FEATURE ARTICLE
The Design and Function of Cochlear Implants
Fusing medicine, neural science and engineering, these devices transform human speech into an electrical code that deafened ears can understand
Michael Dorman, Blake Wilson
Ludwig van Beethoven was 28 years old when he first noticed a ringing
and buzzing in his ears. Soon he was unable to hear high notes from
the orchestra; speech became indistinct. By 1802, four years after
the first symptoms, he was profoundly deaf.
Beethoven fell into a deep depression. He describes this period in
his Heiligenstadt Testament, meant to be read after his death:
For me there can be no relaxation in human society; no
refined conversations, no mutual confidences. I must live quite
alone and may creep into society only as often as sheer necessity
demands.... Such experiences almost made me despair, and I was on
the point of putting an end to my life—the only thing that
held me back was my art ... thus I have dragged on this miserable
existence.
In 2001, Scott N. was 34 and had lost all of his hearing. A surgeon
inserted 16 tiny electrodes into his inner ear, or cochlea,
and connected them to a small package of electronics implanted under
the skin. A year later, Scott came to author Dorman's laboratory at
Arizona State University to test his understanding of speech. The
results were extraordinary: Scott recognized 100 percent of more
than 1,400 words, either in sentences or alone, without any prior
knowledge of the test items.
As impressive as this performance was, the cochlear implant did not
restore normal hearing to Scott. The electrode array produced a
stimulus that was only a crude mimicry of the signals in a normal
cochlea. But as this example shows, a very high level of
functionality can be restored by a neural prosthesis that does not
recreate the normal system. For the thousands of people who have
received a cochlear implant, even an imperfect restoration of
hearing reconnects them to the world of sound. And it allows many of
them to use that most critical toy of modern life, the cell phone.
Although cochlear implants have a 40–year history culminating
in the current generation of high–performance devices, hearing
restoration is not universally welcomed. Among members of the Deaf
community, the absence of hearing is not necessarily viewed as a
disability. Some deaf parents refuse implants for their deaf
children, triggering an impassioned debate between those who agree
and those who challenge the decision. This article avoids that
controversy to focus on the science of cochlear implants. But recent
findings have influenced the temperature, if not the substance, of
the debate. As we point out, hearing must be restored at a very
early age if speech and language skills are to develop at a normal
rate. The decision to use or forgo the implant cannot wait until the
child—who must bear the consequences—reaches the age of consent.
Hear Here
In normal hearing, sound waves traveling through air reach the
tympanic membrane (ear drum) via the ear canal, causing
vibrations that move the three small bones of the middle ear. This
action produces a piston–like movement of the oval
window, a flexible membrane in the bony shell of the cochlea.
Inside the fluid–filled cochlea, oscillations from the oval
window initiate a traveling wave along the basilar membrane
(one that divides the cochlea along its length). Another flexible
membrane, the round window, moves in a complementary way to
maintain the volume of the incompressible fluid in the cochlea.
The basilar membrane has graded mechanical properties. At the base
of the cochlea, near the oval and round windows, it is narrow and
stiff. At the other end of the cochlea, the apex, the basilar
membrane is wide and flexible. These mechanical properties give rise
to a traveling wave of displacement and to points of maximal
response according to the frequency or frequencies of the pressure
oscillations. For a wave with a single frequency, displacement
increases up to a particular point along the membrane and then drops
precipitously. High frequencies produce maxima near the base of the
cochlea, whereas low frequencies produce maxima near the apex.
Movements of the basilar membrane are sensed by a line of hair
cells, which are attached to the top of the membrane in a matrix
called the organ of Corti. Each hair cell has fine rods of protein,
called stereocilia, emerging from one end. When the basilar
membrane moves, these rods bend as if they were hinged at their
bases. The deflection initiates a chain of electrochemical events
that causes electrical spikes, or action potentials, in
cells of the spiral ganglion. These cells conduct the signal to a
relay station in the brainstem called the cochlear nucleus.
The information ascends through multiple other nuclei on its way to
the auditory cortex, the portion of the forebrain that
processes auditory information.
Within this circuit of cells, a sound's frequency is encoded by two
mechanisms. The first is a place code, which indicates the spot
along the tapered basilar membrane that moves the most. Stereocilia
on the hair cells respond to this displacement and cause action
potentials among the closest spiral–ganglion neurons. The
second mechanism is a temporal code that is produced when neurons
become synchronized, or phase–locked, to the period
of an acoustic wave. In normal hearing, neural responses can easily
match frequencies up to about 1,000 hertz. This phase–locking
ability declines progressively at higher frequencies. The perception
of frequency is probably based on some combination of place and
temporal codes, with the temporal code being effective for low
frequencies only.
