896
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 51, NO. 6, JUNE 2004
Chronic Neural Recording Using Silicon-Substrate
Microelectrode Arrays Implanted in Cerebral Cortex
Rio J. Vetter, Justin C. Williams, Member, IEEE, Jamille F. Hetke, Elizabeth A. Nunamaker, Student Member, IEEE,
and Daryl R. Kipke*, Member, IEEE
Abstract—An important aspect of the development of cortical
prostheses is the enhancement of suitable implantable microelectrode arrays for chronic neural recording. The objective
of this study was to investigate the recording performance of
silicon-substrate micromachined probes in terms of reliability
and signal quality. These probes were found to consistently and
reliably provide high-quality spike recordings over extended
periods of time lasting up to 127 days. In a consecutive series of
ten rodents involving 14 implanted probes, 13/14 (93%) of the
devices remained functional throughout the assessment period.
More than 90% of the probe sites consistently recorded spike
activity with signal-to-noise ratios sufficient for amplitudes and
waveform-based discrimination. Histological analysis of the tissue
surrounding the probes generally indicated the development of
a stable interface sufficient for sustained electrical contact. The
results of this study demonstrate that these planar silicon probes
are suitable for long-term recording in the cerebral cortex and
provide an effective platform technology foundation for microscale
intracortical neural interfaces for use in humans.
Index Terms—Brain-computer interface (BCI), brain-machine
interface (BMI), microelectrode, neuroprosthesis.
I. INTRODUCTION
C
ORTICAL recording interfaces show promise for controlling brain-machine interface (BMI) systems. At one
extreme, locked-in patients with no voluntary muscle control
have been able to control a brain-computer interface (BCI)
system by volitional modulation of a small number of spike
trains recorded using an invasive neurotrophic microelectrode
permanently implanted in motor cortex [1]. At the other
extreme, intact subjects have been trained to control a simple
computer interface through noninvasively recorded EEG signals [2]. Additional studies in humans [3], nonhuman primates
[4]–[7], and rodents [8]–[10] have further investigated aspects
of cortical control. At this point, the effects of signal sources,
Manuscript received July 17, 2003; revised February 2, 2004. This work was
supported in part by the Defense Advanced Research Projects Agency (DARPA)
under Grants N66001-02-8059 and MDA972-00-1-0027 and by NIH/NIBIB
under Grant P41 EB-00230-10. Asterisk indicates corresponding author.
R. J. Vetter and E. A. Nunamaker are with the Department of Biomedical
Engineering, University of Michigan, Ann Arbor, MI 48109-0506 USA (e-mail:
[email protected]; [email protected]).
J. C. Williams is with the Department of Biomedical Engineering, University of Wisconsin, Madison, WI 53706-1609 USA (e-mail: jwilliams@
engr.wisc.edu).
J. F. Hetke is with the Department of Electrical Engineering and Computer
Science, University of Michigan, Ann Arbor, MI 48109-2122 USA (e-mail:
[email protected]).
*D. R. Kipke is with the Department of Biomedical Engineering, University of Michigan, 2200 Bonisteel Blvd., Room 1107, Gerstacker Building, Ann
Arbor, MI 48109-0506 USA (e-mail: [email protected]).
Digital Object Identifier 10.1109/TBME.2004.826680
plasticity, decoding algorithms, training protocols, and user
interfaces remain important open questions.
Implantable microelectrodes for chronic neural recording are
receiving renewed attention by a number of groups because of
the interest in using ensemble spike activity and multichannel
local field potentials as control signals in cortical neuroprostheses. Numerous types of neural probes have been developed,
including bundles of microwires [11]–[13], microwires embedded in neurotrophic assemblies [1], polymer substrate
probes [14], [15], and several types of silicon-substrate probes
[16]–[20].
This paper reports on the use of a particular type of silicon micromachined probe, the so-called “Michigan” probe, for chronic
neural recording in the cerebral cortex. The various components of this probe system have been developed over the past
decade [16], [21] (see recent reviews in [20] and [22]). It has
several favorable attributes including batch fabrication, high reproducibility of geometrical and electrical characteristics, easy
customization of recording site placement and substrate shape,
small size, and the ability to integrate it with a silicon ribbon
cable as well as to incorporate on-chip electronics for signal
conditioning [23]. The extensibility of the platform technology
enables custom designs for diverse applications. For example,
planar probes have been assembled into multiplane arrays for
precise three-dimensional placement of recording sites in the
brain [24], and the probes have been combined with a polymer
ribbon cable to form a hybrid assembly to provide additional
mechanical flexibility [25]. Probe designs include microchannels along the shanks for microscale and controlled fluid delivery through the blood-brain barrier [26]. These silicon probes
are currently used by many investigators and are now well validated for recording both spike activity and field potentials in diverse brain structures over acute and semichronic time durations
up to several weeks [27]–[29]. Csicsvari et al. recently demonstrated high-density unit and field activity recordings (96 sites)
in the hippocampus with on-chip active circuitry [23].
The objective of this study was to investigate the use of
Michigan silicon probes for chronic neural recording in cerebral
cortex. This work extends the results of a previous study [29]
by more closely examining the performance of the probes over
a four-month postoperative assessment period. The probes were
found to be reliable for recording multichannel spike activity
and local field potentials over this time period. The results of
this study demonstrate the feasibility of developing this probe
technology into an implantable microscale chronic intracortical
neural interface for medical applications, including cortical
prostheses.
0018-9294/04$20.00 © 2004 IEEE
VETTER et al.: CHRONIC NEURAL RECORDING USING SILICON-SUBSTRATE MICROELECTRODE ARRAYS
897
Fig. 2. Surgical implantation illustrations. (A) Dura is reflected from the
center of the craniotomy outward and carefully placed over the bone. (B) Each
probe is hand inserted with a pair of Teflon coated microforceps. (C) Implanted
Michigan probe assembly.
