A Novel method of QRS Detection Using Adaptive Multilevel Thresholding with Statistical False Peak Elimination

Abstract

Heart is the vital organ of a Human Body, because of its involvement in various works and processes in the entire body such as blood pumping etc., so recording a heart function is also a great thing, it is done through ECG signals. ECG signal records the electrical signals and activity of a Human Heart based on the electrical signals released by the Heart. ECG signal consists of PQRST waves, which are the reference points on an ECG signal. But, recording them is much easy than extracting and analyzing them, so, as to extract them, we are applying an advanced adaptive multi-level thresholding (AAMT) along with a selective statistical false peak elimination for the detection of QRS peaks of an ECG signal. Initially, median and moving average filters are applied for removing noise as well as terms. After AAMT is implemented on the complete dataset of ECG signals. Then selective statistical false peak elimination (SSFPE) is implemented for removing noise terms that might be missed out during filtering. At last, a search back stage will be implemented to search for low amplitude useful peaks. We used MIT-BIH arrhythmia and fantasia databases and got a higher accuracy in detecting and sensitivity is increased to 99.3 for MIT-BIH arrhythmia and 99.7 for fantasia database compared to existing previous methods.

Introduction

I. INTRODUCTION

The Heart of a Human body is an essential and most vital organ. The human heart is responsible for various processes inside our body. Heart’s main purpose is to pump blood to various parts of the body which is the most important task for our body. The human heart emits electrical signals at a very low volts of the order of 60mV, so, we have to record those electrical signals for checking and validating the functions of a healthy heart, the recording of the electrical activity of the human heart is done by Electro-Cardio-Gram (ECG) signals. But, the problem with those ECG signals is their complexity during plotting and analyzing.

An ECG signal consists of PQRST waves of which QRS are principal components. QRS waves are also difficult to detect, but we can detect R peaks easily.

There exists a number of techniques for detecting R peaks, such as Fourier Transform (FT), Direct Cosine Transform (DCT), Fast Fourier Transform (FFT) and application of Wavelets such as Continuous Wavelet, Discrete Wavelet etc. Transforming a signal into frequency domain and taking the instantaneous peaks of the signal will result in R peaks. R peaks can be extracted using statistical analysis with thresholds of different lengths.

The extraction of remaining waves such as, Q or S can also be done by statistical analysis. That involves thresholding at different instants. Q and S peaks along with R forms a complex trio of QRS peak which is a quite different and difficult to be extracted. There exist several techniques in extraction and detection of QRS peaks but, here, we are using an advanced adaptive multi-level thresholding (AAMT) combined with selective statistical false peak elimination (SSFPE) for detecting or extracting QRS peaks from an ECG signal record.

An ECG signal contains PQRST waves of which QRS are the most important waves, but, the extraction of a Q, R, S or any wave is quite difficult and challenging, for that we are implementing an advanced adaptive multi-level thresholding (AAMT) combined with selective statistical false peak elimination (SSFPE) for detecting or extracting QRS peaks from an ECG signal record. After, all the statistical analysis of the segments it uses a search back stage for detecting any missed out true peaks in the ECG signal segment., we have used MIT-BIH arrhythmia and fantasia databases for the implementation.

A. Types of Noises in ECG Signals

There exists a number of noises that corrupts the ECG signal in a different manner. Some degrade the signal’s strength, some completely the spectral characteristics of the signal and some completely change or destroys the signal. Such as, Baseline Wander, Power line interference and motion artifacts. The motion artifacts are due to the motion of the person during the process of fetching the ECG from the person. Person’s movement can affect the signal’s quality from a low level to a considerable level or even higher. The power line interference is a typical type of noise that occurs due to the interference of electrical signals that does not belongs to the activity of the Heart. The interference of the signals that are captured by the ECG during the initial process. Baseline Wander is a type of noise produced in ECG signals because of the poor equipment or presence of any disturbances in the middle and presence of any electrical equipment near to the person who’s taking the ECG test.

B. Median and Moving Average Filter

The usage of a Median filters cascaded with Moving Average filter is the whole new approach of this work. Median filters are generally used for removing noise from the signals. The median filter uses the median values to replace the noisy terms. Usually, the noisy terms deviate from the remaining terms in a locality. They might be too high or too low than the regular values in the particular location. Median filter calculates the median of the particular location and check for its neighbors to replace the deviated values. Thereby, reducing or minimizing the noise from the signal.

Moving Average filter is also a noise removing filter from the signals. Moving average filter is a type of FIR filter which eliminates noise terms. Moving average filter by its name, this calculates the series of averages from the signal locations or instants that moves over a series of instants on the signal. Later, it replaces the moving averages with the noisy terms to eliminate the noise from the original signal. Moving average filter followed by two median filters gives more robust signal than the existing methods or techniques. Later, the de-noised or noise free (almost) signal is sent to the Advanced Adaptive Multi-Level Thresholding (AAMT) for the detection of QRS peaks or waves from the noise removed ECG signal.

