Longevity and Reliability of Chronic Unit Recordings using the Utah, Intracortical Multi-electrode Arrays

Abstract

Objective.

Microelectrode arrays are standard tools for conducting chronic electrophysiological experiments, allowing researchers to simultaneously record from large numbers of neurons. Specifically, Utah electrode arrays (UEAs) have been utilized by scientists in many species, including rodents, rhesus macaques, marmosets, and human participants. The field of clinical human brain-computer interfaces currently relies on the UEA as a number of research groups have FDA clearance for this device through the investigational device exemption pathway. Despite its widespread usage in systems neuroscience, few studies have comprehensively evaluated the reliability and signal quality of the Utah array over long periods of time in a large dataset.

Approach.

We collected and analyzed over 6000 recorded datasets from various cortical areas spanning almost 9 years of experiments, totaling 17 rhesus macaques (Macaca Mulatta) and 2 human subjects, and 55 separate microelectrode Utah arrays. The scale of this dataset allowed us to evaluate the average life of these arrays, based primarily on the signal-to-noise ratio of each electrode over time.

Main Results.

Using implants in primary motor, premotor, prefrontal, and somatosensory cortices, we found that the average lifespan of available recordings from UEAs was 622 days, although we provide several examples of these UEAs lasting over 1000 days and one up to 9 years; human implants were also shown to last longer than non-human primate implants. We also found that electrode length did not affect longevity and quality, but iridium oxide metallization on the electrode tip exhibited superior yield as compared to platinum metallization.

Significance.

Understanding longevity and reliability of microelectrode array recordings allows researchers to set expectations and plan experiments accordingly and maximize the amount of high-quality data gathered. Our results suggest that one can expect chronic unit recordings to last at least two years, with the possibility for arrays to last the better part of a decade.

1. Introduction

First developed in 1992, a microelectrode array recording platform now called the “Utah Electrode Array” (UEA) has provided the opportunity for researchers to record from populations of neurons simultaneously, in-vivo, in a chronic preparation (1). Today, UEAs are a standard data collection method for systems neuroscience and represent a common choice for chronic multi-electrode array recordings, particularly in animal models such as rodents and non-human primates (NHPs), including macaques and marmosets (2-4). With the popularity of such a tool comes increased scrutiny on the reliability and failure modes of the technology. UEAs, like all chronic neural implants that interface with brain tissue, usually degrade in their recording capabilities over time (4); in some cases, these implants can experience more acute issues, either during surgical implantation or shortly thereafter, whereas in many other cases these implants can reliably record unit activity for several years (4). These implants have been in widespread use for many years, yet there have been few attempts to quantify the reliability of this technology in-vivo (4). Most previous efforts to quantify the signal quality of UEAs over time, while laudable, have focused heavily on the consistency of neuronal characteristics in a small sample size of implants and subjects (5-9). Other studies have examined the longevity and reliability of other chronic electrophysiological recording technologies, to help researchers manage expectations and select the right tool for the job (10-14). However, one previous study has examined the failure modes and reliability of UEAs from a large sample size, albeit with UEAs undergoing significant changes in manufacturing processes (4). Additionally, other studies evaluate the differences and capabilities of various recording arrays, but not with non-human primate subjects, making cross-species comparison difficult (14-17). The current work aims to extend this previous study by considering more recent UEAs whose manufacturing process has become relatively stable and considers UEAs with iridium oxide metallization optimized for stimulation as well as platinum metallization on the electrode tips.

UEAs currently remain the only FDA-cleared intracortical implant for chronic human neuroscientific studies under Investigational Device Exemptions (IDE), an allowance which is held by several groups across the United States. There have been at least 18 human subjects who have received a chronic UEA implant for brain computer interface (BCI) studies reported in the literature as of 2019 (18), and we estimate that at least another 7 people have been implanted since then (19). Human studies require ongoing consideration of the risk-benefit ratio of implanting human subjects, which will be informed by more information on the expected reliability and longevity of these implants over time. A majority of human studies expect participants to remain in the study for at least 1 year, which is supported by a significant body of evidence demonstrating the safety and efficacy of these devices for long term recording (2-16). Here we extend this earlier work by examining a large number of array implants over extended durations up to 9 years in both humans and non-human primate subjects.

