Method for detecting abrupt change of electroencephalogram high frequency oscillation energy based on single variable instantaneous energy entropy

Abstract

The present application belongs to the technical field of signal processing, and particularly relates to a brain electrical high-frequency oscillation energy mutation detection method based on single-variable instantaneous energy entropy, aiming to improve the detection accuracy of epilepsy diagnosis. The method comprises the following steps: preprocessing electroencephalogram (EEG) data, dividing the preprocessed EEG data into sliding window segments based on the preprocessed EEG data, calculating wavelet coefficients of each sliding window segment to extract time-varying energy and Hjorth complexity. The wavelet coefficient refers to a coefficient obtained by wavelet transform to represent the time-frequency characteristics of a signal; the time-varying energy refers to a signal energy characteristic varying with time, and the Hjorth complexity is used to describe the waveform complexity of a signal. The time-varying energy and the Hjorth complexity are fused to obtain a fusion feature. The single-variable instantaneous energy entropy is calculated according to the fusion feature, and the epilepsy onset area is detected according to the mutation of the single-variable instantaneous energy entropy. The single-variable instantaneous energy entropy refers to an instantaneous entropy value calculated based on the energy distribution of a single-variable signal, and is used to represent the energy mutation of a signal.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal processing, in particular to a brain electrical high-frequency oscillation energy mutation detection method based on single variable instantaneous energy entropy. BACKGROUND

[0002] As a nonlinear multivariate complex signal recording brain electrical activity, the analysis of electroencephalogram (EEG) has important significance for epilepsy diagnosis. Entropy, as an index for quantifying the complexity and order mutation of neural electrical activity, is widely used to analyze the dynamic characteristics of abnormal discharge of epilepsy due to its strong anti-noise ability.

[0003] However, the traditional entropy-based complexity analysis method relies on the global stationarity assumption and has insufficient time resolution, making it difficult to analyze the transient nonlinear dynamic characteristics and millisecond-level dynamic changes in the process of epilepsy discharge. For example, the instantaneous mutation characteristics of epilepsy discharge are easily masked by background noise, and the traditional analysis method relies on the steady-state assumption of the macro time window, which cannot capture the instantaneous mutation characteristics of epilepsy discharge, resulting in limited detection accuracy of high-frequency oscillation energy mutation in the epilepsy initiation zone. SUMMARY

[0004] The purpose of the present application is to provide a brain electrical high-frequency oscillation energy mutation detection method based on single variable instantaneous energy entropy, aiming to improve the detection accuracy of epilepsy diagnosis.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is as follows: the present application provides a brain electrical high-frequency oscillation energy mutation detection method based on single variable instantaneous energy entropy, comprising: S1: preprocessing the electroencephalogram (EEG) data to obtain preprocessed EEG data; S2: based on the preprocessed EEG data, performing sliding window division to obtain sliding window segments, calculating the wavelet coefficients of each sliding window segment to extract time-varying energy and Hjorth complexity; wherein the sliding window segment refers to a time segment used to intercept continuous EEG data; the wavelet coefficient refers to a coefficient obtained by wavelet transform to represent the time-frequency characteristics of the signal; the time-varying energy refers to the energy characteristics of the signal varying with time, and the Hjorth complexity is used to describe the waveform complexity of the signal; S3: fusing the time-varying energy and Hjorth complexity to obtain fused features; S4: calculating the single variable instantaneous energy entropy according to the fused features, and detecting the epilepsy initiation zone according to the mutation of the single variable instantaneous energy entropy; wherein the single variable instantaneous energy entropy refers to the instantaneous entropy value calculated based on the energy distribution of the single variable signal, and is used to represent the energy mutation of the signal.

[0006] The preprocessing includes data cleaning, band-pass filtering, data standardization, channel selection and integration, and mirror extension, wherein the data cleaning refers to removing noise and artifacts in electroencephalogram (EEG) data, the band-pass filtering refers to retaining signal components in a specific frequency range, the data standardization refers to adjusting data to a uniform magnitude, the channel selection and integration refer to selecting relevant recording channels and integrating data, and the mirror extension is used to reduce boundary effects by symmetrically expanding data.

