scispace - formally typeset
Search or ask a question
Journal ArticleDOI

A wireless and artefact-free 128-channel neuromodulation device for closed-loop stimulation and recording in non-human primates

TL;DR: An artefact-free wireless neuromodulation device that enables research applications requiring high-throughput data streaming, low-latency biosignal processing, and simultaneous sensing and stimulation and may help advance neuroscientific discovery and preclinical investigations of stimulation-based therapeutic interventions.
Abstract: Closed-loop neuromodulation systems aim to treat a variety of neurological conditions by delivering and adjusting therapeutic electrical stimulation in response to a patient's neural state, recorded in real time. Existing systems are limited by low channel counts, lack of algorithmic flexibility, and the distortion of recorded signals by large and persistent stimulation artefacts. Here, we describe an artefact-free wireless neuromodulation device that enables research applications requiring high-throughput data streaming, low-latency biosignal processing, and simultaneous sensing and stimulation. The device is a miniaturized neural interface capable of closed-loop recording and stimulation on 128 channels, with on-board processing to fully cancel stimulation artefacts. In addition, it can detect neural biomarkers and automatically adjust stimulation parameters in closed-loop mode. In a behaving non-human primate, the device enabled long-term recordings of local field potentials and the real-time cancellation of stimulation artefacts, as well as closed-loop stimulation to disrupt movement preparatory activity during a delayed-reach task. The neuromodulation device may help advance neuroscientific discovery and preclinical investigations of stimulation-based therapeutic interventions.

Summary (2 min read)

Introduction

  • UC Berkeley UC Berkeley Previously Published Works Title A wireless and artefact-free 128-channel neuromodulation device for closed-loop stimulation and recording in non-human primates.
  • In fact, most attempts to close the loop for DBS treatments have been done only for short duration using systems that were not fully implantable4,5,9–11.
  • Effectively and efficiently cancelling artefacts requires careful co-design of the stimulators and signal acquisition chains.
  • The wireless artefact-free neuromodulation device (WAND) introduced here incorporates all the key features needed to continuously monitor neural biomarkers in the presence of stimulation artefacts and deliver closed-loop stimulation.

System design

  • Integrated circuits are required to minimize area and power for a large number of recording and stimulation channels.
  • This is followed by a long, post-stimulus voltage decay (indirect artefact) determined by the mismatch of stimulation phases and electrode properties.
  • To date, even the best results in front-end artefact mitigation do not demonstrate complete artefact removal, necessitating back-end digital cancellation for residual artefacts.
  • Different techniques may achieve better results than others, depending on the level of mitigation achieved by the front-end amplifiers.
  • Overall system resiliency to stimulation artefacts depends heavily on the co-design of the stimulator, signal processing blocks and recording front end.

Results

  • WAND components and architecture are shown in Fig.  1. Lines represent means, while shaded areas represent s.d. NAtuRe BioMediCAL eNGiNeeRiNG | VOL 3 | JANUARY 2019 | 15–26 | www.nature.com/natbiomedeng18 ArticlesNaTurE BIomEdICal ENgINEErINg known to have elevated power during sleep states relative to wake states69,70.
  • The authors calculated averaged templates of single- and double-sample artefacts (Fig. 5c) for varying stimulation amplitudes and pulse widths, and confirmed a linear relationship between these parameters and the artefact amplitude (Fig.  5d).
  • Neural activity was recorded throughout the task, and while stimulation turn-off can be based on neural signature, the authors chose to adhere to the durations used in previous work to demonstrate reproducibility of an established result.
  • The architecture of WAND makes it amenable to function as a general-purpose research device, requiring only minor modifications to be reoptimized for new applications.

Methods

  • The WAND board (Fig. 1b–d) consists of a SoC FPGA with a 166 MHz Advanced RISC Machine (ARM, where RISC stands for reduced instruction set computer) Cortex-M3 processor (SmartFusion2 M2S060T; Microsemi) acting as a master module.
  • The Cortex-M3 processor selects which channels are streamed or used for closed-loop, and runs the artefact cancellation and closed-loop algorithms.
  • Each recording channel has a selectable input voltage range of 100 or 400 mV, allowing simultaneous amplification and digitization of the electrode offset, neural signal and stimulation artefact within the linear range.
  • Because some unflagged samples may still be affected by the shorting phase, their artefact cancellation always interpolates over the maximum number of consecutive flagged samples possible after detection of the first artefact sample (Fig. 5b).
  • After the delay period, the centre target disappears from the screen, signalling the ‘go cue’, and the subject is cued to reach to the peripheral target.

Author contributions

  • J.M.R., J.M.C and R.M. are co-principal investigators.
  • S.R.S. and J.M.C. designed the in vivo experiments.

