An Adaptive Kalman Filter for ECG Signal Enhancement
read more
Citations
A survey on ECG analysis
An Efficient Threshold Prediction Scheme for Wavelet Based ECG Signal Noise Reduction Using Variable Step Size Firefly Algorithm
A comprehensive survey of wearable and wireless ECG monitoring systems for older adults
ECG signal enhancement based on improved denoising auto-encoder
Optimization of signal quality over comfortability of textile electrodes for ECG monitoring in fog computing based medical applications
References
A New Approach to Linear Filtering and Prediction Problems
Numerical Recipes in C: The Art of Scientific Computing
Numerical Recipes, The Art of Scientific Computing
PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals.
Related Papers (5)
Frequently Asked Questions (20)
Q2. How is the estimation of the measurement noise covariance performed?
The estimation of the measurement noise covariance is performed by exploiting the spatial correlation of simultaneously recorded multichannel ECG signals.
Q3. What is the effect of omitting the preprocessing?
The effect of omitting the preprocessing, in order to yield as little distortion of the filtered ECG signals as possible, on the performance of the filter is a subject of future research.
Q4. What is the effect of the high-pass filter on the susceptible ST segment?
With regard to the preprocessing of the TWA (and fetal and neonatal ECG) signals, the high-pass filter is expected to slightly distort the susceptible ST segment.
Q5. How is the performance of the filter evaluated?
As mentioned earlier, the performance of the filter is assessed as a function of both N (i.e., the number of residuals ρ averaged for robust estimation of the process noise covariance) and the SNR, using the TWA signals of 23 different patients.
Q6. Why is the noise in the ECG signals minimal for small N?
The fact that for high-SNR ECG signals is minimal for small N stems from the fact that, with almost no noise present, most variations in the ECG signals are of physiological origin.
Q7. What is the effect of the measurement noise covariance on the filter?
underestimation of the measurement noise covariance leads to overestimation of the process noise covariance, causing the filter to also ascribe more weight to ECG complexes that are corrupted by measurement noise.
Q8. What is the effect of the overestimation of the measurement noise covariance?
When this covariance is overestimated, all ECG signal variations will be ascribed to measurement noise, and hence, the process noise covariance will be underestimated, rendering the filter less capable of quickly adapting to dynamical signal variations.
Q9. What is the importance of the adaptive Kalman filter?
The accurate estimation of the measurement noise covariance is rather critical for the performance of the adaptive Kalman filter.
Q10. How many independent ECG signals are needed to estimate the fourth?
The main limitation of this method for estimating the measurement noise is that, at any time, at least four ECG signals have to be recorded: three independent ones to estimate the fourth.
Q11. How long does the adaptive Kalman filter need to adapt to the new morphology?
After the movement of the fetus, the fixed Kalman filter needs about 10 s to completely adapt its output to the new ECG morphology [see Fig. 7(b)], whereas adaptation by the Kalman filter with adaptive noise covariance is more than twice faster.
Q12. How is the performance of the ECG filter quantified?
The performance is quantified by calculating , the normalized MSE between the filtered ECG signals x̂ and the original ECG signals x used (i.e., the signals without additive noise) as follows:= ∑ k (xk − x̂k ) T (xk − x̂k )∑k x T k xk(15)where the summation indicates that is averaged over all heartbeats in the TWA signals.
Q13. What is the effect of the reassembling of the filtered ECG complexes?
Upon reassembling the filtered ECG complexes into a filtered ECG signal that is composed of a multitude of heartbeats, the redundant parts of the filtered ECG complexes can be omitted.
Q14. What is the effect of the mathematical simplification on the filter?
Besides providing a rather elegant solution to the filter problem, the mentioned mathematical simplification also relaxes the computational complexity of the filter, rendering an implementation ofthe filter in MATLAB (The Mathworks, Inc.) capable of filtering at least 12 ECG signals simultaneously in real time.
Q15. How is the SNR of the ECG signals enhanced?
To facilitate this detection, the SNR of the ECG signals is a priori enhanced by linearly combing the signals in such a way as to maximize the variance [principal component analysis (PCA)] [22].
Q16. What is the performance of the fixed Kalman filter?
From Fig. 6, it can be seen that for the Kalman filter with fixed process noise covariance, for simplicity, from here on referred to as the fixed Kalman filter, the performance improves with decreasing λ2 until λ2 = −20 dB; from here on, the performance slightly deteriorates.
Q17. What is the effect of the adaptive Kalman filter on the morphology of the ECG?
After the movement epoch, the fetus has taken a slightly different orientation with respect to the electrodes on the maternal abdomen, affecting the morphology of the ECG signal.
Q18. What is the purpose of the maximization of p(k+1)?
The maximization of p(ρk+1) can be simplified, if the authors return to the intended purpose of the Kalman filter, to adaptively vary the number of averages n used in the enhancement of the ECG complexes, depending on the dynamic variations in signal morphology.
Q19. What are the main assumptions in the derivation of the adaptive Kalman filter?
In the derivation of the adaptive Kalman filter, several assumptions are made for mathematical simplicity, but that might limit the applicability of the filter.
Q20. How many ECG signals are required for noninvasive fetal ECG?
For noninvasive fetal ECG recordings performed on the maternal abdomen, the aforementioned requirement of at least four ECG signals, of which three are linearly independent, could be troublesome.