[0007] In step S2, the length of the sliding window division is determined based on an input high frequency oscillation band and a sampling frequency, and is calculated according to the following formula: ; wherein, is a floor function, is a sampling frequency, is the minimum frequency extracted in the high frequency oscillation band, the high frequency oscillation band refers to a high frequency signal frequency range related to the initiation of epilepsy, and the sampling frequency refers to the number of samples of electroencephalogram data collected per unit time.

[0008] In step S2, the frequency range is determined based on the high frequency oscillation band, and the wavelet parameters are determined in combination with the sampling frequency; the wavelet parameters include a wavelet name and a scale array; the wavelet coefficients are obtained by performing wavelet transform on the sliding window segment according to the wavelet name, the scale array, and the sampling period corresponding to the sampling frequency; wherein the wavelet name refers to a wavelet function identifier used for wavelet transform, the scale array refers to a transform scale sequence corresponding to the frequency range, and the sampling period refers to the time interval between adjacent two sampling points.

[0009] Extracting the time-varying energy and Hjorth complexity includes generating an energy map from the wavelet coefficients and integrating the energy map along the frequency axis to obtain the time-varying energy; wherein the energy map refers to a matrix representing the energy of signals at different frequencies and time points.

[0010] Extracting the time-varying energy and Hjorth complexity also includes obtaining the Hjorth complexity based on the second-order difference of the sliding window segment; wherein the second-order difference refers to the second difference between adjacent three data points of the signal.

[0011] In step S3, the fusion features are obtained by integrating the standardized time-varying energy and Hjorth complexity according to a preset weight; the standardized processing of the time-varying energy is performed according to the following formula: ; the standardized processing of the Hjorth complexity is performed according to the following formula: ; wherein, , are the mean and standard deviation of the time-varying energy, respectively, , These are the mean and standard deviation of the Hjorth complexity, respectively; standardization refers to adjusting the features to the same distribution range, and preset weights refer to the pre-set proportional coefficients used to integrate the two features.

[0012] The preset weights are an equal-weighted combination of time-varying energy and Hjorth complexity, which are then used to fuse features. .

[0013] In step S4, the fused feature is divided into multiple subsequences, the energy distribution ratio of data points in each subsequence is calculated, and the instantaneous energy entropy of a single variable is calculated based on the definition of information entropy; where subsequence refers to non-overlapping segments of the fused feature, and energy distribution ratio refers to the proportion of the energy of each data point in the subsequence to the total energy of the subsequence.

[0014] When a significant mutation occurs in the instantaneous energy entropy of a single variable, the EEG data segment corresponding to the mutation is determined to be a high-frequency oscillatory energy mutation region related to the epilepsy initiation zone.

[0015] Compared with the prior art, the beneficial effects of this application are as follows:

[0016] 1. The present application provides a method for detecting high-frequency oscillation energy mutations in EEG based on univariate instantaneous energy entropy. By reducing noise and artifact interference through preprocessing, it provides high-quality data for subsequent analysis. The sliding window length is determined based on the high-frequency oscillation band and sampling frequency, which can ensure that the window covers key high-frequency cycles and provide a suitable time scale for capturing instantaneous mutations. This solves the problem of poor adaptability of traditional fixed windows to dynamic signals.

[0017] 2. Wavelet parameters (name, scale array) are determined by the high-frequency oscillation band, and wavelet coefficients are calculated. Time-varying energy is obtained by integrating this with the energy map. Simultaneously, Hjorth complexity is extracted based on second-order difference, achieving a dual characterization of signal energy changes and waveform complexity. This overcomes the information limitations of traditional methods that rely solely on a single feature, and more comprehensively reflects the high-frequency oscillation characteristics of the epilepsy initiation zone. The time-varying energy and Hjorth complexity are standardized (based on mean and standard deviation) and fused according to preset weights, eliminating differences in feature magnitudes, strengthening the synergistic effect of key information, and solving the problem of weak anti-interference ability of single features. This makes the fused features more prominent in highlighting the abrupt change characteristics of abnormal signals.