ECG signal noise elimination actually made easy through the process of applying a median filter along with the moving average filter. Two median filters are employed to get noise suppressed at two levels making the signal to noise ratio improvement. Later, an additional filter known as, moving average filter is applied to remove the noise at another level, making the noise eliminated at even low levels. Signal to noise ratio is further more improvised for the ECG signal and making it easier for detection by the processes involved in the work. De-noised signal at each level is made easy, through this type of application of filters at multiple levels. This type of approach makes the signal to noise ratio improving even better than the previously existing works.

C. AAMT

The Advanced Adaptive Multi-Level Thresholding (AAMT) is extension of the existing technique known as Adaptive Multi-level Thresholding (AMT).

The AAMT for detecting or extracting QRS peaks of an ECG signal are quite impressive, because the current technique takes beats as well as amplitude thresholds than the existing AMT technique. The AAMT consumes less time than the AMT. The peaks are first extracted and their locations along with their amplitudes are recorded and their average value is taken. It uses two levels called signal level and noise level which are initially taken as zero or 0 for reference. Later, the averages are used to replace the locations where the noise terms will be present that means deviation from the remaining neighbors at that particular location.

Later, updating these levels at every stage when we find a new true peak or a discarded noise peak. Here, we use another two thresholds both for noise and true peaks, R a threshold of lower amplitude below which peaks are the noise peaks and are discarded and T a threshold of high amplitude above are true peaks and updated. The resultant gets the pseudo peaks with a much more accuracy in detection after being filtering of noise and interference terms in the previous level. If all the pseudo peaks are extracted the signal is segmented with less than 50K samples and with each segment containing two times the samples taken by existing AMT technique. After that, the segments are sent to the next stage for removing noise terms.

AAMT uses thresholds at multiple levels or stages to detect or extract peaks or waves even more accurately. AAMT implements a new type of a combination of median filter along with a moving average filter to de-noise the ECG signal. A combination of two median filters cascaded with a moving average filter makes the noise removal more. Three stages of noise removing from the ECG signal is the new approach used by this paper. Later, the de-noised signal is thresholded at each level and the peaks or waves are detected using multi-level thresholding. The obtained ECG signal is segmented into several segments of length maximum of 50K samples and sent to the next stage known as Selective Statistical False Peak Elimination (SSFPE).

D. Selective Statistical False Peak Elimination (SSFPE)

The Selective Statistical False Peak Elimination (SSFPE) is a statistical analysis of the extracted ECG segments. The technique uses statistical thresholds and selection to eliminate any false peaks present in the segments. After, the segmented ECG signals are extracted, the SSFPE uses the pseudo true peaks for analysis and calculates the difference between the peaks that lie adjacent and a mean peak to peak difference as well as their weights are also calculated. After, all the statistical analysis of the segments it uses a search back stage for detecting any missed out true peaks in the ECG signal segment. The thresholding technique yields several true positive peaks at this stage. Later, a search back stage is employed for detecting the missed-out peaks. Search back works like feedback for the system where it checks once again for the remaining missed out true peaks

II. PROPOSED METHOD

The proposed work starts with a Pre-processing stage, detection of peak stage followed by a post-processing stage. The initial stage extracts the suitable information from an ECG signal by suppressing noise or interferences. The detection is done in two times to make the detection robust; first detection is done through AAMT using R-peaks with an adaptive threshold. The second detection is done by SSFPE through segmented and analysis of previously detected peaks. At last, the search back stage tries to detect any missed out true peaks. 

Pre-processing stage involves the elimination of the noise and interference terms from the signal through two Median filters and a Moving Average filter. Median filters aim to filter the median values that are higher through a sample of ECG signal. After that, the moving average filter also aims for average valued peaks in the signal.  Median filters are the one which calculates the median values of the signal in a location where the noise will be present. After the calculation of median for the locality verifies the neighbors of the particular location, if any value has more deviation or not. If any value has a high deviation that may be very high value or very low value that value will be replaced by the previously calculated median value. Likewise, the median filtering is applied two times to reduce or nullify the noise at more levels. This type of approach makes the signal to noise ratio improving much better than the previously existing works.

The moving average filter is also a noise removing filter mainly chooses to filter the high frequencies. Moving average filter calculates the average of a particular location and that average calculation is moved over the entire signal. This way, the moving average filter calculates the series of averages of the locations over the signal. Later, the averages are used to replace the locations where the noise terms will be present that means deviation from the remaining neighbors at that particular location. Thereby, improvising the signal to noise ratio to a new level that is much better than the previous works.