Additional information

  • Supplementary information is available for this paper at https://doi.org/10.1038/.
  • S41551-018-0323-x. Reprints and permissions information is available at www.nature.com/reprints.
  • When statistical analyses are reported, confirm that the following items are present in the relevant location .
  • This statement should provide the following information, where applicable: - Accession codes, unique identifiers, or web links for publicly available datasets - A list of figures that have associated raw data - A description of any restrictions on data availability.
  • Thus, multiple repeated experiments on a single animal subject were sufficient.

Did you find this useful? Give us your feedback

Content maybe subject to copyright    Report

UC Berkeley
UC Berkeley Previously Published Works
Title
A wireless and artefact-free 128-channel neuromodulation device for closed-loop
stimulation and recording in non-human primates.
Permalink
https://escholarship.org/uc/item/0mt6k2cm
Journal
Nature biomedical engineering, 3(1)
ISSN
2157-846X
Authors
Zhou, Andy
Santacruz, Samantha R
Johnson, Benjamin C
et al.
Publication Date
2019
DOI
10.1038/s41551-018-0323-x
Peer reviewed
eScholarship.org Powered by the California Digital Library
University of California

Articles
https://doi.org/10.1038/s41551-018-0323-x
1
Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, CA, USA.
2
Helen Wills Neuroscience Institute,
University of California, Berkeley, Berkeley, CA, USA.
3
Cortera Neurotechnologies, Inc., Berkeley, CA, USA.
4
Chan Zuckerberg Biohub, San Francisco, CA, USA.
5
These authors contributed equally: Andy Zhou, Samantha R. Santacruz, Benjamin C. Johnson.
6
These authors jointly supervised this work: Jan M. Rabaey,
Jose M. Carmena, Rikky Muller. *e-mail: rikky@berkeley.edu
C
losed-loop neuromodulation improves open-loop therapeu-
tic electrical stimulation by providing adaptive, on-demand
therapy, reducing side effects and extending battery life in
wireless devices
1,2
. Closing the loop requires low-latency extraction
and accurate estimation of neural biomarkers
35
from recorded sig-
nals to automatically adjust when and how to administer stimula-
tion as feedback to the brain. Recent studies have shown responsive
stimulation to be a viable option for treating epilepsy
2,6
, and there is
evidence that closed-loop strategies could improve deep brain stim
-
ulation (DBS) for treating Parkinsons disease and other motor dis-
orders
7,8
. However, there is presently no commercial device allowing
closed-loop stimulation for DBS in patients with movement dis
-
orders, and strategies for implementing such stimulation are still
under investigation. In fact, most attempts to close the loop for DBS
treatments have been done only for short duration using systems
that were not fully implantable
4,5,911
. To enable advanced research in
closed-loop neuromodulation, there is a need for a flexible research
platform, for testing and implementing these various closed-loop
paradigms, that is also wireless, compact, robust and safe.
Designing such a device requires unification of multi-channel
recording, biomarker detection and microstimulation technologies
into a single unit with careful consideration of their interactions.
Wireless, multi-channel recording-only devices capture activity
from wide neuronal populations
12,13
, but do not have the built-in
ability to immediately act on that information and deliver stimu
-
lation. Several complete closed-loop devices have been proposed
and demonstrated, but are limited by low channel counts
1417
and
low wireless streaming bandwidth
1418
. Most recently, variations
of the fully integrated and optimized closed-loop neuromodula
-
tion system-on-a-chip (SoC) have been presented, but full system
functionality has not yet been adequately demonstrated in vivo
1925
.
While future versions may be paired with miniaturized external bat
-
tery packs and controllers, current systems built around these SoCs
require large, stationary devices to deliver power inductively from a
close range
19,21,23
. This limits studies to using small, caged animals.
Furthermore, any device for concurrent sensing and stimula
-
tion must be able to mitigate or remove stimulation artefacts—
the large voltage transients resulting from stimulation that distort
recorded signals and obscure neural biomarkers. Signals recorded
concurrently with stimulation may contain relevant information
for closed-loop algorithms or offline analysis, yet existing devices
disregard these affected windows of data, or fail to reduce arte
-
facts to an acceptable level for recovery of many potentially useful
biomarker features. Effectively and efficiently cancelling artefacts
requires careful co-design of the stimulators and signal acquisition
chains. Additionally, computational reprogrammability is needed
for application-dependent algorithm design in both artefact cancel
-
lation and closed-loop control.
The wireless artefact-free neuromodulation device (WAND)
introduced here incorporates all the key features needed to con
-
tinuously monitor neural biomarkers in the presence of stimulation
artefacts and deliver closed-loop stimulation. WAND combines: (1)
2 custom-designed, 64-channel neural interface application-specific
integrated circuits supporting simultaneous low-noise recording
and high-current stimulation, specifically designed to minimize
stimulation artefacts; (2) flexible and reprogrammable back-end
processing on a SoC field-programmable gate array (FPGA) for
cancelling residual artefacts, computing neural biomarkers, run
-
ning closed-loop algorithms and controlling stimulation; and (3) a
robust, bidirectional wireless link to a graphical user interface (GUI)
A wireless and artefact-free 128-channel
neuromodulation device for closed-loop
stimulation and recording in non-human primates
AndyZhou
1,5
, SamanthaR.Santacruz
1,2,5
, BenjaminC.Johnson
1,3,5
, GeorgeAlexandrov
1
, AliMoin
1
,
FredL.Burghardt
1
, JanM.Rabaey
1,6
, JoseM.Carmena
1,2,6
and RikkyMuller
1,3,4,6
*
Closed-loop neuromodulation systems aim to treat a variety of neurological conditions by delivering and adjusting therapeutic
electrical stimulation in response to a patient’s neural state, recorded in real time. Existing systems are limited by low channel
counts, lack of algorithmic flexibility, and the distortion of recorded signals by large and persistent stimulation artefacts. Here,
we describe an artefact-free wireless neuromodulation device that enables research applications requiring high-throughput
data streaming, low-latency biosignal processing, and simultaneous sensing and stimulation. The device is a miniaturized neu-
ral interface capable of closed-loop recording and stimulation on 128 channels, with on-board processing to fully cancel stimu-
lation artefacts. In addition, it can detect neural biomarkers and automatically adjust stimulation parameters in closed-loop
mode. In a behaving non-human primate, the device enabled long-term recordings of local field potentials and the real-time
cancellation of stimulation artefacts, as well as closed-loop stimulation to disrupt movement preparatory activity during a
delayed-reach task. The neuromodulation device may help advance neuroscientific discovery and preclinical investigations of
stimulation-based therapeutic interventions.
NATURE BIOMEDICAL ENGINEERING | VOL 3 | JANUARY 2019 | 15–26 | www.nature.com/natbiomedeng
15

