DE-NOISING OF ECG USING WAVELETS AND MULTIWAVELETS

May 2, 2023

 

PRESENTED BY,

Dr.S.BALAMBIGAI ASSISTANT PROFESSOR(SRG)

DEPARTMENT OF ECE /KECPh.no :9443895494Email: sbalambigai@gmail.com

May 2, 2023

Need for ECG processing:2

Electrocardiogram and heart rate are vital physiological signals to monitor cardiovascular diseases(CVD).

A report by World Health Organization (March 2013) estimates cases of CVD (including heart attack, stroke, angina) will increase from 17.3 million in 2008 to 23.3 million by 2030 .

Hence, frequent monitoring is required for those under great risk for cardiovascular diseases.

This signifies the importance of research in the recording and processing of electrocardiogram signals.

May 2, 2023

Introduction - Electrocardiogram3

The electrocardiogram is a graphic record of the direction and magnitude of the electrical activity of the heart.

One cardiac cycle consists of the P-QRS-T wave.

The clinically useful information is found in the intervals and amplitudes of an electrocardiogram.

ECG

Recorded with surface electrodes on the limbs or chest.

ECG is used to measure the rate and regularity of heartbeats, the presence of any damage to the heart.

THE HEART It is a 4 chambered

muscular organ consist of 2 atriums and 2 ventricles.

It function in a regular fashion to pump blood thought the body.

Average heart rate of a human being is 72beats/min

May 2, 2023

Uses of ECG7

To find the orientation of the heart. To determine heart rate and to analyze the working mechanism of

the heart. To determine the extent of damage in various parts of the heart

muscle. To diagnose an impending occurrence of heart attack or CVD. To determine unusual electrical activity in patients with abnormal

cardiac rhythms. To determine the thickness of the heart muscles. To determine blockage areas and restricted blood flow areas in the

heart muscles.

May 2, 2023

Normal ECG waveform (Chinchkhede et al 2011)

8

May 2, 2023

Types of Noises in Electrocardiogram10

• Power line interference• Base line drift• Motion artifacts• Muscle contraction• Electrode contact noise• Instrumentation noise generated by electronic devices

Baseline Wander Baseline wander, or extragenoeous low-frequency high-

bandwidth components, can be caused by: Perspiration (effects electrode impedance) Respiration Body movements

Can cause problems to analysis, especially when examining the low-frequency ST-T segment

Power Line Interference Electromagnetic fields from power lines can cause 50/60 Hz sinusoidal

interference, possibly accompanied by some of its harmonics Such noise can cause problems interpreting low-amplitude waveforms and

spurious waveforms can be introduced.

May 2, 2023

Source of ECG13

MIT-BIH Arrhythmia Database obtained from the Beth Israel Hospital Arrhythmia Laboratory

Consists of 48-half-hour ECG recordings which were digitized at 360 Hz having 11-bit resolution

May 2, 2023

Performance measures14

Signal to Noise Ratio

where P is average power and A is RMS amplitude Mean Square Error MSE = where N represents the total number of samples in the given signal,

x (i) is the original ECG and d (i) is the de-noised ECG.

What is a Transformand Why Do we Need One ?

Transform: A mathematical operation that takes a function or sequence and maps it into another one

Transforms are good things because… The transform of a function may give additional /hidden

information about the original function, which may not be available /obvious otherwise

The transform of an equation may be easier to solve than the original equation (recall your fond memories of Laplace transforms in DFQs)

The transform of a function/sequence may require less storage, hence provide data compression / reduction

An operation may be easier to apply on the transformed function, rather than the original function (recall other fond memories on convolution).

Why Wavelet? Time domain analysis, e.g. averaging (Not suitable for

non- stationary signals). Frequency domain analysis (Not suitable for non-

stationary signals) Time-frequency domain analysis Statistical methods (SVD,EMD) Time-scale domain analysis, e.g. wavelet (Variably-

sized regions for the windowing operation which adjust to signal components).

May 2, 2023

Block Diagram For WAVELET Transform Method

17

DWT Analysis

DWT of original signal is obtained by concatenating all coefficients starting from the last level of decomposition.

DWT will have same number of coefficients as original signal.