Detection of Peak stage involves multi-level thresholding by AAMT, it starts with initially setting the min peak interval to 280ms which avoids elimination of true positive peaks that consists of Ectopic beats. After that, the locations and amplitudes of the peaks are calculated, through them the average will also be calculated. Noise and signal values are used to count the true as well as ignored false peaks of noise. Two thresholds are also initialized one of them is a low amplitude threshold for eliminating false noise peaks and other is upper threshold for true peak identification. Later, the ECG signal is segmented with utmost of 50K samples per signal. The segmented signal is sent to the next stage for the detecting of any peaks that are missed out during the current process.  

After, all the statistical analysis of the segments it uses a search back stage for detecting any missed out true peaks in the ECG signal segment. If any value has a high deviation that may be very high value or very low value that value will be replaced by the previously calculated median value. Likewise, the median filtering is applied two times to reduce or nullify the noise at more levels. After, all the statistical analysis of the segments it uses a search back stage for detecting any missed out true peaks in the ECG signal segment. The thresholding technique yields several true positive peaks at this stage. Later, a search back stage is employed for detecting the missed-out peaks.

Post-processing stage involves a statistical analysis of the segmented ECG samples and based on the analysis thresholds are calculated. Those thresholds are used to extract the true peaks that are left out during the previous process. The thresholding technique yields several true positive peaks at this stage. Later, a search back stage is employed for detecting the missed-out peaks. Search back works like feedback for the system where it checks once again for the remaining missed out true peaks. After, the segmented ECG signals are extracted, the SSFPE uses the pseudo true peaks for analysis and calculates the difference between the peaks that lie adjacent and a mean peak to peak difference as well as their weights are also calculated.

III. RESULTS

At the initial stage which is a pre-processing stage the original ECG signal is plotted.

A Search back stage which is like feedback for the post-processing stage, where the technique once again checks for any missed-out peaks during the entire processes which makes the technique even more robust. Finally, the true positive peaks are detected, that are QRS peaks of an ECG signal. Three terms are used to see the performance of the proposed method, and these are sensitivity (Se), positive predictivity (PC), and detection error rate (DER) which are represented by,

where, TP is the true positive that is the number of annotated beats detected properly, FN is the false negatives that are the number of missed beats from the annotations, FP is the false positive that is the number of beats that are not annotated but still detected by the algorithm and TB is the total number of beats annotated in the record.

TABLE 1.

Results from the MIT-BIH arrhythmia database.

--------------------------------------------------------------------------------------------------------------------------------------------

    Rec. No.               TB               DB               TP               FP                FN                Se (%)         P+ (%)           DER(%)

--------------------------------------------------------------------------------------------------------------------------------------------

       100                  2273           2273          2273              0                   0                    100                    100                      0

       101                  1867           1864          1864              0                   3                    100                    99.84                   0.16

       102                  2187           2184          2184              0                   3                    100                    99.86                   0.14

       103                  2083           2083          2083              0                   0                    100                    100                      0

       104                  2229           2222          2222              0                   7                    100                    99.69                   0.31

       105                  2589           2594          2586              8                   3                    99.69                 99.88                  0.42

       106                  2033           2023          2023              0                   10                  100                    99.51                  0.49

       107                  2134           2133          2133              0                   1                    100                    99.95                  0.05

       108                  1757           1746          1745              1                   12                  99.94                 99.32                 0.74

       109                  2532           2526          2526              0                   6                    100                     99.76                 0.24

       111                  2130           2115          2115              0                   15                  100                     99.3                   0.7

       112                  2539           2539          2539              0                   0                    100                     100                    0

       113                  3550           3480          3480              0                   70                  100                     98.03                 1.97

       114                  1984           1983          1982              1                   2                    99.95                  99.9                   0.15

       115                  1952           1953          1952              1                   0                    99.95                  100                    0.05

       116                  2394           2388          2388              0                   6                    100                     99.75                 0.25

       117                  2546           2486          2486              0                   60                  100                     97.64                 2.36

       118                  2279           2279          2279              0                   0                    100                     100                     0

       119                  1988           1988          1988              0                   0                    100                     100                     0

       121                  1862           1861          1861              0                   1                    100                     99.95                  0.05

       122                  2476           2476          2476              0                   0                    100                     100                     0

       123                  1519           1520          1519              1                   0                    99.93                  100                     0.07

       124                  1618           1615          1615              0                   3                    100                     99.81                  0.19

       200                  2597           2589          2588              1                   9                    99.96                  99.65                  0.39

       201                  1977           1958          1952              6                   25                  99.69                  98.74                  1.57

       202                  2147           2131          2131              0                   16                  100                     99.25                  0.75