[0018] 3. By calculating the instantaneous energy entropy of a single variable through the energy distribution of subsequences and combining it with mutation judgment rules, it can keenly capture the instantaneous changes in energy distribution. This overcomes the feature ambiguity caused by the smoothing of time windows in traditional entropy methods, and achieves precise localization of millisecond-level mutations in the high-frequency oscillation energy of the epilepsy initiation zone, greatly improving the accuracy and reliability of detection. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for detecting high-frequency oscillation energy mutations in the epilepsy initiation zone based on univariate instantaneous energy entropy, provided in an embodiment of this application.

[0020] Figure 2 This is a simulation experiment diagram provided in an embodiment of this application to test whether a single variable instantaneous energy entropy can detect a millisecond-level HFO signal;

[0021] Figure 3 This is a schematic diagram illustrating the noise immunity of a measurement method provided in an embodiment of this application;

[0022] Figure 4 This is a schematic diagram illustrating the advantages of a sliding window as provided in an embodiment of this application;

[0023] Figure 5 This is a schematic diagram comparing the original signal with the instantaneous energy entropy of a single variable, provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0026] In embodiments of the invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, article, or apparatus that includes that element.

[0027] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0028] This application provides a method for detecting high-frequency oscillation energy mutations in electroencephalography (EEG) based on univariate instantaneous energy entropy. For example,... Figure 1 As shown. The method includes:

[0029] S1: Preprocess the EEG data to obtain preprocessed EEG data.

[0030] As one possible implementation, preprocessing includes: data cleaning, bandpass filtering, data standardization, channel selection and integration, and mirror extension. Among these, data cleaning refers to removing noise and artifacts from EEG data; bandpass filtering refers to retaining signal components within a specific frequency range; data standardization refers to adjusting the data to a uniform magnitude; channel selection and integration refers to selecting relevant recording channels and integrating the data; and mirror extension is used to reduce boundary effects by symmetrically expanding the data.

[0031] S2: Based on the preprocessed EEG data, a sliding window is divided to obtain sliding window segments, and the wavelet coefficients of each sliding window segment are calculated to extract time-varying energy and Hjorth complexity.

[0032] Among them, the sliding window segment refers to the time segment used to capture continuous EEG data; wavelet coefficients refer to the coefficients that characterize the time-frequency features of the signal obtained through wavelet transform; time-varying energy refers to the signal energy characteristics that change with time; and Hjorth complexity is used to describe the characteristics of signal waveform complexity.

[0033] In step S2, the length of the sliding window is determined based on the input high-frequency oscillation band and the sampling frequency, and is calculated according to the following formula: ;in, This refers to rounding down. Sampling frequency, The minimum frequency extracted from the high-frequency oscillation band refers to the range of high-frequency signals associated with the onset of epilepsy, and the sampling frequency refers to the number of EEG data samples collected per unit time.

[0034] Extract the minimum frequency in the high-frequency oscillation band. and maximum frequency , Calculating the sliding window segment ensures that the sliding window covers at least two of the highest frequency oscillation cycles. Covering two of the highest frequency oscillation cycles means that the basic waveform of the high-frequency signal (such as rising edge, falling edge, peak value, and other key features) can be fully captured, avoiding the truncation or loss of high-frequency oscillation features due to an excessively short window. This ensures that the sliding window contains sufficient high-frequency oscillation feature information, providing a reliable data foundation for subsequent feature extraction and mutation detection.