Fig. 1. Michigan 16-channel probe. (A) General schematic of the four-shank,
16-channel probe used in this experiment. Some probes contained a 20 by
60-m “well” between the first recording site and the tip. (B) Four shank,
16-channel probe with 20-m diameter recording sites. (C) Implanted probe
in auditory cortex. Thin layer of ALGEL® covers the implanted probe and the
surface of the brain. (D) Complete probe assembly. Implantable shanks of the
probe are integrated with a 1-cm long 5-m-thick ribbon cable, which is then
integrated with a silicon substrate containing the bond pads (approximately
1.5 mm by 0.8 mm). This is bonded to the Omnetics® connector, which
interfaces to the instrumentation system. Silicone rubber is applied to the
probe/cable junction to provide a grasping region.
II. MATERIALS AND METHODS
A. Silicon Probe Fabrication
The Michigan silicon probe process begins with selective
diffusion of boron into a silicon wafer to define the shape
and thickness of the device [21]. Lower dielectric layers of
silicon dioxide and silicon nitride are next deposited using
chemical vapor deposition. These layers provide insulation
from the conductive boron-doped substrate. The conductive
interconnect material (polysilicon for most applications) is
deposited, patterned, and then insulated with upper dielectrics
as just described. In order to establish recording sites, selective
access through the upper dielectrics to the polysilicon layer is
achieved using a combination of wet and dry etching. Metal is
next deposited and patterned to form the electrode sites (typically iridium) and bond pads (gold) using a lift-off technique.
Dielectrics in the field area are removed using a dry etch. After
the wafer is thinned, it is placed in ethylenediamine-pyrocatechol, which selectively etches away the undoped silicon. This
process uses eight photolithographic masks and enables the
design of a wide variety of planar probes, with specific site
configuration and physical layout customized to the targeted
neural structure.
The probes used in this study consisted of four penetrating
shanks, each with four uniformly spaced recordings sites of the
same size [Fig. 1(A) and (B)]. The probes were provided by
the Center for Neural Communication Technology, University
of Michigan, a National Institute of Health National Resource
Center.1 The two probe designs used in this study were designated “flag 1” and “4 4 3 mm100chron.” Both probes are
15 m thick. The “flag 1” probe has 60- m wide shanks that
are 1935 m long with 150- m separation between shanks. The
intersite spacing along a shank is 100 m and site areas are
312 m . The “4 4 3 mm100chron” probe has 55- m wide
shanks that are 3000 m long with 125- m shank separation.
The intersite spacing along a shank is 100 m and site areas
are 177 m . Several of the probes in this study included a 20
by 60 m oval hole through the silicon substrate near the tip
of the shank. Fig. 1(C) shows the external aspects of a probe
that was implanted in the cerebral cortex of a rat. The complete
probe assembly included a flexible 1-cm long, 5- m thick integrated silicon ribbon cable [21] that was terminated on a flexible
printed circuit board mounted to a commercially available microconnector2 [Fig. 1(D)].
B. Surgical Implant Procedure
Probes were implanted in Sprague Dawley rats in either an
auditory or motor area of the cerebral cortex. The animals were
administered general anesthesia (mixture of 50 mg/ml ketamine,
5 mg/ml xylazine, and 1 mg/ml acepromazine), using intraperitoneal injection with an initial dosage of 0.1 ml/100 g body
weight and maintained in an areflexive state throughout the surgical procedure using regular anesthesia supplements at 20%
of the initial dosage. Each animal was attached to a standard
stereotaxic frame and three stainless steel bone screws were inserted into the skull. A craniotomy (approximately 3 by 2 mm)
was made over the target cortical area and the connector assembly was attached to the skull and bone screws [Fig. 2(C)]
1www.engin.umich.edu/facility/cnct/.
2Nano
connector, Omnetics Inc., Minneapolis, MN.
898
using dental acrylic.3 A cruciate incision was made in the dura to
visualize the cortical surface and the resulting dural flaps were
folded back over the edges of the bone [Fig. 2(A)]. The cortical
surface was irrigated using 0.9% sterile saline. A stainless steel
ground wire attached to the connector was connected to a bone
screw to provide a ground point and temporary mechanical support until the connector was permanently cemented to the bone
using dental acrylic.
The region at which the 15- m-thick silicon probe thins to
the 5- m-thick silicon cable was coated with a small bead of
silicone elastomer to facilitate handling with Teflon-coated forceps. The probes were manually advanced through the pia into
the cortex [Fig. 2(B)] until the recording sites spanned a region
approximately 0.5 to 1.5 mm below the cortical surface. The incised dura was folded away from the probe. The bond pad and
several millimeters of the cable were encased in a polymer dural
sealant, ALGEL4 [30]. Small pieces of Gel Foam5 saturated in
saline were then used to cover the remainder of the cable. A head
cap was fabricated using dental acrylic to further seal the craniotomy and connector assembly. All procedures complied with
the United States Department of Agriculture guidelines for the
care and use of laboratory animals and were approved by the
University of Michigan Committee on Use of Animals.
C. Data Collection/Analysis
Neural recordings were obtained using a commercial 12-bit
multichannel recording system.6 Recordings were made daily
in each subject for the first four weeks and then at least once
weekly thereafter. The extracellular signals were sampled
at 40 kHz and bandpass filtered from 450–5000 Hz. The
recording system provided differential amplification, sampling,
and channel-specific processing of the signals in order to store
the waveforms and time-of-occurrences of the single-unit
and multiunit activity from each channel. Action potentials
were discriminated using an interactive procedure based on
amplitude thresholds and waveform templates. Single channels
were selectively monitored on an oscilloscope and audio
monitor. The animals were awake or lightly sedated during
regular postoperative recording sessions lasting up to 2 h. The
animals were typically placed in a shielded recording booth,
but this was not necessary for achieving low-noise recordings.