       203                  2846           2669          2665              4                   181                99.85                 93.64                   6.5

       205                  2645           2642          2642              0                   3                    100                     99.89                   0.11

       207                  2091           2047          2037              10                 54                  99.51                 97.42                   3.06

       208                  2938           2929          2929              0                   9                    100                     99.69                   0.31

       209                  3006           3006          3005              1                   1                    99.97                  99.97                   0.07

       210                  2634           2622          2619              3                   15                  99.89                  99.43                   0.68

       212                  2749           2749          2749              0                   0                    100                      100                      0

       213                  3249           3238          3238              0                   11                  100                      99.66                  0.34

       214                  2262           2256          2255              1                   7                    99.96                   99.69                  0.35

       215                  3364           3347          3347              0                   17                  100                      99.49                  0.51

       217                  2205           2205          2203              2                   2                    99.91                   99.91                  0.18

       219                  2170           2153          2153              0                   17                  100                      99.22                  0.78

       220                  2045           2048          2045              3                   0                    99.85                   100                     0.15

       221                  2428           2427          2426              1                   2                    99.96                   99.92                  0.12

       222                  2527           2501          2494              7                   33                 99.72                    98.69                  1.58

       223                  2603           2594          2594              0                   9                   100                        99.65                  0.35

       228                  2042           2018          2017              1                   25                 99.95                     98.78                 1.27

       230                  2256           2256          2256              0                   0                   100                         100                    0

       231                  1577           1573          1572              1                   5                   99.94                      99.68                0.38

       232                  3049           2898          2876              22                173               99.24                      94.33                6.4

       233                  3076           3069          3069              0                   7                   100                         99.77                0.23

       234                  2753           2753          2753              0                   0                   100                         100                   0

--------------------------------------------------------------------------------------------------------------------------------------------

      Total            113757      113009    112934          75               823               99.93                    99.35              0.72

-------------------------------------------------------------------------------------------------------------------------------------------

IV. DISCUSSION

The ECG signal is taken from two databases namely, MIT-BIH Arrhythmia and Fantasia databases. The ECG signal is corrupted from noise already, so, removing the noise by the application of a combination of filters. Two Median filters along with a Moving Average filter helps in reducing the noise to a greater extent. Noise removing is done at three times, by two median filters and a moving average filter, so noise is reduced to lower level than the previously existed processes. Later, the noise removed ECG signal is moved to the AAMT for detecting the QRS Peaks. After, the QRS Peaks are detected it segmented into 50K samples of ECG signal each. These segments are sent to SSFPE for further detection of QRS true Peaks, if any missed out. Later, a search back stage is implemented to again search for the QRS Peaks, this stage makes our work even more robust in determining the QRS Peaks.

Conclusion

We can finally conclude that, the QRS peaks are extracted and detected using AAMT which employs a median filtering along with a moving average filter to filter out noise and interference terms from the original ECG signal and left out true peaks are again detected with SSFPE, later, a search back stage is employed to make the detection even more robust. The present technique of advanced adaptive multi-level thresholding (AAMT) gives better results compared with the previous technique of AMT where, it uses more data and thresholds for processing.

References

[1] T. Sharma and K. K. Sharma, ‘‘QRS complex detection in ECG signals using locally adaptive weighted total variation denoising,’’ Comput. Biol. Med., vol. 87, pp. 187–199, Aug. 2017. [2] B.-U. Kohler, C. Hennig, and R. Orglmeister, ‘‘The principles of software QRS detection,’’ IEEE Eng. Med. Biol. Mag., vol. 21, no. 1, pp. 42–57, Aug. 2002R. Nicole, “Title of paper with only first word capitalized,” J. Name Stand. Abbrev., in press. [3] R. M. Rangayyan, Biomedical Signal Analysis, 33, John Wiley & sons, 2015X. Lu, M. Pan, and Y. Yu, ``QRS detection based on improved adaptive threshold,\'\' J. Healthcare Eng., vol. 2018, pp. 18, Mar. 2018. [4] J. Pan, W.J. Tompkins, A real-time QRS detection algorithm, IEEE Trans. Biomed. Eng. (3) (1985) 230-236. [5] J.D. Drake, J.P. Callaghan, Elimination of electrocardiogram contamination from electromyogram signals: an evaluation of currently used removal techniques, J. Electromyogram. Kinesiology. 16 (2) (2006) 175-187. [6] P. S. Hamilton and W. J. Tompkins, ``Quantitative investigation of QRS detection rules using the MIT/BIH arrhythmia database,\'\' IEEE Trans. Biomed. Eng., vol. BME-33, no. 12, pp. 1157 1165, Dec. 1986.

Copyright

Copyright © 2022 A. Rajani, V. Sandeep. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.