A grant from the National Science Foundation (NSF) to Dr. Bonny Banerjee, assistant
professor of electrical and computer engineering and Institute for Intelligent Systems
at the University of Memphis, will not only foster interdisciplinary research by bringing
together fields as diverse as computer engineering and audiology, but will eventually
allow severely hearing-impaired people lead the lives of their normal-hearing counterparts.
The $298,203 grant from NSF will fund research that aims to automatically tune cochlear
implants (CIs) for individuals with severe-to-profound hearing loss. Dr. Lisa Lucks
Mendel, associate professor in the School of Communication Sciences and Disorders,
worked with Banerjee in obtaining the grant and will handle the clinical aspects of
the project. The project will also involve electrical and computer engineering students,
audiology students and clinicians, who will work together to design and deploy the
learning algorithms for customized tuning of the cochlear implant devices for each
individual with severe-to-profound hearing loss.
Statistics reveal that hearing loss is the most common birth defect in the United
States; it affects 12,000 newborns every year. CIs are an effective intervention for
adults and children with severe-to-profound sensorineural hearing loss who fail to
benefit from acoustic hearing aids. However, without proper tuning of cochlear implants
to each individual’s hearing deficiencies, optimal access to sound cannot be delivered,
even in the case of good candidate selection, surgery and rehabilitation support.
At present, no universal standards for tuning CIs exist making it challenging and
time-consuming for audiologists when working with their patients. With more than 200,000
CI users worldwide and an annual increase of over 30,000, lack of proper tuning is
a severe bottleneck to the usage of available life-changing technology.
Individuals with severe-to-profound hearing loss may benefit from the research U of
M scientists and graduate students are doing to improve the efficiency of cochlear
implants, as shown above. Cochlear implants can dramatically change the lives of hearing-impaired
individuals. (Photo by Rhonda Cosentino)
Before joining the U of M, Banerjee spent more than three years leading research in
the company Audigence Inc., which develops automated software solutions for tuning
digital hearing devices (cochlear implants, hearing aids). The company launched a
commercial software product called Clarujust for the end-user at the Academy of Doctors
of Audiology Annual Convention in Tampa three years ago. Last year, Audigence's IP
was acquired by Cochlear Corp., the world’s leading CI manufacturer. Most importantly,
in a pilot study at the University of Florida, 17 of the 20 CI recipients preferred
to continue using the Clarujust-tuned setting over traditionally-tuned settings.
Banerjee said, “While Clarujust was quite successful in many ways, one thorn in the
bud was its test-retest variability. That is, consecutive tests done on the same patient
in exactly the same way sometimes had very different outcomes. This makes the data
from tests unreliable and hence, the goodness of the tuned device parameters is in
doubt.” He attributes two factors for this behavior – lack of adequate test data and
the analysis of an individual’s stimulus-response errors in terms of hand-coded features.
“These hand-coded features fail to capture the context or norms in the hearing abilities
of each individual.”
These observations led Banerjee to seek out Mendel to co-submit a proposal to NSF
within his six months of joining the U of M. Their working hypothesis is that the
deficiencies in hearing for individuals with significant hearing loss are reflected
in their speech. Banerjee and his team hope to address the shortcomings of Clarujust
with an entirely different approach: by learning features from day-to-day speech around
the clock in an unsupervised and online manner. Informally, features in this project
will be snippets of sound of very short duration (e.g., one millisecond) that recur
in speech and can be used to reconstruct an utterance. The learning algorithms will
be installed in the implanted cochlear implant device. Since the algorithms will learn
online, the speech will not be recorded or stored and privacy will not be compromised.
The learned feature hierarchy from the speech of a severely hearing-impaired individual
will be compared to those learned from the speech of a comparable normal hearing population.
Deficiencies in the patient’s hearing will be ascertained by identifying the missing
or distorted features. This information will guide audiologists to better tune cochlear
implants to enhance the audibility and perceptibility of speech.
Besides CI tuning, the algorithms will be applicable to a variety of monitoring applications
within healthcare and beyond. Continuous monitoring with wearable and implantable
body sensors will increase early detection of emergency conditions and diseases in
at-risk patients and also provide a wide range of healthcare services for people with
various degrees of cognitive and physical disabilities. The project will transform
the traditional ways in which the clinical needs of continuously-monitored patients
are met. Its success will open up avenues for around the clock medical attention focused
on the specific needs of individual patients at minimal cost.
The grant will be distributed over a three-year period. It will support graduate students
and cover the cost of equipment and travel needed for the project.
For more information, contact Banerjee at 678-4498 or email@example.com.