The recorded data sets included sorted spike waveforms and
continuous segments of neural recordings up to 30 s in duration.
The neural recordings were analyzed offline using custom
software. Objective measures of signal quality were computed
from 20–30-s segments of continuous recordings using algorithms to estimate ratio of the background noise level to the
signal amplitude [signal-to-noise ratio (SNR)]. In this case, the
background noise was considered as the sum of sensor noise,
instrumentation noise, and small amplitude signals from diffuse
neural sources. The discrimination of spikes (signal) from background noise was computed by establishing amplitude thresholds based on the probability distribution of all of the sample
3Co-Oral-Ite
Dental Mfg. Co.
trademark, Neural Intervention Technologies, Ann Arbor, MI.
5Registered trademark, Pharmacia Co.
6Plexon, Inc., Dallas, TX.
4Registered
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 51, NO. 6, JUNE 2004
values in the continuous record. An amplitude threshold window
was arbitrarily set at three standard deviations above and below
the mean of the sample distribution. Peaks in the recorded potential that exceeded this discrimination window (positive or negative) were defined as “signal” peaks and those that did not exceed the threshold were defined as “noise” peaks. The statistics
of the peak-to-peak signal features and noise features were then
separately computed. The standard error was used to express
sample variability
where
is the total number of data values,
is the data value
of series and the th point, and is the number of series for
point .
The SNR algorithm provided a reasonable method to assess
the quality of spike recordings in terms of spike amplitude relative to the underlying noise and was similar to published work
in this area [11], [17]. The calculated SNR provided an objective measure of signal quality that was consistent with subjective
assessments made by experienced researchers. The SNR as defined can potentially be biased by high rates of spiking from several moderate amplitude units, which would widen the amplitude distribution and, thus, increase the discrimination window.
However, in practice, this was not found to be an issue.
D. Impedance Measurements
The impedance of each recording site was measured using
an HP4284 LCR meter (Hewlett Packard) in constant voltage
mode, with the probe referenced to the animal ground. A constant voltage of 5-mV amplitude resulted in a current of less than
50 nA at 1 kHz. Impedance measurements were collected each
day for the first four weeks and thereafter once a week.
E. Histology
Five subjects with functioning probes were harvested for
histological analysis. Procedural complications precluded the
analysis of two of the subjects. Each animal was perfused
with 4% paraformaldehyde following a standard protocol [31].
Immediately following perfusion, the skull was placed in 4%
paraformaldehyde for an additional 48 h and then placed in
phosphate buffered saline, both at 4 C, until final analysis of
the tissue could be performed. The acrylic head cap was dissolved away using vapors from a 1, 2-Dichloroethane solution7
to isolate the probe assembly from the head cap. The skull
was then carefully removed from around the brain allowing
the probe assemblies to remain implanted for subsequent
histological evaluation.
Two types of histological evaluations were performed. The
first method created coronal cross sections of the tissue after
the probes were explanted. The implanted hemisphere was
sectioned using a vibratome into 20- m coronal sections and
stained with hematoxylin and eosin (H&E). The 50- m coronal
sections were stained with a primary against glial fibrillary
acidic protein
(Advanced ImmunoChemical, Inc.)
7Fisher
Scientific, Inc.
VETTER et al.: CHRONIC NEURAL RECORDING USING SILICON-SUBSTRATE MICROELECTRODE ARRAYS
899
TABLE I
SUMMARY OF IMPLANTED PROBES AND CORRESPONDING ANIMALS
III. RESULTS
This study included results from 14 silicon probes implanted
into a consecutive series of ten rats (Table I). Seven of the probes
were implanted in the auditory cortex and seven were implanted
in the motor cortex. Neural recordings and probe impedance
measurements were obtained from each subject daily for the
first four postoperative weeks and at least once (typically five
times) weekly thereafter. The longest implant in this series was
127 days. With one exception, each case was terminated due to
factors other than probe failure (e.g., no discriminable spikes
on any probe), including an adverse reaction to anesthesia (one
case) and failure of the acrylic head cap (four cases). Five subjects were terminated with functioning probes in order to obtain
endpoint histology.
Fig. 3. Average percent of active electrodes on all implanted arrays. Time bars
at the bottom of the figure represent each animal’s contribution to mean results.
and a secondary antibody labeled with Alexa 578 (Molecular
Probes, Inc.). Implanted and nonimplanted cortical sections
from each animal were visually analyzed with a MZFLIII
microscope8 or a FITC/Rhodamine filtered fluorescent inverted
microscope.9
The second method allowed the probe to remain in the brain
in order to image the implanted probe surface and surrounding
tissue [32]. This technique required 150- m sections and
confocal imaging. The region of interest was blocked and
sectioned parallel to the probe shank with a vibratome. The
section was then stained with a primary antibody against glial
fibrillary acidic protein
, a secondary antibody
labeled with Alexa 578, and mounted in 95% buffered glycerol.
The
was imaged using a 603-nm filter set and the
probe was imaged using backscattered laser light through a
confocal microscope.10
8Leica,
Inc.
Inc.
10Leica, Inc.
9Leica,
A. Spike Recording Yield of Silicon
The silicon probes were effective for recording discriminable
single and multiunit spike activity. The spike amplitudes ranged
from 50–800 V peak-to-peak with background noise typically
below 20 V peak-to-peak. A brief recording segment across all
of the sites in one subject illustrates the typical range and extent
of signals (Fig. 4). In this case, multiunit and single-unit spike
activity was recorded on all 16 of the sites.
Over all subjects and the entire assessment period, the number
of active sites, or recording yield, was between 88% and 94%
(average of 92%) (Fig. 3). The yield was also consistent from
day to day, as indicated by just a 3.5% standard error (variation).
