0:00:14and then a um next page
0:00:16it is a a a a to B and that
0:00:19the and in um and waveform form estimation hmmm C two signals
0:00:24um so we present by channel
0:00:54so morning everyone
0:00:57but to this presentation entitled to yeah she waved iteration and form estimation in actual hardware signals
0:01:05but using a block keep are
0:01:06and a name in
0:01:08this it a joint work
0:01:09where is my
0:01:12pitch to live there is a professor joint one of my
0:01:15and that's only my
0:01:16are the universal thing false
0:01:19and also with Q O K O and professor for a large
0:01:23are are we is to be and the university of technology also
0:01:27so here is to you
0:01:30i'll of from
0:01:32for that we would give you a brief introduction of
0:01:35the chance you move initial problem
0:01:38and the seems we propose to use a bayesian inference to deal with this problem
0:01:43and was then you should use use the proposed to but more
0:01:47and we also propose
0:01:48to use but you the keeps mess or to estimate a and parameters of the proposed
0:01:55i told and that would give you details of the
0:01:59the proposed schemes that or
0:02:01and i sure use this measure is our to the case that a where is the operation with
0:02:09oh here have a star
0:02:11oh have you about in some
0:02:14but you creation problem
0:02:18as we all know that that our ram for C G for short
0:02:22it it for of
0:02:23the hard to lead to a to be G after uh by the actual position on the screen server
0:02:31the automatic
0:02:34and the lights is of just signal as on a lot of interest in the biomedical engineering from
0:02:41oh problem this
0:02:44as we can see
0:02:52but because see
0:02:53uh a C D signal
0:02:55consists of three distinct waves
0:02:58she way
0:02:59and uh the most significant part of to a caress contrast
0:03:03and that's she way
0:03:05oh problem
0:03:06this figure
0:03:07as we can see the most
0:03:09useful clinically useful information
0:03:12i be file
0:03:12the wave boundaries
0:03:14and the interval durations
0:03:17oh these C D C
0:03:18but that's why the deviation
0:03:21which means
0:03:22the um
0:03:23determination of the with peaks with bob are used and the estimation of we've one
0:03:28it's a very important step to the interpretation of the C D C
0:03:34so things that or as complex as
0:03:36are a more significant part of the each signal
0:03:39is detection is relatively easier
0:03:43it can be used as a reference
0:03:46to the C and she you deviation
0:03:49and um
0:03:51the C and C waves
0:03:52generally they have a low should
0:03:55and due to the presence of the noise
0:03:58and a baseline one three
0:04:00if the are deviation it's more complicated than the as complex as
0:04:04the H
0:04:06so basically we will
0:04:09do firstly to dress complex this and defined a search each both before and after
0:04:14locate look S you but
0:04:17so we can find in the three sure
0:04:23existing methods
0:04:25the first
0:04:26existing mess are the first class is based on reading techniques
0:04:29the president
0:04:30it is to use for techniques to remove the noise
0:04:33and the base a wiring
0:04:35and it
0:04:36and then
0:04:38we you apply by read and it uses specialist
0:04:41term where are the weights
0:04:43and we are are problems
0:04:46is that in
0:04:47class of existing that's sort is based on
0:04:49basis is expansion that makes
0:04:51i mean we we use a basis is especially at least for the functions
0:04:55and the way to use
0:04:57some thresholds to determine where are wave peaks and where the bar
0:05:02a certain class of existing mess or is based on a a piece and that pattern recognition that source
0:05:08and the we can also
0:05:10a by some realistic signal model to feed the C D signal and to estimate
0:05:17to parameters of the model of to do the nation
0:05:20of the D C D C
0:05:22so some particular patient has been given to a part of you
0:05:26you our paper
0:05:28we we propose
0:05:29to use
0:05:30uh uh easy inference
0:05:32together with is
0:05:33some similar to ours
0:05:36to deal with this problem
0:05:38so now i would introduce you
