okay so um the next speaker is
and robert rubber smith
oh she will be presenting
a a a a method
for robust minimum variance beamformer and its application
to um and E G and
a little potential
i can start
a
the um that top like a to see it
um
okay a i it
that's that some work which we do doing in oxford
between a of us
um
so i i mean miss amy
is the map rotation any he's done most of the work on the details the beamforming
so i look a fast for that
we also working with um the pulse like a tree
um i an engineering
that's made
and speaker i Z is is one of the D sections
work in they uh that if possible source just or not
in the brain stimulation
this is a the based in a sense
we've got a clinical problem
um which is to do with the brain stimulation
and we are trying to image it's using a make and and stuff a
so i'll start with a base
discussion about um
and then all want to the beamforming and then out of use some results
a just a fish
that that stimulation is technology
web i
oh they plan electrodes into the brain
so that the here
you can see a
that that coming in a top
coming coming down
it's used
white and not your the sources
in particular we you it's not with problems
see
for the most common
uh those that eating with M are that is press it's channel
and eating with a a a a a motion problems slide stand here like can easy easier
the
case not can be talking about that is pain
and we say that a uh is that you've a person
and
well pain is for the the perception of pain is all sixteen is subjective
some people see it could very well with pain or not find a right
handle
um it's also use some of the
um areas such as much as its is not in your
so no one yeah
stands for it works
what you do is you plot the election in the brain
and will be here brain stimulation the of signals actually coming out of N be very small
the electrodes implant
they then put on a five vote
sine wave normally five votes like very bits
at a frequency to about fifty hz and the functions that a
um and then a a a go through period of this a a a lecture it's by search
that's then a period where the external eyes
well you might access to the stimulates to
and S the signals coming off the implanted electrodes the local field potentials
and during that time they tried rate
so they one the site that's signal of the patient
and then off to that they employ a battery
and everything missing the was down inside everything becomes internal
the what we do is very trying to get a of that text last period
so we can actually the feel potential of the electrodes
and use the
to try to improve a
so mad ninety two and stuff lot of a
a thirteen new technology
um
picking up
a magnetic fields of you were have to um
gauss
so that that's
about three more three um eight or the magnitude smaller than the spec
feel
so essentially put a number of my comment isn't but you might as rather here
got
doesn't
a she
you can see what here
and for these very small signals
then we tried to um we can sort the sources within the brain
and a point of that may is to so
uh one is to improve
the surgery so they it that what in four
and the other one is to try to understand more about the you it's
so that's what are difficult to use
with D B S
one is way looking at that very small signals
we expect most of the
excitation of the brains to come from the a region around the a lecture
well fortunately
that's quite good separation between the frequency at which are the electrodes
and the frequency that which principle
so it is possible to try to access that's small signals want to me like sing
so we looking at about say fifty to a hundred thirty hz stimulation
and the low cell
um a a range between about five and fifteen a
for the range from
yeah that that quite big it's you
is when you bring the wires out
they are have to do a lot and skull
uh the wires a magnetic you got cost of wires are yeah the whole of the skull than not done
inside
but those double for distortion
a that region in terms of your source
or
so that think in this paper a is where
um got two things it using a placement of an now now that a whole
we
a a well on that show how to improves
the recovery of the of the
spatial source
what we think to take that in this paper
it's it's used in the idea that want all simulation is on
then it's
you would expect most of your signal
so yeah the stimulation can see
be coming back from a region
and which stimulating
and hence by splicing the cross correlation
between the signal you putting in
and the signal you're thing
at my comment is
then any fact you can improve your beamformer
especially one region of interest
which is found the region whether you like to tune
so the idea here is way looking at stimulation on and trying to get the best stuff beam as we
can
and then but using
time and stimulation is all
to try to verify that
so what i do i'm elements is they stimulate the time
um
people don't seem to get a
uh used to that's
T where king
but there are it's use a at that time
so that just cost about ten thousand pounds
um
and they lost from a couple of years that's what made thing not to take infection
the um
put in you new battery