Frequencies most prominent (appear as high amplitudes) are retained and others are discarded without loss of information.

Applications of Wavelets Compression De-noising Feature Extraction Discontinuity Detection Distribution Estimation Data analysis

Biological data Financial data

Types of thresholding In hard thresholding, a sudden change occurs, but in soft

thresholding, a change occurs linearly which gives good result.

May 2, 2023 1

Thresholding contd… Hard-thresholding method: In hard thresholding, the remaining

co-efficients above the threshold remains unchanged as given by

Soft-thresholding method: In soft thresholding, the remaining coefficients are reduced by an amount equal to the value of the threshold.

May 2, 2023 1

May 2, 2023

Steps in Wavelet Filtering22

Step 1: Decomposing of the noisy signals using wavelet transform by using an appropriate mother wavelet and the noisy signal is decomposed, at a suitable decomposition level to obtain approximate coefficients aj and detail coefficients dj.

Step 2: Thresholding of the obtained wavelet coefficients yields the estimated wavelet coefficients dj. For each level, a

threshold value is found, and it is applied to the detailed coefficients.

Step 3: Inverse transformation to obtain the cleaned signal The reconstruction of the de-noised ECG signal x (n) is done from the values of dj and aj obtained by inverse discrete wavelet transform (IDWT).

May 2, 2023

Three level decomposition of ECG signal using wavelets

23

May 2, 2023

Three level decomposition of ECG signal using wavelets

24

The useful information in ECG is contained in the frequency of 0.5 Hz-45 Hz ( Nagendra et al 2011).

The high frequencies contain the noise and the low frequencies of the ECG contains the required data for the ECG analysis (Kumar & Agnihotri 2010). Thus, this three level decomposition of the ECG signal guarantees that the dominant frequency components of the actual signal is not lost during this omission of coefficients of few levels

The Daubechies family (Db 4) gives the best de- noising results for the ECG data (Balasubramaniam & Nedumaran 2009). The Rigrous Sure thresholding (Garg et al 2011) gives a better result than min-max and universal thresholding.

Denoising Implementation in Matlab

First, analyze the signal with appropriate wavelets

Hit Denoise

(Noisy Doppler)

Denoising Using Matlab Choose thresholdingmethod

Choose noise type

Choose thresholds

Hit Denoise

May 2, 2023

(a) Input ECG (b) De-noised ECG and (c) error obtained using wavelet de-noising for ECG100

27

May 2, 2023

SNR and MSE values for Wavelet based ECG de-noising

28

ECG Record number

SNR(in dB)

MSE

100 18.6231 4.3456 e -04

103 20.3215 3.6112 e -04

113 20.6286 3.6093 e -04

114 19.9831 3.7621 e -04

119 21.7861 3.0038 e -04

121 21.3048 3.2912 e -04

201 26.3668 2.8917 e -04

203 14.5414 4.5626 e -04

234 19.6314 4.0953 e -04

ECG01 10.4014 4.8591e -04

ECG04 16.8415 4.3802 e -04

May 2, 2023

Moving average filter29

Moving average is a simple mathematical technique which is set to remove the baseline wander noise removal.

It replaces each data point with the average of the neighboring data points.

Advantages of moving average filter (Smith S W 1997) are given below: Moving average filter is considered as a optimal filter to reduce

random white noise while preserving the sharpest step response. It is computationally fast as it requires addition and subtraction

operations rather than multiplication operations. It is a recursive operation. It has high execution speed.

May 2, 2023

Method II: ECG De-noising using Wavelet with Moving Average Filter

30

Wavelet Transform

Moving Average filter

ECG(BL+PL)

CleanECG

ECG (BL)

BL – Baseline noisePL – Power line noise

May 2, 2023

ECG De-noising using Wavelet with Moving Average Filter ( contd..)

31

Algorithm: Step 1: The Daubechies wavelet 4 (Db 4) wavelet is used to de-

noise the input ECG. The RigSure threshold is used for thresholding and the ECG signal is decomposed into three levels of decomposition.

Step 2: After thresholding , the approximate co-efficients of the last and detail co-efficients of all levels after soft thresholding are used to reconstruct the ECG.