Hearing is lost when hair cells become so damaged that they cannot
stimulate cells of the spiral ganglion. Without regular activity,
the portion of that ganglion cell that receives signals, the
dendrite, may atrophy and the cells may die. Fortunately, even in
the case of complete hearing loss, some spiral–ganglion cells
survive and remain connected to the appropriate
frequency–receiving areas in the cochlear nucleus. If
electrical current from the implanted electrodes can cause action
potentials among the remaining cells, then hearing can be restored.
And if multiple groups of neurons (think of these as "neural
channels") can be made to respond in low, middle and high
frequency parts of the cochlea, then perception of speech can be
restored as well.
Slicing the Spectrum
How many channels, these slices of the frequency spectrum, are
necessary to restore speech understanding? Following the early work
of Robert Shannon and others at the House Ear Institute in Los
Angeles, one of us (Dorman) answered this question with collaborator
Philip Loizou of the University of Texas at Dallas. We used
so–called bandpass filters to divide the spectrum of speech
into a relatively small number of frequency bands or channels. A
microprocessor measured the energy in each band every few
milliseconds and transformed the signal into
amplitude–modulated sine waves, each centered on one of the
frequency bands. When we played these simplified audio signals to
normal–hearing listeners, they understood 90 percent of the
words in simple sentences, even when we used as few as four
channels. Eight channels allowed them to identify 90 percent of
isolated words. Under noisy conditions, more channels were needed to
match this performance, and the more channels used, the better the
comprehension. These observations show that in a quiet environment,
speech can be well understood with a relatively small number of
channels—a fact that is central to the success of cochlear implants.
The variability of normal speech also helps. As our brain decodes
speech sounds, it uses cues, called formants, which
identify consonants and vowels by specific concentrations of energy
in the frequency spectrum. But formants do not have a fixed
frequency, even for the same sound, because vocal–tract
geometry varies from speaker to speaker. So instead of being
discrete points, the acoustic signatures of speech sounds can be
thought of as ellipses in frequency space. Even in infants, the
system that perceives speech is designed to be flexible so that it
can "hear through" variations in the signal to identity a
consonant or vowel. This flexibility allows a very reduced
description of speech to be recognized with accuracy.
Figure of Speech
The timing and frequency of consonants and vowels in a spoken word
determine its acoustics. For example, slow changes in overall
amplitude indicate the timing of syllables, phonetic transitions
within syllables and boundaries between silence and sound. In terms
of frequency, the vocal tract produces multiple concentrations of
energy between 300 and 5,000 hertz as it produces speech sounds.
The slow amplitude variations of speech are referred to as the
speech envelope, an aspect that conveys a surprising amount of
information. Victor Zue at the Massachusetts Institute of Technology
classified the envelope shapes of 126,000 words by applying a series
of only six shape variations. He found that, on average, only 2.4
word candidates matched a given sequence. This observation suggests
that implant patients could understand speech much better if their
implants conveyed the shape of the envelope, thereby constraining
the number of word possibilities. However, envelope shape by itself
does not provide enough information to understand speech. To
identify specific words, frequencies in the 300–5,000 hertz
range must be extracted from the signal.
In any vowel or consonant, the frequencies of the first two energy
concentrations comprise the essential signature of the sound. For
example, in Dorman's voice the vowel in bat has energy peaks at 624
and 904 hertz. The vowel in bought has peaks at 620 and 1,055 hertz.
Because a very small difference in the acoustic pattern—150
hertz in this case—can significantly alter the meaning of the
word, investigators initially assumed that a neural prosthesis for
hearing would need a very large number of channels. As we have seen,
this did not turn out to be the case, at least for low–noise environments.
Hardware For Hearing
In a deafened ear, hair–cell failure severs the connection
between the peripheral and central auditory systems. Cochlear
implants restore the link, bypassing hair cells to stimulate
directly the cell bodies in the spiral ganglion.
A cochlear implant has five main components, only two of which are
inside the body. Above the outer ear, an external microphone picks
up sounds in the environment and directs them to a sound processor,
which sits inside a case behind the ear. The processed signals are
conveyed to a high–bandwidth radio–frequency
transmitter, which beams the information through a few millimeters
of skin to a receiver/stimulator that has been surgically implanted
in the temporal bone above the ear. The signals then pass to an
array of electrodes inside the cochlea. Target cells in the spiral
ganglion are separated from the electrodes by a bony partition.
