Data processing method and system for intracranial pressure trend analysis based on electroencephalogram signal spectrum entropy

Abstract

This invention discloses a data processing method and system for intracranial pressure trend analysis based on the spectral entropy of electroencephalogram (EEG) signals, belonging to the field of medical monitoring and biosignal processing technology. The method includes: acquiring EEG signals, setting a sampling frequency of 200Hz–300Hz, a sampling window length of 2s–4s, and a window overlap rate of 50%–60%; using a customized wavelet basis function for denoising, and then removing artifacts through independent component analysis; dividing the signal into five fixed frequency bands and calculating the power of each band; calculating the weighted spectral entropy using the frequency band-weighted modified Shannon entropy formula; calculating the entropy change rate using a variable sliding window combined with a time decay coefficient, and simultaneously calculating the power ratio of β waves to δ waves; inputting the relevant features into a long short-term memory network with residual connections, and outputting a probability distribution of intracranial pressure change trends. This invention improves the repeatability and anti-interference ability of EEG trend-assisted analysis through complex calculations and data processing, and the output results are used for auxiliary analysis without directly providing disease diagnosis conclusions or treatment plans.

Description

Technical Field

[0001] This invention belongs to the field of medical monitoring and biological signal processing technology, specifically relating to a non-invasive intracranial pressure trend analysis data processing method and system based on the spectral entropy of electroencephalogram (EEG) signals. Background Technology

[0002] Elevated intracranial pressure (ICP) is a core pathophysiological process in acute and critical illnesses such as traumatic brain injury, spontaneous intracerebral hemorrhage, cerebral edema, and central nervous system infections. Sustained elevation of ICP leads to decreased cerebral perfusion pressure and impaired cerebral autoregulation, and in severe cases, can induce brain herniation. ICP is an independent risk factor for patient death or severe neurological dysfunction. Therefore, continuous and accurate monitoring of ICP is of irreplaceable value in guiding clinical treatment and improving patient prognosis.

[0003] Currently, clinical intracranial pressure monitoring techniques are divided into two main systems: invasive and non-invasive. Invasive monitoring techniques, represented by intraventricular catheter manometry, are recognized as the gold standard for intracranial pressure monitoring. They have advantages such as high accuracy and the ability to simultaneously drain cerebrospinal fluid to reduce intracranial pressure. However, this type of technique requires drilling a hole in the skull to insert a catheter into the ventricle, which is an invasive procedure and carries risks of complications such as intracranial infection, intracranial hemorrhage, and brain tissue damage. It cannot be used for patients with coagulation disorders, severe agitation, or serious infections, thus greatly limiting its clinical application.

[0004] Given the limitations of invasive monitoring, non-invasive intracranial pressure monitoring technology has become a research hotspot. Existing non-invasive monitoring methods mainly include transcranial Doppler ultrasound, optic nerve sheath diameter ultrasound measurement, tympanic membrane displacement method, and skull impedance method. Transcranial Doppler ultrasound indirectly estimates intracranial pressure through the blood flow velocity in the middle cerebral artery, but it is greatly affected by operator experience, vascular anatomy variations, cerebral vasospasm, and blood pressure fluctuations, resulting in poor accuracy and repeatability, and it cannot achieve continuous dynamic monitoring. Optic nerve sheath diameter measurement can only reflect the trend of increased intracranial pressure, cannot provide quantitative values, and lacks specificity due to eye diseases and body position. The tympanic membrane displacement method and skull impedance method are significantly affected by individual anatomical differences and environmental interference, have narrow clinical applications, and have not yet formed a widely accepted standardized protocol.

