CN-122004902-A - Electroencephalogram signal reconstruction method and system based on artifact removal

CN122004902ACN 122004902 ACN122004902 ACN 122004902ACN-122004902-A

Abstract

The invention discloses an electroencephalogram signal reconstruction method and system based on artifact removal, wherein the method comprises the steps of obtaining multichannel electroencephalogram signals, conducting filtering pretreatment, conducting blind source separation through independent component analysis, obtaining statistically independent signal source components and a mixed matrix through decomposition, constructing an equivalent current dipole model, calculating a spatial artifact index, evaluating signal complexity through time sequence sample entropy, identifying artifact components and nerve activity components through weighted fusion and self-adaptive threshold judgment, retaining nerve activity components and setting the artifact components, and reconstructing pure electroencephalogram signals through inverse transformation of the mixed matrix. The invention realizes the accurate identification and effective removal of the brain electrical artifacts and provides high-quality data for brain electrical analysis.

Inventors

LIU ZHE
Tang Congneng
YUAN ZAIXIN

Assignees

湖南万脉医疗科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (7)

1. The electroencephalogram signal reconstruction method based on artifact removal is characterized by comprising the following steps of: B1, acquiring original multi-channel brain electrical signal data streams from a plurality of scalp electrodes, performing time domain pretreatment by setting a high-pass filter and a low-pass filter, filtering out ultrahigh frequency noise and direct current drift, and obtaining a pretreated multi-channel brain electrical signal; B2, inputting the preprocessed multichannel electroencephalogram signals into an independent component analysis algorithm to perform blind source separation, and decomposing to obtain statistically independent signal source components and corresponding mixing matrixes; b3, constructing an equivalent current dipole model for each independent signal source component, calculating a spatial artifact index, calculating a time sequence sample entropy, evaluating signal complexity, obtaining a comprehensive artifact score through weighted fusion, carrying out self-adaptive threshold judgment, and identifying artifact components and nerve activity components; And B4, reserving neural activity components and setting artifact components, and performing linear recombination by using a mixing matrix to reconstruct pure multichannel brain electrical signals.
2. The method for reconstructing an electroencephalogram based on artifact removal according to claim 1, wherein the step B1 comprises: The method comprises the steps of establishing a multichannel electroencephalogram signal acquisition system, acquiring original multichannel electroencephalogram signal data streams through a plurality of electrodes worn on the scalp, wherein the original multichannel electroencephalogram signals comprise effective signals from brain nerve activities and various physiological artifacts and environmental artifacts, performing time domain preprocessing on the original multichannel electroencephalogram signals, setting the cut-off frequency of a high-pass filter to be 0.5Hz for filtering direct current drift and ultralow frequency noise, setting the cut-off frequency of a low-pass filter to be 50Hz for filtering ultrahigh frequency interference signals irrelevant to the brain nerve activities, and obtaining the preprocessed multichannel electroencephalogram signals through band-pass filtering processing.
3. The method for reconstructing an electroencephalogram based on artifact removal according to claim 2, wherein the step B2 comprises: Inputting the preprocessed multichannel EEG signals into an independent component analysis algorithm for blind source separation processing, wherein the independent component analysis algorithm is based on statistical independence assumption, decomposing the mixed multichannel EEG signals into a group of mutually independent signal source components in statistics, setting an objective function of independent component analysis to maximize the statistical independence among the independent components, and solving a separation matrix through an iterative optimization algorithm The output independent components meet the statistical independence condition, and a group of independent component time sequences are obtained And corresponding mixing matrix Wherein Indicating the number of individual components that are present, Is the first The individual components of the composition are selected from the group consisting of, Representing a transpose of a matrix, the hybrid matrix Record how each independent component is linearly combined to form original multi-channel electroencephalogram signals, and satisfy the relation Wherein Representing the preprocessed multichannel electroencephalogram signals.
4. The method for reconstructing an electroencephalogram based on artifact removal according to claim 3, wherein the step B3 comprises: constructing an artifact component self-adaptive identification algorithm based on fusion of an equivalent current dipole model and time sequence entropy, and aiming at each independent component Making artifact identification decisions, first computing spatial artifact indices, including using independent components In a hybrid matrix Corresponding weight vector in (a) Fitting an equivalent current dipole source in a preset three-dimensional human brain head model, and calculating the space deviation distance between the position of the equivalent current dipole source and the brain parenchyma region And fitting residual variance The spatial artifact index The calculation formula of (2) is as follows: ; Wherein, the Representing the spatial deviation weight coefficient(s), Representing a preset maximum deviation distance threshold, Represents the residual variance weight coefficient and satisfies , Represent the first Spatial artifact indices of individual components; Secondly, calculating the complexity characteristic of the time sequence, and aiming at independent components Is used for calculating sample entropy The sample entropy The calculation formula of (2) is as follows: ; Wherein, the Represent the first The sample entropy of the individual components, Representing individual components The pattern length in the time sequence is And the distance is less than the tolerance Is used for the number of pattern matches of (a), Representing individual components The pattern length in the time sequence is And the distance is less than the tolerance Is used for the number of pattern matches of (a), A mode length parameter is represented and is used to indicate, A tolerance parameter is indicated and a tolerance parameter is indicated, Representing a natural logarithmic function; and finally, comprehensive artifact score calculation and self-adaptive threshold judgment are carried out.
5. The artifact removal-based electroencephalogram reconstruction method of claim 4, wherein the comprehensive artifact score and adaptive threshold decision comprises: computing a composite artifact score : ; Wherein, the Represent the first The composite artifact score of the individual components, The weight coefficient of the spatial feature is represented, The weight coefficient of the time characteristic is represented, Represents the maximum value of the sample entropy in all independent components and satisfies Setting adaptive threshold When (when) Time-determining independent components As an artifact component, when Time-determining independent components Is a component of neural activity.
6. The method for reconstructing an electroencephalogram based on artifact removal according to claim 4, wherein said step B4 comprises: constructing a clean independent component matrix according to the artifact identification result in the step B3 Zeroing out all independent components determined to be artifact components, retaining all independent components determined to be neural activity components, said cleaning the independent component matrix Wherein Represent the first The reconstructed pure multichannel electroencephalogram signal comprises a plurality of cleaned independent components The calculation formula of (2) is as follows: ; Wherein, the And representing the reconstructed pure multichannel brain electrical signals.
7. An electroencephalogram signal reconstruction system based on artifact removal is characterized by comprising: The preprocessing module is used for acquiring original multichannel electroencephalogram signal data streams from a plurality of scalp electrodes, performing time domain preprocessing by setting a high-pass filter and a low-pass filter, and filtering out ultrahigh frequency noise and direct current drift; The blind source separation module inputs the preprocessed multichannel electroencephalogram signals into an independent component analysis algorithm to perform blind source separation, and decomposes the multichannel electroencephalogram signals to obtain statistically independent signal source components and corresponding mixing matrixes; The artifact identification module is used for constructing an equivalent current dipole model for each independent component, calculating a space artifact index, calculating the entropy of a time sequence sample, evaluating the complexity of a signal, obtaining a comprehensive artifact score through weighted fusion, carrying out self-adaptive threshold judgment, and identifying an artifact component and a neural activity component; the signal reconstruction module is used for reserving neural activity components and setting artifact components, performing linear recombination by using an inverse matrix of the mixed matrix, and reconstructing pure multichannel electroencephalogram signals; To implement the artifact removal-based electroencephalogram signal reconstruction method according to any one of claims 1 to 6.