yeah that is you
which is
coming more to like that does seem some evidence that stimulation at
that some of the of the range emotional areas
and that's some showing that actually and not to stimulate send march
because of things that a
strange
so a of my and fast which should welcome foundations support
is trying to actually um
find an adaptive method where we actually you know way to simulate when
and
once tried in former to you
very quickly um Z say that's very high meets work and i don't think that yeah
that you
and then began to demonstrate using training data
from a a a a a patient
pain
with P B S i mean have them
so a techniques
yeah shows um is look something like
so that that's uh how a bit of each
and to
um
a few centimetres long
they have a number or electrodes or may have about four electrodes on where you can stimulate
and
these are the to that the field of getting around each one is in fact a let's
approximately
so we can to assume the N F P data comes from a small cool volume
a bound the implant electrode we know that where that is from looking at um
ct a oh i want to implement
you can use
see
um we can we that now
a a a a a whole of skull
and then make a to use a robust minimum variance beamformer
um what's the to work by element set out in two thousand and five
but the obvious didn't use the particular aspects which relevant to um make imaging
yes or forward model
so the first time here
so so white state wide C R be of measurements
at skull
um S
is the stimulation
so that's from the electrode
and a and is the forward model leave but a vector
then we have a second term
which includes the
uh oh S to try to now but i
that's a N
as T
and we assuming a um but the N F P data is going to a small volume only got here
i can the location of the whole is known
so we can deal with that
and then got some noise
there's also in fact times day
going from be um lot because the instrumentation
between the
um excitation signal
and am signals with a
so just think they are a problem formulation
is to optimize are beam
for
um so we've got a why we want to estimate S
use optimisation on it
or or first time here is the difference between that's essentially
thank expectations
um with a value you out for a penalty factor
um which there is we be and one
so you put note you want you take in and that is
of of the product of a correlation
between the source
i'm the measurements
um
and then subject to all source vector
so that's well
and it to it's all
so we know we got source that
is the only condition
okay and then we find that
so that if we take the um the variances of the and take the expectations
then a first term does becomes
this one down here
so a why S yeah is the first good that is
between the source
the excitation we put on and the measurement
it's a a a a solution essentially
well we're trying to maximise or or correlations so
something to minimizing constraints such that constraints
you can find exact
use the W
this for increments as work that E
which comes to stop one
yeah yes one of the parameters which says what which is the described in that
where you putting on the excitation
now to me a of once month i
constant
um if we substitute are right if W back into the a constraint
and in fact
we find you
this is but
we then that no as that's
using the
normalization low
so be yeah so i put in this
i
um we
oh of the um diarization is with that i've the factor
are a it's a half sheet
a a heart sounds
pose
so that's the vector which later
the region of what we know the excitation is
to the
um close correlation between the measurements and the excitation
we can do that i was that get a secular equation
yeah
which late
a a a a me we found this relates number of than our on unknown
to the eigenvalues yeah i
of all normalized
matrix
or a of how cute
so the details of that are in the paper
um and the was in the paper and so i am of a fast of it that that's
time is and results
um and i have got some papers is here no once
but a big advantage the way that actually be normalized this
so we've normalized here
um
talk about V before
so E and
uh
i
with the to some of those covariance matrices
and also the parameters of the of is what we put X
i
is actually uses as a nice solution which then amenable able to an effective
solution to have find
um and uh
or as some previous work has a a few a given bounds for them the
this kind of normalization as she allows as to optimize and you know
yeah
a out a which allows us to do that
which are not been to go to i think
time
that that used to some themselves
so we've done this in simulation one of the problems of course as we all know made doing things well
mel by medically
um it is very hard to try to validate
um
so we done validation using a simple simulation of a spherical head model
put in the deep source
simulates the excitation station
we put see in the source of the but of interference
um and with that's noise
and to sure that we can to look up on and all conditions
we've allowed to all source
um which would only be
i dominate dominated
by the stimulation we on
or we should be seen was once the brain when off