Articles
NaTurE BIomEdICal ENgINEErINg
for device configuration and control, as well as data logging. These
features are tightly integrated into a small form factor, low-power
device, enabling many proposed closed-loop and responsive neuro
-
modulation applications, as well as offering a platform for develop-
ing new ones. To demonstrate the various functions of WAND, we
performed a series of in vivo experiments that validate long-term,
high-fidelity and wireless multi-channel recording; real-time, com
-
plete removal of stimulation artefacts for accurate recovery of neural
signals; and on-board biomarker detection for closed-loop control.
System design
WAND is designed to be a general-purpose tool with immediate
applicability in various research environments. Inclusion of a wide
feature set is balanced by limitations in device size and power.
Integrated circuits are required to minimize area and power for
a large number of recording and stimulation channels. We custom-
designed a neuromodulation integrated circuit (NMIC) to deliver
stimulation pulses ranging from subthreshold currents (down to
20 µ A) to those required by DBS (5 mA), and to record local field
potentials (LFPs) with a bandwidth of up to 500 Hz
26
. We chose
LFPs as the signals of interest for their usefulness in medical applica
-
tions as an indicator of disease
8,2730
. There is also evidence that LFPs
can be used for motor decoding in brain–machine interfaces
3135
,
with comparable accuracy and better longevity than spike decod
-
ers
35
. This signal is also extremely useful to understanding neural
processing, and is incredibly relevant for a variety of basic neuro
-
science studies, from investigating how neural oscillations coordi-
nate movement
3638
to cued transitions between dynamic states in
corticobasal ganglia circuits
39
and working memory
40,41
. Finally, the
lower (1 kHz) sampling rate required for LFP recording utilizes a
lower wireless bandwidth for real-time streaming, allowing the use
of low-power, off-the-shelf radios.
While numerous high-channel-count recording circuits have
been designed
42,43
, state-of-the-art circuits cannot tolerate—and
often exacerbate—the effects of stimulation artefacts. Electrical
stimulation generates a large voltage transient (direct artefact), con
-
comitant with stimulation current and charge delivery to neural tis-
sue and nearby electrodes. The direct artefact may be many orders of
magnitude larger that the underlying neural signal (mV compared
with µ V, respectively). This is followed by a long, post-stimulus volt
-
age decay (indirect artefact) determined by the mismatch of stimu-
lation phases and electrode properties. Conventional low-noise,
low-power neural amplifiers are sensitive to both direct and indirect
artefacts, saturating from both
44,45
. They recover slowly from satura-
tion due to long time constants of analogue feedback, causing data
loss during—and many milliseconds after—a stimulation pulse.
State-of-the-art methods for mitigating the indirect artefact try to
prevent saturation of the front ends. Saturation can be prevented by
increasing the amplifier linear input range and tolerance to d.c. cur
-
rent offset
21,26,46
, or by subtracting the large amplitude components
of the artefact
47,48
. Artefact duration can be reduced by rapidly clear-
ing charge built up on circuit elements from stimulation
25,26,49,50
. We
have designed the NMICs with improved stimulation and recording
architectures to both prevent large indirect artefacts and minimize
their effects on the front-end circuits.
To date, even the best results in front-end artefact mitigation do
not demonstrate complete artefact removal, necessitating back-end
digital cancellation for residual artefacts. Computationally efficient
back-end methods include subtractive methods (where artefact
templates are subtracted from the waveform to reveal the under
-
lying signal
48,51,52
) and reconstructive methods (where segments of
corrupted data are replaced with interpolated values
5355
). Different
techniques may achieve better results than others, depending on the
level of mitigation achieved by the front-end amplifiers.
Overall system resiliency to stimulation artefacts depends heav
-
ily on the co-design of the stimulator, signal processing blocks and
recording front end. In this work, we demonstrate how the spe
-
cific artefact prevention and mitigation techniques utilized in the
NMICs motivate our implementation of back-end linear interpola
-
tion in an on-board SoC FPGA to completely remove stimulation
artefacts in real time. In particular, the reduced artefact duration
allows for a computationally inexpensive but effective back-end
cancellation solution at the cost of losing only one or two samples.
These innovations allow for online, real-time biomarker computa
-
tion for closed-loop stimulation.
Results
WAND architecture. WAND components and architecture are
shown in Fig.1. For this work, the form factor was designed to
fit into the polyetherimide housing for a custom-built chronically
implanted microelectrode array (Gray Matter Research). The device
has a board area of 10.13 cm
2
and weighs 17.95 g together with a
rechargeable 500 mAh lithium-ion battery pack, allowing 11.3 h of
continuous, wireless operation (Fig.1a). The main components of
WAND are the pair of custom-designed NMICs, a SoC FPGA, a
radio SoC and support circuitry for power regulation and program
-
ming (Fig.1b–d).
Each NMIC consists of 64 recording channels and 4 stimulators
that can address any of the 64 channels, meaning that stimulation
can occur simultaneously on up to 8 channels by leveraging stimula
-
tors across the 2 on-board NMICs. Multi-site stimulation is desirable
for implementing specific spatiotemporal patterns of stimulation,
with many recent studies performing stimulation on two to eight
channels simultaneously
5661
. Using WAND, the stimulation chan-
nels can be dynamically assigned, thus allowing this device to be uti-
lized for multi-site stimulation on up to eight channels concurrently
in a highly flexible manner. Ultimately, stimulation can be delivered
using an open-loop paradigm or a closed-loop approach that relies
on continuous sensing of on-board computed biomarkers.
Stimulation parameters are rapidly reprogrammable by writing to
registers on the NMIC through commands transmitted from a GUI,
or automatically based on calculations performed on board. All stim
-
ulation settings listed in Supplementary Fig.1, as well as the selec-
tion of stimulation sites and triggering of pulses, can be set through
the same interface. A new setting can be preloaded while stimulating
with a previous setting, potentially reducing latency between bio
-
marker state detection and the resulting stimulation update.
Unlike conventional neural interface integrated circuits, the
NMICs enable simultaneous low-noise, low-power neural record
-
ing of LFP with high-compliance electrical stimulation (Fig.1c).
The NMIC prevents large amplitude indirect artefacts by employ
-
ing stimulators with highly accurate charge balancing
26
. Accurate
charge balancing is achieved by reusing the same current source
for both phases of a biphasic pulse and a return-to-ground
stimulator architecture (Supplementary Fig. 2 and the section
System artefact resiliency’ in the Supplementary Information).
To address both direct and indirect artefacts, the NMIC recording
front ends are designed simultaneously for low-noise (1.6 μ V
rms
(root-mean-square voltage) mean channel noise) recording and
a large linear input range of 100 mV. The input range is over ten
times larger than conventional designs
42,43
and avoids saturation
in the presence of large stimulation artefacts (tens of mV) while
still being able to resolve µ V-level neural signals. This large range
of resolvable signals is achieved with a mixed-signal architecture
that integrates the analogue-to-digital converter (ADC) into the
feedback loop, thereby reducing the required gain and signal
swings. The architecture also resets at every sample, enabling
memoryless sampling and rapid recovery from stimulation arte
-
facts (Supplementary Fig.4). Therefore, stimulation artefacts do
not persist beyond the samples when stimulation is occurring,
and minimal data are lost when using reconstructive back-end
cancellation methods such as interpolation.
NATURE BIOMEDICAL ENGINEERING | VOL 3 | JANUARY 2019 | 15–26 | www.nature.com/natbiomedeng
16

