QRS Detection Based on Multiscale Mathematical ... - Semantic Scholar
QRS Detection Based on Multiscale Mathematical Morphology for Wearable ECG Devices in Body Area Networks Fei Zhang; Yong Lian IEEE Transactions on Biomedical Circuits and Systems, Vol. 3, Issue: 4, pp. 220228, 2009.
Presenter: Chia-Jui Chang Advisor: Yeou-Jiunn Chen
Outline Introduction Multiscale Mathematical Morphology Theory Methods Result Conclusion
Introduction
Analyzing the ECG signal is not an easy task due to the presence of large noises. ◦ Impulsive noise caused by muscle contraction ◦ Power-line interference ◦ Baseline drift
Many researchers treated QRS detection as a mature topic ◦ Body area networks(BANs) ◦ Body sensor networks (BSNs)
Captured ECG signal is severely distorted with baseline drift and motion artifacts.
Introduction
The BAN devices are generally restricted ◦ Size ◦ Power consumption ◦ Computational resources.
This calls for a simple and accurate QRS detection algorithm for BAN-based ECG devices.
Introduction
Multiscale mathematical morphology(3M) filtering concept into QRS detection. ◦ Rich theoretical framework ◦ Low computational complexity ◦ Simple hardware implementation
Avoiding the frequency band overlapping of QRS complexes and other components, such as P/T waves
Introduction
Three Question? ◦ A power-efficient implementation scheme for the 3M-based QRS detection algorithm ◦ What is the impact of the structure element ◦ The 3M applicable to wearable ECG devices in BAN applications
This paper will address these questions one by one.
MULTISCALE MATHEMATICAL MORPHOLOGY THEORY
Mathematical morphology is a powerful methodology for the quantitative analysis of geometrical structures. Broad and coherent collection of theoretical concepts Nonlinear signal operators Algorithms aiming to extract information related to the shape and size from images or other geometrical objects.
MULTISCALE MATHEMATICAL MORPHOLOGY THEORY
Methods Block diagram of the proposed QRS detection algorithm
Multiscale Mathematical Morphology Filtering
The multiscale opening and closing operations outperform the multiscale dilation and erosion filtering ◦ Space independence
The combination of the opening and closing operation ◦ Peak or valley extraction
The top-hat operator produces an output consisting of the signal peaks. The bottom-hat operator extracts the valleys.
Multiscale Mathematical Morphology Filtering
The weighted sum of the top-hat and bottom-hat operations forms a peak-valley extractor
Multiscale Mathematical Morphology Filtering
Multiscale Mathematical Morphology Filtering The ideal QRS detection solution should avoid the use of multiplier(s) in order to reduce the power. To be implemented by either microprocessor or a digital signal processor (DSP). The proposed 3M filter does not requires any multiplier.
Multiscale Mathematical Morphology Filtering
The proposed -scale filter is very similar to the linear-phase FIR ◦ The delay elements in the FIR filter are replaced by the opening or closing operators.
It is obvious that any standard linearphase FIR filter can be modified to realize the 3M filtering ◦ Changing the delay elements to either the opening or closing operators.
Multiscale Mathematical Morphology Filtering
Differential Operation
Differential Operation The output ECG sequence is differentiated in order to remove motion artifacts and baseline drifts.
Enhancing ECG by Modulus and Combination
Enhancing ECG by Modulus and Combination The absolute value of the differential output is combined by multiple-frame accumulation
Threshold and Decision
Threshold and Decision Most QRS detectors use similar methods to determine the threshold.
Result
The structure element plays an important role in the 3M filter. ◦ Shape ◦ Amplitude ◦ Length
Different structure elements to an ECG signal with additional noises ◦ 50-Hz power-line interference and random noise.
Result
Result
The structure elements with a smaller slope ◦ To perform better in terms of removing noise ◦ A larger reduction in signal amplitude.
The slope of the structure elements should be kept small in the first stage to maximize the filter performance.
Result
The MIT/BIH Arrhythmia Database is used to evaluate our algorithm. ◦ It contains 48 half-hour recordings of twochannel ◦ 360 samples per second per channel
Result Detection of the QRS by using tape 105 of the MIT/BIH database with baseline wander. Zoom-in segments for tape 105 of the MIT/BIH database in (a) over a 10-mV range.
Result
Result
It indicates that the proposed algorithm correctly detects the QRS of the resting and exercise ECG ◦ Noise ◦ Baseline drift ◦ Large P/T waves.
Result
Result Average QRS detection rate of 99.64%, Sensitivity of 99.83% Positive prediction of 99.82%
Result
Result
3M technique achieves the lowest DER and outstanding false positive and false negative values, except the ANN adaptive filtering ◦ The cost of high hardware complexity.
Result
Conclusion
The proposed algorithm utilizes a multiscale mathematical morphology filter and multiframe differential modulus accumulation to reduce the noise in the ECG signal. ◦ Doesn’t require prior knowledge of the frequency spectrum ◦ Easily implemented by a modified linear-phase FIR filter structure without multipliers. ◦ Make it very attractive for wearable devices in BAN applications.
