and then a um next page

it is a a a a to B and that

the and in um and waveform form estimation hmmm C two signals

um so we present by channel

or

slow

so morning everyone

but to this presentation entitled to yeah she waved iteration and form estimation in actual hardware signals

but using a block keep are

and a name in

this it a joint work

where is my

pitch to live there is a professor joint one of my

and that's only my

are the universal thing false

and also with Q O K O and professor for a large

are are we is to be and the university of technology also

so here is to you

i'll of from

addition

for that we would give you a brief introduction of

the chance you move initial problem

and the seems we propose to use a bayesian inference to deal with this problem

and was then you should use use the proposed to but more

and we also propose

to use but you the keeps mess or to estimate a and parameters of the proposed

i told and that would give you details of the

the proposed schemes that or

and i sure use this measure is our to the case that a where is the operation with

man

oh here have a star

oh have you about in some

but you creation problem

and

as we all know that that our ram for C G for short

it it for of

the hard to lead to a to be G after uh by the actual position on the screen server

so

the automatic

and the lights is of just signal as on a lot of interest in the biomedical engineering from

oh problem this

estimation

as we can see

sorry

but because see

uh a C D signal

consists of three distinct waves

she way

and uh the most significant part of to a caress contrast

and that's she way

oh problem

this figure

as we can see the most

useful clinically useful information

i be file

the wave boundaries

and the interval durations

oh these C D C

but that's why the deviation

which means

the um

determination of the with peaks with bob are used and the estimation of we've one

it's a very important step to the interpretation of the C D C

so things that or as complex as

are a more significant part of the each signal

is detection is relatively easier

and

it can be used as a reference

to the C and she you deviation

and um

the C and C waves

generally they have a low should

and due to the presence of the noise

and a baseline one three

if the are deviation it's more complicated than the as complex as

the H

so basically we will

do firstly to dress complex this and defined a search each both before and after

locate look S you but

so we can find in the three sure

different

existing methods

the first

existing mess are the first class is based on reading techniques

the president

it is to use for techniques to remove the noise

and the base a wiring

and it

sorry

and then

we you apply by read and it uses specialist

term where are the weights

and we are are problems

is that in

class of existing that's sort is based on

basis is expansion that makes

i mean we we use a basis is especially at least for the functions

and the way to use

some thresholds to determine where are wave peaks and where the bar

a certain class of existing mess or is based on a a piece and that pattern recognition that source

and the we can also

a by some realistic signal model to feed the C D signal and to estimate

to parameters of the model of to do the nation

of the D C D C

so some particular patient has been given to a part of you

you our paper

we we propose

to use

uh uh easy inference

together with is

some similar to ours

to deal with this problem

so now i would introduce you

or a mathematical model

the grass complex the our application are assumed to be detected

so a which will not be considered here

and that then

we propose to use a deep each no overlap

window

to cover the whole dct T stick

so basically you we have the same number of the

no if are asking to roles

we now hour

process we know

and for each not here S components

can be seen as a combination of all to me the to mean is

which are present you way and if you with

plus uh local based

so take the that parts as an example

this is she weights with this

a uh processing window can be seen as a a convolution

oh

but

bernoulli gaussian sequence

a on the one would got the sequence

with a

on the one

impulse response

so that

oh no impulse response present a waveform

and this

on the one really gaussian

to represents

the with locations and we want to

and then we can do same to the T waves

so basically we have some conclusions here

and we have a this

but the medical signal more

and this sick it represents the sequence of the baseline a local base fine

and this W P A a noise

which use that students be gone

and for some more we also read

the proposed to use

or a basis expansion techniques

to represent a on no one point four

so this

technique has to be used

oh C to denoising and also on C compression

the advantage of

this technique

i is

that is that we can have a used to mention though

on on parameters

and also we

we can seen that noise some most the version

oh the with that

for them more we also propose to use of force degree polynomial model

local baseline

but this is a patients what previous work

which assumes that the local

baseline which in each no cresting in ball is a cost

so

with all this we have to a better representation of

the the not is in the roles in the processing window

so

red once are all time

so we have the we form we've location is up to two

all the what she ways

and the

baseline like is the noise was far

so

here

i will introduce you if you model

that we don't know the you for realise of the product

all of the that it functions and priors

to represent the posterior distribution

and to estimate and have

so

by assuming the noise in our model is zero-mean gaussian

we can have a just like would function

and concerning the priors of the unknown parameters

they can be assigned according to the T C you signals back to this

just we only expect

at most one

she wave in each to thirty the both

we can assign a block cost rate

to this

indicator

and a by assuming the independence

oh of each individual

thirteen thirteen both

a

priors of the

the the the whole

she indicator vector can be to by a product of each about divided so

and the since we have used a run gaussian sequence

so when there is a way it it

we assign a zero-mean gaussian prior to the

and you

a store C to this petition

um report weeks back and not only the positive

i each but also the negative to every is a zero mean gaussian prior

and for the bit form is also a zero-mean gaussian

but we from partitions

a course we expect to not only make your of what to parts but also to part of the waveform

and the for the baseline for she's is also a zero-mean gaussian at the noise variance prior is you for