These results were consistent with those reported in an earlier
study that looked at probe performance in less detail, but over
longer periods of time [29].
B. Signal-to-Noise Ratio
The SNR and noise level were calculated to further quantify
the recording performance of the probes. Fig. 5(A) and (C)
shows 3-s segments selected from 10-s to 30-s continuous
recordings from two different recording sites. Fig. 5(B) and (D)
shows several spike waveforms that were discriminated from
900
Fig. 4.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 51, NO. 6, JUNE 2004
A 2-s snapshot of data collected simultaneously on 16 channels implanted in motor cortex on day 113 from animal BMI-4.
Fig. 5. Two simultaneous recordings from two different channels implanted in
animal BMI-6. (A) and (C) Show 3-s snapshots of “all inclusive” data consisting
of all the signals of interest and the intrinsic noise where discrimination of
single units has not yet been performed. (B) and (D) represents a 1-ms snapshot
of the sorted waveforms from the corresponding sets of data, (A) and (C). In
(B), five waveform traces are shown representing one single neuron that was
discriminated from the noise in (A). In (D), two sets of five waveform traces
are shown representing two different neurons that were discriminated from the
noise in (C). These figures represent the spectrum of SNRs found in this series
of animals. (A) SNRs at the lower end of the spectrum and (B) SNRs at the
upper end of the spectrum. SNRs shown were calculated from a 10-s duration
of data.
these data records. The recording in Fig. 5(A) had discriminable
multiunit spike clusters, but was among the worst in signal
quality with a SNR of 4.5 and an average noise level of 9.7 V.
The recording in Fig. 5(C) had well-isolated single units and
was among the best in signal quality with SNR of 16.6 and a
noise level of 17.7 V.
Fig. 6 shows the trend in signal quality, as indicated by the
SNR, as a function of postimplant days. The data set includes
224 separate recording sites and 19 assessment times. The mean
SNR (solid trace) varied between the relatively high levels of 6
and 8.6, which reflects a signal quality in which spike activity
is discriminable using standard amplitude thresholds and waveform templates.
“Within probe” SNR variability (dotted lines) was significantly less than “across probe” SNR variability (error bars)
).
during the assessment period (two-sample t-Test,
The former metric was defined as the mean standard error
calculated from the set of 16 signals recorded on each probe
and then averaged over all probes at each of the weekly time
points. The latter metric was defined in terms of the mean
standard error calculated over all recording sites and all probes.
By definition, the calculated SNR varies directly with spike
amplitudes and varies inversely on noise level. The measured
average noise amplitude in the immediate postoperative time
period was approximately 6 V and increased steadily to approximately 11–13 V by postoperative week 5 (Fig. 7). The
overall average noise amplitude for the probes was 11 0.79 V
(within probes) and 1.18 V (across probes). The average
signal amplitudes were found to remain between 50 and 100 V
VETTER et al.: CHRONIC NEURAL RECORDING USING SILICON-SUBSTRATE MICROELECTRODE ARRAYS
Fig. 6. Mean SNR across all implanted probes over time. Time bars indicate
the contribution from each animal. Dotted lines represent the average standard
error within each array and the error bars represent the standard error between
each array.
901
Fig. 8. Mean impedance values across all implanted probes over time. Data are
grouped by the size of the recording sites. Time bars indicate the contribution
from each animal. Dotted lines represent the average standard error within each
array and the error bars represent the standard error between each array.
Fig. 7. Mean baseline noise level and signal amplitude across all implanted
probes over time. Time bars indicate the contribution from each animal. Dotted
lines represent the average standard error within each array and the error bars
represent the standard error between each array.
during the assessment period, without any underlying trend.
Thus, the overall slight decrease in SNR as a function of postoperative time (Fig. 6) reflects an increase in average noise level
with relatively steady signal levels. There were no significant
differences in SNR as a function of the two recording site sizes
).
used (two-sample t-Test,
C. Probe Impedance
Probe impedance measurements provided insight into the
recorded noise levels. Fig. 8 plots average probe impedance
versus postoperative week grouped by the two recording site
sizes used in this study. For both sizes of recording sites (177
and 312 m ), the 1-kHz impedance increased during the
first two to three weeks before flattening, with the smaller
sites having a higher value. The average impedance for the
larger sites gradually decreased during the latter part of the
observation period.
Fig. 9. Histology images from selected implanted arrays. (A) 20-m H&E
stained section from a four-shank device implanted in animal BMI-4 for 127
days. (B) 50-m immuno-stained section (GFAP and Laminen) showing a
close-up of two shanks from a four-shank device implanted in animal MI-7 for
36 days. (C) 150-m section taken parallel to a probe such that a single shank
remains intact within the section. This probe was implanted in MI-10 for 38
days. Scale bars represent 100 m.
D. Histology
Histological analysis from three of the implanted subjects indicated modest tissue reactions to the implanted probes. Procedural difficulties precluded tissue analysis of the fourthand fifth
subjects. Fig. 9(A) and (B) shows coronal cross sections of explanted probe tracks in two different subjects at postoperative
times of 127 days and 36 days, respectively. The probe sites used
to obtain recordings were on the side of shank facing the bottom
902
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 51, NO. 6, JUNE 2004
of the image. The two coronal sections were approximately 1
mm below the cortical surface. Fig. 9(A) suggests that the cortical region around this shank contained normal-appearing neurons. Although there appeared to be minimal fibrous encapsulating tissue around the track, there was a tight glial sheath in
this area. The two right-most tracks in this section appeared to
be rotated approximately 45 relative to the plane of the probe,
although the underlying cause could not be determined.