References
QRS Detection Based on Multiscale Mathematical Morphology for Wearable ECG Devices in Body Area Networks ,Fei Zhang; Yong Lian,Biomedical Circuits and Systems, IEEE Transactions on Volume: 3 , Issue: 4 Digital Object Identifier: 10.1109/TBCAS.2009.2020093,Publication Year: 2009 , Page(s): 220 - 228
http://www.robotworld.org.tw/data/for_trade1-d.php?News_ID=4527
Presenter: Chia-Jui Chang Advisor: Yeou-Jiunn Chen
Outline Introduction Multiscale Mathematical Morphology Theory Methods Result Conclusion
Introduction
Analyzing the ECG signal is not an easy task due to the presence of large noises. ◦ Impulsive noise caused by muscle contraction ◦ Power-line interference ◦ Baseline drift
Many researchers treated QRS detection as a mature topic ◦ Body area networks(BANs) ◦ Body sensor networks (BSNs)
Captured ECG signal is severely distorted with baseline drift and motion artifacts.
Introduction
The BAN devices are generally restricted ◦ Size ◦ Power consumption ◦ Computational resources.
This calls for a simple and accurate QRS detection algorithm for BAN-based ECG devices.
Introduction
Multiscale mathematical morphology(3M) filtering concept into QRS detection. ◦ Rich theoretical framework ◦ Low computational complexity ◦ Simple hardware implementation
Avoiding the frequency band overlapping of QRS complexes and other components, such as P/T waves
Introduction
Three Question? ◦ A power-efficient implementation scheme for the 3M-based QRS detection algorithm ◦ What is the impact of the structure element ◦ The 3M applicable to wearable ECG devices in BAN applications
This paper will address these questions one by one.
MULTISCALE MATHEMATICAL MORPHOLOGY THEORY
Mathematical morphology is a powerful methodology for the quantitative analysis of geometrical structures. Broad and coherent collection of theoretical concepts Nonlinear signal operators Algorithms aiming to extract information related to the shape and size from images or other geometrical objects.
MULTISCALE MATHEMATICAL MORPHOLOGY THEORY
Methods Block diagram of the proposed QRS detection algorithm
Multiscale Mathematical Morphology Filtering
The multiscale opening and closing operations outperform the multiscale dilation and erosion filtering ◦ Space independence
The combination of the opening and closing operation ◦ Peak or valley extraction
The top-hat operator produces an output consisting of the signal peaks. The bottom-hat operator extracts the valleys.
Multiscale Mathematical Morphology Filtering
The weighted sum of the top-hat and bottom-hat operations forms a peak-valley extractor
Multiscale Mathematical Morphology Filtering
Multiscale Mathematical Morphology Filtering The ideal QRS detection solution should avoid the use of multiplier(s) in order to reduce the power. To be implemented by either microprocessor or a digital signal processor (DSP). The proposed 3M filter does not requires any multiplier.
Multiscale Mathematical Morphology Filtering
The proposed -scale filter is very similar to the linear-phase FIR ◦ The delay elements in the FIR filter are replaced by the opening or closing operators.
It is obvious that any standard linearphase FIR filter can be modified to realize the 3M filtering ◦ Changing the delay elements to either the opening or closing operators.
Multiscale Mathematical Morphology Filtering
Differential Operation
Differential Operation The output ECG sequence is differentiated in order to remove motion artifacts and baseline drifts.
Enhancing ECG by Modulus and Combination
Enhancing ECG by Modulus and Combination The absolute value of the differential output is combined by multiple-frame accumulation
Threshold and Decision
Threshold and Decision Most QRS detectors use similar methods to determine the threshold.
Result
The structure element plays an important role in the 3M filter. ◦ Shape ◦ Amplitude ◦ Length
Different structure elements to an ECG signal with additional noises ◦ 50-Hz power-line interference and random noise.
Result
Result
The structure elements with a smaller slope ◦ To perform better in terms of removing noise ◦ A larger reduction in signal amplitude.
The slope of the structure elements should be kept small in the first stage to maximize the filter performance.
Result
The MIT/BIH Arrhythmia Database is used to evaluate our algorithm. ◦ It contains 48 half-hour recordings of twochannel ◦ 360 samples per second per channel
Result Detection of the QRS by using tape 105 of the MIT/BIH database with baseline wander. Zoom-in segments for tape 105 of the MIT/BIH database in (a) over a 10-mV range.
Result
Result
It indicates that the proposed algorithm correctly detects the QRS of the resting and exercise ECG ◦ Noise ◦ Baseline drift ◦ Large P/T waves.
Result
Result Average QRS detection rate of 99.64%, Sensitivity of 99.83% Positive prediction of 99.82%
Result
Result
3M technique achieves the lowest DER and outstanding false positive and false negative values, except the ANN adaptive filtering ◦ The cost of high hardware complexity.
Result
Conclusion
The proposed algorithm utilizes a multiscale mathematical morphology filter and multiframe differential modulus accumulation to reduce the noise in the ECG signal. ◦ Doesn’t require prior knowledge of the frequency spectrum ◦ Easily implemented by a modified linear-phase FIR filter structure without multipliers. ◦ Make it very attractive for wearable devices in BAN applications.
References
QRS Detection Based on Multiscale Mathematical Morphology for Wearable ECG Devices in Body Area Networks ,Fei Zhang; Yong Lian,Biomedical Circuits and Systems, IEEE Transactions on Volume: 3 , Issue: 4 Digital Object Identifier: 10.1109/TBCAS.2009.2020093,Publication Year: 2009 , Page(s): 220 - 228
http://www.robotworld.org.tw/data/for_trade1-d.php?News_ID=4527