0:05:40or a mathematical model
0:05:43the grass complex the our application are assumed to be detected
0:05:47so a which will not be considered here
0:05:50and that then
0:05:51we propose to use a deep each no overlap
0:05:56to cover the whole dct T stick
0:05:59so basically you we have the same number of the
0:06:03no if are asking to roles
0:06:04we now hour
0:06:06process we know
0:06:07and for each not here S components
0:06:10can be seen as a combination of all to me the to mean is
0:06:15which are present you way and if you with
0:06:18plus uh local based
0:06:21so take the that parts as an example
0:06:24this is she weights with this
0:06:27a uh processing window can be seen as a a convolution
0:06:32bernoulli gaussian sequence
0:06:34a on the one would got the sequence
0:06:35with a
0:06:36on the one
0:06:37impulse response
0:06:39so that
0:06:40oh no impulse response present a waveform
0:06:43and this
0:06:44on the one really gaussian
0:06:45to represents
0:06:47the with locations and we want to
0:06:50and then we can do same to the T waves
0:06:52so basically we have some conclusions here
0:06:55and we have a this
0:06:57but the medical signal more
0:06:59and this sick it represents the sequence of the baseline a local base fine
0:07:04and this W P A a noise
0:07:06which use that students be gone
0:07:10and for some more we also read
0:07:12the proposed to use
0:07:14or a basis expansion techniques
0:07:16to represent a on no one point four
0:07:19so this
0:07:20technique has to be used
0:07:21oh C to denoising and also on C compression
0:07:26the advantage of
0:07:27this technique
0:07:29i is
0:07:30that is that we can have a used to mention though
0:07:34on on parameters
0:07:36and also we
0:07:38we can seen that noise some most the version
0:07:41oh the with that
0:07:43for them more we also propose to use of force degree polynomial model
0:07:49local baseline
0:07:50but this is a patients what previous work
0:07:53which assumes that the local
0:07:56baseline which in each no cresting in ball is a cost
0:08:02with all this we have to a better representation of
0:08:06the the not is in the roles in the processing window
0:08:11red once are all time
0:08:14so we have the we form we've location is up to two
0:08:18all the what she ways
0:08:19and the
0:08:20baseline like is the noise was far
0:08:26i will introduce you if you model
0:08:29that we don't know the you for realise of the product
0:08:32all of the that it functions and priors
0:08:36to represent the posterior distribution
0:08:39and to estimate and have
0:08:43by assuming the noise in our model is zero-mean gaussian
0:08:47we can have a just like would function
0:08:50and concerning the priors of the unknown parameters
0:08:55they can be assigned according to the T C you signals back to this
0:08:59just we only expect
0:09:00at most one
0:09:02she wave in each to thirty the both
0:09:05we can assign a block cost rate
0:09:07to this
0:09:09and a by assuming the independence
0:09:11oh of each individual
0:09:14thirteen thirteen both
0:09:16priors of the
0:09:18the the the whole
0:09:20she indicator vector can be to by a product of each about divided so
0:09:27and the since we have used a run gaussian sequence
0:09:31so when there is a way it it
0:09:33we assign a zero-mean gaussian prior to the
0:09:36and you
0:09:37a store C to this petition
0:09:40um report weeks back and not only the positive
0:09:43i each but also the negative to every is a zero mean gaussian prior
0:09:48and for the bit form is also a zero-mean gaussian
0:09:51but we from partitions
0:09:53a course we expect to not only make your of what to parts but also to part of the waveform
0:10:00and the for the baseline for she's is also a zero-mean gaussian at the noise variance prior is you for
0:10:08not that you have to all these conjugate priors because it can
0:10:13some five
0:10:14it it can be the computation
0:10:16so now we have a lower posterior distribution which is the
0:10:20the product of all this priors with the likelihood function
0:10:23but uh it's a topic
0:10:26distribution which