Scott N.'s device uses the continuous interleaved sampling, or CIS,
strategy to convert acoustic signals into a code for stimulating the
auditory nerve. One of us (Wilson), along with colleagues at the
Research Triangle Institute and Duke University, developed the CIS
strategy. It starts by filtering a signal into frequency bands (16
for Scott N.). For each band, the CIS algorithm converts the slow
changes of the sound envelope into amplitude–modulated trains
of biphasic (having negative and positive components) pulses at the
electrodes. The processor sends information from low–frequency
channels to electrodes in the apex and information from
high–frequency channels to electrodes in the base of the
cochlea. This organization maintains the logic of the frequency map
in a normal cochlea.
Adult Results
Scott N.'s ability to understand speech demonstrates that an implant
can restore a normal level of speech recognition in quiet
environments. However, Scott's case is exceptional. Speech is
neither as clear nor as easy to understand for most patients.
Although average scores range between 80 and 100 percent correct on
tests of sentence understanding, the comprehension of isolated words
lies between 45 and 55 percent. The gap between scores shows that
average patients fail to hear the details of many spoken words.
Sentence context allows the missing elements to be reconstructed.
What is the difference between Scott's auditory system and that of
a patient with average or below–average speech understanding?
A patient's performance probably depends on many factors, including
the number and location of surviving cells in the spiral ganglion,
the spatial pattern of current flow from the electrodes, and the
degree to which neurons in the brainstem and cortex can encode
frequency by phase–locking their firing patterns. When only a
few cells survive in the spiral ganglion—for example, after a
long period of deafness—the electrode stimulation is less able
to convey frequency–specific information to the cochlear
nucleus and cortex. And if the surviving cells are clustered at one
end of the ganglion, then the signal that does arrive at the cortex
will lack the range of frequencies needed to understand speech. Even
if there are neurons along the length of the cochlea, individual
electrode currents need to be highly focused to provide independent
channels of stimulation (and therefore, information). If these
currents overlap, either because the signal spreads too far through
the conductive cochlear fluid or because of individual differences
in cochlear anatomy, then the number of functional channels will be
less than the number of electrodes.
In the Ears of a Child
Adults who lose their hearing and later receive a cochlear implant
can associate the new stimulation patterns with their memories of
what speech should sound like. Children born deaf do not have this
luxury. Yet a team led by Richard Miyamoto and Mario Svirsky at
Indiana University has found that congenitally deaf children who
receive a cochlear implant during their first or second years can
learn spoken language at a normal or near–normal rate. These
children can enter first grade with age–appropriate language
skills—a testament to the adaptive ability of young neural
systems. This plasticity undoubtedly plays a major role in the
success of implants at an early age.
Scientists can observe the neural changes in young children fitted
with implants using the tools of neurophysiology. Author Dorman and
his colleagues at Arizona State University, in collaboration with
Anu Sharma and her team at the University of Texas at Dallas, found
that the brains of deaf children under the age of four are quickly
reconfigured in response to the signals from an implant. Using
electrodes on the scalp, we were able to record sound–evoked
electrical activity in the cortex. Within a week after the implant
was activated, we saw changes in the latency of neural responses to
sound. Within six months, children who had heard nothing for up to
three and a half years showed age–appropriate timing of
cortical activity in response to sound.
Children who receive the implant after their seventh birthday have
less success than younger patients in developing speech and oral
language. We saw corresponding evidence for this age limit in the
cortical–latency experiments. After an initial change, the
delay of cortical activity in response to sound remained abnormally
long in older children, even after considerable experience with the implant.
Sadly, the same property that helps the implant work so well in
preschoolers limits its effectiveness for older children. During the
extravagant growth of neural connections during the first years of
life, areas of the brain that lack stimulation can be usurped or
recruited to process active signals that usually go to other parts
of the brain. In this case, regions that would normally analyze
auditory inputs might be appropriated by the spread of visual or
other sensory connections as the child gets older. And once an area
is allocated to a different task, returning to the original task is
difficult or impossible, depending on age. This narrow window of
opportunity has also been observed in animal experiments.
