Data standardization method based on different electroencephalogram acquisition systems

Abstract

The invention discloses a data standardization method based on different electroencephalogram acquisition systems, which comprises the following steps: firstly, denoising EEG signals acquired by different electroencephalogram acquisition systems by adopting a spectral subtraction method to obtain denoised signals, and then, carrying out amplitude-frequency response correction on the denoised signals to obtain a data standardization result. According to the method, signals acquired by different electroencephalogram acquisition systems are corrected from the angles of noise and amplitude-frequency response, so that standardized processing of electroencephalogram data acquired by different electroencephalogram acquisition systems is realized, and the influence of data inconsistency caused by performance difference of the electroencephalogram acquisition systems is effectively reduced; and a new technology for standardizing data acquired by different electroencephalogram acquisition systems is provided for solving the difficulty of multi-center electroencephalogram acquisition in a cross-center standardization process.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electroencephalogram data collection, and particularly relates to a design of a data standardization method based on different electroencephalogram collection systems. BACKGROUND

[0002] Researches exploring the development process of the brain often need large-sample electroencephalogram data. At present, a multi-center research strategy is often used to collect data samples from the population. This strategy puts forward requirements for electroencephalogram collection systems. If electroencephalogram collection systems with large performance differences are used, it often brings the problem of large data differences and difficulty in standardization.

[0003] To avoid the data differences caused by using different electroencephalogram collection systems, the following solutions are adopted in the known multi-center electroencephalogram research: (1) The EEGManyLabs project assigns research tasks to cooperating laboratories that all use the same type of electroencephalogram equipment and electrodes, ensures the consistency of equipment for each sample group, and performs Meta analysis internally according to the type of equipment. However, this scheme requires a sufficient number of sites to participate in the project, which increases the management and operation cost of the project; (2) The HBCD project requires 27 participating sites to use the same electroencephalogram collection system (MagStim EGI Net Station Amps 400). This scheme requires certain requirements for participating sites, which reduces the number of participating sites and reduces the statistical effectiveness of samples.

[0004] In summary, the above two mainstream schemes essentially avoid the problem of equipment differences by limiting the type of equipment, rather than directly standardizing the data of different equipment. This limits the scale and flexibility of the research. Therefore, it is urgent to develop a method that can directly standardize the data obtained by different electroencephalogram collection systems to truly release the potential of multi-center research. SUMMARY

[0005] The purpose of the application is to provide a data standardization method based on different electroencephalogram collection systems. The signals obtained by different electroencephalogram collection systems are corrected and processed from the perspectives of noise and amplitude-frequency response, so as to realize the standardization processing of the electroencephalogram data obtained by different electroencephalogram collection systems.

[0006] The technical scheme of the application is a data standardization method based on different electroencephalogram collection systems, comprising the following steps: S1, performing denoising processing on the EEG signals collected by different electroencephalogram collection systems by using a spectrum subtraction method to obtain denoised signals.

[0007] S2, performing amplitude-frequency response correction on the denoised signals to obtain a data standardization result.

[0008] Further, step S1 comprises the following sub-steps: S11, segmenting the noise signal data, and obtaining a noise power spectrum through fast Fourier transform and average processing.

[0009] S12, segmenting the EEG signal collected by the different EEG acquisition systems through a sliding window, to obtain a first sliding window segmentation result.

[0010] S13, connecting the first sliding window segmentation result to the tail of the EEG signal after mirror processing, to construct a first symmetric signal.

[0011] S14, mapping the first symmetric signal into a frequency domain signal through fast Fourier transform, to obtain a first to-be-processed power spectrum.

[0012] S15, calculating a spectral subtraction result according to the first to-be-processed power spectrum and the noise power spectrum.

[0013] S16, calculating a phase spectrum of a true response EEG signal according to the spectral subtraction result.

[0014] S17, obtaining a first time domain recovery signal through inverse fast Fourier transform according to the spectral subtraction result and the phase spectrum of the true response EEG signal.

[0015] S18, performing average deoverlap processing on the first time domain recovery signal, to obtain a denoised signal.

[0016] Further, in step S13, the phase of the first symmetric signal is set to 0.

[0017] Further, in step S15, the calculation formula of the spectral subtraction result is: wherein represents a true response EEG signal power spectrum, i.e., the spectral subtraction result, represents an EEG signal power spectrum collected by the EEG acquisition system, i.e., the first to-be-processed power spectrum, represents the noise power spectrum, represents a frequency.

[0018] Further, in step S16, the calculation formula of the phase spectrum of the EEG signal is: wherein represents the phase spectrum of the true response EEG signal, represents an imaginary unit, represents a phase angle, represents an EEG signal phase spectrum collected by the EEG acquisition system.