[0035] In step S2, the frequency range is determined based on the high-frequency oscillation band, and the wavelet parameters are determined in combination with the sampling frequency. The wavelet parameters include the wavelet name and the scale array. According to the wavelet name, the scale array, and the sampling period corresponding to the sampling frequency, the sliding window segment is subjected to wavelet transform to obtain the wavelet coefficients. Among them, the wavelet name refers to the wavelet function identifier used for wavelet transform, the scale array refers to the transform scale sequence corresponding to the frequency range, and the sampling period refers to the time interval between two adjacent sampling points.

[0036] For example, wavelet parameters are determined based on the high-frequency oscillation band, specifically by first determining the minimum frequency. and maximum frequency and combined with sampling frequency According to the formula Obtain the normalized center frequency. Determine the wavelet name (wavelet_name) based on the normalized center frequency. At the minimum frequency... and maximum frequency Generate frequency values ​​at n = 50 logarithmic intervals. Then through:

[0037]

[0038] frequency Convert to scale Finally, the input signal is scaled using an array of parameters.

[0039] Wavelet name (wavelet_name) and sampling period Perform a continuous wavelet transform to obtain the wavelet coefficients W.

[0040] In some embodiments, extracting time-varying energy and Hjorth complexity includes: generating an energy map using wavelet coefficients, and integrating the energy map along the frequency axis to obtain the time-varying energy. Here, the energy map refers to a matrix representing the signal energy at different frequencies and time points.

[0041] For example, based on the result of continuous wavelet transform, a wavelet transform is performed on the input signal X to obtain the wavelet coefficient matrix W, the dimension of which is... ;in, For the number of frequency scales, This represents the number of time points. Then, the energy map is calculated. , elements As shown below:

[0042] ;

[0043] The energy diagram is processed according to the following formula. The time-varying energy is obtained by averaging along the frequency axis, that is, by taking the average of each column of elements.

[0044]

[0045] Extracting time-varying energy and Hjorth complexity also includes obtaining Hjorth complexity based on second-order differences using a sliding window segment. Here, second-order difference refers to the quadratic difference between three adjacent data points of the signal.

[0046] For example, calculate the second-order difference for the input signal X. ; and then according to

[0047] Calculate the complexity of Hjorth .

[0048] S3: The fusion feature is obtained by fusing time-varying energy and Hjorth complexity.

[0049] In step S3, the time-varying energy and Hjorth complexity are standardized and then integrated according to preset weights to obtain the fused features; the time-varying energy is standardized according to the following formula: The Hjorth complexity is standardized according to the following formula: ;in, , These are the mean and standard deviation of the time-varying energy, respectively. , These are the mean and standard deviation of the Hjorth complexity, respectively; standardization refers to adjusting the features to the same distribution range, and preset weights refer to the pre-set proportional coefficients used to integrate the two features.

[0050] It should be understood that when performing standardization and calculating the mean and standard deviation of time-varying energy and Hjorth complexity, the nan value needs to be ignored to ensure that only valid data is used in the calculation and to avoid standardization failure due to missing values.

[0051] The preset weights are an equal-weighted combination of time-varying energy and Hjorth complexity, which are then used to fuse features. .

[0052] The standardized ZEZE and ZCZC are linearly combined with equal weights (50% each) to obtain the fused features. The equal-weight fusion calculation is more efficient and suitable for large-scale real-time processing. In practical applications, the weight ratio can be adjusted according to the needs.

[0053] S4: Calculate the univariate instantaneous energy entropy based on the fusion features, and detect the epilepsy initiation zone based on the mutation of the univariate instantaneous energy entropy; where, the univariate instantaneous energy entropy refers to the instantaneous entropy value calculated based on the energy distribution of the univariate signal, which is used to characterize the mutation of signal energy.

[0054] In step S4, the fused feature is divided into multiple subsequences, and the energy distribution ratio of data points in each subsequence is calculated. The univariate instantaneous energy entropy is then calculated based on the definition of information entropy. Here, a subsequence refers to a non-overlapping segment of the fused feature, and the energy distribution ratio refers to the proportion of the energy of each data point in a subsequence to the total energy of the subsequence. For example, based on the definition of Shannon entropy, the univariate instantaneous energy entropy is calculated as follows:

[0055] Divide the discrete-time series into several non-overlapping subsequences of length K. For each subsequence, calculate its total energy. And determine the proportion of energy at each data point to the total energy. Then, based on the definition of information entropy

[0056] Calculate the instantaneous energy entropy of a single variable.

