Editor: This article summarizes the presentation given by Robert Shannon PhD, at the CSPP Cochlear Implant Workshop on February 12, 2000. In his presentation Dr. Shannon described what we know about cochlear implant performance, how we know it, and how we hope to use that knowledge to improve the performance of implants in the future.

This article is a bit technical. My apologies to those of you who have a tough time getting through it. I've tried to present conclusions in something close to plain English, so our non-technical readers can get the main points even if they miss some of the details.

~~~~~~~~~~~~~~~~~~~
CSPP's Cochlear Implant Workshop Sponsored
San Diego CA, February 12, 2000
Robert Shannon PhD
Director, Auditory Implant Research Laboratory
House Ear Institute

Three Mysteries

Some of the general statements that define the current state of cochlear implant include:
- Cochlear implants perform better than we ever dreamed they would.
- All current multi-channel systems allow equivalent performance despite large differences in the number of channels and signal processing strategies.
- There is great diversity in the individual performances of people using cochlear implants.

Scientists don't currently know why any of these statements is true. Their research is focusing on the answers to these mysteries in the belief that the answers will produce the next big advances in cochlear implant technology.

Scientific Background

Acoustic scientists characterize sound by its amplitude, temporal and spectral attributes. These terms refer to its loudness (amplitude), how it changes with time (temporal), and how it is characterized in frequency (spectral). Current research indicates that cochlear implant users perform about as well as can be expected on the amplitude and temporal aspects of a sound, but that they don't do as well on spectral attributes.

Researchers test the effects of sound amplitude by changing the amplitude and measuring the results. One particularly revealing test involves the compression or expansion of the amplitude range of a sentence and recording people's ability to understand the sentence. Their findings are that extremely large amplitude distortions (compressions or expansions) must be applied before normally hearing people can no longer understand the sentence. Their conclusion is that amplitude mappings are relatively unimportant, and that the amplitude fidelity provided by current technology is plenty good.

The second group of attributes is temporal attributes, which describe how a sound varies with time. Scientists have devised two tests to determine the importance of a sound's temporal attributes.

The first temporal test is called the Cross Spectral Asynchrony test. It involves breaking sound into frequency bands and shifting the bands in time by different amounts. This would be like listening to a concert in which the violins (high frequencies) were a little ahead of the cellos (mid frequencies), which were a little ahead of the bass (low frequencies). You might imagine that the music wouldn't sound very good, but a normally hearing person could probably recognize the distorted song almost as easily as the undistorted song. And that's exactly what happens. As the delays increase from 0 to a quarter of a second, the sample sentences become more and more "echoy", but they remain understandable.

The second temporal test is called Temporally Reversed Speech. In this experiment, scientists take a normal sentence and break it up into fragments. Each fragment is then played backwards, but the fragments are played in normal order. You might imagine that, for very short fragments, the sentence sounds normal. For example, if the sound of the "t" in the word "ten" persists over ten fragments, playing each fragment backwards would be unnoticeable. As the length of the fragments increases, however, and significant portions of words are played backwards, the sentence becomes unintelligible. The fragment length at which the sentences became hard to understand was about a tenth of a second.

So what does all this mean? In the two temporal tests, distortions of a quarter and a tenth of a second were required to deteriorate understanding. Because the temporal performance of cochlear implants is far better than that, cochlear implant users don't suffer from the effects of temporal distortions. And we've already shown that amplitude distortion is not the culprit. So that leaves spectral attributes as the promising area in the search for the solutions to the mysteries.

The Importance of Spectral Attributes

You've all seen the picture of Abraham Lincoln on the five-dollar bill. It's a famous picture, and most of you can probably recall that picture right now. That picture contains a lot of information. There's a lot of detail there. A fully detailed picture requires millions of picture elements (called pixels), where each pixel is a tiny dot of color.

There's a somewhat less famous version of that picture in which the pixel size has been greatly increased. Instead of having thousands of pixels in each row and column of the picture, this picture has about twenty rows and ten columns, for a total of about 200 big pixels. Each of the big pixels is displayed as the average of the intensities of all the tiny pixels that are contained in the big pixel. The overall effect is a blocky image containing squares of various shades of gray (using a black and white picture). But even at as low a resolution as about ten by twenty, many people are able to recognize the picture. As the resolution is increased (the picture contains more and smaller blocks) everyone recognizes the picture long before we return to the detailed resolution of the actual image.

Frequency (spectral) channels

This somewhat lengthy introduction to the topic of spectral attributes is to present the concept of channels, where each channel provides different information. You can think of each pixel as a channel, and the picture that contained 200 "channels" obviously had far less information and was harder to understand than the picture that contained millions of "channels".