Step 3: The wavelet de-noised ECG is given as input to the moving average filter of order 7

Step 4: The de-noised ECG is obtained at the final output after de-noising by the wavelet and moving average filter.

May 2, 2023

Method III: ECG de-noising using wavelet with moving average filter

32

(a) Noisy ECG signal, (b) De-noised ECG signal after using wavelet and moving average filter and (c) Superimposition of input noisy ECG signal of (a) and reconstructed signal after applying wavelet and moving average filter of (b)

May 2, 2023

SNR for the methods Wavelet and MAF33

ECG Record NumberWavelet and Moving average

filter(SNR in dB)

100 27.652

103 24.781

113 26.409

114 21.671

119 29.813

121 26.965

201 26.024

203 17.318

234 25.928

ECG01 17.801

ECG04 24.414

May 2, 2023

MSE for the method: Wavelet and MAF34

ECG Record Number Wavelet and Moving average filter(MSE)

100 1.9213 e -04

103 2.3947 e -04

113 2.0712 e -04

114 2.4812 e -04

119 1.8914 e -04

121 1.9566 e -04

201 2.1231 e -04

203 3.8942 e -04

234 2.3497 e -04

ECG01 3.4217 e -04

ECG04 1.8112 e -04

May 2, 2023

Summary35

The possible reasons for these wavelet based methods to be effective only for power-line noises in ECG are given below :

The power line interference is narrow-band noise centered at 50 Hz or 60 Hz with a bandwidth of less than 1Hz and the useful information in ECG is contained in the frequency of 0.5 Hz - 45 Hz

The use of wavelets to de-noise ECG by decomposing the into three levels removes the power line frequency of 50 Hz or 60 Hz.

Wavelet based de-noising requires the selection of suitable wavelet denoising parameters for the success electrocardiogram signal filtration in wavelet domain

May 2, 2023

Summary36

Wavelet Transform and MAF have removed the baseline and power line noise effectively in the ECG signal.

Wavelet Transform removes the power line and MAF removes the baseline noise from the ECG signal.

May 2, 2023

(a) Input ECG (b) De-noised ECG and (c) error obtained using wavelet de-noising for ECG100

37

May 2, 2023

SNR and MSE values for Wavelet based ECG de-noising

38

ECG Record number

SNR(in dB)

MSE

100 18.6231 4.3456 e -04

103 20.3215 3.6112 e -04

113 20.6286 3.6093 e -04

114 19.9831 3.7621 e -04

119 21.7861 3.0038 e -04

121 21.3048 3.2912 e -04

201 26.3668 2.8917 e -04

203 14.5414 4.5626 e -04

234 19.6314 4.0953 e -04

ECG01 10.4014 4.8591e -04

ECG04 16.8415 4.3802 e -04

May 2, 2023

Multiwavelets (contd..)39

Instead of one scaling function and one wavelet, multiple scaling functions and wavelets are used.

Leads to more degree of freedom i.e more number of independent samples to be used for reconstruction.

May 2, 2023

Multiwavelets (contd..)40

Multiwavelets have properties such as compact support (value of wavelet is 0 after a time interval a-b), orthogonality, symmetry and vanishing moments (decay towards low frequency) when compared to scalar wavelets.

The increase in the degree of freedom in multiwavelets is obtained at the expense of replacing scalars with matrices, scalar functions with vector functions and single matrices with block of matrices

May 2, 2023

De-noised signal ECG record 100 using multiwavelet

41

May 2, 2023

Error after de-noising ECG record 100 using multiwavelet

42

May 2, 2023

SNR and MSE values for Multiwavelet based ECG de-noising

43

ECG Record Number

SNR(in dB)

MSE

100 24.9531 2.7802e -04

103 29.9125 2.3381 e -04

113 30.9861 1.9903 e -04

114 33.0423 1.4710 e -04

119 24.3102 2.8632 e -04

121 29.9025 2.3499 e -04

201 26.5337 2.4119 e -04

203 18.8351 3.4671 e -04

234 26.5183 2.5004 e -04

ECG01 15.3892 3.5226 e -04

ECG04 22.7168 2.9284 e -04

May 2, 2023

Statistical analysis of SNR for de-noising ECG signals using wavelets and multiwavelets

44

There is an average increase of 6.6064 dB in terms of SNR with a 34.53% increase in the performance of multiwavelets over that of wavelets for the various ECG records.