Articles
NaTurE BIomEdICal ENgINEErINg
All 128 channels of neural data from both NMICs are sampled,
digitized (15 bits; 1 kS s
1
) and serialized on chip, then transmit-
ted to the on-board FPGA and microcontroller SoC via a custom-
designed bidirectional interface implemented in the FPGA fabric
62
.
The same interface is used for downlink commands to control
NMIC circuitry and update stimulation parameters (Fig. 1d).
Software running on the included Cortex-M3 microprocessor then
aggregates neural and other sensor data, cancels stimulation arte
-
facts, selects a subset of data to be streamed back to the base sta-
tion and runs closed-loop neuromodulation algorithms (Fig.1d).
The FPGA fabric and Cortex-M3 software are reprogrammable
through a serial wire debug interface, allowing customization for
different applications. A 2.4 GHz Bluetooth Low Energy (BLE)
radio SoC allows for robust bidirectional wireless communication
up to 2 m from the subject. BLE offers low-power telemetry, and
customization of the BLE protocol enables data streaming rates
close to the 2 Mbp modulation rate. WAND can stream up to 96
uncompressed LFP recording channels in real time to a personal
computer (PC) running a custom-built GUI for system configura
-
tion and data visualization.
High-fidelity multi-channel wireless recording. To evaluate the
quality of recordings made using WAND, we recorded 96 chan
-
nels of LFP activity from a non-human primate (NHP) using a
chronically implanted microdrive electrode array with access to
both cortical and subcortical nuclei (Fig.2a). We compared WAND
recordings with sequentially recorded neural data from a wired,
state-of-the-art, commercial neurophysiology system (Tucker-
Davis Technologies). Respective recordings from each system have
qualitatively similar signal properties, as assessed by computing
the power spectral densities of the recorded data (Fig.2b,c). The
WAND recordings exhibit lower 60 Hz interference due to the lack
of long interface cables and better-isolated recording references.
To demonstrate robust detection of biomarkers in WAND record
-
ings and establish a baseline for neuromodulation experiments, we
recorded LFP activity during a standard self-paced, centre-out joy
-
stick task (Fig.3a,b). During this behaviour, ongoing beta and high-
gamma rhythms are inversely modulated by task-related periods of
movement (Fig.3c,d). Beta band oscillations are found to emerge
during specific motor actions and notably before instructed reaches
or movements
37,63,64
. In premotor and motor areas, this rhythm has
been linked to neural activity related to motor preparation
6568
. The
subject had an average reaction time of 183.3 ± 4.8 (s.e.m.) ms across
400 trials. For LFP signals recorded from premotor and motor areas,
we found that the reaction time was significantly correlated with the
average power of beta band activity around the go cue (see Methods;
Pearsons correlation coefficient, r = 0.12; P = 0.03).
To validate long-term, wireless system functionality, we per
-
formed unconstrained, overnight recordings for five nights in the
subjects home cage, recharging the battery between each session
WAND circuit
board
Radio SoC
Antenna
Power regulation
64×
Fast Fourier
transform
Threshold
detection
Stimulation
command
Data
aggregation
SoC FPGA
Battery
Microdrive
Antenna
Power
regulation
4×
64×
ADC
Stimulator
NMIC 1
64×
WAND board
Artefact
removal
Base
station
Commands
Data
3 V
2.5, 1.8, 1.2 V
1.8 V
To PC
Serial peripheral
interface–universal
serial bus bridge
a
Battery pack
Microdrive
headstage
SoC FPGA
NMIC 2NMIC 1
Digital core
d.c.–d.c. converters
Stimulation core Bandgap
64 front ends and stimulation multiplexer
bc
d
Radio SoC
System
controller
Closed-loop
control
Radio SoC
35.6 mm
33 mm
4×
64×
ADC
Stimulator
NMIC 2
Chip
controller
Chip
controller
Commands
Data
Fig. 1 | WAND system architecture. a, Three-dimensional computer-aided design model of WAND with a primate headstage and battery pack, shown
without the polyetherimide case. b, Top- (left) and bottom-view (right) photographs of the WAND circuit board, showing its relevant subsystems and
dimensions (33 mm (H) ×  35.6 mm (W)). c, Micrograph of an NMIC with annotated subcircuits. d, Functional diagram of the WAND system, showing data
and power connections on the main device board, and connections to the microdrive electrode array, battery and a wireless base station.
NATURE BIOMEDICAL ENGINEERING | VOL 3 | JANUARY 2019 | 15–26 | www.nature.com/natbiomedeng
17