[0005] Electroencephalogram (EEG) signals directly reflect the electrical activity of neurons in the brain. Increased intracranial pressure (ICP) leads to insufficient brain perfusion and inhibited neuronal metabolism, resulting in characteristic changes in the spectral distribution of EEG signals, theoretically possessing the potential for non-invasive ICP monitoring. However, current research on ICP monitoring based on EEG signals has not effectively solved the problem of reproducible modeling when using EEG signals for trend-assisted analysis. The fundamental reason is the complex clinical environment for EEG signal acquisition, which is severely affected by power line interference, electromyography (EMG) artifacts, and eye movement (EMG) artifacts. This results in unstable correlation models between extracted spectral features and ICP trends, difficulty in eliminating individual differences, and poor algorithm robustness, making it difficult to achieve reliable continuous trend assessment in complex clinical environments. This core problem severely restricts the clinical application value of EEG signals in the field of non-invasive ICP monitoring.

[0006] Therefore, how to achieve the stability of EEG signals in intracranial pressure trend analysis under conditions of artifact interference and individual differences, and how to establish a stable and quantifiable correlation model, are the core issues that urgently need to be solved in the field of non-invasive intracranial pressure monitoring technology. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to overcome the shortcomings of the prior art and provide a non-invasive intracranial pressure trend analysis data processing method and system based on the spectral entropy of electroencephalogram (EEG) signals.

[0008] To achieve the above objectives, one of the present invention provides the following technical solution: The data processing method based on intracranial pressure trend analysis of electroencephalogram (EEG) signal spectral entropy includes the following steps: Step S101: Acquire EEG signals: sampling frequency is 200Hz~300Hz, sampling window length is 2s~4s, and window overlap rate is 50%~60%; Step S102, Dual Preprocessing: The EEG signal is decomposed into multiple layers using a custom wavelet basis function, and denoising is performed using an adaptive threshold. Then, blind source separation and artifact removal are performed using independent component analysis with physiological constraints. Step S103, Frequency Band Division and Power Calculation: The preprocessed EEG signal is divided into five fixed frequency bands. The corresponding wavebands and frequency ranges for each frequency band are: delta wave 0.5-4Hz, theta wave 4-8Hz, alpha wave 8-13Hz, beta wave 13-30Hz, and gamma wave 30-100Hz. Among them, delta wave and beta wave are the core response wavebands to changes in intracranial pressure, theta wave and alpha wave represent the basic resting brain activity level, and gamma wave captures the characteristics of abnormal high-frequency synchronous discharge of neurons. All five wavebands are involved in the weighted spectral entropy joint calculation, and the power of each frequency band is calculated to form the frequency band power distribution. S104. Global normalization processing: Normalize the power of each frequency band to obtain the probability distribution of the power proportion of each frequency band. S105. Calculate the weighted spectral entropy: Calculate the weighted spectrum H using the band-weighted modified Shannon entropy. The calculation formula is as follows: , in, These are frequency band weighting coefficients. The proportion of power in the i-th frequency band to the total power. The total number of frequency bands is given; the time-series change rate of the weighted spectral entropy is calculated using a variable sliding window combined with a time attenuation coefficient; the power ratio between the δ-wave and β-wave frequency bands is calculated based on the aforementioned division. S106. Calculate the time series change rate: The time series change rate of the weighted spectral entropy is calculated using a variable sliding window combined with the time decay coefficient; S107. Calculation of the band power ratio: The formula for calculating the band power ratio of β wave to δ wave is: [Formula omitted for brevity] ,in For beta band power, Power in the delta band; S108. Construct a comprehensive brain functional state index S: Couple the spectral entropy change rate using an adaptive weight fusion method. A comprehensive brain functional state index S was constructed using the β / δ power ratio, and the weights were determined by principal component analysis or linear discriminant analysis. S109. Perform intracranial pressure trend prediction: Use a long short-term memory network with residual connections to perform intracranial pressure trend prediction.

[0009] Furthermore, the range of values ​​for the frequency band weighting coefficients corresponding to the δ wave and β wave is greater than that for the weighting coefficients of the other frequency bands.