  Name of the techniqueWavelet Multiwavelet

Parametersof SNR( in dB)

Mean 19.1299 25.7363Variance 15.1536 28.3428Standard Deviation

4.1417 5.3238

May 2, 2023

Summary45

The possible reasons for these wavelet based methods to be effective only for power-line noises in ECG are given below:

The power line interference is narrow-band noise centered at 50 Hz or 60 Hz with a bandwidth of less than 1Hz and the useful information in ECG is contained in the frequency of 0.5 Hz - 45 Hz ( Nagendra et al 2011).

The use of wavelets to de-noise ECG by decomposing the into three levels removes the power line frequency of 50 Hz or 60 Hz.

May 2, 2023

Summary46

Wavelet based de-noising requires the selection of suitable wavelet de-noising parameters for the success electrocardiogram signal filtration in wavelet domain

Multiwavelets are faster in decomposition and have good symmetry properties.

Even though the performance of wavelets and multiwavelets are greater than the adaptive filter method, wavelet transforms ignore polynomial components of the signal up to the approximation order of the basis.

It is also found that the wavelet and multiwavelet de-noising removes power line noise alone from the ECG, but not the baseline noise.

EEG The electroencephalogram (EEG) is a recording of the electrical

activity of the brain from the scalp. The first recordings were made by Hans Berger in 1929 The systematic approach of recognition, source identification, and

elimination of artifact is an important process to reduce the chance of misinterpretation of the EEG and limit the potential for adverse clinical consequences

EEG Waves Alpha wave -- 8 – 13 Hz. Beta wave -- >13 Hz. (14 – 30 Hz.) Theta wave -- 4 – 7.5 Hz. Delta waves – 1 – 3.5 Hz.

Different types of brain waves in normal EEG

EEG Recording From Normal Adult Male

Alpha wave rhythmic, 8-13 Hz mostly on occipital lobe 20-200 μ V normal, relaxed awake rhythm with eyes closed

Beta wave irregular, 14-30 Hz mostly on temporal and frontal lobe mental activity excitement

Theta wave rhythmic, 4-7 Hz Drowsy, sleep

Delta wave slow, < 3.5 Hz in adults normal sleep rhythm

Different types of brain waves in normal EEGRhythm Frequency

(Hz)Amplitude(uV)

Recording& Location

Alpha(α) 8 – 13 50 – 100 Adults, rest, eyes closed.Occipital region

Beta(β) 14 - 30 20 Adult, mental activityFrontal region

Theta(θ) 5 – 7 Above 50 Children, drowsy adult, emotional distressOccipital

Delta(δ) 2 – 4 Above 50 Children in sleep

Requirements EEG machine (8/16 channels). Silver cup electrodes/metallic bridge electrodes. Electrode jelly. Rubber cap. Quiet dark comfortable room. Skin pencil & measuring tape.

Computerized EEG Machine

EEG Electrodes

Sliver Electrodes Electrodes Cap

10 /20 % system of EEG electrode placement

EEG Electrodes Each electrode site is labeled with a letter and a

number. The letter refers to the area of brain underlying the

electrode e.g. F - Frontal lobe and T - Temporal lobe. Even numbers denote the right side of the head

and Odd numbers the left side of the head.

Montage Different sets of electrode arrangement on the scalp by 10 – 20

system is known as montage. 21 electrodes are attached to give 8 or 16 channels recording.

Factor influencing EEG Age

Infancy – theta, delta wave Child – alpha formation. Adult – all four waves.

Level of consciousness (sleep) Hypocapnia(hyperventilation) slow & high

amplitude waves.HypoglycemiaHypothermiaLow glucocorticoids

Slow waves

Eye opening Alpha rhythm changes to beta on eye opening (desynchronization /

α- block)

Thinking Beta waves are observed

Use of EEG Epilepsy

Generalized seizures.

