2 groups of 8 electrodes that evoked horizontal and vertical line percepts. To examine the reproducibility of the evoked percepts, and to monitor the subject’s ability to discriminate between horizontal and vertical evoked percepts, we presented these 2 patterns in random order and asked the subject to indicate the perceived line orientation using a 2-AFC approach. The subject’s performance was always above chance level and across 22 sessions on 4 different days, the overall success rate was 81.4%. Moreover, her ability to correctly discriminate horizontal and vertical patterns increased with training, and in later sessions accuracy reached 100%.

Encouraged by these results, in the last month of the experimental period we investigated several spatial stimulation patterns that we expected to induce the perception of letters using up to 16 simultaneously stimulated electrodes (the maximum number of simultaneous channels possible with our neurostimulator). Our subject spontaneously reported the perception of some characters and reliably discriminated some letters such as i, L, C, V, and O. Over a series of 5 days, we performed 24 sessions using a 2-AFC approach (30 stimulations/trial) to differentiate between several complex 2D patterns. The characters tested were i versus L, i versus O, L versus C, C versus V, and L versus V. The overall success rate was always greater than 70%. Figure 7, F–H illustrate some of the electrode combinations that elicited the perception of letters. Although several combinations of electrodes evoked perceptions that resembled letters (such as an inverted U or a rotated T), we were unable to induce the perception of all the letters of the alphabet. Furthermore, many stimulation patterns were not recognized as letters by our subject (see Supplemental Figure 2); hence, we only tried them a few times before switching to other stimulation patterns that were able to induce letter percepts. Additionally, on some occasions the perceptions did not match with our expectations based on the stimulated pattern of electrodes. An example is shown in Figure 7G, where we simultaneously stimulated a group of 4 electrodes and another separate group of 3 electrodes, and the participant’s perception resembled the letter L.

To increase motivation and facilitate the learning of different stimulus patterns, the subject played several games that in reality were 2-AFC experiments (20 stimulations/trial). The games used patterned stimulation, and the main aim was to help the subject to discriminate between different patterns and/or group of electrodes. One of the games was to guess if Maggie Simpson (a fictional character in the animated television series “The Simpsons”) was shooting to the left or to the right. Each shooting direction was associated with the stimulation of a particular set of electrodes, and shooting sequences were randomized. After each shooting, the subject had to a press the appropriate button (left or right) on a wireless keypad to stop the bullets (see Supplemental Video 2).

To further investigate the influence of learning, we selected electrode sets that evoked 2 different alphanumeric characters, a lowercase o and an uppercase O, which were easily recognized by the subject even on the very first trials (Figure 8A). When we asked the subject to characterize these percepts, she clearly and spontaneously described the perception as a “big O” or a “small o.” Across 20 sessions over several days, involving a total of 400 trials, there was an improvement process that reached 90% to 95% after 70 trials and 100% accuracy in the latest sessions (Figure 8B). Furthermore, the effects of training were lasting. These improvements in the identification of evoked perceptions over time indicate that the subject learned to reliably recognize specific patterns.

Open in a separate window

Figure 8

Evolution of the performance in a task in which the subject discriminated an uppercase O from a lowercase o.

(A) Stimulated electrodes for the perception of each pattern. (B) Learning process that reached an accuracy of 100% in the last session. Error bars denote SEM.

Preliminary experiments with a bio-inspired retinal encoder.

As the experimental period neared completion, we performed some preliminary tests using a bio-inspired retina-like visual encoder to dynamically stimulate different combinations of electrodes according to the visual scene in front of the subject. The visual encoder contained a head-mounted camera and spatial and temporal image processors (Figure 9, A–C). Using the visual encoder, the subject was able to “head scan” objects in front of her.

Open in a separate window

Figure 9

Bio-inspired artificial retina.

(A) Schematic organization of the sight restoration concept. (B) The image acquisition system. The input images are captured by a video camera attached to a spectacle frame for subsequent bio-inspired processing. (C) Signal processing module. Input images are processed by a combination of several spatial and temporal filters that enhance specific features of captured information. The weighting module re-encodes this information into a neuromorphic stream of electrode addresses and sends it to the CereStim 96, which generates the stimulation signals applied to the intracortical microelectrodes. (D) Discrimination of the border between black and white bars using the bio-inspired artificial retina (frame extracted from Supplemental Video 1). (E) Evolution of the time required to perform the object location task (4 possible locations) over several days.