a different frequencies you sign "'cause" on way
so yeah that was with the start of wiener filter a
and we've compared them with the filter so in both whole now without correlation
so yeah it is just the interference and their noise
i i mean is
um because it's that we've
simulate interference long that thing else be be estimated it
um we can see the wiener filter
is that you not
um a well
um um
as well
improves things the the S I
but that but
using the be
so if that the S noise as well as a fair comparison
that we can see a method that she does give us advanced just
oh the the other two men said putting in this post correlation so um
um between of the source
and the measurements since we know sources is doing that frequency as she does hell
a technical data is most interesting
um of the forty or or one with a body pain wanting pain had separate yes
um in but in the pack
which is close down to them
oh and fifty hz
so the it and sell them on magnetic
so they are a problem
for the mac
uh
but in fact
the
say whether whether wow clusters then you can still
do you have an all i can just and that's
before four we assessed the lectures
for sequences like to
um so i are you can use for but not often
you got inside you
and that just can't of is that um
using the he was first right
this my
the things we do
first of all
we
recall yeah that P data in all conditions
so that's that she recording straight of the electrodes
the that are too long used a beamformer
um
and then we were what happened all
so we use the beam forming one number you know
we then be constructed using a P point how much is what we would expect to see from
the yeah it is maybe a filter
yeah it is
using a a a a as well
a use of the um i three
not very much
and here it is
using you a
B
and you be
so you can see you press that i think that that i
um i think this is a much better
the U B
finally
the a point of this
try to find out
a in the brain
so this is rate just the fit it interest as suppose
and this is a get the difference
between the
stimulation of
one
so this woman that
a once it was turned off
and you didn't
on on
um not an entirely goods
um experiment because she stopped yes you knows this awful
do
you a you do get some differences between the of a young condition
and the way in which are response
this a you looking at part should known to be sensitive to pain
um um rate standard high
a i'm the A C C is the and tear single call
yeah right
and the three and i is the um of the email
so this technique can see that point
a
spatial layers grain
um where we getting sports
to be um nipple stimulation
so we we've also a of using correlations that all closely and identify more closely spatial regions in right
um i i a and the results if five using elliptical volume of the yeah that's P
to improve things of using circular volume
still provide
solution
and we've shown some results
um
improvements using simulation
i mean and you
and you
yeah
and
yeah
well
G
as much
so
or
or a more you are still going on
or you are you
two more
yeah
well
and
i
oh
or like to call
or
yes may just the problem
say that's that
for my my kinetic model of the brand such a simple
"'cause" you can assume that is a of a two D is
is one is what
to
so
the
oh well as for a um but it's not a context could you what a much more complex model right
so my a standard are not experts what they don't E G my understanding is make use gives you what
oh
that is to discuss that with you for are actually that if anyone and any information on that
that's a my hands and you i mean it is a recent technology is come about lost five to ten
years
well a you also
to machine
and not let me
oh
what should you signals
tricks
of
what was is
are you sure from your speech and so
or
but
for a the S T N was a packet is next to it
is
but pain yeah
the use S the end from it
oh so you're right
trying to be your with peace
for four
what what you saying
for the final are we measure yeah that peaks as X to i and then we tried to estimate as
well
yes using the mac
in a compared
so
you
for point
well
yep
so what is
right
for
she
a
all
the marks is
i five to
i i mean that sense that to the signal
we receiving see in the of
or you
a frequency
a point where we stephen it is that you want place that
is quite hard to know how you try to optimize
world i mean would really strong
so i
one Q one
so
still
from
and
estimate
um
right
which still
what we are we are doing a spectral analysis as well so we looking
so a beamformer is that you
i i think that if
for
so we all by some that to look at
maybe to right
the B two
of what's it's yeah was either
and yeah the
source you right hmmm part
to
ooh
yeah a just that even that's one
but
that's what if you they get
sure one
uh_huh four
do
we would
you
hmmm
it's a that's not in a sense
that are not quite so what you're looking for
seven seven sense i mean it is that's what we try to estimate we getting and then measure we get
"'em" and i suppose way we can possibly indicate
still
one thing to get a good results or
but it's good question and very hard to know how about it
from
see
you try to the um
so i mean
okay well thank you everyone
uh i think you again maybe