 A warm welcome to this second session of interaction in this course on wavelets and multirate digital signal processing, in which for variety we now bring in again a student presentation. We have to record now a series of tutorial sessions and student presentations and we shall intersperse these. Largely speaking we shall have tutorial sessions, here and there very infrequently we could have some presentations or exposition or expressions of thought by students of the course on the theme of the course. In that spirit today I am going to request one of the students who has attended this course namely Tony Sebastian to make a presentation on the excellent application presentation that he worked upon in this course. I shall not take away from him the thunder of explaining what he did, safe to give an introduction in a few lines. I shall just introduce the broad theme of his application and the broad theme relates to what is called denoising. Now denoising as the name suggests means an operation of separation of wanted and unwanted in a mixture of signal and noise. As expected normally the noise or the perturbation is unwanted and it is often the case that when one goes into the wavelet domain particularly in the context of biomedical signals it is easier to separate the wanted signal from the unwanted noise. We could have several instances of this but what we have today is essentially a suppression of respiratory artifacts which Tony would explain to you on his own. So, now I introduce Tony Sebastian and I request him to make a presentation on his work related to suppression of these kinds of artifacts and wavelet based denoising. Hello everybody myself Tony Sebastian of second year MTAG student of biomedical department IIT Bombay. Welcome to my application assignment presentation on wavelet based denoising for the suppression of respiratory artifacts in impedance cardiogram signals. Now the technique I am going to present here was originally developed by Dr. Vinod K Pandey and Procept PC Pandey of E department IIT Bombay. Now what is impedance cardiography? Impedance cardiography is a non-inversive technique for monitoring stroke volume and other cardiovascular indices and thereby obtaining diagnostic information on cardiovascular functioning by sensing the variations in thoracic impedance due to the change in blood volume. Now since most of you are from engineering background many of the technical terms in this definition are not familiar for you. So, it is better to have some basic understanding of the heart structure for getting into this definition and hence and fence for the better understanding of the project. Now let us look into the structure of the heart. Now this is the structure of the heart most of you may be familiar with but this is not how the heart looks like. Let us look into the actual structure of the heart. Now this is how the structure of the heart look like. Thanks to my friend Ajay Thijore for drawing such a wonderful diagram for this presentation and now let us look into the structure of the heart with an ingenious perspective. We can visualize heart as a combination of four chambers. Here you can see one chamber that is the right atrium. Here you can see the second chamber that is the left atrium. Now here is the right ventricle and here is the left ventricle. Now the right side of the heart that means the left hand side of the picture deals with the deoxygenated blood and the left side of the heart that is the right side of the picture deals with the oxygenated blood. Now this four chambers can be visualized as a combination of as four pumps nor it is very similar to the mechanical pumps. The function is just to pump the blood. Now blood from the different part of the body will enter into the heart to the right atrium. Now here you can see the two major vessels that is this one and this one. These are the superior venacova and the inferior venacova. Superior venacova will be bringing blood from the upper part of the body to the heart and inferior venacova will be bringing blood from the lower part of the body to the heart. Now the name superior and inferior is not because of its functioning it is just because of the position. Now as soon as this right atrium filled with blood this blood will be right atrium will pump the blood to the right ventricle. There is a valve separating the right atrium and the right ventricle that is this one that is the tricuspid valve. Now from this right ventricle this right ventricle will pump blood to the lungs for getting oxygenated. As you know blood which is coming from different part of the body to the heart has carbon dioxide in it and we need oxygen in the blood for the body functioning. Now in the lungs this blood will exchange carbon dioxide and oxygen for that this right ventricle will be pumping blood to the lungs through the pulmonary artery. This is the pulmonary artery so right ventricle will be contracting in this direction by that time this tricuspid valve will be closing and this particular valve will be opening. This particular valve is known as semilunar valve because it is opening to the pulmonary artery it is known as pulmonary semilunar valve. So blood from this right ventricle will pump to the pulmonary artery and it will go to the lungs and from the