0:10:27uh we can not
0:10:28compute a form estimators
0:10:30so that why we propose
0:10:32to use a map model can estimate of a source
0:10:36to generate samples
0:10:38a sounded sick leave for this posterior distribution
0:10:41and to estimate other parameters of the um
0:10:45the way to model
0:10:47so this is a proposal a keep were
0:10:50exactly is uh
0:10:52there is a simple and S and C and C
0:10:55simulation is or
0:10:57uh but that's to this brought constraint
0:11:00we have a uh as
0:11:02a slight modification to
0:11:05class two kids that we're
0:11:07which that instead of
0:11:09that but that whole the to cater
0:11:13we can that generate
0:11:15brought by block
0:11:17we have only that had a size that it is only
0:11:20last oh the each block that's one
0:11:26you to the company
0:11:27patient no efficient
0:11:30so basically is that after generate samples for C where cage or we be see if the is
0:11:37you way
0:11:38with is that what that dude
0:11:39and for for you it is the same
0:11:42and once we have a
0:11:44that putting the
0:11:46indicator are
0:11:48we was that of the waveform for block
0:11:50or the process we do and the the baseline baselines and the noise
0:11:54noise bar
0:11:56and the the estimates can be a obtained by using
0:11:59for this
0:12:00this straight
0:12:01i meters
0:12:02can be
0:12:03all all T and by using a a sample based the map estimator
0:12:07and for this
0:12:09the rest of the other having to get a job in by a and M
0:12:15before drawing use the solution is so mention that the
0:12:19the pre-processing step yeah our vacation
0:12:22it's been done by has are region which is very that's
0:12:26to you your as complex
0:12:28and the processing window not set in set to ten part eight
0:12:32this is a force
0:12:34we cannot have a very large
0:12:37uh that that thing we know last course you C you has it
0:12:40so so though
0:12:42stationary nature
0:12:44and we can not have a a better little last either course
0:12:48we have to use several observations
0:12:50to be able to estimate a we've a problem was that
0:12:55a if you have a christ if you are interested you can find more
0:12:58details now journal paper sure here
0:13:02and now
0:13:03oh use so use some to we examples
0:13:06for a that to database
0:13:08the first example is from
0:13:11uh to that it's is there a it's a three six
0:13:14uh this example has been shown to be force it had some
0:13:17but are we of the reasons and a long beach
0:13:22a state the of the posterior distribution no that C N you will be cater pages
0:13:27which means
0:13:28the probability of having that she or to location
0:13:33so as we can see
0:13:37to to give sample right handed it is to locate the way so that's the interesting
0:13:42probably a of a proposed are them
0:13:44instead that of using a read it and you to determine whether it is a way of one
0:13:49i one map
0:13:51that is the map estimator
0:13:53can tell us
0:13:54the most probable position of having a Q a here
0:13:58and C a small to have the peaks that the wave
0:14:02it just tell us that peaks
0:14:03and the well there is can be
0:14:06turn it by
0:14:08this is estimates of we four
0:14:12but using different creature
0:14:14and this is the reconstructed signal
0:14:17and you read and estimated a baseline of be baseline glass
0:14:21and all it no signal a don't at because you are very close
0:14:26and and the
0:14:27a second rate shows us to detect if it useful
0:14:33so here is an example with a premature ventricular contraction now
0:14:40the she where is the scene
0:14:42and follows by a giant
0:14:43if your as
0:14:45as follows it follows by first
0:14:47Q way
0:14:49as we can see
0:14:51most most our written can handle this situation
0:14:54the posterior distribution of this
0:14:56i is very low
0:14:58so there would be no false alarms
0:15:01and this in
0:15:03you work that she way is about detected
0:15:06with the battery of five was to reduce pollution
0:15:09and this this you just is a cost that it's signal
0:15:12and the the detected if it was one
0:15:17another example of