To investigate the spatial accuracy of this approach, the subject was first trained to use head scanning in order to discriminate the location of the borders between black and white bars printed on cardboard. The subject was able to locate the black/white border in all the tests (see Figure 9D and Supplemental Video 3). Then the subject was trained to discriminate the location of a large white square (20 × 20 cm) appearing randomly at either the left or right half of a 21-inch computer monitor. The subject was seated 50 cm directly in front of the monitor and quickly learned to perform head scanning to successfully localize the white square (100% accuracy across 70 trials). We then presented a smaller white square (14 × 14 cm) randomly in 1 of 4 possible locations — upper left and right and lower left and right —and she had to point out the location of the white square. The subject was able to correctly point to the white square 100% of the time (150 trials across 5 days), and her reaction time decreased from 8.6 ± 0.6 seconds for the first trials to 5.4 ± 0.3 seconds (38% less on average) after only 2 training days. The effects of training were long lasting and once she learned to perform the task, the required time remained stable (Figure 9E). Furthermore, after the short training period, the subject reported that the task became easier to carry out.

Implant location.

The UEA was implanted in the subject’s right occipital cortex, near the occipital pole and close to the border between V1 and V2 (Figure 1, B and C). During preliminary studies we found that transcranial magnetic stimulation of the right visual cortex elicited reproducible phosphenes in our subject, especially when the coil was located near the occipital pole. Therefore, we selected her right occipital cortex for implantation. For fine presurgical planning, we performed a 3D reconstruction of the surface anatomy and the neurovascular structures of the subject’s primary visual cortex to optimize the implantation location of the electrode array. We tried to avoid as much as possible the major blood vessels that supplied the region. Anatomical images of the occipital cortex and other areas of the visual system were acquired on a 3-T magnetic resonance scanner (Siemens MAGNETON Skyra) using the MPRAGE protocol (192 sagittal slices, 256 × 256 matrix, 1 × 1 × 1 mm³, TR = 1900, TE = 2.49, FS = 3). We used Horos, a free and open-source computer software that is distributed under the LGPL license (https://horosproject.org/), to create 3D reconstructions of the cerebral surface showing cortical anatomy as well as the major cortical vessels. The Freesurfer image analysis suite v6.0 (https://surfer.nmr.mgh.harvard.edu/) was used for cortical reconstruction and volumetric segmentation. Furthermore, we examined the predicted retinotopic organization of the occipital cortex of this blind subject using cortical surface anatomy to predict her visual areas V1, V2, and V3 according to the procedures described by Benson et al. (22, 23). Based on this information, we selected a region at the border of V1 and V2, which could be accessed easily while avoiding major blood vessels.

We selected this region because the striate cortex is located in the medial surface of the occipital lobe, and hence more difficult to reach, and further, contains the calcarine artery, a branch of the posterior cerebral artery that could be damaged by the implantation. Moreover, our previous studies in people suffering from epilepsy and brain tumors who had to undergo surgical resection of a brain region show that electrical stimulation of extrastriate cortex (Brodmann area 18 and all Brodmann area 19) still induces visual perceptions with consistent retinotopic organization (37).

Surgery.

The surgical procedures for UEA implantation followed standard neurosurgical procedures that have been described elsewhere (34, 37). Briefly, after the scalp was prepared with an antiseptic, a supratentorial occipital right paramedial craniotomy centered over the desired location of array implantation was performed. After opening the dura toward the midline, the surface of the brain was exposed and the UEA was inserted using a pneumatic inserter (Blackrock Microsystems). The electrode’s external connector was attached to the skull using six 5-mm titanium microscrews. The final array location was documented with intraoperative photographs and a postoperative computed tomography (CT) scan that was fused with the preoperative planning on 3D MRI.

After 3 months, the microscrews securing the external connector to the skull loosened, causing a minor displacement of the connector relative to the skull. The connector was remounted using new microscrews under local anesthesia. After 6 months, the subject was reanesthetized and a craniotomy was performed in order to explant the UEA. There was negligible fibrosis around the implanted UEA, and the UEA and the external connector were easily explanted without complications. The wound was closed in a standard way using subcutaneous reabsorbable sutures and stainless skin staples. The subject has been followed periodically after the explantation and reports no complications.