lungs blood will come back to the heart to the left atrium. So blood will come back from the from the lungs to the heart through this pulmonary vein to the left atrium. Left atrium is on the top side from this left atrium blood will be pumping into the left ventricle. There is a valve separating left atrium and the left ventricle that is known as the mitral valve or the or bicuspid valve. Now comes the most important part of the heart. Left ventricle out of these four chambers I say left ventricle is the most important part because left ventricle will be pumping blood to the different part of the body. As you know since our body parts are far away from the heart it has to do a lot of work for pumping blood to the different part. So this left ventricle will be contracting with maximum force and when it is contracting this particular valve will be opening. This valve is again a semilunar valve which is the iotic semilunar valve because it is opening to the iota and from this left ventricle the blood will be pumping to the iota. Iota will be taking away blood to the different part of the body. Here you can see different branches of this iota. Now here comes here you can see the beauty of this cardiac design. I say the beauty because see compared to all the other chambers left ventricle has to do maximum work for that this particular muscle has maximum thickness compared to other valves other other other other chambers. Now another important factor here is why left ventricle is important is see even if this atrial muscles have some problems or even if this atrial pumps are not working because of the gravitational force and because of the weight this tricuspid valve and bicuspid valve will automatically open and 70 percentage of the blood will automatically fall into the ventricle even if this atrial pumps are not contracting. So we can say disorder is related to atrium are not comparatively that much danger compared to the disorder is related to ventricle. Now let us look into some of the waveforms related to heart. This upper dotted line indicates the iotic blood pressure that is when we are measuring our blood pressure by using normal pressure meter we will be getting this iotic blood pressure that is when we are going to hospital the doctor is measuring our iotic blood pressure by using his normal pressure senses and this blood pressure what we are normally getting for a healthy man 80 to 120 that is the systolic and diastolic pressure of the iota. Here is the iota you can see because normally our doctors are measuring blood pressure in the hand this blood pressure is slightly whatever pressure we are measuring is slightly lesser than the iotic pressure but it gives an approximate measure of iotic blood pressure. Now the second waveform is the ventricular blood pressure and the third one is the iotic pressure. Now at the initial phase you can see iotic pressure is little bit higher than ventricular blood pressure. Now when this iotic pressure is higher than ventricular pressure this tricuspid valve as well as the bicuspid valves are open and at this time this left atrium and the right atrium are contracting and blood will be falling to the ventricles. Now as soon as this ventricles start contracting pressure inside ventricle will increase start rising and at some particular point it will overcome iotic blood pressure. At that particular moment this iotic valve this tricuspid valve and semilunar valve will be closing because the pressure inside the ventricle is higher. So this valve will be closing and it prevents the bike flow and the pressure inside ventricles will be start again and again and building. At some particular point it will overcome the pressure inside iota and at that particular moment this semilunar valve will be opening and ventricle will be pumping blood into the iota. This will be continue for a moment when this ventricle start relaxing the pressure inside the ventricle start reducing. When this ventricular pressure becomes lower than the iotic pressure this iotic valve will be again closing and ventricle starts relaxing the blood flow stops. Now this ventricle pressure again will come down and when it comes down below the iotic the atrial blood pressure this tricuspid valve again will tricuspid valve and bicuspid valve again will open and blood will again come to ventricle and this cycle will be continuing. Now next waveform is iotic flow as you can see this particular point is the point at which the iotic valves are opening the semilunar valves are opening and at this particular moment blood flow to the iota starts and it will be a pulsatile flow you can see the if you try at our radial atria or some particular points you can see the pulsatile flow you can feel the pulsatile flow of the blood. So, this will be pulsatile and when this particular valve is opening blood flow will be for a certain blood flow will happen and it will continue for a moment and once the valve closes flow again stops again in the next cycle it will be continuing. The next two waveforms will be looking a bit later and the next waveform this particular waveform is the most common biosignal and most of you might be familiar with this particular waveform this is the electrocardiographic waveform in most of the films the climax scene will