INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF NEURAL ENGINEERING
J. Neural Eng. 1 (2004) 72–77 PII: S1741-2560(04)77892-1
Using human extra-cortical local field
potentials to control a switch∗
Philip Kennedy1, Dinal Andreasen1,2, Princewill Ehirim3, Brandon King4,
Todd Kirby5, Hui Mao6 and Melody Moore7
1 Neural Signals Inc., 3688 Clearview Avenue, Atlanta, GA 30340, USA
2 Georgia Institute of Technology, Atlanta, GA, 30318, USA
3 Gwinnett Medical Center, Lawrenceville, GA, USA
4 Abbott Laboratories, Abbott Park, IL, USA
5 Respironics Inc., 1001 Murry Ridge Lane, Murrysville, PA 15668, USA
6 Department of Radiology, Emory University, 1364 Clifton Road, Atlanta, GA 30322, USA
7 Department of Computer Information Systems, Georgia State University, Atlanta, GA 30303, USA
E-mail: phlkennedy@neuralsignals.com
Received 16 March 2004
Accepted for publication 19 May 2004
Published 14 June 2004
Online at stacks.iop.org/JNE/1/72
doi:10.1088/1741-2560/1/2/002
Abstract
Individuals with profound paralysis and mutism require a communication channel. Traditional
assistive technology devices eventually fail, especially in the case of amyotrophic lateral
sclerosis (ALS) subjects who gradually become totally locked-in. A direct brain-to-computer
interface that provides switch functions can provide a direct communication channel to the
external world. Electroencephalographic (EEG) signals recorded from scalp electrodes are
significantly degraded due to skull and scalp attenuation and ambient noise. The present
system using conductive skull screws allows more reliable access to cortical local field
potentials (LFPs) without entering the brain itself. We describe an almost locked-in human
subject with ALS who activated a switch using online time domain detection techniques.
Frequency domain analysis of his LFP activity demonstrates this to be an alternative method of
detecting switch activation intentions. With this brain communicator system it is reasonable to
expect that locked-in, but cognitively intact, humans will always be able to communicate.
In the field of brain–computer interfacing there is much interest
in recording multi-unit and individual-unit activity for control
of devices in monkeys and humans. Studies from Nicholelis’
group [1] showed that monkeys can control both the direction
and force of a virtual object from several areas of cortex.
This work was preceded by that of Schwartz’s group [2]
and Donoghue’s group [3] who showed that monkeys can
control virtual targets, virtual arms or robot arms in three-
dimensional (3D) space. Andersen et al [4] demonstrated
that monkeys can control multiple cognitive variables using
signals recorded from parietal cortex. Observations from our
group [5] indicate that locked-in humans could slowly control
∗ Financial disclosure. Authors PK and DA may derive some financial
gain from the sale of this device. A patent has been applied under US and
international law: 10/675,703.
a computer cursor to drive a spelling device to restore synthetic
speech. During these recordings, we noted that high-amplitude
local field potentials (LFPs) recorded intra-cortically were
available for possible prosthetic purposes. We described these
observations in two patients who drove the cursor across
the screen into a target or drove the digits on a cyber
hand [6]. These LFPs have such large amplitudes and
were so readily available that they raised the possibility of
recording LFPs extra-cortically to provide at least a switch
function. If so, then extra-cortical recordings might allow
locked-in patients to communicate using digital control of
presently available application software for spelling, speech
production, Internet access and so on. The system was
seen as a simplified intra-cranial implant that would avoid
the logistical problem of implanting electrodes into the
brain.
1741-2560/04/020072+06$30.00 © 2004 IOP Publishing Ltd Printed in the UK 72

Using human extra-cortical local field potentials to control a switch
With appropriate permissions [7], we implanted
electrodes into two amyotrophic lateral sclerosis (ALS)
patients. The first patient (GT) refused ventilatory support
and died in his sleep after some useful data were acquired
as discussed below. The second patient (RR), whose data
form the bulk of this paper, initially underwent a functional
magnetic resonance image to determine his bilateral upper and
lower extremity cortical representations (figure 1(a)). Because
he had some slight movement of his right foot and routinely
used this to operate a call bell, we chose to implant above
this part of area 4 motor cortex as well as the hand area on
the same side, as shown in figure 1(a), right. This skull x-
ray shows the two pairs of stainless steel skull screws that
were implanted through the full thickness of the skull. The
second screw of each pair was implanted rostral to area 4
and positioned over cortex that was not expected to be active
during movement or imagined movement. Recordings were
differential and were transmitted transcutaneously from an FM
transmitter. Other implanted electronics included the power
induction system and calibration signal as previously described
[5]. Other screws seen in the figure (arrow) were retaining the
implanted electronics and one was for grounding the power
supply.