Fig. 9(C) is a confocal image of an intact probe at a postoperallowed the imaging
ative time of 38 days. Staining for
and laminen were
of reactive astrocytes. In Fig. 9(B),
localized around the probe track and relatively sparsely distributed. This image further documents minimal encapsulation
of the probe shanks. In Fig. 9(C), the probe shank can easily be
resolved due to its reflective silicon surface. The cellular elements surrounding the probe represent astrocytes as identified
immunofluorescence.
by their
IV. DISCUSSION
This study establishes the basic functionality of Michigan
silicon probes for chronic long-term neural recording in cerebral cortex. The probes were found to provide reliable highquality multichannel spike recordings during approximately
four-month assessment periods. In summary, 207 of the 224
(92.4%) individual electrode sites recorded spike activity (Table
II). The average spike amplitude was 77 V and the average
noise level was 11 V, which reflects an average SNR or 7.1. The
average site impedance of the 312 and 177 m sites was 0.715
and 1.647 M , respectively. The average SNR did not significantly differ for these two sites. The average SNR was largely
consistent during the assessment period with an increasing trend
in background noise balanced by a corresponding increasing
trend in signal level. The increasing background noise over
time is at least partially attributed to an increase in the average
probe impedance. The performance of the silicon probes is
difficult to directly compare with other types of implantable
microelectrodes (e.g., [12], [33], and [34]) because of notable
differences in animal models, surgical techniques, spike sorting
techniques, and measurements of signal quality.
However, the results of this study can be directly compared
to a previous study by our group that used implanted microwire
arrays in rats [11]. In that study, 84% of the microwire implants
that lasted more than nine weeks recorded spike activity. However, half (4/8) of the devices implanted did not last nine weeks,
and in those cases the microwire recordings tended to fail within
approximately 3–4 weeks. In this study, 92% of the silicon probe
sites recorded spike activity with few sites dropping out during
the assessment period. Spike activity on the microwire arrays
was typically observed on one to five sites within 24 h post-implant, with additional sites becoming active over time. On average, it took 33 days before spike activity plateaued across the
arrays, with spike amplitudes ranging from 50–250 V. With
the silicon probes, spike activity was typically present on all 16
channels immediately upon implantation or within the first 2 h
postimplant, and spike amplitudes were significantly higher at
50–800 V. In this direct comparison of our microwires to our
silicon probes, the silicon probes were found to out perform microwires in all metrics.
TABLE II
SUMMARY STATISTICS OF PROBE CHARACTERISTICS
One possible explanation for the differences of the two
electrode types is related to the properties of the microenvironment around the devices and the amount of tissue disrupted
by each implanted device. With a thin planar probe (15 m
thick by 60 m wide or 9.0 10 mm ) inserted 2 mm into
the cortex, each shank displaces 0.0018 mm of cortical tissue
(or 0.0072 mm with four shanks, 16 sites). With the 50- m
diameter microwires used in our previous study, (cross sectional
area of 19.63 10 mm ) inserted 2 mm into the cortex, each
wire displaces 0.0039 mm of cortical tissue (or 0.0624 mm
with 16 wires, 16 sites). Making a first-order assumption of
cell densities in human cortex [35], this tissue displacement
would involve approximately 50 neurons and 40 000 synapses
for this type of silicon probe and approximately 1600 neurons
and 1 280 000 synapses for this type of microwire array.
This analysis may even underestimate the situation because
it neglects additional modalities of tissue disruption such as
microhemorrhages resulting from punctures of the relatively
dense microvasculature [31] and shear forces imposed on the
nearby tissue during probe insertion. In general, the size of the
implanted device is a large contributor to the ability of the probe
to successfully function and should be minimized as much as
possible, while still maintaining electrical viability [31].
While detailed histological analysis of the region around the
probes is beyond the scope of this study, the limited histological results that were available indicate reasonably modest tissue
reactions to the implanted probe shanks during the assessment
period. The combined histological and electrophysiological results indicate that the tissue responses were not severe enough
to isolate the recording sites from active and presumably normal
nearby neurons. These data suggest that the local environment
surrounding the probe and the recording site surfaces may reach
a functional steady state at extended time periods. The slight rotation of two of the shanks from the nominal plane of the device
is not often observed. Detailed analysis of the tissue responses to
these probes and comparison to other types of electrodes (e.g.,
[31]) remains an open issue. The results of our study found that
these silicon probes resulted in substantially less adverse tissue
response than did the microwire probes previously implanted in
our lab [unpublished results].
A more intensive histological evaluation of implanted probes
was reported by the Shain group [31], in which an assessment
of the immune response to two different types of silicon probes
was reported. A 25–50- m layer of vimentin-positive cells was
found to surround the implanted devices. Vimentin expression is
increased in reactive astrocytes and is not typically present in mature astrocytes. Analyses at postimplant time durations of 1 day
and 1, 2, 4, and 6 weeks were performed. This paper reported that
probes with the smallest cross-sectional area and the smoothest
corners resulted in the smallest degree of adverse tissue response.
VETTER et al.: CHRONIC NEURAL RECORDING USING SILICON-SUBSTRATE MICROELECTRODE ARRAYS
The average standard error (variability) in SNR between
probes was significantly higher than the average standard error
within probes. This is an interesting observation because it
suggests that changes in the tissue surrounding the probe tend
to affect the entire probe more than individual electrodes on
a probe. There were similar differences in variability for both
the noise and impedance measurements. These results may
be at least partially attributable to differences between the
probes and variability in surgical techniques and require further
investigation.
The average noise and the average site impedance were each
in their maximum ranges during the 2- to 6-week postoperative period. It is reasonable to speculate that the immune response occurring in the microenvironment around the probes
contributes directly to measured impedance [32], [36] and thus,
indirectly to noise level. In addition, the larger sites exhibit more
impedance variability than the smaller sites and the initial increase of impedance is greater in the larger sites than in the
smaller sites. However, the site impedances did not continually
increase over time. If a probe shank was being progressively encapsulated, a gradual increase in the impedance trends would be
expected. The impedance data are consistent with the idea that
the short-term and most pronounced immune response plateaus
after about three weeks [31] and then reaches some form of
steady state.