[0019] Further, step S2 comprises the following sub-steps: S21, data cleaning is performed on the amplitude-frequency response data, and a reverse filter is constructed to obtain an amplitude-frequency response of the reverse filter.

[0020] S22, the de-noised signal is segmented by a sliding window to obtain a second sliding window segmentation result.

[0021] S23, the second sliding window segmentation result is connected to the tail of the de-noised signal after mirror processing to construct a second symmetric signal.

[0022] S24, the second symmetric signal is mapped into a frequency domain signal through fast Fourier transform to obtain a second to-be-processed power spectrum and a phase spectrum.

[0023] S25, amplitude-frequency response correction is performed according to the amplitude-frequency response of the reverse filter and the second to-be-processed power spectrum to obtain an amplitude-frequency response correction result.

[0024] S26, the second time domain recovery signal is obtained through inverse fast Fourier transform according to the amplitude-frequency response correction result and the phase spectrum.

[0025] S27, the second time domain recovery signal is subjected to average de-overlapping processing to obtain a data standardization result.

[0026] Further, the phase of the second symmetric signal is set to 0 in step S23.

[0027] Further, the calculation formula of the amplitude-frequency response correction result in step S25 is: Wherein represents the amplitude-frequency response correction result, represents an ideal amplifier, represents a self-set filter, that is, the second to-be-processed power spectrum, represents the amplitude-frequency response of the reverse filter, represents a frequency.

[0028] The present application has the following advantages: (1) The present application effectively reduces the influence of data inconsistency caused by the performance difference of the electroencephalogram acquisition system, and provides a new technology for standardizing the data collected by different electroencephalogram acquisition systems in the cross-center standardization process.

[0029] (2) The present application ensures that the data before and after spectrum subtraction generally has strong correlation or moderate correlation, and the two have consistency in trend, and spectrum subtraction is used to weaken the direct current drift and amplifier noise that may exist in the original electroencephalogram signal.

[0030] (3) The amplitude-frequency response correction algorithm in the present invention can improve the consistency of EEG signals collected by different EEG acquisition systems and restore the data peak value. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The figure shows a flow chart of a data standardization method based on different EEG acquisition systems provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the accompanying drawings are merely exemplary and are intended to illustrate the principles and spirit of the present invention, rather than to limit the scope of the present invention.

[0033] The embodiment of the present invention provides a data standardization method based on different EEG acquisition systems, such as Figure 1 As shown, the following steps S1~S2 are included: S1. Electroencephalogram (EEG) signals collected by different EEG acquisition systems are denoised using spectral subtraction to obtain denoised signals.

[0034] During long-term use, the noise characteristics of the amplifier will gradually deteriorate. The mechanism causing this phenomenon is similar to that of white noise and 1 / f The formation mechanism of noise is related to this. At the white noise level, the main reason for the increase of white noise is the increase of internal resistance caused by material aging and the drying of capacitor electrolyte or loss of dielectric. f Noise level, causing 1 / f The reason for the increase in noise is that the density of trap states increases due to the aging of the oxide layer, and the capture-release process of carriers becomes more frequent, which is macroscopically manifested as 1 / f Increased noise. Furthermore, the accumulation of stress from environmental factors (such as temperature and humidity) can lead to increased device noise. Therefore, EEG signals collected by different EEG acquisition systems require noise correction.

[0035] Step S1 includes the following sub-steps S11 to S18: S11. Segment the noise signal data and obtain the noise power spectrum through Fast Fourier Transform (FFT) and averaging.

[0036] In view of the fact that the noise of some electroencephalogram acquisition systems does not have consistency or similarity, the embodiment of the present application formulates the following scheme for the noise signal data in step S11: for the same model or same brand of electroencephalogram acquisition system, a common noise template is generated for the electroencephalogram acquisition system with relatively consistent noise level, and a separate noise template is generated for the electroencephalogram acquisition system with relatively significant difference in noise level; for the electroencephalogram acquisition systems of the same model with inconsistent or similar noise level, a template is separately generated for each electroencephalogram acquisition system.

[0037] S12, the EEG signals collected by different electroencephalogram acquisition systems are segmented by a sliding window to obtain a first sliding window segmentation result.

[0038] S13, the first sliding window segmentation result is connected to the tail of the EEG signal after mirror processing to construct a first symmetric signal.

[0039] In the embodiment of the present application, the phase of the first symmetric signal is set to 0 to avoid artifacts caused by different starting point and terminal level of the signal.

[0040] S14, the first symmetric signal is mapped to a frequency domain signal by fast Fourier transform to obtain a first to-be-processed power spectrum.

[0041] S15, the spectral subtraction result is calculated according to the first to-be-processed power spectrum and the noise power spectrum.

[0042] The EEG signal collected by the electroencephalogram acquisition system The real response EEG signal modeled as a deterministic component And a random noise component The model can be expressed in mathematical form as follows: Wherein represents time.