be some ECG waveforms will be coming some ECG waveform will be coming and once the waveform will be stopping that will be the end of the story. So, this waveform particular waveform this electrocardiographic signal gives a measure of the electrical activities of the heart this particular waveform I will draw once again and I will explain. Now, a basic electrocardiographic signal will be looking something like this. So, this is the p wave then q r s and a t wave this p wave indicates the beginning of contraction of the atrial pumps. So, as I show in the previous picture the atrial pumps will be start contracting at this axis is the time axis. So, in each cardiac cycle at this particular moment the atrial pumps will be start contracting and this q r s complex indicates the contraction of ventricles. So, atrial contracted at this particular moment the ventricular pumps are start contracting when the ventricular pumps are start contracting the upper pumps again start relaxing at t wave the ventricular pumps are relaxing. So, by looking at an electrocardiographic signal a cardiologist can tell how good a heart is working or what are the major disorders of a heart we can tell. Now, coming back to the previous picture. So, this gives a measure of electrical activities of the heart. Now, the last one is the phonocardiogram that actually gives the cardiac sounds cardiac sounds mean our doctors basically by stethoscope they are hearing these sounds these sounds are actually sounds of closing of these walls first sound is the closing of this iotic walls and the second one is the closing of this semilunar walls. Now, we will go back to impedance cardiography our technique. Now, for impedance cardiography we will place 4 electrodes like this in the body and we will be injecting current through the upper 2 electrodes. That means, we will be basically injecting a high frequency current of very low amplitude and we will be measuring the potential voltage difference between these 2 particular moments and we will find out the impedance variation across these 2 particular point this area is the thoracic region. Now, we will go back to the previous picture and understand the 2 waveforms which we kept aside. Now, these are the 2 waveforms we kept this one is the delta z that means, the impedance variation because of the blood flow because in every body part in the thoracic region there are basically 3 types of conductive tissues. That is first one is the in this area basically we can see 3 different types of tissues 1 a is the rib cage you can see something like this we have ribs this side and this side rib cage bone these are bones these are basically non conductive then here we will be having the lungs. Lungs are muscle tissues soft tissues they are conductive, but not that much conductive compared to blood and next one will be the Venokova and the thoracic iota Venokova will be somewhere here and iota will be here and compared to all the other body fluids or body parts blood is more conductive if for example, if you say a conductivity of blood is 150 then conductivity of lungs and other muscles are 1500 or 10 times lower compared to blood. So, whenever the heart is pumping what will be happening this iota will be changing its dimension now because of that the voltage we are measuring changes and we can sense we can find out the impedance variation from that now this is how the impedance variation looks like whenever the iota is open that means blood is pumping to the iota its dimension changes and the impedance go down resistance go down. So, here you can see a fast change in the impedance and it will again come down again in the next cycle the same variations will happen and the time derivative of this particular waveform is known as the impedance cardiogram signal. By using certain parameters measured from this impedance cardiogram signal we can find out stroke volume and as we told other cardiovascular parameters the three major points in this particular waveform are the B point the C point and the X point these three points are the major points in this particular waveform and this is the time derivative of this impedance waveform and this is known as the impedance cardiogram signal. Now, coming back to the slides for finding out how to estimate the stroke volume. Now, for estimating the stroke volume we have a special formula developed by Cubisect et al that formula is stroke volume here in the slide you can see stroke volume is equal to rho into l square divided by z naught square into negative of d z by d t max into t l v e t. Now, let us spend a minute in understanding this formula rho is the resistivity of the blood that is a constant that is somewhere around 150 then l is the distance between the electrodes as we show in the previous picture the distance between these two electrodes are the l that is the we are actually modeling this thorax as a conductor. So, the distance between these two points is l. Now, the next parameter is the z naught that from the previous picture this delta z will be above a particular base value. So, that particular base value is z naught normally it is around 25, but it varies from patient to patient and the next parameter is d z by d t max. d z by d t max this is the d z by d t waveform d z by d t max is