Description

Electroencephalogram signal reconstruction method and system based on artifact removal Technical Field The invention relates to the technical field of signal processing, in particular to an electroencephalogram signal reconstruction method and system based on artifact removal. Background The electroencephalogram technology is used as a noninvasive nerve electrophysiological detection means, can record the electrical activity of cerebral cortex neuron groups in real time, has millisecond-level time resolution, and plays an important role in the fields of neuroscience research, clinical diagnosis, brain-computer interfaces and the like. With the deep development of brain science research and the wide application of brain-computer interface technology, the demand for high-quality brain electrical signals is urgent, however, the brain electrical signals are inevitably polluted by various physiological and non-physiological artifacts in the acquisition process, and the signal quality and the accuracy of subsequent analysis are seriously affected. Artifacts in the electroencephalogram signals mainly comprise ocular artifacts, myoelectric artifacts, electrocardiographic artifacts, external electromagnetic interference and the like, the amplitude of the artifact signals is often far greater than that of the real electroencephalogram signals, and useful neural activity information can be completely covered. The ocular artifacts are produced by eye movements and blinking actions, and are mainly manifested by low frequency, large amplitude potential changes, most pronounced at the forehead electrode. Myoelectric artifacts originate from contractile activity of the head and neck muscles, often manifested as high frequency, abrupt electrical signals. Electrocardiographic artifacts are caused by the propagation of an electric field generated by the beating of the heart to the scalp electrode, exhibiting periodic spike waveforms. These artifacts not only affect the visual observation of the brain electrical signal, but more importantly can lead to erroneous results of brain electrical feature extraction and pattern recognition. The traditional electroencephalogram artifact removal method mainly comprises frequency domain filtering, time domain filtering and artifact removal technology based on template matching. The frequency domain filtering method suppresses the artifact signals of the specific frequency band by designing a band-pass filter or a notch filter, but the method has obvious limitation, and when the artifact signals overlap with useful nerve signals in the frequency domain, important nerve activity information is inevitably lost in the filtering process. Time domain filtering methods such as moving average filtering and median filtering, while capable of smoothing signals and removing part of noise, have limited effect on artifact signals with large amplitude and are prone to signal distortion and loss of time resolution. The artifact removal method based on template matching requires pre-establishing a standard template of an artifact signal, and removing an artifact component through pattern matching and subtraction operation. However, such methods face problems of difficult template establishment, significant individual differences, poor adaptability, and the like, and are difficult to handle complex and variable actual artifact conditions. In addition, the traditional method mostly adopts fixed parameter setting and unified processing strategies, and cannot adapt to individual characteristics of different subjects and signal changes under different experimental conditions, so that the artifact removal effect is unstable. The independent component analysis method which is rising in recent years provides a new technical approach for removing the brain electrical artifacts, and the method can decompose the mixed brain electrical signals into a plurality of mutually independent components based on the statistical independence assumption of the signals. However, the existing artifact removal method based on independent component analysis mainly depends on artificial vision judgment or simple statistical characteristics in a component identification link, lacks an objective and quantitative automatic identification algorithm, is low in efficiency, is easily influenced by subjective factors, and is difficult to ensure in identification accuracy. Meanwhile, the existing method only considers single characteristic dimension when the artifact component is identified, for example, the method is only based on frequency domain characteristics or only based on spatial distribution characteristics, ignores multi-dimensional characteristics of artifact signals, and is low in identification accuracy. More importantly, the prior art lacks an adaptive threshold judgment mechanism, and mostly adopts a fixed threshold or an empirical threshold to judge artifact components, so that the method cannot adapt to the characteristic