It is likely that the UEA will continue to be the dominant neural recording platform for human intracortical BCI studies in the near future. FDA clearance for investigational approval requires the collection of significant longitudinal safety and efficacy data along with the institutional knowledge for a successful clinical trial design (20). A number of efforts are currently underway to develop more advanced tools for chronic implants across non-human and human primates alike (21-25), but given the prevalence of the UEA for chronic recording and stimulation, it is imperative to characterize the reliability of these arrays across and within species, to take full advantage of their capabilities.

Here, using data from over 6000 recording sessions in 55 array implants across human (n=2) and non-human primate (n=17) subjects, we assess the reliability of long-term unit recordings in primary motor, dorsal and ventral premotor, prefrontal, and primary somatosensory cortices. We show that nearly fifty percent of implants exhibit year-long recordings with a yield greater than 40% of the total available electrodes displaying satisfactory quality (SNR>1.5). Moreover, long-term recordings from ~1000 days and up to nine years are shown to be possible in some cases. We also examine the influence of electrode length and electrode-tip metallization on the longevity and quality of recordings.

2. Methods

2.1. Non-Human Primate Subjects

Recordings collected during the period of 2003-2020 from a total of 55 UEAs implanted in 17 rhesus macaques (Macaca Mulatta) (10 female, 7 male) were analyzed. This project analyzed data from the maximum possible number of recordings throughout the history of our research group. All animals were implanted with at least one (often two or more) UEAs (Blackrock Microsystems, Inc. Salt Lake City, UT). The earliest implants were manufactured by Cyberkinetics Neurotechnology Systems, Inc., Foxboro MA whose research business was sold to Blackrock Microsystems, Inc. in 2008. The metallization, number of electrodes, and electrode length of these arrays varied depending on brain area, scientific need, and implantation year, and individual array details are included in Table 1. Briefly, arrays were either implanted as an 8x8 or 10x10 grid, with either 1.5 mm or 1.0 mm length electrodes (although never mixed within a given array), with electrode tips metallized with either platinum or iridium-oxide. Since our research group focuses primarily on upper limb motor tasks, UEAs were most often implanted in primary motor (M1), dorsal premotor (PMd), or ventral premotor cortices (PMv). Other implanted areas included somatosensory (S1), orofacial primary motor (M1o), and prefrontal cortex (PFC). We saw no appreciable difference in signal longevity or quality between the two areas which had large enough numbers of implants to compare (dorsal/ventral premotor versus primary motor areas). All implants were connected to CerePort connectors with analog headstages produced by Blackrock Microsystems. All arrays were wired in a consistent manner; none were hand soldered.

Table 1.

List of microelectrode arrays included in the current study. Column A indicates the abbreviation used for the identity of the animal and brain area. Many animals were implanted with multiple arrays. Column B indicates the sex of the animal in which the array was implanted. Column C indicates the brain area in which the array was implanted. Column D indicates the date of initial array implantation in YYMMDD format. The rightmost column is reserved for reasons for explant. “Medical” encompasses infections, skin retraction, or other medical issues which required the implant to be removed, unrelated to signal quality. “Signal Quality” refers to poor signal quality or yield. “Study End” indicates that the arrays were explanted not due to poor signal quality, but due to the end of a given experiment. “Hardware failure” indicates non-array components failing such that recording ability was compromised. The final four rows of Table 1 are reserved for the human participant arrays which were included in the study.