[0057] When a significant mutation occurs in the instantaneous energy entropy of a single variable, the EEG data segment corresponding to the mutation is determined to be a high-frequency oscillatory energy mutation region related to the epilepsy initiation zone.

[0058] To evaluate the detection capability of the univariate instantaneous energy entropy proposed in this invention for high-frequency oscillation energy mutations in the epileptic initiation region, embodiments of this application also provide detection experiments, exemplarily, such as... Figure 2 As shown. This application embodiment generates an EEG sequence with a single high-frequency oscillation signal, according to... Figure 2 It can be seen that at the high-frequency oscillation signal, that is... Figure 2 The region with the largest peak in (a) corresponds to Figure 2 (b) The univariate instantaneous energy entropy also increased significantly, thus high-frequency oscillations in the EEG signal could be clearly detected.

[0059] For example, such as Figure 3As shown. To further analyze the noise resistance of the method provided in this application embodiment, this application embodiment also provides a noise resistance detection experiment. The experiment generated 10s simulated signals of four typical high-frequency oscillation events (e.g., 20ms steep edge, 150ms traditional high-frequency oscillation, etc.), with a base noise of 1 / f noise superimposed with electromyography artifacts, and set nine signal-to-noise ratio (SNR) levels from 40dB to 0dB. For each SNR, Gaussian white noise was first added to the original signal, and then electromyography artifacts were superimposed to simulate real physiological interference.

[0060] Three algorithms are used to process the noisy signal: the IE algorithm, which calculates intrinsic entropy based on EMD decomposition; the IF algorithm, which extracts instantaneous frequency using Hilbert transform; and the univariate instantaneous energy entropy algorithm proposed in this embodiment. Figure 3 The UEE algorithm in [the text is incomplete and requires further context]. By calculating the F1 score and AUC value, the differences in performance between noise-free and noisy conditions are compared, and the p-value is obtained through a paired t-test. [The remaining text appears to be a separate, unrelated sentence fragment.] Figure 3 The F1 comparison curve and AUC comparison curve shown are based on Figure 3 It can be seen that the F1 score and AUC value of the single-variable instantaneous energy entropy proposed in this invention are higher than those of other algorithms at all nine signal-to-noise ratio (SNR) levels.

[0061] To evaluate the performance difference of the proposed method for detecting high-frequency oscillating HFO signals under dynamic and fixed window conditions, experimental simulations were also performed in the embodiments of this application. For example, as shown... Figure 4 As shown, the experiment generated a simulated EEG signal containing a 10Hz fundamental frequency component, a 250Hz high-frequency oscillation occurring at 0.5 seconds (lasting 20ms), and Gaussian white noise (standard deviation 0.1). UIEE features were calculated using two windowing strategies. The dynamic window algorithm automatically adjusted the window length based on the signal frequency range (ensuring coverage of at least two of the highest frequency cycles), while the fixed window algorithm used a preset sample length of 200.

[0062] The experiment visualized the original signal, dynamic window UIEE results, and fixed window UIEE results, for reference. Figure 4 . Figure 4 (a) In addition to the simulated high-frequency oscillation original signal, burst is a typical feature of high-frequency oscillation signals, characterizing epilepsy-related high-frequency activity. Periodic oscillations can be seen in the waveform, and burst segments with sudden changes in energy / amplitude appear at specific locations (highlighted in dark in the figure), which are the "target signals" for subsequent analysis. Figure 4 (b) and Figure 4 (c) shows the dynamic window algorithm and the fixed window algorithm, respectively, used to verify the algorithm's ability to detect abnormal high-frequency events. Figure 4In curve (b), the burst segment of the high-frequency oscillation signal corresponds to a significant peak, clearly highlighting the abnormal energy change, proving that the dynamic window algorithm can keenly capture instantaneous changes, preserve high-frequency details, and is suitable for detecting time-varying high-frequency oscillation characteristics. Figure 4 In (c), the fixed window length is not adjusted according to the signal characteristics, which will result in the loss of instantaneous details due to the "smoothing effect": the peak of the high-frequency oscillation burst is weakened and delayed, and the interference of background noise / normal oscillation is not effectively distinguished (the curve is generally flat and abrupt changes are blurred). Therefore, the UIEE results of the dynamic window are more sensitive to high-frequency oscillation signals, and the dynamic window used in this algorithm is more suitable for physiological signal characteristics.