Articles
NaTurE BIomEdICal ENgINEErINg
(Fig.4a). We recorded on average of 10.2 consecutive hours per
night, with a mean packet error rate below 0.5% and a median
packet error rate below 0.1%, where each transmitted wireless packet
contained 1 ms of neural data from all streamed channels. Offline
analysis of the data revealed useful sleep-related biomarkers. Delta
(0–4 Hz; Fig. 4b) and theta (4–7 Hz; Fig.4c) frequency bands are
PSD (µV
2
Hz
–1
)PSD (µV
2
Hz
–1
)
b
c
Frequency (Hz)
050 100 150
200
050 100 150 200
10
4
10
3
10
2
10
1
10
0
10
–1
10
–2
WAND
TDT
10
4
10
3
10
2
10
1
10
0
10
–1
10
–2
10
–3
Time (s)
Simultaneous LFP recording channels from one NHP
1
96
1230
a
7
20
Frequency (Hz)
WAND
TDT
Fig. 2 | Wireless, multi-channel recording. a, Representative 3 s segments of 96 channels of simultaneous LFP recordings taken from one NHP during
freely moving behaviour. b,c, Comparison of mean power spectral densities (PSD) (Welch’s method, n=  1,170 windows, one 5 min recording, 512 ms
windows, 256 ms overlap) from channels 20 (b) and 7 (c) for recordings taken from WAND and subsequent recordings taken from a commercial wired
neurophysiology system (Tucker-Davis Technologies (TDT)). Lines represent means, while shaded areas represent s.d.
–29
–27
–25
–23
–21
Time (s)
–1 0
123
400
300
200
100
0
Trials
–48
–36
–24
–12
0
Beta power (dB)
Go cue
Target
hold
Reward on
500 ms
100 μV
–1
01
23
Beta power (dB)
–26
–27
–28
–29
High-gamma power (dB)
Time (s)
Centre
hold
500 ms
Reach to peripheral
target
Go cue
Target
hold
500 ms
Reward
High gamma
Beta
a
b
c
d
Reward off
Fig. 3 | LFP recordings during the joystick task. a, Diagram of the centre-out joystick task with a timeline of the task periods for movement and reward.
The orange patch on the NHP’s head represents the location of the WAND device and headstage implant. b, Representative LFP recordings from three
channels during the centre-out task. c, Trial-averaged (n=  400) beta (13–22 Hz) and high-gamma (70–200 Hz) power aligned to the go cue during the
centre-out task. Lines represent means, while shaded areas represent s.e.m. d, Beta power aligned to the go cue. Each row represents activity from a single
trial. Trials are organized by the time to target hold following the go cue.
NATURE BIOMEDICAL ENGINEERING | VOL 3 | JANUARY 2019 | 15–26 | www.nature.com/natbiomedeng
18

