1287_ch13.pdf

Section IV

Reconstruction of Movements

Using Brain Activity

Advances in

Brain–Machine

Interfaces

Jose M. Carmena and Miguel A.L. Nicolelis

CONTENTS

13.1 Introduction

13.1.1 Invasive and Noninvasive BMIs

13.2 BMI Design

13.2.1 Chronic, Multisite, Multielectrode Recordings

13.2.2 Data Acquisition and Telemetry

13.2.3 Bidirectional BMIs: Decoding and Encoding

13.3 Reaching and Grasping with a BMI

13.3.1 Neuronal Variability

13.4 Future Directions

13.4.1 Continuous Shared-Control BMIs

13.4.2 Neuroprostheses: Translating BMIs to the Clin ic

Acknowledgments

References

13.1 INTRODUCTION

Throughout history, the introduction of new technologies has signiﬁcantly impacted

human life in many different ways. Until now, however, each new artiﬁcial device

or tool designed to enhance human motor, sensory, or cognitive capabilities has

relied on explicit human motor behaviors (e.g., hand, ﬁnger, or foot movements),

often augmented by automation, in order to translate the subject’s intent into concrete

goals or ﬁnal products. The increasing use of computers in our daily lives provides

a clear example of such a trend. Yet, the realization of the full potential of the “digital

revolution” has been hindered by its reliance on low bandwidth and relatively slow

user–machine interfaces (e.g., keyboard, mouse). Because these user–machine inter-

faces are far removed from how the brain normally interacts with the surrounding

environment, the potential of such a tool is limited by its inherent inability to be

assimilated by the brain’s multiple internal representations as a continuous extension

of our body appendices or sensory organs.

0-8493-1287-6/05/$0.00+$1.50

Two decades ago, an alternative method was proposed

for restoring motor behav-

This method proposed a bypass of the spinal

cord and started a new paradigm, namely the interface between brains and machines,

or brain–machine interfaces (BMIs). This approach contends that paralyzed patients

could enact their voluntary motor intentions through a direct interface between their

brains and artiﬁcial actuators in virtually the same way that we see, walk, or grab

an object. The studies in macaque monkeys conducted by Fetz and collaborators

2–5

were the ﬁrst experimental support for a cortical driven BMI. In fact, recent BMI

research in animals and humans has supported the contention that we are at the brink

of a technological revolution, where artiﬁcial devices may be “integrated” in the

multiple sensory, motor, and cognitive representations that exist in the primate brain.

These studies have demonstrated that animals can learn to utilize their brain activity

to control two-dimensional displacements of computer cursors,

6,7

8,9

and, more

In addition to the current

research performed in rodents and primates, there are also preliminary studies using

human subjects.

11–13

The ultimate goal of this emerging ﬁeld of BMIs is to allow human subjects to

interact seamlessly with a variety of actuators and sensory devices through the

expression of their voluntary brain activity, either for augmenting or restoring sen-

sory, motor, and cognitive function — e.g., after a traumatic lesion of the central

nervous system. Moreover, by providing ways to deliver sensory (visual, tactile,

auditory, etc.) feedback from these devices to the brain, one could establish a

reciprocal (and more biologically plausible) interaction between large neural circuits

and machines, hence fulﬁlling the requirements for artiﬁcial actuators to be recog-

nized as simple extensions of our bodies.

In addition to the potential clinical application, BMIs also serve as a unique tool

for systems neuroscience research. The combination of multiple-site, multiple-elec-

trode recordings

13.1.1 I

NVASIVE

AND

ONINVASIVE

BMI

The noninvasive approach in BMIs utilizes features of brain activity, such as event-

related responses or continuous electroencephalogram (EEG) rhythms, to control a

computer-based device. These devices, commonly known as brain–computer inter-

faces (BCIs), record brain activity from surface electrodes positioned on the scalp.

They typically consist of a computer screen on which a subject, after training, can

control the selection of characters by moving a cursor up, down, left, and right to

operate simple word processing programs or indicate a particular action to a care-

giver.

iors in severely paralyzed patients.

one-dimensional

to three-dimensional movements of simple and elaborate robot arms

recently, reaching and grasping movements of a robot arm.

with the BMI paradigm provides the experimenter with a new way

to quantify neurophysiological modiﬁcations occurring in cortical networks, as ani-

mals learn motor tasks of various complexities.