The different outcomes of implants in younger and older children
reflect different patterns of neural organization in the children
prior to implantation. Using positron–emission tomography
(PET), Dong Soo Lee and his colleagues at the Seoul National
University found extremely low activity in the auditory cortex and
surrounding brain areas in children who were deaf for a relatively
short period—which is what one would expect given that there
was no auditory input. This group of children adapted well to
cochlear implants.
However, in children deprived of sound for more than 7 years, PET
scans before the implant surgery showed a more normal level of
activity in the auditory cortex and language areas. Because this
cortex was not activated by auditory input, it must have received
input from some other sense—probably vision. It is reasonable
to suppose that the encroachment of other functions into brain areas
normally devoted to auditory processing is one reason that older
children have a much more difficult time acquiring speech and oral
language skills after receiving the cochlear implant. This
biological reality adds an important codicil to the debate over
cochlear implants for the deaf children of deaf parents. By the time
a deaf child reaches the age at which he or she might elect to have
an implant, it will be too late to achieve the best outcome.
An Elusive Pleasure
Scott and others like him who achieve high levels of word
recognition report that speech sounds natural and clear through the
implant. No patient (in our experience) has described music in this fashion.
This result points to a fundamental difference in the requirements
for speech understanding and music appreciation. Implants do not
need to reproduce the precise frequencies of speech to preserve
meaning. But precision is absolutely essential for music. An octave,
for example, cannot be stretched in the way that frequency
components for speech can be stretched. If the A above A440 is to be
heard as an octave higher, then an implant must convey a signal at
880 hertz. A small error yields a different note. Although we have
had a small amount of success using octave intervals to tune signal
processors for a few patients with extensive musical backgrounds,
creating pleasant—or even tolerable—musical experiences
for the majority of cochlear–implant patients remains an
elusive goal.
The Next Verse
One advance that we will see shortly is the union of electric and
acoustic stimulation, or combined EAS. Many hearing–impaired
people have some ability to hear low frequencies but retain little
or no hearing at higher frequencies. If an electrode array can be
inserted about two–thirds of the way into the cochlea, then
hearing at 1 kilohertz and above can be restored by electrical
stimulation. And if the surgery doesn't damage the distal third of
the cochlea, then electrical and acoustic hearing can together
provide access to the range of frequencies necessary for speech understanding.
Christoph von Ilberg and his colleagues at the University Clinic at
Frankfurt were the first to demonstrate the feasibility of this
approach. Recent studies have shown that acoustic hearing can be
preserved in 75 to 90 percent of patients in whom a 20
millimeter–long electrode array is inserted into the cochlea,
which is normally 28 to 35 millimeters long. Experiments from author
Wilson's lab have shown that just a small region of acoustic hearing
below 500 hertz greatly improves the performance of electrical
hearing, even when acoustic speech comprehension is near zero. For
example, one patient who understood only 10 percent of words via
acoustic stimulation and 60 percent by electric stimulation
recognized 90 percent with the combined stimulation.
We suspect that auditory nuclei in the brainstem, which sort signals
from noise, recognize patterns of neural discharge that are unique
to acoustic stimulation. The output from even a small region of
normal hearing may engage these nuclei in a way that electrically
evoked patterns cannot, thereby allowing more of the signal to reach
higher levels of auditory processing. Thus the combination of
electric and acoustic stimuli can have a synergistic effect on
speech understanding, especially in noisy environments.
Combined EAS has produced some remarkable results for patients with
residual hearing in the low frequencies, and patients with residual
hearing up to 1,000 hertz may one day become candidates for the
procedure. The popularity of this approach as a treatment for
severe, but not total, hearing loss will depend on how reliably the
remaining hearing can be preserved. Such preservation might be
improved with shorter electrode insertions or with
pre–treatment of the cochlea with certain drugs. However,
shorter arrays also reduce the performance of electric
stimulation—leaving the patient with few options if the
remaining hearing is lost. These trade–offs—electrode
insertion depth versus preservation of unaided hearing, combined EAS
performance versus the performance of electric stimulation
alone—remain to be fully explored.
Better Hearing Through Chemistry
In the near future, drug–delivery systems will be integrated
into the design of a cochlear implant. These systems will attempt to
do two things: arrest the shriveling or demise of remaining hair
cells and neural structures in the cochlea, and promote the growth
of neural tentacles called neurites from
spiral–ganglion cells toward the electrodes. If neurons in the
vicinity of each electrode can be kept alive, and especially if they
are brought closer to the electrodes with the growth of neurites,
then each electrode is more likely to function as an independent
channel of stimulation.