[0010] Furthermore, the customized wavelet basis function is a biorthogonal wavelet basis, and the adaptive threshold is dynamically adjusted according to the noise range of the EEG signal.

[0011] Furthermore, in the physiologically constrained independent component analysis, the physiological constraints include frequency range constraints of EEG signals and threshold constraints of waveform parameters.

[0012] Furthermore, the window duration of the variable sliding window is 1s to 5s, and the window duration is adaptively adjusted according to the signal stability.

[0013] Furthermore, the formula for calculating the comprehensive brain function state index S is as follows: Where S is the comprehensive brain function state index, The rate of change of spectral entropy. The ratio of beta wave power to delta wave power. and The preset weighting coefficients, and .

[0014] Furthermore, the rate of change of the spectral entropy The calculation formula is: ,in The entropy value at the current moment. Let τ be the entropy value for the first t seconds, and τ be the duration of the sliding window in seconds.

[0015] Furthermore, the frequency band weighting coefficients The value range is 0.1 to 0.8, where the weighting coefficient for the δ wave ranges from 0.4 to 0.8, and the weighting coefficient for the β wave ranges from 0.3 to 0.7.

[0016] The second aspect of this invention provides the following technical solution: A non-invasive intracranial pressure trend analysis data processing system based on EEG signal spectral entropy is used to execute a data processing method as described in one of the present invention, including an EEG signal acquisition module, a signal preprocessing module, a spectral entropy calculation module, a time series analysis module, a feature fusion module, a trend prediction module, and an output module.

[0017] The EEG signal acquisition module is used to acquire EEG signals from the subject's scalp. It uses 4 to 16 channels of scalp electrodes and controls the sampling impedance within the range of 5k to 50k ohms. The module has a built-in sampling parameter adjustment unit that can adjust the sampling frequency, sampling window length, and window overlap rate as needed.

[0018] The signal preprocessing module is connected to the EEG signal acquisition module and is used to perform wavelet decomposition denoising and independent component analysis (ICA) artifact separation. The signal preprocessing module incorporates a customized biorthogonal wavelet denoising unit and an ICA unit with physiological constraints. The wavelet denoising unit performs multi-level decomposition and adaptive threshold denoising on the raw EEG signal; the adaptive threshold is dynamically adjusted according to the noise range of the EEG signal. The ICA unit performs blind source separation and artifact removal, where the physiological constraints include EEG signal frequency range constraints and waveform parameter threshold constraints.

[0019] The spectral entropy calculation module is connected to the signal preprocessing module and is used to calculate the multi-band power distribution and weighted spectral entropy. The spectral entropy calculation module includes a built-in band power calculation unit and a weighted spectral entropy calculation unit. The band power calculation unit is used to divide the valid EEG signal into frequency bands and calculate the power of each band. The weighted spectral entropy calculation unit is used to store the calibrated band weighting coefficients and run the band weighted corrected Shannon entropy algorithm.

[0020] The timing analysis module is used to calculate the time-series change rate of spectral entropy and the frequency band power ratio. The module includes a built-in timing change rate calculation unit and a power ratio calculation unit. The timing change rate calculation unit uses a variable sliding window combined with a time decay coefficient to calculate the weighted spectral entropy's timing change rate. The window duration of the variable sliding window is 1-5 seconds, adaptively adjusted according to signal stability. The power ratio calculation unit is used to calculate the frequency band power ratio of the β wave and the δ wave.

[0021] The feature fusion module is connected to the time series analysis module and is used to construct a comprehensive brain functional state index through adaptive weights. The feature fusion module has a built-in adaptive weight fusion unit, which is used to couple the temporal change rate of spectral entropy and the frequency band power ratio through adaptive weight fusion.

[0022] The trend prediction module incorporates a Long Short-Term Memory (LSTM) network with residual connections. This LSM network contains 2-4 hidden layers, each containing 32-128 neurons. The trend prediction module receives input features such as the comprehensive brain function state index, weighted spectral entropy, and frequency band power ratio, mines the temporal correlations of multiple features, and outputs a probability distribution of intracranial pressure change trends and anomaly warning signals.