[0063] This application embodiment uses the publicly available CHB-MIT dataset to obtain long-term scalp electroencephalogram (EEG) recordings of six pediatric patients diagnosed with refractory epilepsy. Univariate instantaneous energy entropy features were calculated using the raw signals and the method proposed in this invention. Support Vector Machine (SVM) was then used to classify the preictal and interictal phases within the subjects. For example,... Figure 5 As shown. According to Figure 5 It can be seen that the univariate instantaneous energy entropy characteristics calculated by the method proposed in this invention have higher accuracy than the original signal.

[0064] This application provides a method for detecting the epilepsy initiation zone based on univariate instantaneous energy entropy. By reducing noise and artifact interference through preprocessing, it provides high-quality data for subsequent analysis. The sliding window length is determined based on the high-frequency oscillation band and sampling frequency, which ensures that the window covers key high-frequency cycles and provides a suitable time scale for capturing instantaneous changes. This solves the problem of poor adaptability of traditional fixed windows to dynamic signals.

[0065] Wavelet parameters (name, scale array) are determined by high-frequency oscillation bands, and wavelet coefficients are calculated. Time-varying energy is obtained by integrating this with energy maps. Simultaneously, Hjorth complexity is extracted based on second-order difference, achieving a dual characterization of signal energy changes and waveform complexity. This overcomes the information limitations of traditional methods that rely solely on single features, providing a more comprehensive reflection of the high-frequency oscillation characteristics of the epilepsy initiation zone. Standardization of the time-varying energy and Hjorth complexity (based on mean and standard deviation) and fusion according to preset weights eliminates differences in feature magnitudes, strengthens the synergistic effect of key information, and solves the problem of weak anti-interference ability of single features, making the fused features more prominent in highlighting the abrupt change characteristics of abnormal signals.

[0066] By calculating the instantaneous energy entropy of a single variable through the energy distribution of subsequences and combining it with mutation judgment rules, it can keenly capture the instantaneous changes in energy distribution. This overcomes the feature ambiguity caused by the smoothing of time windows in traditional entropy methods, and achieves precise localization of millisecond-level mutations in the high-frequency oscillation energy of the epilepsy initiation zone, greatly improving the accuracy and reliability of detection.