Citations
More filters
Journal ArticleDOI
TL;DR: Neuralink's approach to BMI has unprecedented packaging density and scalability in a clinically relevant package and has achieved a spiking yield of up to 70% in chronically implanted electrodes.
Abstract: Brain-machine interfaces hold promise for the restoration of sensory and motor function and the treatment of neurological disorders, but clinical brain-machine interfaces have not yet been widely adopted, in part, because modest channel counts have limited their potential. In this white paper, we describe Neuralink’s first steps toward a scalable high-bandwidth brain-machine interface system. We have built arrays of small and flexible electrode “threads,” with as many as 3072 electrodes per array distributed across 96 threads. We have also built a neurosurgical robot capable of inserting six threads (192 electrodes) per minute. Each thread can be individually inserted into the brain with micron precision for avoidance of surface vasculature and targeting specific brain regions. The electrode array is packaged into a small implantable device that contains custom chips for low-power on-board amplification and digitization: The package for 3072 channels occupies less than 23×18.5×2 mm3. A single USB-C cable provides full-bandwidth data streaming from the device, recording from all channels simultaneously. This system has achieved a spiking yield of up to 70% in chronically implanted electrodes. Neuralink’s approach to brain-machine interface has unprecedented packaging density and scalability in a clinically relevant package.

432 citations

Posted ContentDOI
18 Jul 2019-bioRxiv
TL;DR: Neuralink’s approach to BMI has unprecedented packaging density and scalability in a clinically relevant package and has achieved a spiking yield of up to 85.5 % in chronically implanted electrodes.
Abstract: Brain-machine interfaces (BMIs) hold promise for the restoration of sensory and motor function and the treatment of neurological disorders, but clinical BMIs have not yet been widely adopted, in part because modest channel counts have limited their potential. In this white paper, we describe Neuralink’s first steps toward a scalable high-bandwidth BMI system. We have built arrays of small and flexible electrode “threads”, with as many as 3,072 electrodes per array distributed across 96 threads. We have also built a neurosurgical robot capable of inserting six threads (192 electrodes) per minute. Each thread can be individually inserted into the brain with micron precision for avoidance of surface vasculature and targeting specific brain regions. The electrode array is packaged into a small implantable device that contains custom chips for low-power on-board amplification and digitization: the package for 3,072 channels occupies less than (23 × 18.5 × 2) mm3. A single USB-C cable provides full-bandwidth data streaming from the device, recording from all channels simultaneously. This system has achieved a spiking yield of up to 85.5 % in chronically implanted electrodes. Neuralink’s approach to BMI has unprecedented packaging density and scalability in a clinically relevant package.

275 citations

Journal ArticleDOI
02 Apr 2020-Cell
TL;DR: This review presents recent research on devices that are relevant to neuromodulation, with an emphasis on multimodal operation, miniaturized dimensions, biocompatible designs, advanced neural interfaces, and battery-free, wireless capabilities.