Localize brain tumors. Sleep disorders (Polysomnography- EEG activity together with

heart rate, airflow, respiration, oxygen saturation and limb movement) Sleep apnea syndrome Insomnia

Helpful in knowing the cortical activity Determination of brain death.

Flat EEG(absence of electrical activity) on two records run 24 hrs apart.

1. Cardiac artifacts2. Electrode artifacts3. External device artifacts4. Muscle artifacts5. Ocular artifacts

EEG Artifacts

The artifact occurs with maximum amplitude and clearest QRS morphology over the temporal regions and often is better formed and larger on the left side.

The R wave is most prominent in channels that include the ear electrodes.

Cardiac artifact

Cardiac artifact

ECG artifact may occur inconsistently by not being present with every contraction of the heart and may have an irregular interval when a cardiac arrhythmia is present.

In either situation, it may be identified by its temporal association with the QRS complexes in an ECG channel.

Cardiac pacemakers produce a different electrical artifact. it is distinct from ECG artifact in both distribution and morphology. Pacemaker artifact is generalized across the scalp and comprises high

frequency

Pacemaker artifact

Transients comprising very fast activity recur in channels with the A1 and A2 electrodes.

Types: Electrode pop Electrode contact Electrode/lead movement Perspiration Salt bridge Movement artifact

Electrode artifact

Electrode artifacts usually manifest as one of two disparate waveforms, brief transients that are limited to one electrode and low frequency rhythms across a scalp region.

The brief transients are due to either spontaneous discharging of electrical potential that was present between the electrode or its lead.

The spontaneous discharges are called electrode pops, and they reflect the ability of the electrode and skin interface to function as a capacitor and store electrical charge across the electrolyte paste or gel that holds the electrode in place.

With the release of the charge there is a change in impedance, and a sudden potential appears in all channels that include the electrode.

Poor electrode contact or lead movement produces artifact with a less conserved morphology than electrode pop.

The poor contact produces instability in the impedance, which leads to sharp or slow waves of varying morphology and amplitude.

These waves may be rhythmic if the poor contact occurs in the context of rhythmic movement, such as from a tremor.

Lead movement has a more disorganized morphology that does not resemble true EEG activity in any form and often includes double phase reversal

Lead movement

Multiple channels demonstrate the artifact through activity that is both unusually high amplitude and low frequency and also disorganized without a plausible field

The smearing of the electrode paste between electrodes to form a salt bridge or the presence of perspiration across the scalp both produce artifacts due to an unwanted electrical connection between the electrodes forming a channel.

Perspiration artifact is manifested as low amplitude, undulating waves that typically have durations greater than 2 sec; thus, they are beyond the frequency range of cerebrally generated EEG.

Slat bridge artifact differs from perspiration artifact by being lower in amplitude, not wavering with low frequency oscillation, and typically including only one channel

It may appear flat and close to isoelectric.

Sweat artifact

Sweat artifact

This is characterized by very low-frequency (here, 0.25- to 0.5-Hz) oscillations. The distribution here (midtemporal electrode T3 and occipital electrode O1) suggests sweat on the left side

Salt bridge artifact

Activity in channels that include left frontal electrodes is much lower in amplitude and frequency than the remaining background. The lack of these findings when viewed in a referential montage confirms that an electrolyte bridge is present among the electrodes involved.

TYPES 50/60 Hz ambient electrical noise Intravenous drips Electrical devices: intravenous pumps, telephone Mechanical effects: ventilators, circulatory pumps

External device artifacts

60 Hz artifact

The very high frequency artifact does not vary and is present in the posterior central region, which does not typically manifest muscle artifact. This example was generated by eliminating the 60Hz notch filter.

Electrical noise may also result from falling electrostatically charged droplets in an IV drip.

A spike like EEG potential results, which has the regularity of the drip.

Intravenous drips

Electrical devices may produce other forms of noise. Anything with an electric motor may produce high amplitude,

irregular or spike like artifact. This is due to the switching magnetic fields within the motor. The artifact occurs with the motor’s activity; thus, it may be constant

or intermittent, as is the case with infusion pumps. Mechanical telephone bells are the classic source for a more

sinusoidal form of this artifact but are increasingly a less common source of the intermittent form of this artifact.