[0067] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0068] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting high-frequency oscillation energy mutations in electroencephalography based on univariate instantaneous energy entropy, characterized in that, include: S1: Preprocess the EEG data to obtain preprocessed EEG data; S2: Based on the preprocessed EEG data, a sliding window is divided to obtain sliding window segments. The wavelet coefficients of each sliding window segment are calculated to extract time-varying energy and Hjorth complexity. Here, the sliding window segment refers to the time segment used to extract continuous EEG data; the wavelet coefficients refer to the coefficients that characterize the time-frequency features of the signal obtained through wavelet transform; the time-varying energy refers to the signal energy characteristics that change with time; and the Hjorth complexity is used to describe the characteristics of signal waveform complexity. S3: The fusion feature is obtained by fusing time-varying energy and Hjorth complexity; S4: Calculate the univariate instantaneous energy entropy based on the fusion features, and detect the epilepsy initiation zone based on the mutation of the univariate instantaneous energy entropy; where, the univariate instantaneous energy entropy refers to the instantaneous entropy value calculated based on the univariate signal energy distribution, which is used to characterize the signal energy mutation; In step S2, the frequency range is determined based on the high-frequency oscillation band, and the wavelet parameters are determined in combination with the sampling frequency. The wavelet parameters include the wavelet name and the scale array. According to the wavelet name, the scale array, and the sampling period corresponding to the sampling frequency, the sliding window segment is subjected to wavelet transform to obtain wavelet coefficients. Here, the wavelet name refers to the wavelet function identifier used for wavelet transform, the scale array refers to the transform scale sequence corresponding to the frequency range, and the sampling period refers to the time interval between two adjacent sampling points. Extracting time-varying energy and Hjorth complexity includes: generating an energy map using wavelet coefficients, and integrating the energy map along the frequency axis to obtain the time-varying energy; where the energy map refers to a matrix representing the energy of signals at different frequencies and time points; In step S4, the fused feature is divided into multiple subsequences, the energy distribution ratio of data points in each subsequence is calculated, and the instantaneous energy entropy of a single variable is calculated based on the definition of information entropy; where subsequence refers to non-overlapping segments of the fused feature, and energy distribution ratio refers to the proportion of the energy of each data point in the subsequence to the total energy of the subsequence.

2. The method for detecting high-frequency oscillation energy mutations in electroencephalograms based on univariate instantaneous energy entropy according to claim 1, characterized in that, The preprocessing includes: data cleaning, bandpass filtering, data standardization, channel selection and integration, and mirror extension. Among them, data cleaning refers to removing noise and artifacts from EEG data; bandpass filtering refers to retaining signal components within a specific frequency range; data standardization refers to adjusting the data to a uniform magnitude; channel selection and integration refers to selecting relevant recording channels and integrating the data; and mirror extension is used to reduce boundary effects by symmetrically expanding the data.

3. The method for detecting high-frequency oscillation energy mutations in electroencephalography based on univariate instantaneous energy entropy according to claim 1, characterized in that, In step S2, the length of the sliding window is determined based on the input high-frequency oscillation band and the sampling frequency, and is calculated according to the following formula: in, This refers to rounding down. Sampling frequency, The minimum frequency extracted from the high-frequency oscillation band refers to the range of high-frequency signals associated with the onset of epilepsy, and the sampling frequency refers to the number of EEG data samples collected per unit time.

4. The method for detecting high-frequency oscillation energy mutations in electroencephalograms based on univariate instantaneous energy entropy according to claim 1, characterized in that, Extracting time-varying energy and Hjorth complexity also includes obtaining Hjorth complexity based on second-order difference of sliding window segment; where second-order difference refers to the quadratic difference between three adjacent data points of the signal.

5. The method for detecting high-frequency oscillation energy mutations in electroencephalograms based on univariate instantaneous energy entropy according to claim 3, characterized in that, In step S3, the time-varying energy and Hjorth complexity are standardized and then integrated according to preset weights to obtain the fused features; The time-varying energy is standardized according to the following formula: The Hjorth complexity is standardized using the following formula: in, , These are the mean and standard deviation of the time-varying energy, respectively. , These are the mean and standard deviation of the Hjorth complexity, respectively; standardization refers to adjusting the features to the same distribution range, and preset weights refer to the pre-set proportional coefficients used to integrate the two features; Time-varying energy, It has Hjorth complexity.

6. The method for detecting high-frequency oscillation energy mutations in electroencephalograms based on univariate instantaneous energy entropy according to claim 5, characterized in that, The preset weights are an equal-weighted combination of time-varying energy and Hjorth complexity, which are used to fuse features. .

7. The method for detecting high-frequency oscillation energy mutations in electroencephalograms based on univariate instantaneous energy entropy according to claim 1, characterized in that, When a significant mutation occurs in the instantaneous energy entropy of a single variable, the EEG data segment corresponding to the mutation is determined to be a high-frequency oscillatory energy mutation region related to the epilepsy initiation zone.

Images