Name	Sex	Brain Area	Implant Date	Electrode Length (mm)	Size	Metallization	Number of Recordings	Reason for Termination
AtPmv	F	PMv	080123	1.0	96	Platinum	47	Signal Quality
AtM1b	F	M1	130305	1.0	128	Iridium Oxide	40	Signal Quality
AtM1o	F	M1o	080123	1.0	96	Platinum	13	Medical
BiM1	M	M1o	100824	1.0	96	Iridium Oxide	10	Signal Quality
BoPMda	M	PMd	030128	1.0	96	Platinum	317	Signal Quality
BoM1b	M	M1	080408	1.5	96	Iridium Oxide	68	Signal Quality
BoPMdb	M	PMd	080408	1.0	96	Platinum	68	Signal Quality
BoM1a	M	M1	030128	1.0	96	Platinum	6	Signal Quality
CoPMd	F	PMd	090413	1.0	96	Platinum	82	Medical
CoM1	F	M1	090413	1.5	96	Platinum	68	Medical
CoPMv	F	PMv	090413	1.5	96	Platinum	54	Medical
JaPMv	M	PMv	121002	1.5	96	Iridium Oxide	27	Signal Quality
JaPMd	M	PMd	121002	1.0	96	Iridium Oxide	17	Signal Quality
KiM1ips	F	M1ipsi	120604	1.0	96	Iridium Oxide	327	Study End
KiM1con	F	M1con	120604	1.0	96	Iridium Oxide	51	Study End
LeM1c	M	M1l	150720	1.0	128	Iridium Oxide	202	Signal Quality
LeM1b	M	M1m	150720	1.0	128	Iridium Oxide	63	Signal Quality
LeM1a	M	M1	110711	1.5	96	Iridium Oxide	38	Signal Quality
LePMd	M	PMd	110711	1.5	96	Platinum	32	Signal Quality
LePMv	M	PMv	110711	1.5	96	Platinum	25	Signal Quality
MkM1c	M	M1m	071128	1.0	96	Iridium Oxide	501	Hardware Failure
MkM1b	M	M1l	060906	1.0	96	Platinum	32	Hardware Failure
MkM1d	M	M1l	071128	1.0	96	Platinum	32	Hardware Failure
MkM1a	M	M1m	060906	1.0	96	Platinum	28	Hardware Failure
MkPMd	M	PMd	060906	1.0	96	Platinum	14	Hardware Failure
NkM1b	F	M1contra	140303	1.0	96	Iridium Oxide	42	Study End
NkPFC	F	PFCcontra	140303	1.0	96	Iridium Oxide	13	Study End
NkM1a	F	M1ipsi	140303	1.0	96	Iridium Oxide	12	Study End
NiPMv	F	PMv	041103	1.0	96	Platinum	91	Signal Quality
NiM1	F	M1	041103	1.0	96	Platinum	86	Signal Quality
NiPMd	F	PMd	041103	1.0	96	Platinum	21	Signal Quality
OrM1a	F	M1	070501	1.5	96	Platinum	52	Signal Quality
OrM1b	F	M1o	070501	1.0	96	Platinum	52	Signal Quality
OrPmv	F	PMv	070501	1.0	96	Platinum	18	Signal Quality
RjPMd	M	PMd	031027	1.0	96	Platinum	156	Signal Quality
RjM1a	M	M1	031027	1.0	64	Platinum	81	Signal Quality
RjM1b	M	M1	050804	1.0	96	Platinum	21	Hardware Failure
RjPMv	M	PMv	050804	1.0	96	Platinum	10	Hardware Failure
RoM1a	M	M1	040602	1.0	96	Platinum	211	Signal Quality
RoPMd	M	PMd	040602	1.0	96	Platinum	34	Signal Quality
RxM1a	F	M1	050328	1.0	96	Platinum	97	Signal Quality
RxPMva	F	PMv	050328	1.0	96	Platinum	80	Signal Quality
RxM1b	F	M1	060712	1.5	96	Platinum	55	Signal Quality
RxPMdb	F	PMd	060712	1.0	96	Platinum	42	Signal Quality
RxM1c	F	M1	090729	1.5	96	Iridium Oxide	33	Signal Quality
RxPMvb	F	PMv	060712	1.0	96	Platinum	27	Signal Quality
RxPMda	F	PMd	050328	1.0	96	Platinum	10	Signal Quality
RxS1a	F	S1	090729	1.5	96	Iridium Oxide	6	Signal Quality
VePMv	F	PMv	050809	1.0	32	Platinum	132	Medical
VeM1a	F	M1	050809	1.0	96	Platinum	110	Medical
VePMd	F	PMd	050809	1.0	96	Platinum	48	Medical
ZiM1a	F	M1contra	120730	1.0	96	Iridium Oxide	384	Study End
ZiM1c	F	M1ipsi	120730	1.0	96	Iridium Oxide	5	Study End
P1_A	F	M1	120210	1.5	96	Platinum	350	Medical
P1_P	F	M1	120210	1.5	96	Platinum	335	Medical
P2_A	M	M1	150504	1.5	88	Platinum	722	N/A
P2_P	M	M1	150504	1.5	88	Platinum	718	N/A