not the difference between the c point and x naught x point, but it is actually the difference between b point and the c point. Now, b point what is the significance of b point and x point see here comes the significance of the other waveform see b point indicates this particular point this particular point is the point at which ventricular pressure is above the iotic pressure that means this is the particular moment at which the iotic valves are opening and the blood is pumping into the iota and x point is this particular point and at this particular point ventricular pressure will be again come down below the iotic blood pressure and at this particular moment this valve will be closing. So, during this period that means the time period from which b point to x point iotic valve is open and this is the particular moment in which iotic valves are opening and this is the duration in which ventricles are pumping blood into the iota. So, T L V E T is basically the point difference between time duration between b point and x point. So, from this waveform impedance calligraphy waveform we will be basically taking d z by d t max as well as T L V E T. Now, the other major parameters related to heart they are stroke volume stroke volume is basically the amount of blood pump by heart in one during one heartbeat that is during which this iotic valve is open what is the amount of blood pump by the heart in one cardiac cycle and the second parameter is cardiac output. Now, cardiac output is the amount of blood pump by heart in one minute that is basically stroke volume multiplied by heartbeat that is we can see by counting just by counting number of C points or C points will be very easy to detect and by counting the number of C points or by counting the QRS complex in the ECG diagram we can get the heart rate and then by multiplying stroke volume into that particular value we will be getting cardiac output. And next parameter is left ventricular rejection time that is basically the time difference between the b point and x point. Next parameter of interest is systemic vascular resistance that is the resistance offered by iota to the blood flow and the next cardiovascular indices is left cardiac work that is as I told you we are basically among all the cardiac parameters we are basically interested in this part of the heart that is left ventricle the work done by this part of the heart for pumping blood into different part of the body that is the left cardiac work. Now, coming back into the artifacts, impedance cardiography signal basically how few types of artifacts what are artifacts? Artifices are actually man made signals which are not necessary which are unnecessary in impedance cardiographic point of view. Now, we have two types of artifacts major artifacts are the respiratory artifact and the motion artifact. Now, respiratory artifacts are very low in frequency that is basically from 0.04 heads to up to 2 heads and motion artifacts are up to 10 heads. Now, if we analyze the impedance cardiographic spectrum we have signals of interest from 0.8 heads up to 20 heads. Now, here you can see that this motion artifact and respiratory artifact has signal components of components in the same band. Now, here in this particular presentation we are looking into respiratory artifact separation. Now, in this particular picture you can see a signal in an impedance cardiographic signal recorded during exercise. So, this is an ICG waveform recorded from a healthy subject during exercise. So, this particular signal has both respiratory and motion artifacts. Now, this particular peaks high peaks are because of motion and this slow varying baseline is because of respiration. Now, what is the disadvantage of this artifacts? Here you can see when we are finding out d z by d t max this particular points will be coming into picture and they will be introducing severe errors and because in the formula we are using this d z by d t max this will introduce severe errors for calculating stroke volume. Also this baseline drift also introduce errors in d z by d t max also in TLVET calculation. So, it is very important to suppress these artifacts for the proper estimation of stroke volume and other cardiovascular indices. Now, the project objective is to investigate different denoising techniques for the suppression of these artifacts and study in detail the wavelet based denoising technique for the respiratory artifact suppression and investigate few different wavelets and see how these wavelets are useful in this denoising applications. Now, there are different methods of artifact suppression. The first technique is breath hold. Now, you can see the respiratory artifact is because of respiration. So, the easiest way or the simplest way of respiratory artifact cancellation is holding the breath. Now, the problem with breath hold is when we are holding the breath cardiac activity will always go down. And another problem is when we are recording the impedance cardiogram after exercise it is difficult to hold the breath. Because after the exercise cardiac activity will be more. So, it is difficult to hold the breath. The second established technique is ensemble averaging. The problem related ensemble averaging was proposed by Howard Settler in 1990. The problem