In a cochlear implant each channel represents a different frequency (also called a spectral range or pitch), so the more channels a device has, the more information is presented to the user, and the easier it is for him to make sense from the audio information. The most important spectral attribute in increasing the effectiveness of cochlear implants is the number of effective channels.

People tend to equate the number of electrodes with the number of channels, but that is far to simplistic a view. In a perfectly functioning device with perfect electrode placement, perfect isolation, perfect signal processing, etc. that may be true. But in the real world, providing a number of channels equal to the number of electrodes is an idealistic goal that we are unable to even approach with today's technology.

An interesting experiment is to take a normal sentence, break it up into frequency channels, and play it for people with normal hearing using different numbers of channels. By recording how many channels are required for a person to understand the sentence, scientists are specifying the minimum number of effective channels a cochlear implant must provide to give the user an opportunity to understand speech.

As you might expect, how many channels are required depends on how well a person knows the contents of the sentence. (Just like with Lincoln's picture. If the subjects of the experiment were Canadians instead of Americans, they would probably recognize the picture before it attained full resolution, but they would probably need more "channels" than Americans.) If the order of presentation started with 32 channels and went down (using the same sentence), people could normally understand the sentence down to two channels. If the presentation started with one channel and went up, people needed about four channels to understand the sentence.

Modern cochlear implants have an effective eight channels, well above the minimum requirement established by these experiments. Note that scientists don't necessarily believe that the quality of the sound provided by modern implants is the same as that provided by the eight channel experiments. But they do believe that the amount of information available is about the same.

Scientists can do these same kinds of experiments with people who have cochlear implants. However, instead of breaking the sound into channels and providing them with a certain number of channels, they are able to control the number of channels by turning electrodes on and off. When they do these experiments, they find that word and sentence discrimination increases up to about eight electrodes, then levels off. For the reasons mentioned previously, and probably for some unknown reasons as well, increasing the number of electrodes beyond eight provides no additional benefit. (This contrasts with the results of analogous experiments conducted on normally hearing people, who continue to improve up to at least 16 channels.)

These findings are comforting in the sense that they are consistent with the observations that the performance of the various modern cochlear implants are essentially identical, despite large differences in number of electrodes, strategies, signal processing characteristics, etc. They are also tantalizing, because a clear path to performance improvement appears to be understanding why the number of effective cochlear implant channels is limited to eight and then increasing this number.

Cochlear Implant Performance Improvement

Scientists believe that a primary contributor to the current performance limits is frequency distortion. A visual analog with our Lincoln picture is to imagine that pixel size changes depending on location. If pixel size increased from the left side of the picture to the right side, and from top to bottom, the entire image would be distorted and would be much more difficult to recognize. You can imagine that increasing the number of "channels" would do little to increase picture recognition, because the picture itself is distorted.

Tonotopic Shift

One cause of frequency distortion is called Tonotopic Shift, which means that the frequencies aren't mapped to their correct location. To understand the source of this distortion, we need to understand something about the cochlea and how the electrode array interfaces to it.

The cochlea is shaped like a spiral staircase in which the spiral gets tighter and tighter as you go up the staircase. It has three and a half turns, and each section of the cochlea corresponds to a particular frequency. The high frequencies are at the bottom and the low frequencies are at the top. The cochlea of a normally hearing person contains about 30,000 "hair cells" growing on the inside wall of the cochlea (called the modiolus). The hair cells are connected to the auditory nerve. When sound hits the eardrum, moving fluid in the cochlea causes the hair cells to move, which causes electrical impulses to be sent up the auditory nerve to the brain. Most people's hearing loss is due to damaged or missing hair cells, and the function of the cochlear implant is to replace those hair cells.

The electrode array consists of a number of electrodes imbedded in a long thin rubber strip. This strip is inserted into the cochlea as far as possible. In a normal cochlea, it is usually inserted about one and a half of the three and a half turns. Recalling that different portions of the cochlea correspond to different frequencies, this insertion level places electrodes at positions corresponding to a frequency range of 1000 to 10,000 Hz.

When a cochlear implant is mapped, the characteristics of the signal going to each electrode are modified. The goal is to match the signal being sent as closely as possible to the ideal signal for that electrode, given its placement in the cochlea and the characteristics of the person using the implant. Because each electrode is at a particular place in the cochlea, it is able to stimulate a signal that is interpreted by the brain as being of a particular frequency. (In the normal ear the hair cells at that location are "tuned" to respond only to that frequency. The electrode, which is replacing the hair cells, should also "respond" only to that frequency.) So the key to a good mapping is to ensure that appropriate frequencies are sent to each of the electrodes.