Data collection was conducted using Blackrock’s Cerebus data acquisition system. Neural data recorded as analog signals were amplified with a gain of 5000, bandpass filtered using the built-in hardware filter between 0.3 Hz and 7.5 kHz and digitized at 30 kHz. Spike waveforms were detected and saved into the “.nev” file format at 46 samples per waveform. Spike events were detected using a global root mean squared signal energy (RMS) threshold set by the experimenter; occasionally the experimenter manually adjusted individual electrodes’ thresholds.

Due to the broad date range of these recordings, training protocols and behavioral tasks varied widely across our subjects. Most subjects were trained to perform upper limb motor tasks, either involving two-dimensional planar reaching tasks using a two-link exoskeletal robot (Kinarm, BKIN Technologies, Ltd., Kingston, Ontario), unconstrained three-dimensional reaching, grasping, and reach-to-grasp tasks, or brain-computer interface (BCI) tasks. One animal also engaged in orofacial behavior.

2.2. Human Subjects

Two 96-channel arrays were implanted in participant P1, and two 88-channel arrays were implanted in participant P2. Arrays were implanted in the hand knob area of motor cortex (precentral gyrus). P2 also had two 32 channel arrays implanted in somatosensory cortex for the purposes of electrical stimulation; recording and stimulation performance for these arrays has been previously described (26). Participant P1 contributed to 350 recording sessions over 2.7 years. Participant P2 contributed to 721 recordings sessions over 5.3 years (8). Data collection with participant P2 is ongoing. Neural data were recorded at the beginning of each BCI testing session while the participants were at rest or speaking with experimenters.

2.3. Array Insertion and Recording Procedure

High-speed insertion (150-285 microseconds) of electrode arrays was performed with the pneumatic Blackrock Electrode Inserter System which consists of a wand with a hammer at the end that translates either 1mm or 1.5 mm depending on the electrode length of the arrays. The inserter pressure was always set to 20 PSI. In all cases, the inserter wand was firmly attached to a micromanipulator that was affixed to a stereotaxic frame or surgical table. The surface of the wand’s hammer was positioned on and parallel to the back end of the array before inserting. All connector pedestals used in this study consisted of 128-channel Cereport-style connectors that were attached to the skull with 8 bone screws per pedestal. Data acquisition was performed with cables attached to the Cereports without wireless transmission.

2.4. Neural Data Processing

Each “.nev” file was processed to evaluate the number of electrodes that displayed spike events over the entire recording. A channel’s spikes were included in an signal-to-noise ratio (SNR) calculation if at least fourteen spike events were detected throughout the entire recording. Each electrode’s SNR was calculated by measuring the average peak-to-trough amplitude of the electrode’s waveforms and dividing it by two times the average standard deviation of the voltage of the waveforms. The average standard deviation was computed by taking the mean of standard deviation values over all 46 samples of the waveform. If the electrode’s SNR exceeded 1.5, the electrode was designated as “good” and counted towards that recording’s array yield. This threshold guarantees that a channel’s signal is 50% higher than the noise of the channel. This threshold may be seen as too high by some, but we preferred to have confidence in a “good” channel, rather than include channels of nebulous quality. It should be noted that neural signals with an SNR of less than 1.5 can still carry information that may be useful for BCI or scientific applications. Moreover, each recording was manually inspected and erroneous “good” channels, often having inaccurate SNR measurements due to artifacts, were removed from the “good” electrode count. No spike sorting was performed so our SNR values were conservative and were lower than would be expected from individually sorted units. Also, due to the lack of spike sorting, we could not address the degree to which array recordings change in their capability to detect separable units within a given electrode’s signal. All statistics and figures displaying SNR were calculated across all electrodes of the array in which spikes were detected

After manual inspection, our search resulted in data from 19 subjects (17 NHPs and 2 humans) and 55 Utah microelectrode arrays, totaling 6132 recording sessions. Due to different array sizes (8x8, 10x10), array yield of arrays was not presented as the absolute number of electrodes but rather as the percentage of total recordable electrodes, thus allowing for comparisons across different electrode configurations. SNR was also calculated as an average across all electrodes within a given recording session.