[0023] The output module is connected to the trend prediction module and uses a visual display screen to show the dynamic trend curve of intracranial pressure and brain metabolic status data in real time. It also actively triggers graded early warning prompts when early intracranial pressure abnormalities are detected. The output module is equipped with a graded early warning triggering unit, which sets graded early warning thresholds based on the probability distribution of intracranial pressure change trends.

[0024] The beneficial effects of this invention are as follows: This method achieves safe and non-invasive intracranial pressure monitoring through scalp EEG acquisition, avoiding the risks of infection and bleeding associated with traditional invasive monitoring. It enables continuous, long-term dynamic monitoring 24 / 7, meeting the clinical needs of intensive care. A dual preprocessing technique combining biorthogonal wavelet transform and physiologically constrained independent component analysis effectively resists various interference signals such as power frequency interference, electromyography artifacts, and eye movement artifacts, ensuring high robustness of signal processing and stable operation in complex clinical environments such as intensive care units. Using frequency-weighted modified Shannon entropy as the core quantitative indicator, a clear physiological correlation with intracranial pressure changes was established, including delta waves and beta waves. The corresponding frequency band weighting coefficient range is larger than that of other frequency bands, ensuring the accuracy and objectivity of the assessment results. An innovative method for calculating the temporal change rate of intracranial pressure is introduced, combining a variable sliding window with a time decay coefficient, enhancing the dynamic capture of the progressive increase in intracranial pressure. A comprehensive assessment index integrating EEG complexity and metabolic state is constructed, employing an adaptive weight fusion method to couple the temporal change rate of spectral entropy with the power ratio of β-waves to δ-waves, and combining it with a long short-term memory network with residual connectivity for trend prediction. This improves the accuracy and specificity of the assessment, enabling intelligent prediction and early warning of intracranial pressure trends, thus gaining valuable time for clinical intervention. Attached Figure Description

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall process of the non-invasive intracranial pressure trend analysis method of the present invention.

[0026] Figure 2 This is a block diagram of the module structure of the non-invasive intracranial pressure trend analysis system of the present invention.

[0027] Figure 3 This is a schematic diagram illustrating the calculation process of the comprehensive brain function state index of this invention.

[0028] Explanation of reference numerals in the attached figures: 100. EEG signal acquisition module; 200. Signal preprocessing module; 300. Spectrum entropy calculation module; 400. Time series analysis module; 500. Feature fusion module; 600. Trend prediction module; 700. Output module. Detailed Implementation

[0029] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0030] like Figure 1As shown in the figure, this embodiment provides a data processing method based on intracranial pressure trend analysis of EEG signal spectral entropy. The overall process of this method includes nine core steps, from the original acquisition of EEG signals to the final output of intracranial pressure trend probability distribution, forming a complete closed-loop processing flow.

[0031] In step S101, the system acquires the subject's scalp EEG signals through the EEG signal acquisition module 100. Specifically, 4-16 channel scalp electrodes are placed on the subject's scalp according to international standard lead positions. The electrode sampling impedance is controlled within the range of 5k-50k ohms to ensure a good signal-to-noise ratio for the acquired EEG signals. The sampling frequency is set to 200-300 Hz; in a preferred embodiment, a sampling frequency of 250 Hz is used. This sampling frequency satisfies the Nyquist sampling theorem, fully preserving information from the highest 125 Hz frequency components in the EEG signal, sufficient to cover the 30-100 Hz high-frequency band required for EEG analysis. The sampling window length is set to 2-4 seconds; in a preferred embodiment, a 3-second window length is used. Each analysis window contains 750 sampling points, providing sufficient spectral resolution to distinguish EEG components in different frequency bands. The window overlap rate is set to 50%~60%, with a preferred embodiment using 55%. There is approximately a 1.65-second data overlap area between adjacent analysis windows, ensuring sufficient temporal continuity and data integrity of the acquired EEG signals. The sampling parameter adjustment unit is built into the EEG signal acquisition module 100 and can dynamically adjust the sampling frequency, sampling window length, and window overlap rate according to actual clinical needs to adapt to different monitoring scenarios.