Impedance measurements.

Prior to each stimulating and recording session, we performed an impedance test of all 96 electrodes at 1 kHz using the “test electrodes” function of the CereStim 96 multichannel microstimulation system (Blackrock Microsystems). Impedance measurements were analyzed off line using custom MatLab software (MathWorks) and Python software (Python Language Reference v2.7 available at http://www.python.org).

Neural recording and stimulation.

Electrical stimulation and multiunit neural recording sessions were generally performed 5 days per week (Monday to Friday), 1 to 2 times per day (morning and afternoon sessions) for up to 4 hours per session over a period of 6 months. Neural signals were recorded using the NeuroPort data acquisition system (Blackrock Microsystems). Constant current stimuli were delivered using the CereStim 96 and customized software was used to deliver multiple stimulation patterns to different subsets of electrodes, thereby creating custom stimulation configurations using interleaved channels. The electrical stimuli used in these experiments were delivered between the selected electrode (or a selected group of electrodes) and a deinsulated platinum wire that was positioned under the dura, near the implanted UEA. Because of limitations in the CereStim 96 stimulator, only 16 electrodes could be stimulated simultaneously. In some experiments, we also simultaneously recorded EEG signals at a sampling rate of 1000 Hz with the help of the Curry 7 platform (Compumedics). The number of hours of stimulation over the 6-month period was approximately 540 hours.

For removing stimulation artifacts, we used a fast independent component analysis (ICA) algorithm (53).

Optimization of the stimulation parameters.

We determined psychometric curves for the detection and discrimination tasks. Each stimulation was accompanied by a “stimulating-now” tone. Different test stimulus amplitudes, ranging from 1 to 128 μA, were presented randomly until 10 responses for each test stimulus amplitude were recorded. In addition, 20% of the trials were randomized sham stimuli (no stimulus). Psychometric data were fit using a Weibull cumulative distribution function to estimate response thresholds (amplitude value at 50% probability of detection). The calculations were performed using the scipy.optimize function in the Python package SciPy (54).

The geometrical surface area of electrode tips was calculated from SEM micrographs of the explanted UEA by assuming conically shaped tips.

Threshold testing.

A binary search procedure was used to measure phosphene thresholds on each of the 96 electrodes. Stimulation spanned 1 to 128 μA with a minimum step size of 1 μA. In addition, we used 10% catch trials in which a null stimulus was presented. This technique allowed us to estimate the lowest stimulus intensity that produced reliable phosphene perceptions for each electrode. Briefly, we first randomly selected one UEA electrode and then we started stimulation using a binary convergence search. Each intensity was tested at least twice and we instructed the subject to consider only clear phosphenes. We started with an amplitude of 1 μA. If the subject did not perceive the stimulus, we increased the amplitude of the stimulus by a factor of 2. If the subject perceived the new intensity, we performed a half-interval search between this value and the last unperceived amplitude until the volunteer was able to correctly identify the stimulus in 2 consecutive trials. This procedure was repeated until the subject was able to perceive 2 successive phosphenes evoked by a given stimulus amplitude or when we reached a maximum stimulation current of 128 μA. We then moved to the next randomly chosen electrode. This protocol allowed us to test all 96 electrodes in approximately 20 to 30 minutes and was used to periodically measure thresholds for all electrodes.

Accommodation of phosphenes to repeated stimulation.

A stimulus of twice the threshold current level was presented 30 consecutive times. For these experiments, we used 50 charge-balanced cathodic first biphasic pulses, with several different pulse widths (100, 170, 400, and 800 μs) at a frequency of 300 Hz, and adjusted the interpulse interval to keep the complete duration of each train to 166 ms. Stimulation trains were presented every 4 seconds, which allowed sufficient time for the subject to respond verbally regarding the features of the perceptions and to perform a keyboard button press. Five electrodes were tested using this paradigm.

Two-point discrimination.