It is important to note that the number of arrays we could examine decreased with time due to loss of signals, hardware failure, infections surrounding the percutaneous connector, and experiment termination (see Table 1). Some analyses took loss of arrays into account to represent the number of arrays that survived past a certain point in time (see Figure 3). Other analyses prioritized calculating the reliability and yield of functioning arrays, considering only arrays from which viable recordings were possible (see Figure 5).

Figure 3.

Proportion of arrays exceeding a certain yield at month-to-month intervals post-implantation. Each line delineates a different percentage yield threshold. If an array is not recorded from in a given time window, it is not counted in the “proportion” estimate. After their last available recording date, arrays are counted in the proportion measurement, to accurately depict the degradation of arrays on average. However, arrays which ceased recording due to medical, hardware failure, or study end reasons (see Table 1) were not included in the proportion measurement after their last available recording date.

Figure 5.

Reliability of viable chronic recordings over time. Shaded regions denote standard error of the mean. a. Average yield (blue) and SNR (red) over arrays with viable recordings (i.e. array recordings that were not terminated) versus number of days post-implantation for NHP implants. Full time range not shown (maximum lifetime is in excess of 3000 days). b. Close-up of average yield and SNR over the first 40 days post-implantation. A significant difference was observed between the first week and fourth week of recordings (p = 0.0122, 2-sample t-test, 2-tailed). c,d. equivalent statistics for human implants. Further details can be found in Downey et al., 2018 (8). The purple vertical line indicates the date at which the R.M.S. spike threshold was set to a different value during data collection (see methods for details). The change in the figure is due to a larger number of recordings becoming possible to average across, reducing variance in the estimate of yield and SNR. The green line indicates the date data collection with P1 ended.

3. Results

3.1. Lifetime of chronic recordings

A total of over 6000 recording sessions over 55 arrays in 17 animals and two human subjects were analyzed by computing SNR values of unsorted waveforms for each electrode on the array (Figure 1). A subset of electrodes was then selected as “good” if their SNR exceeded 1.5 and passed our manual examination from which overall yield was assessed over time. Over the 55 array implants, there was a large variance in lifetime of recordings from 44 days to over 3000 days (Figure 2). Recordings from UEAs were terminated for several reasons including loss of signals, electrode assembly hardware failure (i.e. the titanium connector pedestal detached from the skull), infections surrounding the connector pedestal, or completion of an experiment (Table 1).

Figure 1.

Example waveforms and signal-to-noise ratio (SNR) from a UEA recording session (Subject Mk, array MkM1c). SNR values are presented in the top left corner of each panel. Red SNR values indicate electrodes which exceeded the requisite SNR threshold of 1.5 to be considered as “good” electrodes included in subsequent analyses of array yield. The red line in each panel indicates the threshold across which a waveform must cross to be considered an action potential. Empty panels indicate electrodes in which less than fourteen spikes were detected within the given recording session, and therefore were not included in analysis.

Figure 2.

Summary heat map of signal-to-noise ratio (circle color) and array yield (circle size) over time for all array implants analyzed. Each colored circle denotes a single recording session. The color of the circle denotes the average SNR of the array for that given recording session. The size of the circle denotes the percentage of electrodes in that array that demonstrated a signal-to-noise ratio above a threshold of 1.5.

To gauge the likelihood that a UEA would provide high quality signals over time, we examined the proportion of arrays that exceeded a fixed percentage yield at month-to-month intervals (Figure 3). We observed a slight increase in the proportion of arrays exceeding a given yield in the first month of recording post-implantation followed by a steady decrease over 36 months. Nearly seventy percent of arrays displayed at least a 40% yield in the first three months of recordings and fifty percent of arrays exceeded the same yield threshold after one year of recording. A long tail of reliability was observed with nearly ten percent of arrays exceeding the same yield threshold out to at least 36 months post-implantation. Based on a linear regression fit to the 40% threshold data in Figure 3, arrays typically experience a ~2% drop in yield every 30 days (slope = −0.00058 per day).