The surgical techniques developed in the course of this study
are sufficient for use in rats, but are not optimal in the general
sense. Refining the techniques is a critical process that involves
a number of considerations [37], including surgical tools and
techniques, packaging, interconnects, and probe insertion procedures [38]–[44]. This is especially relevant in the course of
developing the silicon probe, as it can be used safely by neurosurgeons for implantation into human patients.
It is problematic to directly extrapolate these results in rats to
larger species such as cats, rhesus monkeys, and even humans.
While these findings powerfully demonstrate that the silicon
probes are effective at establishing long-term interfaces with
active and viable neurons, physical and biological differences
among species may require additional refinements in the probes
and their use. For example, larger brains may undergo increased
relative motion between intracortical regions and the overlying
skull in response to normal movement and physiological processes. This necessitates refinement in probe interconnect design and packaging, although it does not necessarily affect the
recording sites themselves. Differences among species in immune responses and wound healing are more subtle and may
necessitate additional probe refinements. In any case, the findings of this study provide a solid performance benchmark from
which further probe developments can precede.
ACKNOWLEDGMENT
The authors would like to acknowledge the contributions of
P. Rousche and members of the Neural Engineering Lab., most
notably D. Pellinen and K. Otto. They also thank M. Holecko
for assistance in the histology, N. Gulari for probe fabrication,
and R. Oweiss for packaging. They also acknowledge that the
probes were provided by the University of Michigan Center for
903
Neural Communication Technology sponsored by NIH/NIBIB
Grant P41 EB-00 230-10.
REFERENCES
[1] P. R. Kennedy, R. A. Bakay, M. M. Moore, K. Adams, and J. Goldwaithe,
“Direct control of a computer from the human central nervous system,”
IEEE Trans. Rehab. Eng., vol. 8, pp. 198–202, June 2000.
[2] J. R. Wolpaw, N. Birbaumer, D. J. McFarland, G. Pfurtscheller, and T. M.
Vaughan, “Brain-computer interfaces for communication and control,”
Clin. Neurophysiol.: Official J. Int. Fed. Clin. Neurophysiol., vol. 113,
pp. 767–791, 2002.
[3] S. P. Levine, J. E. Huggins, S. L. BeMent, R. K. Kushwaha, L. A. Schuh,
M. M. Rohde, E. A. Passaro, K. V. Elisevich, and B. J. Smith, “A direct
brain interface based on event-related potentials,” IEEE Trans. Rehab.
Eng., vol. 8, pp. 180–185, June 2000.
[4] D. Taylor, S. Tillery, and A. B. Schwartz, “Direct cortical control of 3D
neuroprosthetic devices,” Science, vol. 296, pp. 1829–1832, 2002.
[5] M. D. Serruya, N. G. Hatsopoulos, L. Paninski, M. R. Fellows, and J. P.
Donoghue, “Instant neural control of a movement signal,” Nature, vol.
416, pp. 141–142, 2002.
[6] J. M. Carmena, M. A. Lebedev, R. E. Crist, J. E. O’Doherty, D. M. Santucci, D. F. Dimitrov, P. G. Patil, C. S. Henriquez, and M. A. Nicolelis,
“Learning to control a brain-machine interface for reaching and grasping
by primates,” Public Libr. Sci., vol. 1, pp. 193–208, 2003.
[7] K. V. Shenoy, D. Meeker, S. Cao, S. A. Kureshi, B. Pesaran, C. A. Buneo,
A. P. Batista, P. P. Mitra, J. W. Burdick, and R. A. Andersen, “Neural
prosthetic control signals from plan activity,” NeuroReport, vol. 14, pp.
591–596, 2003.
[8] R. J. Vetter, K. J. Otto, T. C. Marzullo, and D. R. Kipke, “Brain-machine
interfaces in rat motor cortex: Neuronal operant conditioning to perform
a sensory detection task,” presented at the 1st Int. IEEE EMBS Conf.
Neural Eng., Capri, Italy, 2003.
[9] K. J. Otto, R. J. Vetter, T. C. Marzullo, and D. R. Kipke, “Brain-machine
interfaces in rat motor cortex: Implications of adaptive decoding algorithms,” presented at the 1st Int. IEEE EMBS Conf. Neural Eng., Capri,
Italy, 2003.
[10] J. K. Chapin, K. A. Moxon, R. S. Markowitz, and M. A. Nicolelis,
“Real-time control of a robot arm using simultaneously recorded neurons in the motor cortex,” Nat. Neurosci., vol. 2, pp. 664–670, 1999.
[11] J. C. Williams, R. L. Rennaker, and D. R. Kipke, “Long-term neural
recording characteristics of wire microelectrode arrays implanted in
cerebral cortex,” Brain Res. Brain Res. Protoc., vol. 4, pp. 303–313,
1999.
[12] M. A. Nicolelis, D. Dimitrov, J. M. Carmena, R. Crist, G. Lehew, J. D.
Kralik, and S. P. Wise, “Chronic, multisite, multielectrode recordings in
macaque monkeys,” Proc. Nat. Acad. Sci., vol. 100, pp. 11041–11 046,
2003.
[13] I. Ulbert, E. Halgren, G. Heit, and G. Karmos, “Multiple microelectrode-recording system for human intracortical applications,” J. Neurosci. Methods, vol. 106, pp. 69–79, 2001.
[14] P. J. Rousche, D. S. Pellinen, D. P. Pivin Jr., J. C. Williams, R. J. Vetter,
and D. R. Kipke, “Flexible polyimide-based intracortical electrode arrays with bioactive capability,” IEEE Trans. Biomed. Eng., vol. 48, pp.
361–371, Mar. 2001.