124 citations

Journal ArticleDOI
TL;DR: A lightweight, wireless, battery-free injectable microsystem that combines soft microfluidic and microscale inorganic light-emitting diode probes for programmable pharmacology and optogenetics, designed to offer the features of drug refillability and adjustable flow rates, together with programmable control over the temporal profiles is presented.
Abstract: Pharmacology and optogenetics are widely used in neuroscience research to study the central and peripheral nervous systems. While both approaches allow for sophisticated studies of neural circuitry, continued advances are, in part, hampered by technology limitations associated with requirements for physical tethers that connect external equipment to rigid probes inserted into delicate regions of the brain. The results can lead to tissue damage and alterations in behavioral tasks and natural movements, with additional difficulties in use for studies that involve social interactions and/or motions in complex 3-dimensional environments. These disadvantages are particularly pronounced in research that demands combined optogenetic and pharmacological functions in a single experiment. Here, we present a lightweight, wireless, battery-free injectable microsystem that combines soft microfluidic and microscale inorganic light-emitting diode probes for programmable pharmacology and optogenetics, designed to offer the features of drug refillability and adjustable flow rates, together with programmable control over the temporal profiles. The technology has potential for large-scale manufacturing and broad distribution to the neuroscience community, with capabilities in targeting specific neuronal populations in freely moving animals. In addition, the same platform can easily be adapted for a wide range of other types of passive or active electronic functions, including electrical stimulation.

96 citations

Journal ArticleDOI
01 Apr 2020
TL;DR: The development of neural interfaces, which can provide a direct, electrical bridge between analogue human nervous systems and digital man-made devices, is examined, considering challenges and opportunities created with such technology.
Abstract: Devices such as keyboards and touchscreens allow humans to communicate with machines. Neural interfaces, which can provide a direct, electrical bridge between analogue nervous systems and digital man-made systems, could provide a more efficient route to future information exchange. Here we review the development of electronic neural interfaces. The interfaces typically consist of three modules — a tissue interface, a sensing interface, and a neural signal processing unit — and based on technical milestones in the development of the electronic sensing interface, we group and analyse the interfaces in four generations: the patch clamp technique, multi-channel neural interfaces, implantable/wearable neural interfaces and integrated neural interfaces. We also consider key circuit and system challenges in the design of neural interfaces and explore the opportunities that arise with the latest technology This Review Article examines the development of neural interfaces, which can provide a direct, electrical bridge between analogue human nervous systems and digital man-made devices, considering challenges and opportunities created with such technology.

88 citations

References
More filters
Journal ArticleDOI
TL;DR: This work used Drosophila melanogaster larvae to develop a high-throughput whole organism screen for drugs that modulate food intake and identified the serotonin (5-hydroxytryptamine or 5-HT) receptor antagonist metitepine as a potent anorectic drug.
Abstract: Dysregulation of eating behavior can lead to obesity, which affects 10% of the adult population worldwide and accounts for nearly 3 million deaths every year. Despite this burden on society, we currently lack effective pharmacological treatment options to regulate appetite. We used Drosophila melanogaster larvae to develop a high-throughput whole organism screen for drugs that modulate food intake. In a screen of 3630 small molecules, we identified the serotonin (5-hydroxytryptamine or 5-HT) receptor antagonist metitepine as a potent anorectic drug. Using cell-based assays we show that metitepine is an antagonist of all five Drosophila 5-HT receptors. We screened fly mutants for each of these receptors and found that serotonin receptor 5-HT2A is the sole molecular target for feeding inhibition by metitepine. These results highlight the conservation of molecular mechanisms controlling appetite and provide a method for unbiased whole-organism drug screens to identify novel drugs and molecular pathways modulating food intake.

2,329 citations

Journal ArticleDOI
TL;DR: This work uses a BCI to interpret pathological brain activity in patients with advanced Parkinson disease and to use this feedback to control when therapeutic deep brain stimulation (DBS) is delivered to improve on both the efficacy and efficiency of conventional continuous DBS.
Abstract: Brain-computer interfaces (BCIs) could potentially be used to interact with pathological brain signals to intervene and ameliorate their effects in disease states. Here, we provide proof-of-principle of this approach by using a BCI to interpret pathological brain activity in patients with advanced Parkinson disease (PD) and to use this feedback to control when therapeutic deep brain stimulation (DBS) is delivered. Our goal was to demonstrate that by personalizing and optimizing stimulation in real time, we could improve on both the efficacy and efficiency of conventional continuous DBS. We tested BCI-controlled adaptive DBS (aDBS) of the subthalamic nucleus in 8 PD patients. Feedback was provided by processing of the local field potentials recorded directly from the stimulation electrodes. The results were compared to no stimulation, conventional continuous stimulation (cDBS), and random intermittent stimulation. Both unblinded and blinded clinical assessments of motor effect were performed using the Unified Parkinson's Disease Rating Scale. Motor scores improved by 66% (unblinded) and 50% (blinded) during aDBS, which were 29% (p = 0.03) and 27% (p = 0.005) better than cDBS, respectively. These improvements were achieved with a 56% reduction in stimulation time compared to cDBS, and a corresponding reduction in energy requirements (p < 0.001). aDBS was also more effective than no stimulation and random intermittent stimulation. BCI-controlled DBS is tractable and can be more efficient and efficacious than conventional continuous neuromodulation for PD. Copyright © 2013 American Neurological Association.

