This confirms that the thesis titled “ECG Signal Denoising Using FIR Filter and QRS Detection” was done by Jannatul Robaiat Mou at the Department of Electronics and Communication Engineering, Khulna University of Engineering &. This confirms that the thesis submitted by Jannatul Robaiat Mou entitled “Denoising ECG Signal Using FIR Filter and QRS Detection” has been approved by the Board of Examiners in partial fulfillment of the requirements for the degree of M.Sc. . Foisal Hossain, Department of Electronics and Communication Engineering, Khulna University of Engineering and Technology (KUET).
But the amplitude and duration of the ECG signal are usually disturbed by various types of noise and interference based on interfaces between ECG machines and the human body. In this thesis, we proposed new algorithms that can de-noise and detect QRS complexes in the ECG signal. In general, noise-free algorithm removes the noisy signal and we used Remez exchange algorithm for the first algorithm, arbitrary size designed with FIR filter for second algorithm, FIR filter with window method for third proposed algorithm. We also proposed a moving average filter and weighted moving average window, and an algorithm based on the forward difference quotient and threshold for corresponding ECG noise elimination algorithms, respectively.
List of Tables
Introduction
- Introduction
- Motivation
- Problem Statement
- Review of Related Research Work on Thesis
- Objectives
- Organization of Thesis
An ECG is very sensitive, different types of noise and disturbances can corrupt the ECG signal as the actual amplitude and duration of the signal can be changed. Most of the low and middle income people die of heart attacks and strokes in the world. These power lines can affect the recording of the ECG and cause interference at the line frequency in the recorded trace.
This section presents an overview of the existing literature in areas related to the work in this thesis on denoising techniques of ECG signal using various methods and QRS detection. Mostly ECG recording is very sensitive and it is affected by different types of noise and it changes the amplitude and duration of the signal. It also gives the detail of the thesis objective, its achievement and what was concluded after the completion of this thesis.
Background and Related Work
Introduction
Electrocardiography
- Uses of ECG
- Types of ECG
- Recording of ECG signal
- Different ECG Waves
- ECG Register
- The ECG Electrodes
- The Extremity Leads
- The Chest Leads
- Ladder Diagram
- Color Coding of the ECG Leads
In Figure 2.3, the QRS complex represents the rapid depolarization of the right and left ventricles. The U-wave voltage is normally <25% of the T-wave voltage: disproportionately large U-waves are abnormal. The depolarization of the heart results in an electrical force that has a direction and magnitude; an electric vector.
An easy rule to remember: lead I + lead III = lead II This is done using the height or depth, independent of the wave (QRS, P of T). In vector electrocardiography, the movement of electrical activity of the P, QRS and T waves is described. The origin of impulse formation (sinus node for the first two beats and AV junction for the third pulse) and the conduction in the heart are shown.
Overview of Basic Noises in ECG Signal
- Power-Line Interference Noise
- Baseline Drift Noise
- Abrupt Baseline Shift Noise
- Electrosurgical Noise
- Noise Generated by Electronic Devices Used in Signal Processing
Some studies have shown that the standard deviation of the noise is typically 10% of the peak-to-peak amplitude of the ECG [19]. The effects of typical EMG noise can be seen in the ECG signal shown in Figure 2.12, and it is particularly problematic in the areas of the P and T complexes. Since the sampling frequency of the ECG signal is from 250 to 1000 Hz, an alias of this signal would be added to the ECG signal.
This chapter discusses basic ECG topics that are very important when analyzing ECG signals. It also discusses how the ECG signal is generated and how to obtain it using electrodes. Various noises in the ECG signal, such as power line interference, EMG noise, baseline drift, abrupt change in baseline, electrosurgical noise, etc.
Proposed Denoising Algorithms
Introduction
Noise Elimination methodology
- Algorithm-2/ Frequency Sampling filter
- Algorithm -4/ M point average with window length FIR filter
- Algorithm-6/ Forward difference quotient and amplitude thresholding
For a given set of specifications (i.e., passband edge frequencies, N, and the ratio of the passband to stopband ripples), the optimal method involves the following key steps. The input parameters are set for an optimal design program to obtain the filter length coefficients and the frequency response of the filter coefficient h[n], as shown in Figure 3.2. From the specification table, it can be designed so that the band edge frequency should be normalized to half the sampling frequency (the vector of normalized band edge frequencies point = and size of frequency sampling points (| ( )|)) as shown in Table 3.1.
A sample of optimal transition band frequency sampling values is given in Table 3.1 for N=22. In the table, bandwidth refers to the number of frequency samples in the filter's passband. The impulse response or filter coefficients ℎ( ) can be obtained as the inverse DFT of the frequency sample.
Using the above parameters as an input, the coefficients h(n) were obtained and h(n) are listed in Table 3.2 and the filter spectrum is shown is Figure 3.3(b) and 3.3(c) represent the frequency response of frequency sampling filter. We can start with the ideal low-pass response shown in Figure 3.4 (b) where the cutoff frequency and the frequency scale are normalized: T=1. Obtain values of ( ) for chosen window function and the values of the actual/practical FIR filter coefficients by .