[0032] In step S102, the signal preprocessing module 200 performs dual preprocessing on the acquired raw EEG signal. This preprocessing process is completed collaboratively by the wavelet decomposition denoising unit and the physiologically constrained independent component analysis unit. First, the customized biorthogonal wavelet denoising unit uses biorthogonal wavelet basis functions to perform 4-8 level multi-scale decomposition of the raw EEG signal. The biorthogonal wavelet basis possesses compact support and linear phase properties. The compact support property enables the wavelet basis to have good localization capabilities in both the time and frequency domains, accurately capturing transient changes in the EEG signal; the linear phase property ensures that the signal does not produce phase distortion during decomposition and reconstruction, which is crucial for preserving the waveform characteristics of the EEG signal. During the multi-level decomposition process, the wavelet coefficients are processed by an adaptive threshold, which dynamically adjusts its value according to the real-time statistical characteristics of the noise interval of the EEG signal. Specifically, when strong power frequency interference is detected in the signal, the adaptive threshold algorithm automatically increases the suppression of power frequency-related wavelet coefficients; when an increase in high-frequency EMG artifacts is detected, the adaptive threshold applies stricter threshold constraints to the high-frequency wavelet coefficients. Through this dynamic adjustment mechanism, the biorthogonal wavelet denoising unit can specifically filter out 50 Hz power frequency interference and its harmonic components, while effectively attenuating high-frequency EMG artifacts, causing minimal attenuation of the effective components of the EEG signal, providing a clean signal basis for subsequent analysis.

[0033] After wavelet denoising, the EEG signal enters a physiologically constrained independent component analysis unit for blind source separation. This unit decomposes the EEG signal into several statistically independent source signal components and filters out signal components that conform to the physiological characteristics of EEG through preset physiological constraints. The physiological constraints include two core elements: first, an EEG signal frequency range constraint, which limits only signal components with frequencies between 0.5 and 100 Hz to be considered potentially valid components of the EEG signal; signal components with frequencies outside this range are identified as artifacts and eliminated. Second, a waveform parameter threshold constraint, which quantitatively evaluates the waveform morphology of independent components, including parameters such as peak-to-peak value, mean, variance, and zero-crossing rate. When the waveform parameters of an independent component exceed the physiological parameter threshold range of a normal EEG signal, that component is identified as an artifact and eliminated. In actual clinical settings, the physiologically constrained independent component analysis unit can effectively eliminate eye movement artifacts, ECG artifacts, and somatic motion interference components. Eye movement artifacts typically manifest as low-frequency, large-amplitude slow-wave components, primarily concentrated in the 0.1–3 Hz range, exhibiting regular sinusoidal or square-wave characteristics, significantly different from the complex waveforms of normal EEG signals. ECG artifacts possess regular waveform characteristics synchronized with the ECG, with frequencies concentrated in the 0.5–2 Hz range and showing a fixed multiple relationship with heart rate. Somatic motion interference components exhibit sudden amplitude abrupt changes and broadband noise characteristics. Through dual screening using frequency range constraints and waveform parameter threshold constraints, only signal components conforming to EEG physiological characteristics are retained, resulting in a pure and effective EEG signal, which is then transmitted to the spectral entropy calculation module 300.