0:15:18but if physique she weights
0:15:20and that we can see
0:15:22we can have a a very of years
0:15:25she we've form estimation
0:15:28and this is reconstructed signal for this dataset
0:15:33the last example is
0:15:36a does that we are
0:15:37there are some as as so you wave fred
0:15:42so you can see you when there is no to it in the thirty two tomorrow
0:15:46we can have we would have a very low
0:15:49posterior distribution
0:15:50and now or map estimator
0:15:52we're not need any
0:15:55for some
0:15:57so this is a reconstructed see no we is that it that the fiducial points
0:16:02for this is that does that
0:16:05and now have that to show is um
0:16:08constative C of cooperation is with are
0:16:11we have implemented the filtering in that one of the
0:16:15filtering techniques and the other is the bit
0:16:19is the basis expansion time
0:16:22as you get C in terms of of the detection is its ability
0:16:27our our would some out performs the classical mess source
0:16:31and also in terms of of D H arrow
0:16:36our our in is
0:16:37slightly better is comparable in that's better
0:16:40and based on the compilation that like to remind you that uh
0:16:45there are two advantages other advantages of our
0:16:50the first is that uh
0:16:52we can provide as well as
0:16:54the we a estimates
0:16:57and if probably is
0:16:59and they are of and he's is that the
0:17:03this is a ms art
0:17:04and we can provide
0:17:06the reliability ability information
0:17:09such as the inter a a confidence interval
0:17:12except for
0:17:13uh oh of the with see which is
0:17:16very interesting for of the medical yep
0:17:20so here the convolution
0:17:22the know what do here is that we have proposed a is model
0:17:26for the not rest in the state D signals which is based on a blind deconvolution problem
0:17:32yeah and that we have proposed a block example are
0:17:35to estimate
0:17:36the on parameters of space model
0:17:40and and the as process back
0:17:42since yeah are more the we have damage use estimation
0:17:45so we can
0:17:47i have to
0:17:48to where at down that don't other nodes
0:17:51you you some problem that's that really a detection problem
0:17:54and it's is we can have the with form estimation
0:17:57we made at at a it's miss detection problem
0:18:00so all that force back
0:18:03and since it is not yet
0:18:05uh online application
0:18:07uh we are
0:18:09currently investigating the sequential mass are
0:18:12uh us to create role what colour methods
0:18:15to this
0:18:15basic mode
0:18:17set for the attention if you a is it we have a a little that that them or a personal
0:18:22web site
0:18:57i test your back
0:19:11holman breeders to to the you nation problem
0:19:16uh yes i think we have a
0:19:18and we have calculated that better at lower our are based on the for example the cs C standard
0:19:24the and
0:19:27because tolerance
0:19:29all of the
0:19:31deviation work
0:19:32and we can
0:19:34feed this
0:19:35stand stand it is that
0:19:36you're what is not
0:19:37like a
0:19:38for you
0:19:39is is that
0:20:43yeah i that
0:20:44if we con
0:20:47use that come back to this
0:20:51i mean yeah and
0:20:54should be some noise present in signal
0:20:56and if the noise
0:20:59is the very likely to that
0:21:01with forms
0:21:02it could be
0:21:04uh i
0:21:06the um how to say
0:21:08our data are within
0:21:09we you we should you pretty per but probably to addition of the way
0:21:15yeah the um
0:21:18why we have a sharp here but the L is not about their point five
0:21:23since we assign the prior that to the but but but it you of having no way is is you
0:21:28pour to the
0:21:30a probability of having a a will be zero and with a map estimator just
0:21:34you would not be a problem
0:21:36yeah i
0:21:38is problem
0:21:41and this also is the don't know which is possible because they here we have a
0:21:47uh uh to be
0:21:49the noise
0:21:50which is
0:21:52similar to a way if
0:21:54sometimes i i just think it's a way
0:22:07yeah yeah i i i have read to this point
0:22:10and the
0:22:11to to to them
0:22:13computational load it's um more
0:22:20so you