The corresponding frequency response shows that ripples and overshoots, characteristic of direct clipping, are greatly reduced. The width of the filter pass is determined by the width of the main lobe of the window. We considered changing the noisy signal to different order ( ) to eliminate the ECG signal.
Note the small increment ∆ = , where T is the sampling interval and the equation of the samples of the noisy ECG signal represents the angle.
Summary
Now we need to create a new error vector that stores the amplitude of the noise value.
Proposed QRS Detection Algorithm
Introduction
QRS Detection Methodology
Summary
Results & Discussion
5 . 1 Introduction
Signal to Noise ratio (SNR)
Correlation
Denoising Simulation Results
The Algorithm-4 can remove the noise from the damaged ECG signal at different noise intensity levels. The filtered noise-free signal of the baseline shift is shown in Figure 5.4. The ECG filtered signal with all the various noise, including powerline, EMG, and baseline shift noise, correctly removed the noise for Algorithm-4. This algorithm removes EMG noise with higher noise density and properly removes baseline drift noise.
To eliminate the abrupt baseline shift noise from the ECG signal, we have used Algorithm-6 to remove the noise at each step of it as shown in the figure. In Figure 5.6 (c) – (d) it shows the output of the first, 3rd and fourth forward different equations. It is observed that Algorithm-6 does not remove the noise properly with higher noise density.
Especially Algorithm-6 is fully supported to remove sudden noise with higher noise density. In Figure 5.8 (b), the SNR values for Algorithm-4 show the highest values than other algorithms with the increment of . The SNR value is gradually increased for algorithm 1,2,4 &5 for powerline noise and baseline shift noise.
So Algorithm-4 & 6 proved the robustness for denoised ECG signals than Algorithm-1,2,3 & 5 for four different noises. For comparison between the ECG signal and the filtered signal by the proposed algorithm, we also considered three processes such as autocorrelation, cross-correlation and power spectral density (PSD) for algorithm robustness. The second subplot is shown while the high peak shown in the filtered signal is present in the ECG signal due to cross-correlation.
This algorithm is used to find out non-uniformity of the ECG signal from the MIT-BIH database. The filtered signal of algorithm-4 & 6 has less power amplitude than noisy ECG signals and more and more resolution of side loops of filtered signal than algorithm 1, 2, 3 & 5. Figure 5.18: Power spectral density of ECG signal and filtered signal by Periodogram , Welch method and Lomb-Scargle algorithm of Algorithm-1 for EMG noise.
PSD for = 0
PSD for = 0
PSD for = 0.25
PSD for = 0.25
PSD for = 0.5
PSD for = 0.5
PSD for = 0.75
PSD for = 0.75
PSD for = 1
PSD for = 1
Figure 5.23: Power spectral density of ECG signal and filtered signal according to periodogram, Welch method and Lomb-Scargle algorithm according to algorithm-6 for abrupt shift noise.
5 . 4 QRS Detection Simulation Results
Summary
The results are shown in the average of SNR, PRD, MSE, correlation coefficients, because each signal of the MIT-BIH database is simulated and averaged over all signals for comparison.
Conclusion
Future work
Chiu, “A novel QRS detection algorithm applied to heart rate variability analysis of patients with sleep apnea,” Biomedical Engineering Applications, Basic & Communications, vol. Tompkins, “Neural network-based adaptive matched filtering for QRS detection,” IEEE Transaction Biomedical Engineering, vol. Mehta, “Detection of QRS complexes in 12-lead ECG using adaptive quantized thresholding,” International Journal of Computer Science and Network Security, vol.
Poon, “Analysis of first derivative based QRS detection algorithms,” IEEE Transactions on Biomedical Engineering, vol. Chooi “Feature extraction and classification for EEG signals using wavelet transform and machine learning techniques”, Australasian Physical & Engineering Sciences in Medicine, Volume 38, Issue 1, pp. Tompkins, “Quantitative investigation of QRS detection rules using the MIT/BIH arrhythmia database,” IEEE Transactions on Biomedical Engineering, vol.
Laguna, “A wavelet-based ECG delineator: evaluation on standard datasets,” IEEE Transactions on Biomedical Engineering, vol. Tai, “Detection of ECG feature points using wavelet transforms,” IEEE Transactions on Biomedical Engineering , vol. Seo, “A novel QRS detection method using waltzes and artificial neural networks,” Journal of Medical Systems , vol.
Tompkins, “Adaptive matched filtering for Neural-network-based adaptive matched filtering for QRS detection,” IEEE Transactions on Biomedical Engineering, vol. Briller, “An approach to cardiac arrhythmia analysis using hidden Markov models,” IEEE Transactions on Biomedical Engineering , vol. Valli, “Genetic design of optimal linear and nonlinear QRS detectors,” IEEE Transactions on Biomedical Engineering, vol.
Rieta, “Application of phase transformation for automatic determination of single-lead ECG fiducial points,” Physiological Measurement , vol.
List of Publications
Journal Paper
IEEE Conference Paper