1,017 citations

Posted Content
TL;DR: In this article, the temporal structure of single unit (SU) activity and simultaneously recorded local field potential (LFP) activity from area LIP in the inferior parietal lobe of two awake macaques during a memory-saccade task were investigated.
Abstract: A number of cortical structures are reported to have elevated single unit firing rates sustained throughout the memory period of a working memory task. How the nervous system forms and maintains these memories is unknown but reverberating neuronal network activity is thought to be important. We studied the temporal structure of single unit (SU) activity and simultaneously recorded local field potential (LFP) activity from area LIP in the inferior parietal lobe of two awake macaques during a memory-saccade task. Using multitaper techniques for spectral analysis, which play an important role in obtaining the present results, we find elevations in spectral power in a 50--90 Hz (gamma) frequency band during the memory period in both SU and LFP activity. The activity is tuned to the direction of the saccade providing evidence for temporal structure that codes for movement plans during working memory. We also find SU and LFP activity are coherent during the memory period in the 50--90 Hz gamma band and no consistent relation is present during simple fixation. Finally, we find organized LFP activity in a 15--25 Hz frequency band that may be related to movement execution and preparatory aspects of the task. Neuronal activity could be used to control a neural prosthesis but SU activity can be hard to isolate with cortical implants. As the LFP is easier to acquire than SU activity, our finding of rich temporal structure in LFP activity related to movement planning and execution may accelerate the development of this medical application.

953 citations

Journal ArticleDOI
TL;DR: This study provides Class I evidence that responsive cortical stimulation is effective in significantly reducing seizure frequency for 12 weeks in adults who have failed 2 or more antiepileptic medication trials, 3 or more seizures per month, and 1 or 2 seizure foci.
Abstract: Objectives This multicenter, double-blind, randomized controlled trial assessed the safety and effectiveness of responsive cortical stimulation as an adjunctive therapy for partial onset seizures in adults with medically refractory epilepsy. Methods A total of 191 adults with medically intractable partial epilepsy were implanted with a responsive neurostimulator connected to depth or subdural leads placed at 1 or 2 predetermined seizure foci. The neurostimulator was programmed to detect abnormal electrocorticographic activity. One month after implantation, subjects were randomized 1:1 to receive stimulation in response to detections (treatment) or to receive no stimulation (sham). Efficacy and safety were assessed over a 12-week blinded period and a subsequent 84-week open-label period during which all subjects received responsive stimulation. Results Seizures were significantly reduced in the treatment (-37.9%, n = 97) compared to the sham group (-17.3%, n = 94; p = 0.012) during the blinded period and there was no difference between the treatment and sham groups in adverse events. During the open-label period, the seizure reduction was sustained in the treatment group and seizures were significantly reduced in the sham group when stimulation began. There were significant improvements in overall quality of life (p Conclusions Responsive cortical stimulation reduces the frequency of disabling partial seizures, is associated with improvements in quality of life, and is well-tolerated with no mood or cognitive effects. Responsive stimulation may provide another adjunctive treatment option for adults with medically intractable partial seizures. Classification of evidence This study provides Class I evidence that responsive cortical stimulation is effective in significantly reducing seizure frequency for 12 weeks in adults who have failed 2 or more antiepileptic medication trials, 3 or more seizures per month, and 1 or 2 seizure foci.

948 citations

Journal ArticleDOI
TL;DR: The temporal structure of local field potential activity and spiking from area LIP in two awake macaques during a memory-saccade task was studied and it was found that LFP activity in parietal cortex discriminated between preferred and anti-preferred direction with approximately the same accuracy as the spike rate.
Abstract: Many cortical structures have elevated firing rates during working memory, but it is not known how the activity is maintained. To investigate whether reverberating activity is important, we studied the temporal structure of local field potential (LFP) activity and spiking from area LIP in two awake macaques during a memory-saccade task. Using spectral analysis, we found spatially tuned elevated power in the gamma band (25-90 Hz) in LFP and spiking activity during the memory period. Spiking and LFP activity were also coherent in the gamma band but not at lower frequencies. Finally, we decoded LFP activity on a single-trial basis and found that LFP activity in parietal cortex discriminated between preferred and anti-preferred direction with approximately the same accuracy as the spike rate and predicted the time of a planned movement with better accuracy than the spike rate. This finding could accelerate the development of a cortical neural prosthesis.

938 citations

Related Papers (5)