3.2. Extended long-term recordings

To examine the day-to-day variability in signal quality among very long-term recordings, we focused on a subset of arrays that continued to record good signals beyond 950 days (Figure 4a). Sixteen out of 55 arrays displayed longevity well over 800 days of which three lasted well into five years of service (two of the three arrays were implanted in human subject P2). All linear regressions are statistically significant (p < .05), save for “BoPMdb”, the only array whose regression displayed an upward trend. One array (Mk) exhibited extended recording capabilities to nearly nine years, only needing to be explanted due to an infection near the connector. This subset of arrays generally displayed a gradual decrease in signal quality over time with variations from recording to recording presumably due to several factors such as micromovements of the array that may have occurred in the home cage such as sudden head accelerations (excluding our human participants) (27), headstage malfunctions, and sources of electrical noise of unknown origin that could not be eliminated. The two arrays implanted in a human subject (P2) are still implanted as of October 2021 with good recording quality.

Figure 4. Extended long-term performance for a subset of arrays.

a. Yield for long-term array implants as a function of days post-implantation. Lines indicate best fit linear regressions to the data for each array. All linear regressions are statistically significant (p < .05), save for “BoPMdb”, the only array whose regression displayed an upward trend. The recording quality metrics exhibit a discontinuity at day 565 for subject P1, due to a change in the spike threshold from −5.25 to −4.5 RMS, indicated by the vertical dashed lines. b. Example spike waveforms from one array implant (MkM1c) at regular intervals over nearly 9 years. SNR values are presented at the top left corner of each panel (red font denotes electrodes that exceeded an SNR threshold of 1.5). c. SNR of Monkey Mk’s recordings, over the lifetime of the implant.

3.3. Reliability among arrays possessing viable recordings

We next examined the temporal evolution of average yield and SNR among arrays from which recordings continued to be available and did not include arrays from which recordings were terminated (Figure 5). Therefore, the number of arrays contributing to the average decreased with time, and standard error values increased. However, with this measure, we were able to confirm that arrays implanted in NHPs typically maintained their electrode yield and SNR over most of their lifetime (Figure 5a). In contrast to long-term yield, short-term yield displayed a rapid increase over the first 40 days post-implantation in NHP implants (Figure 5b). A significant difference was observed between the first week and fourth week of recordings (p = 0.0122, 2-sample t-test, 2-tailed).

Using the same methods used for NHP analyses, we also examined the signal quality over time for UEAs implanted in human participants as part of a brain-machine interface study (28). We found that electrode yield increased slightly and SNR steadily increased in the shortterm, saturating to some overall maximum value after approximately four weeks (Figure 5c,d). Due to the consistent and long-term nature of the human BCI study, signal quality metrics were available for human arrays for almost 1000 and 2000 days for subject P1 and P2, respectively. The electrode yield from these two human participants was generally higher and more consistent over time as compared to our NHP results though a slow decline in SNR was noted over time as in the NHP arrays.

3.4. Performance effects of electrode tip metallization and length

Platinum metallization of the electrode tips was adopted in the early fabrication of UEAs but was not well suited for microstimulation of the cortex. The option to use iridium oxide metallization was made available in 2009 which allowed for chronic stimulation due to its improved charge injection capacity, as well as recording (iridium oxide arrays were used for acute stimulation in our lab but were not used for chronic stimulation) (29). We directly compared recording yield and SNR between these two metallization options and found that average array yield was significantly higher for iridium oxide as compared to platinum tips during chronic recordings in the intermediate time range although there was no difference in the short and long term (Figure 6a). In contrast, there was no significant difference in mean SNR between the two metallization materials at any time (Figure 6b).

Figure 6. Effects of electrode tip metallization and electrode length on array performance in NHP implants.

a, c. The average yield for each six-month date window was calculated for each array; an average was then taken across all arrays with the same metallization or electrode length (short-1.0 mm, long-1.5 mm), respectively. Shaded regions for all plots indicate the standard error of the mean, across arrays. b, d. Average SNR based on metallization and electrode length, respectively. Statistical tests were corrected for multiple comparisons using Bonferroni correction. Stars indicate statistically significant differences between groups, Bonferroni corrected. Averages were calculated over recordings and arrays for which data was available for each time window. Only NHP data were used for this figure.