related to ensemble averaging is it will distort the waveform or it will remove the bit to bit variability in the impedance cardiogram. Actually, we are interested in stroke volume as well as the variation in the stroke volume. In each bit how much the variation is coming in the blood that gives the that makes the sensing diagnosis. Now, when we are doing ensemble averaging it may blur this B point, X point. And so, we will not get after ensemble averaging we will not get this clear B point and X point. And hence it will introduce distortion or errors in calculating the stroke volume. Next technique is adaptive filtering which is developed by a professor P.C. Pandey and Vinod K. Pandey again in 2005 in IIT Bombay. The problem with adaptive filtering is adaptive filtering is always good in biosignal denoising if we have a reference signal. But for adaptive filtering we need to have a reference signal for that we have to keep a pressure sensor in the nostril. So, this is again problem of pressure complaints that is how adaptive filtering works. Then the next technique is the one which we are working on is wavelet based denoising which is known as the level dependent thresholding which is again developed by Pandey and Pandey in 2007. Now, in wavelet based denoising selection of wavelet bases is very important. Why selection of wavelet bases is very important? In denoising applications it is observed that basically in whether it is in ECG denoising or in ICG denoising it is observed that if the wavelet and the waveform has some similar shape then the denoising is better always better. So, in ECG denoising or in ICG denoising there are some particular wavelets which has some shape similarity with biosignals. Those wavelets will be giving better denoising better separation of noise and the signal. Now, let us see some waveforms. This is the experimental setup of impedance cardiography. Here you can see the recording of impedance cardiogram 4 electrodes are placed over here 2 in the upper part and 2 in the lower part. This is the impedance cardiograph and through the upper 2 electrodes current will be injecting high frequency current of low magnitude will be injecting and this will be measuring the this lower 2 electrodes will be used for measuring the voltage and hence we will be calculating the impedance variation. Now, this impedance cardiograph this particular impedance cardiograph is the HIC 2000 impedance cardiograph from bio impedance technology. Now, we have acquired the signal at sampling rate of 500 hertz. Signals are recorded both in resting condition as well as signal recorded by performing different activities one signal which we saw before is while performing some activity. Now, this is the basic wavelet decomposition. The original waveform is sampled at 500 hertz. Now, each detail gives a band pass signal and each approximation gives a low pass signal. Now, when we are decomposing the signals into different levels. For example, if we are decomposing the signal into 8 levels or 9 levels the first decomposition if the sampling rate is 500 it will be having signal components up to 250 hertz because of Nyquist criterion. So, the first detail will be having components from 125 hertz to 250 hertz and the approximation will be having signals from 0 hertz to 125 hertz. Second detail will be having components from 62.5 hertz to 125 hertz and second approximation will be having components up to 62.5 hertz. Similarly, in 8th level detail it will be having components from 0.98 hertz to 1.95 hertz and approximation will be having components from 0 to 0.98 hertz. Now, in this particular method what we are doing is we are decomposing the signal into different level and we are also decomposing the artifact into different components and we are seeing up to what level the signals are present and up to what level the artifacts are present. Now, we use different wavelets for the decomposition and this is the 10 level decomposition of an ICG signal which is shown here by using the coflit 5 wavelet. Here you can see detail 1 and detail 2 they do not have much components detail 3 they do not have much components because we do have components only up to 20 hertz. These are high frequency components we do not have any signal or artifact in this area and from the 4th approximation 4th detail onwards we have signal components 5th, 6th, 7th up to 10 all the approximation all the detail and approximation this is an ICG recorded under breath hold condition. So, it is not having any artifact it is having only purely signal component. So, all the 10 detail and approximation has signal components. So, we cannot use this particular wavelet for the separation because it is not capturing signal components in any particular detail because see 8th level detail or 9th level detail are very low frequency components. So, this will be having even if the signal supposed to have a respiratory artifact this will be having signal as well as d 9 will be having d 9 and d 10 will be having signal as well as artifact. Now, this particular waveform is the 10 level decomposition of same ICG using the watch 6 wavelet. Again you can see all the 10 detail and the approximation has signal component components in it. Now, here comes the one which we are interested in this is