[15] T. Stieglitz, H. Beutel, M. Schuettler, and J. U. Meyer, “Micromachined
polyimide-based devices for flexible neural interfaces,” Biomed. Microdev., vol. 2, pp. 283–294, 2000.
[16] K. L. Drake, K. D. Wise, J. Farraye, D. J. Anderson, and S. L. BeMent,
“Performance of planar multisite microprobes in recording extracellular
single-unit intracortical activity,” IEEE Trans. Biomed. Eng., vol. 35, pp.
719–732, Sept. 1988.
[17] C. T. Nordhausen, E. M. Maynard, and R. A. Normann, “Single unit
recording capabilities of a 100 microelectrode array,” Brain Res., vol.
726, pp. 129–140, 1996.
[18] U. G. Hofmann, A. Folkers, F. Mosch, D. Hohl, M. Kindlundh, and
P. Norlin, “A 64(128)-channel multisite neuronal recording system,”
Biomed Tech. (Berl.), vol. 47, pp. 194–197, 2002. Suppl 1 Pt 1.
[19] T. H. Yoon, E. J. Hwang, D. Y. Shin, S. I. Park, J. O. Seung, S. C. Jung,
H. C. Shin, and S. J. Kim, “A micromachined silicon depth probe for
multichannel neural recording,” IEEE Trans. Biomed. Eng., vol. 47, p.
1082, Aug. 2000.
[20] K. D. Wise, D. J. Anderson, J. F. Hetke, D. R. Kipke, and K. Najafi,
“Wireless implantable microsystems: High-density electronic interfaces
to the nervous system,” Proc. IEEE, vol. 92, pp. 76–97, Jan. 2004.
[21] J. F. Hetke, J. L. Lund, K. Najafi, K. D. Wise, and D. J. Anderson, “Silicon ribbon cables for chronically implantable microelectrode arrays,”
IEEE Trans. Biomed. Eng., vol. 41, pp. 314–321, Apr. 1994.
904
[22] J. F. Hetke and D. J. Anderson, “Silicon microelectrodes for extracellular
recording,” in Handbook of Neuroprosthetic Methods, W. E. Finn and P.
G. LoPresti, Eds. Boca Raton, FL: CRC, 2002.
[23] J. Csicsvari, D. A. Henze, B. Jamieson, K. D. Harris, A. Sirota, P. Bartho,
K. D. Wise, and G. Buzsaki, “Massively parallel recording of unit and
local field potentials with silicon-based electrodes,” J. Neurophysiol.,
vol. 90, pp. 1314–1323, 2003.
[24] Q. Bai, K. D. Wise, and D. J. Anderson, “A high-yield microassembly
structure for three-dimensional microelectrode arrays,” IEEE Trans.
Biomed. Eng., vol. 47, pp. 281–289, Mar. 2000.
[25] J. F. Hetke, J. C. Williams, D. S. Pellinen, R. J. Vetter, and D. R. Kipke,
“3-D silicon probe array with hybrid polymer interconnect for chronic
cortical recording,” presented at the 1st Int. IEEE EMBS Conf. Neural
Eng., Capri, Italy, 2003.
[26] R. Rathnasingham, S. C. Bledsoe, J. D. McLaren, and D. R. Kipke,
“Characterization of implantable microfabricated fluid delivery devices,” IEEE Trans. Biomed. Eng., vol. 51, pp. 138–145, Jan. 2004.
[27] S. Furukawa, L. Xu, and J. C. Middlebrooks, “Coding of sound-source
location by ensembles of cortical neurons,” J. Neurosci., vol. 20, pp.
1216–1228, 2000.
[28] J. Csicsvari, B. Jamieson, K. D. Wise, and G. Buzsaki, “Mechanisms of
gamma oscillations in the hippocampus of the behaving rat,” Neuron,
vol. 37, pp. 311–322, 2003.
[29] D. R. Kipke, R. J. Vetter, J. C. Williams, and J. F. Hetke, “Silicon-substrate intracortical microelectrode arrays for long-term recording of neuronal spike activity in cerebral cortex,” IEEE Trans. Rehab. Eng., vol. 11,
pp. 151–155, June 2003.
[30] R. J. Vetter, T. A. Becker, J. C. Williams, and D. R. Kipke, “The use
of ALGEL as an artificial dura for chronic cortical implant neuroprosthetics,” presented at the 1st Int. IEEE EMBS Conf. Neural Eng., Capri,
Italy, 2003.
[31] D. H. Szarowski, M. D. Andersen, S. Retterer, A. J. Spence, M. Isaacson,
H. G. Craighead, J. N. Turner, and W. Shain, “Brain responses to micromachined silicon devices,” Brain Res., vol. 983, pp. 23–35, 2003.
[32] J. C. Williams, “Performance of chronic neural implants: measurement,
modeling, and intervention strategies,” Ph.D. dissertation, Arizona State
Univ., Tempe, AZ, 2001.
[33] P. J. Rousche and R. A. Normann, “Chronic recording capability of the
utah intracortical electrode array in cat sensory cortex,” J. Neurosci.
Meth., vol. 82, pp. 1–15, 1998.
[34] X. Liu, D. B. McCreery, R. R. Carter, L. A. Bullara, T. G. Yuen, and W. F.
Agnew, “Stability of the interface between neural tissue and chronically
implanted intracortical microelectrodes,” IEEE Trans. Rehab. Eng., vol.
7, pp. 315–326, Sept. 1999.
[35] V. Braitenberg and A. Schuz, Cortex: Statistics and Geometry of Neuronal Connectivity. New York: Springer-Verlag, 1998.
[36] J. C. Williams, G. D. Smith, R. J. Vetter, and D. R. Kipke, “Correlation
analysis of electrical interface properties of chronic neural implants,”
Soc. Neurosci. Abstr., 2001.
