(Reference: Pattern Recognition (ECG Signal Processing and Recognition) )
1. Sources of ECG signal interference
The human body's ECG signal is a nonlinear, non-stationary, and random weak physiological signal . During the signal acquisition process, the ECG signal is easily interfered by various factors such as instruments, human activities, operators, and the surrounding environment. noise. The main noises of ECG (electrocardiogram) signals include: baseband drift, myoelectric interference, power frequency noise, and other noise interference.
Baseline drift is one of the main noises of the ECG signal, mainly due to patient breathing. The frequency is generally lower than 1Hz caused by the sliding of the electrode patch, etc., and it appears as a slow-changing sinusoidal curve, as shown in the following figure:
Myoelectric interference is caused by irregular high-frequency electrical disturbances generated by human muscle tremors. Its frequency range is very wide, generally between 10 and 1000 Hz, and it manifests as irregular and rapidly changing waveforms;
Power frequency noise is another kind of noise of ECG signal. It mainly comes from the power frequency power supply and the electromagnetic field radiated from the surrounding environment of the device. The frequency is 50Hz or 60Hz. It manifests as periodic small ripples, as shown in the figure below:
2. Signal preprocessing
1. Remove baseline drift
The frequency of the baseline noise is relatively low, and the ECG signal itself contains a lot of low-frequency signals, so a low-pass filter cannot be used to remove the baseline drift. The traditional methods for removing baseline drift include: median filter method, wavelet transform method, algorithmic average filter method, EMD decomposition, etc.
Considering that the baseline drift of the ECG signal is relatively smooth, this paper uses the baseline fitting method (the least squares method to fit the baseline function) to filter out the baseline drift . For a detailed introduction of this method and Matlab simulation, see: [20220114] [Signal Processing] What is a baseline Drift/trend item? How to eliminate?
The overall processing flow is as follows:
2. EMG removal
The frequency component of the EMG signal is related to the type of muscle, generally in the range of 30-300 Hz, while the frequency of the ECG signal is mainly concentrated in the 5-20 Hz, so the EMG signal is removed by designing a low-pass filter .
3. Remove power frequency noise
Since the power frequency noise is usually at 50Hz or 60Hz, the power frequency noise can be filtered out by analyzing the frequency domain of the original signal and designing a band stop filter according to the signal spectrum . The overall processing flow is as follows:
(Reference: ECG signal preprocessing based on MATLAB )
3. ECG signal screening
1. Basic characteristics of the ECG signal
As shown in the figure below, each regular waveform cycle of the ECG signal represents a cardiac cycle. The cardiac cycle is about 0.80s, and the cycle of the ECG signal is also about 0.80s. One cycle of the ECG signal contains P waves, QRS complex waves, and T waves. The start, end, peak, trough, and interval of these waveforms record the detailed information of the heart activity state, which provides very important information for the diagnosis of heart diseases. basis for analysis.
2. ECG Signal Screening Conditions
Refer to the three screening criteria in the paper:
(1) Each ECG signal period is between 0.6s~1.0s;
(2) Each ECG signal period has one and only one QRS complex wave;
(3) The deviation of multiple ECG signal cycles should not be too large;
3. Screening algorithm design
Firstly, the peak/trough detection algorithm is designed, which can process the signal characteristics such as signal period and waveform sequence, and perform conditional screening to obtain high-quality ECG signals after screening. The overall processing flow is as follows:
