Transcript Template for poster presentations

Bionic Restoration of
Movement
Kristin Scudder
Converting thought into action…
There are many disorders that disrupt the neuromuscular channels of the brain, which results in the brain being
unable to communicate with its external environment. Brain-machine interfaces (BMI) and brain-computer
interfaces (BCI) provide the brain with a new non-muscular channel through which the brain can send
messages and demands to the external environment. These interfaces are communication devises that extract
electrical signals from the brain, either invasively or non-invasively, and translate them into output actions to
control any number of output devices. Much research has been conducted involving ways of improving noninvasive methods to match the performance of invasive methods. One way of improving non-invasive
methods is to improve the performance of the feature extraction method. Current BCIs are implemented
mainly using the Fast Fourier Transform (FFT) and Autoregressive method (AR). However, research suggests
that the Wavelet Packet Transform (WPT) would work better on the non-stationary EEG signals. The goal of
this project is to test these three signal processing methods and determine which most accurately determines
the user’s intent.
At present, the FFT and AR methods of feature extraction are most commonly used in BCI implementations.
However, both these models are unable to describe signal information in various time windows and frequency
bands. Since, EEG signals are non-stationary, a feature extraction method that would be able to describe
signal information more dynamically is necessary. Accurate feature extraction is essential to the success of
BCI implementations in the future. It is this phase, together with the translation algorithm, which is
responsible for accurately distinguishing a user’s intent. By improving the accuracy of feature extraction,
researchers will be able to implement their BCIs to do much more complicated tasks. In addition, it will bring
the non-invasive BCIs closer to invasive BCIs in terms of capabilities.
Communication devises that extract electrical signals from the brain and
convert them into output devises to control any number of applications.
BMI: signal extracted from
the brain is sent directly to
output device
BCI: signal extracted from the brain is
first sent to a computer, where it can run
a computer-based application, or be sent
to another output device
Utilize electroencephalography (EEG), which is
a measurement of electrical activity produced by
the brain that is recorded from electrodes placed
on the subject’s scalp.
Limitations:
• Low information rate, 20-30 bits per
minute
• signals obtained represent only a field
of potential rather than specific cellular
activity
• Insufficient for controlling artificial
limbs
Electrodes are implanted into a region of the
brain in order to obtain signals via specific
neuron firing patterns
As shown in the above diagram, the signal acquisition module extracts electrical signals. This module then amplifies and
digitizes these signals and sends them over to the signal processing module.
In the first part of signal processing, specific signal features, which encode the users’ commands, are extracted from the
digital signals. In the second part, these signals are sent to a translation algorithm , where the signal features are translated
into desired output actions. Some of the desired movements for motor prosthetics include: movement of a cursor, clicking a
button, and specification of complex time-varying trajectories, such as reaching for an object.
These device commands are sent to the data output module, which uses them to run an output application or an electronic
device. In an open-looped BCI, the output is not accessible to the user. However, in a closed loop BCI, this output is sent
back to the subject’s brain. The brain uses this feedback to maintain and improve the accuracy of the system.
The final module, operating protocol, specifies all the specific details about how the interface will run. It defines how
the system will be turned on and off, whether communication is continuous or not, what feedback will be provided to the
user, and much more
Utilize the VEP to determine the direction of a subject’s
gaze, in order to control the movement of a cursor or to
select a symbol from and 8x8 grid of symbols. The user
concentrations on a location or symbol on the screen.
Subgroups will quickly be highlighted and the VEP will
spike up if the users symbol is in that subgroup.
Refers to the peak of 300 ms reached in the EEG when
frequent or significant auditory, visual, or somatosensory
stimuli are mixed together with frequent or routine stimuli
Compare three signal processing methods for feature
extraction
•Fast Fourier Transform (FFT)
•Autoregressive Model (AR)
•Wavelet Packet Transform (WPT)
Accurate feature extraction is extremely important, because
it is during the signal processing phase of BCI
implementations that the user’s intent is determined.
Currently, accurate determination of the user’s intent is a key
problem in BCI research. Research shows that the WPT
Advantages of WPT:
• Multiple resolutions
• Faster response and higher accuracy
Slow voltage changes generated in the cortex of the brain.
There is a choice located on the top of the screen and another
choice located at the bottom of the screen. The selection
process takes four seconds. During the first two seconds, the
system measures the users initial voltage level. During the
last two seconds, the user selects the choice in the top or
bottom of the screen by increasing or decreasing the voltage
level.
Mu rhythm is an oscillation measurement representing 8-12
Hz of EEG activity in the primary sensory or motor cortical
areas. The beta rhythm is an oscillation measurement
representing 18-26 Hz of EEG activity in the somatosensory
area of the brain. Both these ranges represent EEG activity
when the brain is not engaged in processing input of
producing output. Movement or preparation of movement
decreases these rhythms, while relaxation increases these
rhythms.
 Collect and use sample EEG data for mental tasks
 Use data of subjects completing 10 trials
 EEG samples recorded for 10 seconds during each
mental task
• Subjects performed five mental tasks:
• Baseline task
• Math task
• Geometric figure rotation task
• Mental letter-composing task
• Visual counting task
Collect additional EEG sample data for motor
imagery tasks
Run each method of feature extraction on
sample EEG data
Analyze how well each method
differentiates between mental and motor
imagery tasks
• Accuracy should be determined by
how well the method predicted the
user’s intent