the 10 level decomposition of the same ICG by using d may wavelet that is the discrete may wavelet. Now, here you can see d 1 and d 2, d 3 as such they are high frequency components they do not have we do not have any components in that, but d 4, d 5, d 6, d 7 and d 8 has signals in it. But in d 9, d 10 and 8 and we do not have any signal components. So, this is basically d 9 is the low frequency area that is from 0 to 0.98 hertz. So, here is basically in this frequency band the respiratory artifacts are coming and in this particular session we do not have any signal component. So, this particular wavelet will be very useful for separating the signal and the artifact. So, in our denoising technique we will be basically adding this first 8 details and we will be removing d 9, d 10 and 8 and for the artifact separation. So, the artifacts free ICG signal is obtained by adding d 1, d 2, d 3, d 4, d 5, d 6, d 7 and d 8. Now further studies at SPLAB showed that similar results this obtained by DMA wavelet can be achieved by using one more wavelet that is similar 26 wave also gives similar result as obtained by DMA wavelet. Here you can see one ICG waveform recorded under resting under resting condition, but with respiration. This first waveform is the ICG recorded with respiratory artifact. The slow varying oscillations are because of this respiration sinusoidal oscillations are because of the respiration. Now the second wave is denoist ICG waveform by using DMA wavelet that is we are we have decomposed the signal into 10 levels by using DMA wavelet and reconstructed the signal by using first 8 details. And the third one is same procedure followed by using similar 26 wavelet by visually we can see both the wavelets are giving same performance. Now fourth one is the artifact which is extracted from the ICG waveform by using DMA wavelet. This is basically first waveform minus the second waveform that is the first waveform which has artifact and the second one is the denoist waveform. The difference is basically the artifact signal and the last one is the respiratory artifact removed by using similar 26 wave waveform. Here you can see both the waveform both the wavelets are capturing exactly same artifact and exactly the same signals. Now why this happens? See here we can see the wavelet and scaling function of DMA wavelet and similar 26 wavelet. Here DMA wavelet has 1 note 2 samples in the wavelets compared to similar 26 wavelet has 52 in it. Now here you can see it almost has same shape. We can see a shape similarity in DMA wavelet and similar wavelet both the mother wavelet and the scaling function has similar shape for DMA wavelet and similar 26 wavelet. Because of this similar shape similarity we are getting same result for both the wavelets. Now this is a basic block name of wavelet decomposition by using filter bands. Here you can see we are using we will be having low pass filter decomposition filter and again here after this here we will be getting wavelet coefficients and for getting the details by we have to use the reconstruction filter of same length again. Here you can see the advantage of using similar 26 wavelet because similar 26 wavelet is having only 52 coefficients compared to DMA wavelet having 1 note 2. So, since we are using 10 level decomposition if we use similar 26 wavelet the calculation complexity will be comparatively very less compared to DMA wavelet. Now coming back to the summary of the presentation we can see that both similar 26 wavelet and DMA wavelet are giving better performance in artifact suppression and they are giving almost similar performance compared to other wavelet and we studied different wavelets like coflit wavelet and the bar 6 wavelet they were not comparatively that much good compared to these wavelets. And advantage of similar 26 wavelet is it reduces the calculation complexity because of it is having low number of filter coefficients. Now the future work will be we have we can study the applicability of wavelet in denoising motion artifacts and also we can investigate whether any other wavelets are giving denoising application. What you can try is you can use this technique or similar technique for the denoising of ECG denoising because ECG data files are available in MITB database you can download some ECG database data signals from MITB database and you can try similar kind of technique in denoising of ECG signal. ICG signals are till now it is not that much popular compared to ECG signal. So, we do not have any open database compared to ECG signal. So, you can download some ECG signals from the MITB database and I hope you all will try some of the denoising techniques may be this techniques may not be applicable, but some modification of this may help in denoising ECG signal denoising also. So, you can try that kind of denoising and these are few reference papers we have referred. First paper is QBSA Caterals paper. They propose the impedance cardiographic technique itself and the few other reference papers are denoising techniques and signal processing techniques. If you are interested in further studies you can refer to these papers and get more information on this denoising technique and more about impedance cardiographic technique. That is all about the presentation. Thank you.