Given that the electrode placement is such that the first (deepest) electrode normally corresponds to a cochlear location of 1000 Hz, that electrode is generally mapped to 1000 Hz. If the electrode array is normally inserted, everything matches up and that electrode will provide useful information to the brain. But suppose the cochlea is partially blocked, and the electrode array isn't inserted as far. Then the first electrode as at a cochlear location that is tuned perhaps 2000 Hz. When that electrode fires, the brain interprets it as sound at 2000 Hz. But if the processor has a "normal" program, that electrode fires when a 1000 Hz signal is present. The effect is analogous to having the cello play the bass part, the violin play the cello part, etc. These tonotopic shifts would probably not render a musical selection unintelligible, but they can render the more demanding speech sounds incomprehensible.

Modiolus Hugging

A second key to improving cochlear implant performance is to place it as close to the cochlear inner wall (modiolus) as possible. Recall that the cochlea is shaped like a spiral staircase that gets tighter as one ascends. The auditory nerve runs up the center of that staircase, and the closer the electrode is to the auditory nerve, the more precisely the electrode signal can stimulate nerve endings that correspond to the electrode's frequency, and only those nerve endings. Thus, the electrode array should "hug" the modiolus. Both Cochlear Corporation and Advanced Bionics have recently introduced array improvements directed towards this goal.

Cochlear Corporation's Contour Array

Cochlear Corporation has introduced a new array called the Contour Electrode. Its purpose is to ensure that the electrode array hugs the modiolus. During manufacturing, the array is pre-curled, so that it's natural shape will hug the modiolus. The array is straightened for insertion using a removable pin called the stylette. The array is inserted with the stylette in place. When the stylette is removed, the array resumes it's curled, modiolus-hugging shape.

Experimental results with this new array are just coming in. Early indications, based on the performance of six people in New York and 20 in Australia, are that average performance is significantly better than with the previous electrode arrays.

Advanced Bionics Electrode Positioning System and HiFocus Array

Advanced Bionics has taken a somewhat different approach to obtain the same result. Their solution is a shim called the Electrode Positioning System (EPS). After the array is in place, the EPS is inserted on the outside of the array, forcing it against the modiolus.

This system may provide the additional benefit of inserting the electrode array farther into the cochlea. Imagine the electrode array inserted its normal one and a half turns into the cochlea, and resting against the outer wall. If a shim forces the array against the inner wall, the path length will be shorter than when the array was against the outer wall. (If you walk up a spiral staircase hugging the inside railing, you don't walk as far as if you hug the outside railing.) Because the array length is fixed, this will force the array farther into the cochlea. (A rope will go farther up the staircase if you place it along the inside railing than if you place it along the outside railing.) It's too early to tell if this theoretical advantage can be translated into a performance advantage.

A separate but related Advanced Bionics development is the HiFocus array. Recall that cochlear implants only provide eight effective channels, despite a larger number of electrodes. The HiFocus array is an attempt to increase the number of effective (independent) channels by increasing the electrical isolation between electrodes. This is done by inserting rubber isolators between the electrodes. It's too early to determine if the HiFocus array provides improved performance, but early mapping results indicate that comfortable listening levels are reached at lower electrode currents, which is consistent with the expected performance improvement.

As with Cochlear's Contour Array, early results on the Advanced Bionics innovations are just coming in. Based on a small number of individuals and results only one month after hookup, the early indications are that the Advanced Bionics innovations provide significant improvement over the previous models.

Historical Trends and Future Directions

The history on cochlear implants has been one of continual performance improvement, from the early single channel models to today's newest innovations. The following table shows the performance progression in both sentence and word recognition. The numbers represent the average percentage of sentences and words correctly understood by users of the indicated devices. (Note: these numbers are used to represent the performance trend, and should not be taken as absolute measures of the performance of a particular device.)

Model       Sentence    Word
3M/House       5            0
F0F2              20           8
F0F1F2           52         20
MPEAK            67         25
SPEAK             78         38
Contour/EPS    90         45
(preliminary)

There is every reason to believe that the trend indicated in the previous table will continue in the foreseeable future. In addition to continued performance improvements, the future may hold the following developments:

1. Fully implantable cochlear implant - The technology for this exists today, but is not yet "ready for prime time". This could be a reality within five years. Implantable hearing aids are already available.

2. Implantation of younger children - The FDA is allowing trials on children as young as 12 months, and children as young as 6 months have been implanted in Europe. One difficulty with implanting younger children is being able to ascertain with confidence the degree of their hearing loss.

3. Implantation of people with some residual hearing - A few months ago, surgeons in Iowa implanted people with nearly normal hearing up to 1000 Hz, and virtually nothing above. The result is a hybrid system consisting of normal hearing at low frequencies and a cochlear implant at high frequencies. Early results are encouraging.

4. Dedicated programs (mappings) for specific applications, such as music.