yeah

so

not that you have to all these conjugate priors because it can

some five

it it can be the computation

so now we have a lower posterior distribution which is the

the product of all this priors with the likelihood function

but uh it's a topic

distribution which

uh we can not

compute a form estimators

so that why we propose

to use a map model can estimate of a source

to generate samples

weights

a sounded sick leave for this posterior distribution

and to estimate other parameters of the um

the way to model

so this is a proposal a keep were

exactly is uh

there is a simple and S and C and C

simulation is or

uh but that's to this brought constraint

we have a uh as

a slight modification to

class two kids that we're

which that instead of

generate

that but that whole the to cater

there

we can that generate

brought by block

we have only that had a size that it is only

last oh the each block that's one

so

it

you to the company

patient no efficient

so basically is that after generate samples for C where cage or we be see if the is

you way

with is that what that dude

and for for you it is the same

and once we have a

finish

that putting the

indicator are

we was that of the waveform for block

or the process we do and the the baseline baselines and the noise

noise bar

and the the estimates can be a obtained by using

for this

this straight

i meters

can be

all all T and by using a a sample based the map estimator

and for this

the rest of the other having to get a job in by a and M

estimate

so

before drawing use the solution is so mention that the

the pre-processing step yeah our vacation

it's been done by has are region which is very that's

to you your as complex

and the processing window not set in set to ten part eight

this is a force

we cannot have a very large

uh that that thing we know last course you C you has it

so so though

stationary nature

and we can not have a a better little last either course

we have to use several observations

to be able to estimate a we've a problem was that

a if you have a christ if you are interested you can find more

details now journal paper sure here

and now

oh use so use some to we examples

for a that to database

the first example is from

uh to that it's is there a it's a three six

uh this example has been shown to be force it had some

but are we of the reasons and a long beach

a state the of the posterior distribution no that C N you will be cater pages

which means

the probability of having that she or to location

so as we can see

here

to to give sample right handed it is to locate the way so that's the interesting

probably a of a proposed are them

instead that of using a read it and you to determine whether it is a way of one

i one map

that is the map estimator

can tell us

the most probable position of having a Q a here

and C a small to have the peaks that the wave

it just tell us that peaks

and the well there is can be

turn it by

this is estimates of we four

but using different creature

and this is the reconstructed signal

and you read and estimated a baseline of be baseline glass

and all it no signal a don't at because you are very close

and and the

a second rate shows us to detect if it useful

so here is an example with a premature ventricular contraction now

which

we

the she where is the scene

and follows by a giant

if your as

as follows it follows by first

Q way

as we can see

most most our written can handle this situation

the posterior distribution of this

i is very low

so there would be no false alarms

and this in

you work that she way is about detected

with the battery of five was to reduce pollution

and this this you just is a cost that it's signal

and the the detected if it was one

another example of

but if physique she weights

and that we can see

we can have a a very of years

she we've form estimation

and this is reconstructed signal for this dataset

the last example is

a does that we are

there are some as as so you wave fred

so you can see you when there is no to it in the thirty two tomorrow

we can have we would have a very low

posterior distribution

and now or map estimator

we're not need any

for some

so this is a reconstructed see no we is that it that the fiducial points

for this is that does that

and now have that to show is um

constative C of cooperation is with are

we have implemented the filtering in that one of the

filtering techniques and the other is the bit

right

is the basis expansion time

as you get C in terms of of the detection is its ability

our our would some out performs the classical mess source

and also in terms of of D H arrow

our our in is

slightly better is comparable in that's better

and based on the compilation that like to remind you that uh

there are two advantages other advantages of our

method

the first is that uh

we can provide as well as

the we a estimates

estimates

and if probably is

and they are of and he's is that the

this is a ms art

and we can provide

the reliability ability information

such as the inter a a confidence interval

except for

uh oh of the with see which is

very interesting for of the medical yep

so here the convolution

the know what do here is that we have proposed a is model

for the not rest in the state D signals which is based on a blind deconvolution problem

yeah and that we have proposed a block example are

to estimate

the on parameters of space model

and and the as process back

since yeah are more the we have damage use estimation

so we can

i have to

to where at down that don't other nodes

you you some problem that's that really a detection problem

and it's is we can have the with form estimation

we made at at a it's miss detection problem

so all that force back

and since it is not yet

uh online application

uh we are

currently investigating the sequential mass are

uh us to create role what colour methods

to this

basic mode

so

set for the attention if you a is it we have a a little that that them or a personal

web site

i test your back

i

holman breeders to to the you nation problem

uh yes i think we have a

and we have calculated that better at lower our are based on the for example the cs C standard

the and

to

because tolerance

errors

all of the

deviation work

and we can

feed this

stand stand it is that

you're what is not

like a

for you

is is that

yeah i that

if we con

yeah

use that come back to this

yeah

i mean yeah and

there

should be some noise present in signal

and if the noise

is the very likely to that

with forms

it could be

uh i

the um how to say

our data are within

we you we should you pretty per but probably to addition of the way

right

yeah the um

yeah

why we have a sharp here but the L is not about their point five

since we assign the prior that to the but but but it you of having no way is is you

pour to the

a probability of having a a will be zero and with a map estimator just

you would not be a problem

yeah i

is problem

and this also is the don't know which is possible because they here we have a

uh uh to be

the noise

which is

hmmm

very

similar to a way if

so

sometimes i i just think it's a way

yeah yeah i i i have read to this point

and the

to to to them

compute

computational load it's um more

so you