伪迹移除 | Artifacts reduction algorithms's two main ways

Artifacts reduction algorithms’s two main ways

Regression and Filtering Methods

Regression

The regression model use a function to fit the data to smooth the data.
Disadvantages:
1.this method only works for reference channels that are available.
2.EEG signal being non-linear and non-stationary process,linear regression is not the best choice for analysis in such applications.
3.it can only be used to treat few particular types of artifact,not all types.

Filtering

linear adaptive filters

Note:too sensitive and unstable to adjust the parameters

non-linear adaptive filters(include Volterra filters and neural network based adptive filters)

Note:stronger processing capabilities and complex calculation
Disadvantages:
filters may eliminate useful EEG signals during artifact deletion.

Separate or Decompose EEG Data and Noise Data into Other Domains

EMD(Empiricla Mode Decomposition)

decompose the input signals into multiple empirical modes according to IMF.
Note:EMD is an empirical and data-driven method
Disadvantages:
computationally complex
may not be suitable for online application

ICA(developed from BSS)

separate the ideal signal and noise included in the EEG signal as independent components
Disadvantages:
1.not automatic
2.requiring human intervention makes results subjective and time consuming
3.cannot operate on single-channel data
4.high computational complexity

WT(wavelet transfrom)

maps the signal to the wavelet domain.
According to the wavelet coefficients of signal and noise,they have different properties and mechanisms at different scales,eliminating the wavelet coefficients generated by noise and maximally retaining the coefficients fo real signals.