The surgical procedure was fully described in [6].
Briefly, after permission was obtained, the subject was fully
anesthetized and the scalp prepped in the standard sterile
fashion. The targets were localized and a ‘C’ shaped incision
was made to include the targets and the expected space for the
electronics. The skull screws were driven approximately 1 mm
below the inner table of the skull after drilling and tapping the
hole. The electronics were attached via stainless steel Teflon
insulted wires. Acrylic was used to insulate any exposed wires
and to retain the electronics in place after which the scalp was
closed in layers.
After allowing at least three weeks for healing of the
incision, recording was achieved by powering the electronics
with an external power induction coil and transmitting the
signal via RF to a radio receiver integrated with a computer.
Observations were made as shown in figure 1(b). This diagram
illustrates the subject in the lying position. He was turning on
and off a light switch using electromyographic (EMG) activity
recorded from an electrode placed over the Hallucis Longus
muscle in his right foot. This activity was amplified and a
voltage level detected threshold crossings that were converted
into pulses that activated the switch. The threshold level
was selected by the investigator who observed the increase in
LFP amplitude during EMG-activated switch closures. Once
selected, it was not adjusted. Concurrent with the EMG
activity, the LFPs from the electrode over the foot area were
converted into pulses in the computer. The foot pressure plate
was used only as a monitor of attempted movement or foot
pressure and to provide feedback to the subject. To avoid any
learning or adaptation effects, the subject was unaware when
the output from the computer processor to the light switch was
changed from the thresholded EMG activity to the thresholded
LFP activity. After this change, the subject continued to turn
the light switch on and off using LFP activity alone.
We were able to obtain thresholding data during two
separate recording sessions. During the first session (030502),
LFPs
EMG
Foot
plate
Subject
Light on/off
(a)
(b)
Figure 1. (a) The functional MRI of subject RR is shown on the left
where activity is seen in the left leg area during attempted foot
movements. This activity was used to localize the target for
implantation of skull screw electrodes. These targets were aligned
using a stereotaxic system during surgery. The implanted skull
screws are shown in the x-ray on the right along with the two
amplifiers and three retaining screws. One of the retaining screws
was also used as a ground. The medially placed screws were used to
record LFPs from the leg area and the laterally placed screws from
the arm area. (b) Subject RR lay in bed during recording of LFPs
and EMG activity. His task was to turn the light on and off
sequentially. The foot pressure plate monitors foot pressure and
attempted movements. The LFP and EMG activity was routed to the
processor to control the light switch.
there were 26 switch activations using EMG and then, after
switching the system to thresholded LFP activity, there were
27 switch activations. During the second session (031102),
there were 12 switch activations using LFP only. An example
of raw EMG and LFP data is shown in figure 2 during self-
paced LFP-activated switch closures. The top row shows 5 s
of LFP activity with increases in amplitude and variations in
frequency. The threshold line is shown above the LFP activity
and the pulses resulting from crossing the line are shown below
as vertical dash marks. The EMG burst and tonic activities are
shown in the next row, with the switch output voltage below.
As seen in the top row, LFP amplitude and frequency changes
precede the EMG burst but the switch voltage change (lower
row) was not triggered because only one threshold crossing
occurs. The voltage was triggered when a repeating pattern
of LFP activity was produced as shown by the vertical dash
marks. To activate the switch, two repeated crossings of the
LFP threshold were required.
We compared the EMG activity with LFP activity during
threshold crossings for the 26 and 27 switch closures in session
030502 that were obtained during self-paced activity. Both
EMG (figure 3, left) and LFPs (right) showed increases in
activity prior to and during switch activation. These data
demonstrate that LFP activity was capable of substituting for
73