The main issue with this approach is the small communication bandwidth

currently available. Still, with the aid of artiﬁcial intelligence and robotics, more

complex tasks can be achieved using this approach. Such is the case with autonomous

wheelchair navigation. Millán and colleagues have devised an EEG-based system

that discriminates among the recorded signals that are generated in different mental

This system exploits the

diversity of brain waves during different mental exercises, which has no real corre-

lation with behavioral outcome — i.e., no decoding of motor actions is performed

by the BCI. Nevertheless, the noninvasive approach lacks the spatial resolution

and bandwidth necessary for extracting the kind of time-varying motor signals that

would be necessary to control accurate three-dimensional arm movements in real

time, as would be needed for prosthetic devices.

17,18

The invasive approach typically uses extracellular recordings of individual neu-

rons through chronically implanted microelectrodes in the cortex.

6–10

Other

use local ﬁeld potentials, which offer more resolution than the EEG

recordings on the scalp, but still contain less information than extracellular single-

neuron recordings. In this chapter, however, we will focus on invasive BMIs that

use arrays of microelectrodes chronically implanted in the cortex of macaque monkeys.

13.2 BMI DESIGN

In this section, the state of the art in BMI design will be reviewed, together with

some discussion of speciﬁc issues. Figure 13.1 illustrates the different parts that

form a closed-loop cortical BMI.

13.2.1 C

HRONIC

, M

ULTISITE

, M

ULTIELECTRODE

ECORDINGS

The capability of recording the activity of many single cortical neurons for long

periods of time in awake, behaving macaque monkeys or rodents is a powerful tool

that permits neurophysiological investigation of learning, perception, and sensori-

motor integration. Moreover, BMIs in humans will require electrodes to be implanted

chronically for long periods of time, raising issues on the quality and stability of

the recordings, and on the biocompatibility of the materials.

Recent attempts to obtain long-lasting, single-neuron recordings from macaques

have employed the 100-electrode “Utah array” or arrays of individual sharp micro-

wires.

This technique permits high quality single-unit recordings for long

periods of time in macaque monkeys.

A multineuron acquisition processor (MAP)

(Plexon, Inc., Dallas, TX) cluster, formed by three 128-channel MAPs synchronized

by a common clock signal, was specially built for simultaneous recordings from

hundreds of neurons in real time as reported in this study. This 384-channel recording

system has a theoretical capacity of recording up to 1536 single neurons simulta-

neously (e.g., 4 neurons per channel), at a 25-

s precision. Among other results, the

activities, such as adding numbers, thinking of a family member, imagining geo-

metric shapes, etc. Once identiﬁed, these signals are translated into high-level com-

mands that actually control the navigation of the wheelchair. Lower-level actions,

such as path planning and obstacle avoidance, are performed by the system through

the reading of the sensors attached to the wheelchair.

researchers

However, these studies have provided relatively modest neuronal yields

of uncertain longevity, and, in most cases, they have thus far been limited to just

one or two cortical areas per animal.

Nevertheless, progress in the development of high-density microwire arrays

during the past years has resulted in the standardization of this technique in rodents

6,7,20–23

and primates.

FIGURE 13.1

General architecture of a closed-loop control brain–machine interface.

study demonstrates the simultaneous recording of extracellular activity of 247 single

cortical neurons from 384 microwires implanted in multiple cortical areas of the

brain of a macaque monkey 30 days after the implantation surgery. In a different

monkey, recordings were obtained from up to 58 isolated neurons 18 months after

surgery. The success of this technique has been crucial for our BMI work.

9,10

13.2.2 D

ATA

CQUISITION

AND

ELEMETRY

The next step toward a ﬁnal BMI product will be to perform unsupervised (i.e.,

automatic) spike detection in real time from the extracellular recordings. This detec-

tion should be reliable so that spikes can be separated from the background noise

before they are sorted and transmitted via a transcutaneous wireless telemetry device.

Spike information will then be sent to a processing device implanted or carried

elsewhere on the subject’s body.

Since the critical issues in the components of a BMI system are small size and

low power consumption, the computational bandwidth of the system will be limited.

Obeid and colleagues are working on a wearable multichannel neural telemetry

system that would suit the needs of a BMI. The current version of their system

allows sampling, digitizing, processing, and transmission of 32 channels.

26,27

Further

versions of this system currently under development will increase the number of

channels substantially. Obeid et al. have also investigated different classes of real-

time spike detection algorithms. A computationally cheap method, such as the

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: