A method for analyzing electroencephalogram signals separates periodic and aperiodic components

Abstract

The present application relates to the technical field of electroencephalogram signal analysis, and particularly relates to an electroencephalogram signal analysis method for separating periodic components and aperiodic components, which comprises the following steps: collecting electroencephalogram signals, and calculating power spectrum of the electroencephalogram signals; constructing an aperiodic component model based on the power spectrum, performing logarithmic transformation on the power spectrum, and fitting the aperiodic component model in logarithmic space by using a weighted least square method to output aperiodic component parameters; determining the aperiodic component according to the aperiodic component parameters, converting the power spectrum and the aperiodic component to linear space, and outputting periodic components by subtraction; and extracting electroencephalogram feature parameters based on the periodic components. The present application is used to effectively separate periodic components and aperiodic components in electroencephalogram signals, and extract key electroencephalogram feature parameters on this basis, so as to improve the stability and consistency of electroencephalogram signal analysis.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal analysis technology, and more specifically to an EEG signal analysis method that separates periodic and non-periodic components. Background Technology

[0002] Electroencephalography (EEG) signals, as important physiological signals reflecting brain neural activity, play a crucial role in the research and diagnosis of neurological diseases. Compared to detection methods such as magnetic resonance imaging (MRI), EEG signals offer advantages such as low acquisition cost, high temporal resolution, and ease of operation, thus finding widespread application in research on cognitive impairment diseases such as Alzheimer's disease. Currently, the main methods for analyzing EEG signals include power spectral analysis and brain network-based analysis methods, which characterize brain functional states by extracting energy distributions across different frequency bands or constructing functional connectivity networks.

[0003] Current EEG signal analysis methods typically analyze the overall power spectrum directly without distinguishing between periodic and non-periodic components. This mixing of the two makes the analysis results susceptible to interference from non-periodic background components, affecting the accuracy and stability of feature extraction. Furthermore, while some methods attempt to separate periodic and non-periodic components, there is a lack of unified standards in modeling methods and parameter settings, and many are not optimized for the characteristics of EEG signals from Alzheimer's patients, resulting in unstable separation effects and difficulty in effectively reflecting disease-related EEG changes. In modeling non-periodic components, existing methods often use ordinary least squares for fitting. Since the power spectrum typically has large amplitude values ​​in the low-frequency range, the low-frequency component tends to dominate the fitting process, reducing the accuracy of the fitting results across the entire frequency range. Simultaneously, current methods focus on relative power spectra or single-band energy indicators for feature extraction, lacking in-depth analysis based on the separation of periodic components, making it difficult to reveal the EEG change patterns and complexity at different stages of Alzheimer's disease. Summary of the Invention

[0004] The purpose of this invention is to provide a method for separating periodic and non-periodic components in electroencephalogram (EEG) signals to solve the above-mentioned technical problems. This method effectively separates the periodic and non-periodic components in EEG signals and extracts key EEG feature parameters, thereby improving the stability and consistency of EEG signal analysis.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components, the method comprising: S1. Acquire EEG signals and calculate the power spectrum of the EEG signals; EEG signals are acquired through EEG acquisition equipment, and can be further preprocessed before being input into the power spectrum calculation module. The power distribution of the EEG signals at different frequency points is obtained through frequency domain analysis.

[0006] S2. Based on the power spectrum, an aperiodic component model is constructed. The power spectrum is logarithmically transformed, and the aperiodic component model is fitted in logarithmic space using weighted least squares to output the aperiodic component parameters. The power spectrum is then transformed back to logarithmic space, converting the aperiodic component model into a linear function. A weighted loss function is constructed, assigning different weights to each frequency point, and the model parameters are solved to obtain the spectral slope and broadband offset. Through parametric modeling and fitting of the aperiodic component, the distribution of the aperiodic component in the power spectrum is determined, providing a basis for the separation of periodic components.

[0007] S3. Determine the aperiodic component based on the aperiodic component parameters, transform the power spectrum and the aperiodic component to a linear space, and output the periodic component by subtraction; extract EEG feature parameters based on the periodic component. Transform the power spectrum and aperiodic component in logarithmic space back to linear space, and perform subtraction on them in this linear space to obtain the periodic component; further calculate EEG feature parameters based on the periodic component, including the peak frequency and peak power of the α band, relative power spectral density, and normalized spectral entropy. By separating the periodic component, the extraction of EEG feature parameters is based on the signal after removing the aperiodic background, forming a parameter set used to describe the characteristics of the EEG signal.

[0008] Specifically, step S1 also includes preprocessing the EEG signals, the preprocessing steps of which include: S11. Delete non-target electrodes and locate the remaining channels; select target electrodes according to experimental requirements, remove electrode channels that are not involved in the analysis, and spatially locate the remaining channels in combination with electrode position information, thereby reducing the interference of irrelevant electrodes on the analysis results.

[0009] S12. Perform notch filtering on the EEG signal from 48Hz to 52Hz to suppress the influence of power frequency noise on the EEG signal and improve signal quality.

[0010] S13. Perform bandpass filtering on the EEG signal from 1Hz to 45Hz; limit the analysis frequency range and remove low-frequency drift and high-frequency noise.

[0011] S14. Perform bad channel interpolation and bad segment deletion on EEG signals; by identifying abnormal channels and performing interpolation repair, while deleting time periods containing obvious artifacts, the impact of abnormal data on overall signal analysis is reduced, and the continuity and reliability of data are improved.

[0012] S15. Perform independent component analysis on the EEG signal and remove electrooculography (EOG) and electromyography (EMG) components with a probability greater than 70%; reduce the interference of physiological artifacts such as EOG and EMG on the EEG signal.

[0013] S16. Perform bad segment deletion processing on the EEG signal again; after removing artifact components, further screen out the remaining abnormal segments.

[0014] Specifically, the power spectrum of the EEG signal is estimated using the Welch function. The EEG signal is divided into multiple time segments, each segment is windowed, and the spectrum of each segment is calculated. The power spectra of all segments are then averaged to obtain the power spectrum of the EEG signal. Averaging the spectral results across multiple time segments reduces the impact of random fluctuations on the spectrum estimation, resulting in a more stable power spectrum. The Welch function is configured as follows: window length is set to 4 seconds; window overlap is set to 50%; and frequency resolution is set to 0.1 Hz. The 0.1 Hz resolution is used to accommodate the subsequent calculation of the α peak peak frequency; too high a resolution would be too slow, while too low a resolution would fail to capture the accurate peak frequency.

[0015] Specifically, step S2 includes: S21. Constructing an EEG signal model: in It's an electroencephalogram (EEG) signal. It is the non-periodic component of brainwaves. f is the periodic component of EEG, and f is the frequency. By decomposing the power spectrum into periodic and non-periodic components, a unified mathematical model is established, providing a foundation for the subsequent modeling and separation of non-periodic components.

[0016] S22. Modeling non-periodic components based on EEG signal models: Where A is the broadband offset. It is the spectral slope; Take the logarithm of the power spectrum's horizontal axis and place the non-periodic components into logarithmic space: Taking the logarithm of the previously calculated power spectrum's horizontal axis reveals that the entire power spectrum approximates a straight line. Placing the aperiodic component in logarithmic space (taking the logarithm of both sides of the equation) shows that it is a linear function. This logarithmic transformation converts the aperiodic component model into a linear function, facilitating parameter determination using linear fitting methods.

[0017] S23. Fit the non-periodic components using a weighted loss function. The formula for the weighted loss function is: in It is the weight of the i-th frequency point. It is the power at the i-th frequency point; by introducing weighting coefficients, the influence of different frequency points on the fitting process is adjusted, so that the solution process of model parameters takes into account the differences of each frequency point.

[0018] S24. Output aperiodic component parameters, including spectral slope and broadband offset.

[0019] Specifically, weight The basis for determination is: Here, h represents the weight adjustment coefficient. By calculating the frequency values ​​corresponding to the frequency points, exponential function-based weight distributions are constructed around 6Hz and 20Hz, respectively, and these distributions are superimposed to form the overall weight function. Frequency-related weight coefficients are constructed to ensure that different frequency positions have different degrees of influence during the fitting process.

[0020] Specifically, step S3 includes: S31. Transform the power spectrum and aperiodic components into a linear space. Subtract the aperiodic components from the power spectrum in the linear space to obtain the periodic components. Perform the subtraction operation in the unified space to separate the periodic components from the aperiodic components.

[0021] S32. Calculate the peak power of the α band using the bubble sort algorithm, and obtain the peak frequency through indexing; traverse the power values ​​of each frequency point within the preset α band range, determine the maximum value through comparison operations, and record the corresponding frequency position. Obtain peak parameters reflecting the periodic oscillation characteristics, which are used to characterize the distribution of EEG signals in a specific frequency band.

[0022] S33. Divide the frequency band into several sub-bands and calculate the relative power spectrum: Where B is the target sub-frequency band, and T is all sub-frequency bands being analyzed. This represents the power value corresponding to the frequency. S34. Calculate the spectral entropy: in, This refers to the probability of the i-th frequency component: in The power spectral density is the frequency corresponding to the frequency. , ; Calculate the normalized spectral entropy by dividing the spectral entropy by the theoretical maximum entropy: Here, n is the number of frequency points. By calculating the spectral entropy and normalized spectral entropy, the dispersion of the frequency distribution is quantified, forming parameters used to describe the characteristics of EEG signals.

[0023] Specifically, the frequency bands are divided into 1-4 Hz sub-bands, 4-8 Hz sub-bands, 8-12 Hz sub-bands, 12-30 Hz sub-bands, and 30-45 Hz sub-bands. The higher frequency bands are not analyzed due to excessive high-frequency interference.

[0024] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs an aperiodic component model and fits it in logarithmic space using the weighted least squares method to obtain the aperiodic component parameters. Based on this, the power spectrum and the aperiodic component are transformed into linear space and subtracted, thereby achieving the separation of periodic and aperiodic components, so that subsequent analysis is based on the separated periodic components.

[0025] 2. This invention introduces weighting coefficients in the fitting process of aperiodic components and constructs a weighted least squares loss function, assigning different weights to different frequency points, thereby adjusting the influence of each frequency point on the results during the fitting process, making the fitting results of the aperiodic component model more balanced across the entire frequency range.

[0026] 3. Based on the separated periodic components, this invention extracts EEG characteristic parameters, including the peak frequency and peak power of the α band, and calculates the relative power spectral density and normalized spectral entropy, thus characterizing the EEG signal from multiple dimensions and enriching the expression form of EEG characteristic parameters. Attached Figure Description

[0027] Figure 1 This is a flowchart of an electroencephalogram (EEG) signal analysis method for separating periodic and non-periodic components according to the present invention. Figure 2 This is a comparison of the EEG signal analysis method for separating periodic and non-periodic components according to the present invention with existing technologies in terms of non-periodic parameters. Figure 3 This is a comparison of α-peak characteristics in different populations in the EEG signal analysis method for separating periodic and non-periodic components according to the present invention. Figure 4 This invention relates to a method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components, which illustrates the variation trend of the α peak among different groups. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0029] like Figure 1 The method shown is an EEG signal analysis method for separating periodic and non-periodic components. The method includes: S1. Collect EEG signals and calculate the power spectrum of the EEG signals; S2. Construct an aperiodic component model based on the power spectrum, perform a logarithmic transformation on the power spectrum, and fit the aperiodic component model in the logarithmic space using the weighted least squares method to output the aperiodic component parameters. S3. Determine the aperiodic components based on the aperiodic component parameters, convert the power spectrum and aperiodic components to linear space, and output the periodic components by subtraction; extract EEG feature parameters based on the periodic components.

[0030] Furthermore, step S1 also includes preprocessing the EEG signal, the preprocessing steps of which include: S11. Delete non-target electrodes and locate the channel electrodes in the remaining channels; S12. Perform notch filtering on the EEG signal from 48Hz to 52Hz. S13. Perform bandpass filtering on the EEG signal from 1Hz to 45Hz. S14. Perform bad lead interpolation and bad segment deletion processing on EEG signals; S15. Perform independent component analysis on the EEG signal and remove electrooculography (EOG) and electromyography (EMG) components with a probability greater than 70%. S16. Perform bad segment deletion processing on the EEG signal again.

[0031] Specifically, the power spectrum of the EEG signal is estimated using the Welch function.

[0032] Specifically, configure the Welch function: set the window length to 4 seconds; set the window function overlap rate to 50%; and set the frequency resolution to 0.1Hz.

[0033] Specifically, step S2 includes: S21. Constructing an EEG signal model: in It's an electroencephalogram (EEG) signal. It is the non-periodic component of brainwaves. It is the periodic component of EEG, where f is the frequency; S22. Modeling non-periodic components based on EEG signal models: Where A is the broadband offset. It is the spectral slope; Take the logarithm of the power spectrum's horizontal axis and place the non-periodic components into logarithmic space: S23. Fit the non-periodic components using a weighted loss function. The formula for the weighted loss function is: in It is the weight of the i-th frequency point. It is the power at the i-th frequency point; S24. Output aperiodic component parameters, including spectral slope and broadband offset.

[0034] Specifically, weight The basis for determination is: Where h represents the weight adjustment coefficient. Preferably, the weight adjustment coefficient h is 0.75.

[0035] Specifically, step S3 includes: S31. Transform the power spectrum and aperiodic components to linear space, and subtract the aperiodic components from the power spectrum in linear space to obtain the periodic components. S32. Calculate the peak power of the α band using the bubble sort algorithm, and obtain the peak frequency by indexing; S33. Divide the frequency band into several sub-bands and calculate the relative power spectrum: Where B is the target sub-band and T is all sub-bands being analyzed; S34. Calculate the spectral entropy: in, This refers to the probability of the i-th frequency component: in The power spectral density is the frequency corresponding to the frequency. , ; Calculate the normalized spectral entropy by dividing the spectral entropy by the theoretical maximum entropy: Where n is the number of frequency points.

[0036] Specifically, the frequency band is divided into 1-4 Hz sub-band, 4-8 Hz sub-band, 8-12 Hz sub-band, 12-30 Hz sub-band and 30-45 Hz sub-band.

[0037] like Figure 2 As shown, in one embodiment, based on the baseline EEG signals of 13 patients with mild Alzheimer's disease before transcranial alternating current stimulation, the EEG signal analysis method described above was used to process the signals and obtain non-periodic component parameters, which were then compared with existing methods. The results show that, under the same data conditions, the broadband offset and spectral slope obtained by the method of this invention differ from those of existing methods, reflecting the differences in non-periodic component modeling results between the two methods.

[0038] like Figure 3 As shown, in one embodiment, based on baseline EEG signals collected from 13 patients with mild Alzheimer's disease and 8 patients with mild cognitive impairment before transcranial alternating current stimulation, the periodic components were analyzed using the aforementioned EEG signal analysis method to extract EEG characteristic parameters, including the peak frequency and peak power of the alpha band. Based on the statistical results of these EEG characteristic parameters, differences were found between different subject groups in the peak frequency and peak power.

[0039] like Figure 4 As shown, in one embodiment, based on the aforementioned EEG signal data, the normalized spectral entropy is calculated using the EEG signal analysis method. Without separating the periodic and non-periodic components, the difference in normalized spectral entropy between different subject groups is not significant; after separation and calculation using the method of this invention, differences in normalized spectral entropy appear between different subject groups, indicating that the method of this invention provides different analytical results in terms of EEG feature parameter extraction.

[0040] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0042] The present invention has been further described above with reference to specific embodiments. However, it should be understood that the specific description herein should not be construed as limiting the nature and scope of the present invention. Various modifications made to the above embodiments by those skilled in the art after reading this specification are all within the scope of protection of the present invention.

Claims

1. A method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components, characterized in that, The method includes: S1. Acquire electroencephalogram (EEG) signals and calculate the power spectrum of the EEG signals; S2. Construct an aperiodic component model based on the power spectrum, perform a logarithmic transformation on the power spectrum, and fit the aperiodic component model in the logarithmic space using the weighted least squares method to output the aperiodic component parameters. S3. Determine the aperiodic component based on the aperiodic component parameters, convert the power spectrum and the aperiodic component to a linear space, and output the periodic component by subtraction; extract EEG feature parameters based on the periodic component.

2. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 1, characterized in that, Step S1 further includes preprocessing the electroencephalogram (EEG) signal, the preprocessing step including: S11. Delete non-target electrodes and locate the channel electrodes in the remaining channels; S12. The EEG signal is subjected to notch filtering processing from 48Hz to 52Hz; S13. Perform bandpass filtering on the EEG signal from 1Hz to 45Hz; S14. Perform bad lead interpolation and bad segment deletion processing on the EEG signal; S15. Perform independent component analysis on the electroencephalogram (EEG) signal and remove electrooculogram (EOG) and electromyogram (EMG) components with a probability greater than 70%. S16. Perform bad segment deletion processing on the EEG signal again.

3. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 1, characterized in that, The power spectrum of the EEG signal was estimated using the Welch function.

4. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 3, characterized in that, Configure the Welch function as follows: set the window length to 4 seconds; set the window function overlap rate to 50%; and set the frequency resolution to 0.1 Hz.

5. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 1, characterized in that, Step S2 includes: S21. Constructing an EEG signal model: in It's an electroencephalogram (EEG) signal. It is the non-periodic component of brainwaves. It is the periodic component of EEG, where f is the frequency; S22. Based on the aforementioned EEG signal model, model the non-periodic components: Where A is the broadband offset. It is the spectral slope; Taking the logarithm of the power spectrum's horizontal axis and placing the aperiodic component into logarithmic space: S23. Fit the non-periodic components using a weighted loss function, the formula for which is: in It is the weight of the i-th frequency point. It is the power at the i-th frequency point; S24. Output aperiodic component parameters, including spectral slope and broadband offset.

6. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 5, characterized in that, Weight The basis for determination is: Where h represents the weight adjustment coefficient.

7. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 1, characterized in that, Step S3 includes: S31. Convert the power spectrum and aperiodic components to a linear space, and subtract the aperiodic components from the power spectrum in the linear space to obtain the periodic components. S32. Calculate the peak power of the α band using the bubble sort algorithm, and obtain the peak frequency by indexing; S33. Divide the frequency band into several sub-bands and calculate the relative power spectrum: Where B is the target sub-band and T is all sub-bands being analyzed; S34. Calculate the spectral entropy: in, This refers to the probability of the i-th frequency component: in The power spectral density is the frequency corresponding to the frequency. , ; Calculate the normalized spectral entropy by dividing the spectral entropy by the theoretical maximum entropy: Where n is the number of frequency points.

8. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 7, characterized in that, The frequency band is divided into 1-4 Hz sub-band, 4-8 Hz sub-band, 8-12 Hz sub-band, 12-30 Hz sub-band and 30-45 Hz sub-band.

9. The method for analyzing electroencephalogram (EEG) signals by separating periodic and non-periodic components according to claim 6, characterized in that, The weight adjustment coefficient h is set to 0.75.

Images