[37] E. M. Maynard, E. Fernandez, and R. A. Normann, “A technique to prevent dural adhesions to chronically implanted microelectrode arrays,” J.
Neurosci. Meth., vol. 97, pp. 93–101, 2000.
[38] T. Moon, M. Ghovanloo, and D. R. Kipke, “Buckling strength of coated
and uncoated silicon microelectrodes,” presented at the 25th Annu. Int.
Conf. IEEE Engineering in Medicine and Biology. Soc., Cancun, SE,
Mexico, 2003.
[39] M. D. Johnson, J. C. Williams, and D. R. Kipke, “Chemical sensing
capability of MEMS implantable multichannel neural microelectrode
arrays,” presented at the 25th Annu. Int. Conf. IEEE Engineering in
Medicine and Biology. Soc., Cancun, SE, Mexico, 2003.
[40] R. J. Vetter, R. H. Olsson, J. F. Hetke, J. C. Williams, D. S. Pellinen, K. D.
Wise, and D. R. Kipke, “Silicon-substrate intracortical microelectrode
arrays with integrated electronics for chronic recording,” presented at
the 25th Annu. Int. Conf. IEEE Engineering in Medicine and Biology.
Soc., Cancun, SE, Mexico, 2003.
[41] M. Ghovanloo, K. D. Wise, and K. Najafi, “Toward a button-sized
1024-site wireless cortical microstimulating array,” presented at the 1st
Int. IEEE EMBS Conf. Neural Engineering, Capri, Italy, 2003.
[42] J. Chen, K. D. Wise, J. F. Hetke, and S. C. Bledsoe Jr., “A multichannel
neural probe for selective chemical delivery at the cellular level,” IEEE
Trans. Biomed. Eng., vol. 44, pp. 760–769, Aug. 1997.
[43] R. H. Olsson, M. N. Gulari, and K. D. Wise, “Silicon recording arrays
with on-chip electronics for in-vivo data acquisition,” presented at the
IEEE-EMBS Conf. Microtechnologies Med. Biol., Madison, WI, 2002.
[44] S. Nikles, D. S. Pellinen, D. R. Kipke, and K. Najafi, “Long term measurements of polyimide electrical characteristics for use in chronically
implanted neural probes,” presented at the 25th Annu. Int. Conf. IEEE
Eng. Medicine and Biology Soc., Cancun, SE, Mexico, 2003.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 51, NO. 6, JUNE 2004
Rio J. Vetter received the B.S.E. degree in
bioengineering in 1999 and the M.S. degree in bioengineering in 2002 from Arizona State University,
Tempe. He is currently working toward the Ph.D.
degree at the University of Michigan, Ann Arbor.
His research interests include the development of
long-term neural prosthetic devices and novel signal
processing techniques for brain-machine interface
applications as well as BioMEMS, neuroprostheses,
and systems neuroscience.
Justin C. Williams (M’02) received B.Sc. degrees in
mechanical engineering in 1995 and in engineering
physics in 1996 from South Dakota State University,
Brookings, the M.Sc. degree in bioengineering in
2001, and the Ph.D. degree in bioengineering in
2001 from Arizona State University, Tempe.
From 1996 to 2001, he was a Research Assistant in
the Bioengineering Department, Arizona State University, where his work was in the area of neural engineering and implantable devices. From 2002 to 2003,
he was a jointly appointed Research Fellow in the Department of Biomedical Engineering, University of Michigan, Ann Arbor, and
a Research Scientist in the Department of Neurological Surgery, University of
Wisconsin, where his work focused on neurosurgery applications of neural engineering devices. In 2003, he was appointed Assistant Professor in Biomedical
Engineering and Neurological Surgery at the University of Wisconsin, Madison,
where his research interests include BioMEMS, neuroprostheses, and functional
neurosurgery.
Jamille F. Hetke received the B.S.E.E. degree from
Michigan State University, East Lansing, in 1985 and
the M.S. degree in bioengineering from the University of Michigan, Ann Arbor, in 1987.
She has been involved in neural probe research at
the University of Michigan since the completion of
her degrees and is currently an Assistant Research
Scientist in the Department of Electrical Engineering
and Computer Science. She participates in research in
the Kresge Hearing Research Institute and the Neural
Engineering Laboratory, and is the main contact for
investigators interested in obtaining probes from the Center for Neural Communication Technology, Ann Arbor. Her research interests include the design
of silicon arrays and implantable systems for eventual use in neuroprostheses.
Elizabeth A. Nunamaker (S’03) received the B.S.E.
degree in bioengineering from Arizona State University, Tempe, in 2000 and the M.S. degree in biomedical
engineering from Northwestern University, Evanston,
IL, in 2002. She is currently working toward the Ph.D.
degree in the Department of Biomedical Engineering,
University of Michigan, Ann Arbor.
Her current research interests include localized
drug delivery and microfluidics for BioMEMS
applications.
Daryl R. Kipke (S’87–M’90) received the B.Sc. degree in engineering science in 1985, the M.Sc. degree in bioengineering in 1986, the M.S.E. degree
in electrical engineering in 1988, and the Ph.D. degree in bioengineering in 1991 from the University
of Michigan, Ann Arbor.
From 1991 to 1992, he was a Research Associate
in the Department of Bioengineering and the Institute
for Sensory Research at Syracuse University, Syracuse, NY. In 1992, he was appointed to the Bioengineering faculty at Arizona State University, Tempe.
In 2001, he joined the faculty of the College of Engineering at the University of
Michigan. He is currently an Associate Professor in the Department of Biomedical Engineering and the Department of Electrical Engineering and Computer
Science at Michigan, the Director of the Center for Neural Communication
Technology, and the Leader of the Neural Engineering Laboratory. His primary
research interests include the areas of BioMEMS for neural implants, neuroprostheses, systems neuroscience, functional neurosurgery, and neurotechnologies.