A number of stimulation experiments were conducted with different pairs of electrodes distributed over the whole UEA. We also stimulated electrode pairs with a delay between stimuli to identify the minimum time interval that facilitated the perception of 2 discrete phosphenes. For these experiments, we used delays of 50, 100, 200, 250, 300, 400, 500, 700, 1000, 2000, and 4000 ms between the 2 trains of stimulating pulses.

Phosphene mapping.

For mapping the location of phosphenes evoked by each microelectrode, we used our standard stimulus protocol. The subject was instructed to concentrate on holding her eye position as if she were looking straight ahead. Each stimulus was accompanied with a brief “stimulating-now” tone. If she perceived a phosphene, then she was requested to indicate the location of the phosphene with respect to the center of her visual field. To help with this task, we used a wireless drawing tablet (Wacom Intuos M, Wacom Co. Ltd) on which we added a small post at the center of the tablet for tactile reference and orientation. In addition, we asked her to describe the qualities of her perceptions with respect to the shape, size, color, brightness (subjective ratings from 0 to 5), and possible motion of the evoked perceptions. The digitizing tablet was also used by the subject for drawing the perceptions elicited by the stimulation of multiple electrodes.

In some of these experiments, we used a portable eye tracking system (Tobii Pro Glasses 2) to measure the subject’s eye movements. Our results showed that the phosphenes moved along with eye movements, which complicated the generation of consistent phosphene maps. To avoid this complication, before each electrical stimulation the participant placed her finger on the post and was asked to concentrate on holding her gaze straight at the center of her visual space.

Bio-inspired retinal encoder.

The visual encoder system has been described elsewhere (12, 13, 55–59). In brief, it consisted of a video camera attached to an eyeglass frame for image acquisition using head scanning, and custom hardware/software that performs a real-time analysis of the light patterns that are received by the light sensors in the camera and a multichannel spatiotemporal filtering of the visual information to extract and enhance the most relevant features of the scene. The system also includes an additional stage that converts the output of the bio-inspired retinal encoder into stimulus parameters and reencodes this information into a neuromorphic stream of electrode addresses and stimulation amplitudes. The filtering in the retinal encoder was adjusted to change the pulse train amplitude sent to the electrodes to modulate phosphene brightness. Although the retinal encoder has the capability of communicating with hundreds of electrodes, the CereStim 96 electrode stimulator used in these experiments limited the delivery of simultaneous stimulation to 16 electrodes. These 16 electrodes were selected to stimulate via electrodes distributed over the entire UEA.

Training sessions.

Training sessions lasted 10 to 15 minutes. Usually we performed several trials (20–30 stimulations/condition). The number of training sessions was dependent on the task to learn but as a rule we performed 4 to 6 training sessions per task.

Statistics.

Statistical analysis was performed using IBM SPSS 20.0 statistics software. We used linear regression to determine significant changes in electrode thresholds over time and for investigating accommodation to repeated stimuli. Unless otherwise noted, data are reported as mean ± standard deviation (SD) and averages were compared using t test or ANOVA with Scheffe’s procedure for multiple comparisons. In all cases, a P value of less than 0.05 was considered statistically significant.

We would like to thank the study subject (BG) and her husband for their extraordinary commitment to this study and their patience with experiments. We are very grateful to all the people from IMED Hospital Elche, especially to Carlos Yago, Enrique Ibañez, and Angel Gomez for all their help and support and for providing clinical care and monitoring throughout the study. We would also like to thank Joaquín López (Bidons Egara Research Chair) and Randall J. Olson (John A. Moran Eye Center, University of Utah) for their invaluable help and support during the planning and development of this research, and Noah C. Benson (University of Washington) and Jennifer Sorinas (University of Zurich) for all their help with the retinotopic analysis of the MRI images. Funding was provided by grant RTI2018-098969-B-100 from the Spanish Ministerio de Ciencia Innovación y Universidades, by grant PROMETEO/2019/119 from the Generalitat Valenciana, by the Bidons Egara Research Chair of the University Miguel Hernández, by the John A. Moran Eye Center of the University of Utah, by a career development award from the NIH National Institute for Neurological Disorders and Stroke (K23 NS114178), and by an unrestricted grant from Research to Prevent Blindness to the Department of Ophthalmology and Visual Sciences, University of Utah..