Electrical devices: intravenous pumps, telephone

Electrical motor

The very high frequency activity suggests an electrical source, and the fixed morphology and repetition rate indicate an external device. This was caused by an electric motor within the pump.

Movement during the recording of an EEG may product artifact through both the electrical fields generated by muscle and through a movement effect on the electrode contacts and their leads.

Although the muscle potential fields are the signals sought by electromyographers, they are noise to electroencephaographers.

Indeed, EMG activity is the most common and significant source of noise in EEG.

EMG activity almost always obscures the concurrent EEG because of its higher amplitude and frequency.

EMG

Muscle artifact

The high amplitude, fast activity across the b/l ant. region is due to facial muscle contraction and has a distribution that reflects the locations of the muscles generating it. Typical of muscle artifact, it begins and ends abruptly.

Types Blink Eye flutter Lateral gaze Slow/Roving eye movements Rapid eye movements electroretinogram

Ocular artifact

Blink artifact

Bifrontal, diphasic potentials with this morphology and field are reliably eye blink artifact.

Repetitive blinks usually appear as a sequence of the slow wave ocular artifacts and thus resemble rhythmic delta activity.

Although ocular flutter involves vertical eye movements, it differs from repetitive blinks by being more rapid and having lower amplitude.

Eye flutter artifact

Medium amplitude, low frequency activity that is confined to the frontal poles is identified as ocular artifact through its morphology. Compared to blink artifact, flutter artifact typically has a lower amplitude and a more rhythmic appearance

Lateral eye movement

Although a horizontal, frontal dipole is the key finding with lateral eye movements, the artifact is also distinguished by its morphology, which has a more abrupt transition between the positive and negative slopes that blinks and most flutter.

Artifacts are usually easily recognized by experienced EEGer. The process of visual analysis and digital filtering allow

identification of most physiologic and nonphysiologic artifacts. Digital filters can be applied and removed multiple times, and

can significantly improve interpretation of EEG contaminated by artifacts by allowing specific frequencies to be removed from the digital display.

Artifact detection and rejection

Commonly used methods to remove artifacts

If the analysis is restricted to certain frequency bands, an automated algorithm can be designed to only analyze activity in this frequency band.

For ex., a 1 to 20Hz band pass may be used to remove muscle artifact.

This method is not very useful for analysis of the entire bandwidth of EEG, as artifacts can occur at any frequency.

Even for very narrow frequency bands, there may be significant artifact remaining after band pass filtering.

The process of filtering may significantly alter the appearance of EEG and make subsequent identification of artifacts more difficult.

Use of Band Pass Filters:

In this case, a technologist or EEGer visually reviews the entire EEG recording and marks segments with artifacts.

This is a reliable method, and may detect some artifact that would be missed by automated techniques.

It is time consuming, however, and reader fatigue may become problematic for long or multichannel recordings.

Subtle or brief artifacts may not be identified, and different readers may have different thresholds for rejection.

This method is only possible for offline(not real time) digital analysis.

Manual rejection of artifact segments:

This technique rejects short segments of EEG if the segment exceeds predefined thresholds.

These thresholds can be simple analysis of the EEG channels themselves such as amplitude, numbers of zero crossings, or 60Hz artifact.

If a segment shows very high amplitude, it is eliminated. Some techniques use other special electrodes to identify artifact

signals, such as EOG,EMG, EKG or accelerometers. If the signal in these channels exceeds a threshold, the segment

of EEG will be rejected.

Automatic rejection of artifact segments:

97

EEG DATA

EEG DATA and EEGLAB Toolbox is obtained from

Swartz Center for Computational Neuroscience,Institute for Neural Computation, University of California San Diego

EEG DATA:

http://sccn.ucsd.edu/~arno/fam2data/publicly_available_EEG_data.html

EEGLAB Toolbox:

http://sccn.ucsd.edu/eeglab/

98

STEPS IN DENOISING EEG

Apply Wavelet

Transform

Threshold the Noisy

Wavelet coefficients

Apply InverseWavelet

TransformNoisy EEG

Wavelet coefficients

Signal coefficients

Denoised EEG