One approach is to inject growth–promoting neurotrophins into
the cochlea. In experiments with deafened guinea pigs, Takayuki
Shinohara and his coworkers at the Karolinska Institute in Stockholm
showed that by injecting brain–derived neurotrophic factor and
ciliary neurotrophic factor, they could increase the survival and,
critically, the sensitivity of spiral–ganglion cells. This
outcome hints at future implant designs in which neurites from
spiral ganglion cells grow toward multipurpose electrodes that
deliver electrical and pharmacological stimuli.
A second approach is to block apoptosis, the normal process
of cell death following injury. These self–destruct messages
can be triggered by many events, such as acoustic trauma or ototoxic
drugs, which work through a so–called mitogen–activated
protein kinase (MAPK) signaling pathway. The pathway can be blocked
at various points. One of the links in this chain is the protein
called c–Jun N–terminal kinase (JNK). This enzyme is the
target of a peptide inhibitor developed by a multi–center,
multi–national team that includes Jing Wang of the University
of Montpelier and Thomas Van De Water of the University of Miami. By
blocking JNK, they prevented hair–cell death and hearing loss
following acoustic trauma or administration of the ototoxic
antibiotic neomycin.
This outcome is especially relevant for future applications of
combined electric and acoustic stimulation. Injecting a
MAPK–JNK blocker could buffer existing hair cells from damage
caused by the surgery. In that case, the odds of preserving acoustic
hearing might increase, making combined EAS into a viable therapy
for a very large number of hearing–impaired people.
Imagining Beethoven Today
We wonder how Beethoven might feel if he were alive today and had
received a cochlear implant. We expect he would understand speech
well enough to "relax in human society" and engage in
"refined conversations" and "mutual
confidences." He would avoid the isolation that caused his
despair. The sound of his art, however, would certainly fail to
bring him joy. We will need many more years of hard work and good
luck to make this time–travel story end with an idyllic, or,
if you like, a pastoral tune.
Bibliography
- Dorman, M. F., P. C. Loizou and D. Rainey. 1997. Speech intelligibility as a function of the number of channels of stimulation for signal processors using sine?wave and noise?band outputs. Journal of the Acoustical Society of America 102:2403–2411. [CrossRef]
- Lee, D. S., J. S. Lee, S. H. Oh, S. K. Kim, J. W. Kim, J. K. Chung, M. C. Lee and C. S. Kim. 2001. Cross?modal plasticity and cochlear implants. Nature 409:149–150. [CrossRef]
- Miyamoto, R. T., D. M. Houston, K. I. Kirk, A. E. Perdew and M. A. Svirsky. 2003. Language development in deaf infants following cochlear implantation. Acta Otolaryngologica 123:241–244.
- Sharma, A., M. F. Dorman and A. J. Spahr. 2002. Rapid development of cortical auditory evoked potentials after early cochlear implantation. Neuroreport 13:1365–1368.
- Shinohara, T., G. Bredberg, M. Ulfendahl, I. Pyykko, N. P. Olivius, R. Kaksonen, B. Lindstrom, R. Altschuler and J. M. Miller. 2002. Neurotrophic factor intervention restores auditory function in deafened animals. Proceedings of the National Academy of Sciences of the U. S. A. 99:1657–1660.
- von Ilberg, C., J. Kiefer, J. Tillein, T. Pfenningdorff, R. Hartmann, E. Sturzebecher and R. Klinke. 1999. Electric?acoustic stimulation of the auditory system: New technology for severe hearing loss. ORL: Journal for Oto?Rhino?Laryngology and Its Related Specialties 61:334–340.
- Wang, J., T. R. Van De Water, C. Bonny, F. de Ribaupierre, J. L. Puel and A. Zine. 2003. A peptide inhibitor of c?Jun N?terminal kinase protects against both aminoglycoside and acoustic trauma?induced auditory hair cell death and hearing loss. Journal of Neuroscience 23:8596–8607.
- Wilson, B. S. 2004. Engineering design of cochlear implants. In Auditory Prostheses: Cochlear Implants and Electric Hearing, ed. F.?G. Zeng, A. N. Popper and R. R. Fay. New York: Springer?Verlag.
- Wilson, B. S., C. C. Finley, D. T. Lawson, R. D. Wolford, D. K. Eddington and W. M. Rabinowitz. 1991. Better speech recognition with cochlear implants. Nature 352:236–238. [CrossRef]
- Zue, V. 1985. The use of speech knowledge in speech recognition. Proceedings of the IEEE 73:1602–1615.