[0043] Considering the independence of the two components of the model, the power spectrum expression of the signal can be obtained as follows: Wherein represents the power spectrum of the real response EEG signal, i.e. the spectral subtraction result, represents the power spectrum of the EEG signal collected by the electroencephalogram acquisition system, i.e. the first to-be-processed power spectrum, represents the noise power spectrum, represents frequency.

[0044] Therefore, the calculation formula of the spectral subtraction result in the embodiment of the present application can be expressed as: S16, calculate the phase spectrum of the true response EEG signal according to the spectral subtraction result.

[0045] In order to completely estimate the time domain of the deterministic signal by inverse transform in the frequency domain, the phase part must be recovered, and the calculation formula of the phase spectrum of the EEG signal is: Wherein represents the phase spectrum of the true response EEG signal, represents the imaginary unit, represents the phase angle, represents the phase spectrum of the EEG signal collected by the brain electrical acquisition system.

[0046] S17, according to the spectral subtraction result and the phase spectrum of the true response EEG signal, the first time domain recovery signal is obtained by inverse fast Fourier transform (IFFT).

[0047] S18, the first time domain recovery signal is subjected to average deoverlap processing to obtain a denoising signal.

[0048] In the embodiment of the present application, the spectral subtraction result and the phase spectrum of the true response EEG signal are subjected to IFFT, and the real part is taken to obtain , that is, the denoising signal.

[0049] Through multiple experimental verification, according to the requirement of IEC 80601-2-26:2019 / Amd 1:2024 and GB9706.226 that the total input noise of the brain electrical acquisition system is less than or equal to 6.0 μVpp, the maximum input noise peak level that can be processed by the spectral subtraction method is 9.85 μVpp.

[0050] S2, the amplitude frequency response correction is performed on the denoising signal to obtain a data standardization result.

[0051] For periodic or obviously frequency characteristic signals, amplitude frequency response standardization helps to extract key frequency information, thereby improving the recognition, classification and matching accuracy of the signal. Amplitude frequency response correction has a wide range of applications in the field of communication (such as suppressing passband ripple) and the field of audio signal processing (such as speech analysis and enhancement). In the field of brain electrical research, some scholars believe that the use of high-quality brain electrical acquisition systems will not significantly affect the oscillation amplitude of the signal, and some research based on sweep frequency technology has evaluated the ripple and oscillation of the brain electrical acquisition system in the passband, and concluded that "the response difference between devices is not easy to be detected". In addition, the gain stability of the amplifier will drift with the use time. Therefore, for multi-center research using different brain electrical acquisition systems, amplitude frequency response correction is necessary.

[0052] Step S2 includes the following sub-steps S21 to S27: S21. Clean the amplitude-frequency response data and construct an inverse filter to obtain the amplitude-frequency response of the inverse filter.

[0053] In the embodiment of the present invention, the modified Akima piecewise cubic Hermite interpolation is used to clean the amplitude-frequency response data.

[0054] S22. Perform sliding window segmentation on the denoised signal to obtain a second sliding window segmentation result.

[0055] S23. Mirror the second sliding window segmentation result and connect it to the tail of the denoised signal to construct a second symmetrical signal.

[0056] In the embodiment of the present invention, the phase of the second symmetrical signal is set to 0 to avoid artifacts caused by different levels between the starting point and the end point of the signal.

[0057] S24. Map the second symmetric signal into a frequency domain signal through fast Fourier transform to obtain a second power spectrum and phase spectrum to be processed.

[0058] S25 , performing amplitude-frequency response correction according to the amplitude-frequency response of the inverse filter and the second power spectrum to be processed to obtain an amplitude-frequency response correction result.

[0059] For the standardization process of the amplitude-frequency response of the EEG acquisition system, the amplitude-frequency response of the non-ideal amplifier can be regarded as the ideal amplifier. After passing through a custom filter The amplitude-frequency response after . In this case, an inverse filter can be used to Restore to the ideal state. Among them, the frequency domain response of the inverse filter is for: That is, for At any point on the domain , the frequency domain response of its inverse filter at this point is: The calculation formula for the amplitude-frequency response correction result is: in It represents the amplitude-frequency response correction result. represents an ideal amplifier, Indicates a custom filter. That is the second power spectrum to be processed, represents the amplitude-frequency response of the inverse filter, representing the frequency.

[0060] In the embodiments of the present application, for a specific electroencephalogram acquisition system, equivalent to the amplitude-frequency response of the EEG signal acquired by the electroencephalogram acquisition system. The amplitude-frequency response of the electroencephalogram acquisition system to each frequency point of the standard signal source can be obtained by statistical measurement. However, the specific measurement process cannot realize the statistics of all frequency points, so under the premise of the statistics of the amplitude-frequency response of a specific frequency interval, the amplitude-frequency response of the system to all frequency points is estimated by interpolation of discrete frequency points.