We also compared recording performance between short (1.0 mm) and long (1.5 mm) electrodes. Given our research interests in motor and premotor cortices, our initial bias was to use long electrodes to target as close to layer 5 as possible in the thicker motor cortex. However, we were also concerned about the known observation that UEAs sink over the long term by compressing superficial layers, and so we also used short electrodes in some cases in order to prevent the tips from sinking into white matter particularly in the thinner premotor cortex (unpublished histology). We found no difference in average yield or SNR between short and long electrodes (Figure 6c,d).

During our analysis, questions arose about the potential impact of manufacturer changes or surgeon experience over the entire range of dates during which implants were placed in subjects. To address these concerns, we analyzed the reliability of all NHP implants, based on their implantation date. Despite having been implanted by different surgeons, and having undergone largely minor changes in manufacturing process, we observed no significant relationship between array longevity and date of implant (r = 0.11, p = 0.43).

3.5. Maximum performance over array lifetime

We next addressed the maximum possible performance for these arrays over their lifetimes. We calculated the distribution of maximum yield over arrays and the time post-implantation when the maximum yield occurred. Despite the large variance in maximum yield across arrays, we found that 16 (out of 55) arrays exhibited a maximum yield exceeding 90% (Figure 7a). The maximum yield typically occurred within the first 150 days of a given array’s recording life (Figure 7b). The results in Figure 7b reject a chi-squared test of uniformity at a p-value of (p = .002)

Figure 7. Maximum possible yield.

a. Distribution of the maximum recorded electrode yield over all arrays, from 0% to 100% of possible electrodes. b. Distribution of dates when the maximum number of recorded channels occurred. Most maximum recording days occurred within the first 150 days post-implantation. This data rejects a chi-squared test of uniformity at a p-value of (p = .002)

4. Discussion

UEAs are now a standard recording device for chronic electrophysiology in animals and is the only chronic intracortical recording technology that is FDA-cleared for investigational purposes in humans. Assessing the longevity and reliability of these devices is important for long-term experiments in NHPs and is of particular significance for chronic use in human BCIs. Most previous publications have examined the signal quality of Utah arrays using data from a limited number of subjects or explicitly focused on physical failure modes of the arrays (however, see (4) for one exception); other papers have examined the longevity and/or reliability of electrophysiological recording techniques, while not specifically focusing on the Utah Array (5-13,18). We found that nearly sixty percent of array implants exceeded 40% electrode yield after 6 months post implantation and nearly fifty percent of array implants exceeded 40% yield at one year. We also observed a subset of arrays (n=12) that exhibited moderate yield and SNR beyond 900 days post-implantation and one implant lasted nearly 9 years. We also characterized an important aspect of this recording technology, namely, a rapid increase in electrode yield within the first 40 days post-implantation in the non-human primate implants, most likely because of acute inflammation and subsequent recovery after surgery (8). Early recovery of signal quality post-implantation is likely due to alleviation of acute inflammation and potential small hematomas as swelling and bleeding reduce. The acute immune response to a foreign body is associated with activated microglia and astrocytes which begins at the moment of insertion and lasts approximately 6-8 weeks post-implantation followed by a chronic response associated with glial encapsulation of the foreign body (30). The improved yield in the first 40 days post-implantation may be related to the termination of the acute immune response whereas the long-term decrease in yield may be in part the result of gliotic encapsulation.

We also observed that the yield of human implants was higher than NHP implants. There are many potential reasons for this difference, despite the hardware itself being identical: Utah arrays approved for human implant undergo additional quality control checks prior to implant; surgical procedures are executed by neurosurgeons as opposed to an investigator with a scientific or engineering background; most importantly, human study participants and their caregivers take care of the connector and surrounding skin whereas monkeys can damage their own connector or the skin around the percutaneous connector pedestal leading to infection or failure of the implant. The results directly reported here are limited to two human participants, but successful long term human implants have been commonly reported in the literature (18). It should be noted that, as with the NHPs, the human patients had multiple array implants per subject even in the same cortical area so it seems unlikely that the difference in yield is due to the number of array implants. However, in some NHP cases, arrays were placed very close to one another such that the edge of one array was nearly touching another which was not true of the human implants. That may have played some role given that our best NHP yield (see Figure 4a) were among implants that were more than 2 mm from each other.