[0061] S26, according to the amplitude-frequency response correction result and the phase spectrum, a second time-domain recovery signal is obtained by inverse fast Fourier transform.

[0062] S27, the second time-domain recovery signal is subjected to average de-overlapping processing to obtain a data standardization result.

[0063] In the embodiments of the present application, the average de-overlapping processing is an operation corresponding to the previous sliding window segmentation, which is used to splice the segmented signals into long signals. At the same time, the average value in the overlapping area of adjacent segmented signals is taken to avoid the edge discontinuity in splicing.

[0064] In the embodiments of the present application, the amplitude-frequency response correction algorithm has certain application conditions: when the average gain level of the amplitude-frequency response curve of the current electroencephalogram acquisition system is greater than or equal to 0.47, the algorithm is applicable, otherwise the algorithm is invalid.

[0065] Those skilled in the art will appreciate that the embodiments described herein are intended to help the reader understand the principles of the present application and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspirations disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the scope of protection of the present application.

Claims

1. A data standardization method based on different EEG acquisition systems, characterized in that: The following steps are involved: S1. Using spectral subtraction to denoise the EEG signals collected by different EEG acquisition systems to obtain denoised signals; S2. Perform amplitude-frequency response correction on the denoised signal to obtain a data normalization result.

2. The data standardization method based on different EEG acquisition systems according to claim 1, characterized in that: The step S1 includes the following sub-steps: S11, segmenting the noise signal data and obtaining the noise power spectrum through fast Fourier transform and averaging; S12, performing sliding window segmentation on the EEG signals collected by different EEG acquisition systems to obtain a first sliding window segmentation result; S13, mirroring the first sliding window segmentation result and connecting it to the tail of the EEG signal to construct a first symmetrical signal; S14, mapping the first symmetric signal into a frequency domain signal through fast Fourier transform to obtain a first power spectrum to be processed; S15, calculating a spectrum subtraction result according to the first to-be-processed power spectrum and the noise power spectrum; S16. Calculating the phase spectrum of the true response EEG signal according to the spectrum subtraction result; S17, obtaining a first time-domain restored signal by inverse fast Fourier transform according to the spectrum subtraction result and the phase spectrum of the true response EEG signal; S18. Perform an average de-overlapping process on the first time-domain restored signal to obtain a denoised signal.

3. The data standardization method based on different EEG acquisition systems according to claim 2, characterized in that: In step S13, the phase of the first symmetrical signal is set to 0.

4. The data standardization method based on different EEG acquisition systems according to claim 2, characterized in that: The calculation formula of the spectrum subtraction result in step S15 is: in Represents the true response EEG signal power spectrum, that is, the spectrum subtraction result, It represents the power spectrum of the EEG signal collected by the EEG acquisition system, that is, the first power spectrum to be processed. represents the noise power spectrum, Indicates frequency.

5. The data standardization method based on different EEG acquisition systems according to claim 4, characterized in that: The calculation formula of the phase spectrum of the EEG signal in step S16 is: in represents the phase spectrum of the real response EEG signal, represents the imaginary unit, Indicates the phase angle, Represents the phase spectrum of the EEG signal collected by the EEG acquisition system.

6. The data standardization method based on different EEG acquisition systems according to claim 1, characterized in that: The step S2 comprises the following sub-steps: S21. Cleaning the amplitude-frequency response data and constructing an inverse filter to obtain an amplitude-frequency response of the inverse filter; S22, performing sliding window segmentation on the denoised signal to obtain a second sliding window segmentation result; S23, mirroring the second sliding window segmentation result and connecting it to the tail of the denoised signal to construct a second symmetrical signal; S24, mapping the second symmetric signal into a frequency domain signal through fast Fourier transform to obtain a second power spectrum and phase spectrum to be processed; S25, performing amplitude-frequency response correction according to the amplitude-frequency response of the inverse filter and the second power spectrum to be processed to obtain an amplitude-frequency response correction result; S26. Obtain a second time-domain restored signal by inverse fast Fourier transform according to the amplitude-frequency response correction result and the phase spectrum; S27. Perform average de-overlapping processing on the second time domain restored signal to obtain a data normalization result.

7. The data standardization method based on different EEG acquisition systems according to claim 6, characterized in that: In step S23, the phase of the second symmetrical signal is set to 0.

8. The data standardization method based on different EEG acquisition systems according to claim 6, characterized in that: The calculation formula of the amplitude-frequency response correction result in step S25 is: in Indicates the amplitude-frequency response correction result, represents an ideal amplifier, Indicates a custom filter. That is the second power spectrum to be processed, represents the amplitude-frequency response of the inverse filter, Indicates frequency.

Images