Our findings also suggest that UEAs with iridium oxide metallized tips result in higher recording yield as compared to platinum metallization (in non-human primates). Iridium oxide also has the added benefit that chronic electrical stimulation is possible. However, these results should be interpreted with caution given the fact that our platinum implants were done earlier than the iridium oxide implants over the 17 years of implants examined in this study. The improved yield of the iridium oxide implants may have been due to improvements in surgical techniques as we learned better implantation methods. There have been a few improvements in the UEA manufacturing process over the 17-year period most of which were relatively minor with the exception of transitioning from array to wafer scale manufacturing in 2009-10 that may explain the improved yield of the iridium oxide implants (31). Nevertheless, when considering all our implants regardless of metallization, we found no correlation between array longevity and date of implant.

4.1. Species-specific differences in signal quality

Data from our human subjects tended to display higher array yields and SNR than the NHP data. However, the number of human datasets was vastly smaller (only 4 arrays and 2 subjects) than the datasets in the NHPs, and so it is difficult to draw any strong conclusions regarding differences in signal quality between humans and NHPs. One possibility may be that the surgical ability with which microelectrode arrays were implanted may have had a significant impact on the eventual efficacy and quality of long-term recordings in each subject. With each human implant, trained neurosurgeons and a team of surgical staff supported the careful placement of UEAs; with each NHP implant, the process was conducted by PIs, postdocs, and graduate students. Although veterinarian and animal clinic personnel are sometimes available for NHP implantation procedures in some research institutions, the availability of these resources is far from guaranteed. An added factor that may have contributed to the human implants’ higher signal quality and lifespan was more diligent care. NHPs have a higher likelihood of implant damage due to their activity and behavior, while human participants are generally more careful with their implants. Moreover, the wound margins surrounding the percutaneous connectors tend to become dirtier and more prone to infections in NHPs versus humans.

Although our ability to comment on the consistency of human implant reliability is limited due to our small sample size, previous work indicates that the duration of human implants can also vary (18). Across 18 chronic human implants (including the two participants included here), at least 9 remained implanted for at least 1 year and 7 of these for more than 2 years. Importantly, the authors noted that some of these experiments remain ongoing and that the total reported duration does not indicate that the array failed.

4.2. Study Limitations

One of our unique advantages in investigating questions concerning multi-electrode array signal quality is the massive amount of data available to us over the past 17 years. However, our intentions for implanting these arrays were to address scientific questions regarding cortical function and not to specifically address array longevity and reliability. Therefore, one limitation of this study is that we did not sample recordings evenly in time, and we sometimes terminated recording when we were finished with experiments instead of due to array failure. An additional limitation is the difficulty of cross-species comparisons of results. This challenge is present in any cross-species study, born out here by the substantial differences in anatomy (cortical thickness and relative size of the gyrus) and anatomical landmarks, which may have had an additional effect on the differences in longevity and reliability demonstrated between human and non-human primate results presented here.

Another limitation due to the size and scale of the datasets with which we analyzed was the lack of spike sorting to distinguish individual single units within a given electrode. As a result, all our metrics and analyses are on the scale of individual electrodes and arrays, instead of individual single units. Our decision to forgo spike sorting, therefore, underestimates unit yield and SNR because two high SNR single units of different amplitudes on a single electrode, for example, would result in a lower average SNR. It should be reiterated that meaningful information about the brain may be derived from channels which exhibit poorly isolated units.

The Utah Array continues to be a reliable and valuable tool for systems neuroscience and human brain-computer interfaces. We have demonstrated that most non-human primate implants will reliably last over a year, with the potential for arrays to last for the better part of a decade; human implants also display impressive longevity, lasting multiple years. While the Utah array may be acceptable for current animal research, more research and development must be done to create chronic implants that will reliably last into years and decades of use, for human applications and more advanced animal studies.