[0034] In step S103, the preprocessed valid EEG signal enters the frequency band division and power calculation process. The frequency band power calculation unit divides the valid EEG signal into five fixed frequency bands: 0.5-4 Hz for delta waves, 4-8 Hz for theta waves, 8-13 Hz for alpha waves, 13-30 Hz for beta waves, and 30-100 Hz for gamma waves. The Fast Fourier Transform (FFT) algorithm is used to calculate the signal power within each frequency band. Specifically, the Welch method is used to estimate the segmented power spectrum of the EEG signal. The 3-second window of the EEG signal is divided into several overlapping sub-segments. The Fourier transform results of each sub-segment are calculated separately and then averaged to obtain a smoothed power spectral density estimate. Based on this, the power spectral density within each sub-band is integrated to obtain the absolute power value of each frequency band.

[0035] The frequency band division design is based on findings from neurophysiological research. Delta waves are primarily generated by deep neurons in the cerebral cortex and are closely related to sleep states and the degree of cortical inhibition. When increased intracranial pressure leads to insufficient brain perfusion, delta wave power typically shows a compensatory increase. Beta waves reflect the alertness and metabolic activity of the cerebral cortex, and their power changes are directly related to the excitability level of neurons. Although gamma waves have the highest frequency, they exhibit characteristic changes in certain cognitive tasks and neuropathological states. The five-band division scheme comprehensively considers neurophysiological basis and clinical practicality, retaining the classic frequency band division system of traditional EEG analysis while providing a standardized frequency band structure for subsequent spectral entropy calculations.

[0036] After obtaining the absolute power values ​​of each frequency band in step S104, the power of each frequency band is normalized globally to generate a probability distribution of the proportion of power of each frequency band to the total power. Specifically, the absolute power values ​​of the five frequency bands are summed to obtain the total power, and then the absolute power values ​​of each frequency band are divided by the total power to obtain the normalized power proportion. The sum of the power proportions of each frequency band is strictly equal to 1, satisfying the normalization constraint of the probability distribution. The core function of normalization is to eliminate individual differences in EEG amplitude and baseline drift interference. The EEG signals of different subjects may differ by several times or even tens of times in absolute amplitude. Normalization converts the power of each frequency band into a proportional form, making the calculated spectral entropy comparable among different subjects. Simultaneously, normalization can eliminate baseline interference caused by changes in electrode contact impedance and amplifier gain drift, ensuring the stability of characteristic parameters during long-term monitoring.

[0037] Step S105 calculates the weighted spectral entropy using the band-weighted modified Shannon entropy formula. The weighted spectral entropy calculation unit stores pre-calibrated band weighting coefficients and executes the band-weighted modified Shannon entropy algorithm. The calculation formula is as follows: .in Let be the weighting coefficient for the i-th frequency band. The proportion of power in the i-th frequency band to the total power. The total number of frequency bands, in the embodiments of the present invention The value is equal to 5. The weighting coefficients for the frequency bands range from 0.1 to 0.8, with the specific allocation strategy as follows: the weighting coefficient for delta waves ranges from 0.4 to 0.8, the weighting coefficient for beta waves ranges from 0.3 to 0.7, and the weighting coefficients for theta, alpha, and gamma waves range from 0.1 to 0.3. The weighting coefficient ranges for delta and beta waves are significantly larger than those for the other frequency bands. This design is based on the closer physiological correlation between intracranial pressure changes and these two frequency bands. In the early stages of increased intracranial pressure, the decrease in cerebral perfusion pressure first affects the metabolic state of the cerebral cortex, leading to a decrease in beta wave power. As intracranial pressure further increases, ischemic and hypoxic changes occur in the brain tissue, and the delta wave power shows a compensatory increase. By assigning higher weighting coefficients to delta and beta waves, the sensitivity of the weighted spectral entropy to changes in intracranial pressure is significantly improved, while theta, alpha, and gamma waves, which are relatively insensitive to changes in intracranial pressure, are assigned lower weighting coefficients to avoid these components interfering with the assessment results.

[0038] This embodiment provides a complete set of five-band weighted coefficients for practical operation: δ wave 0.6, β wave 0.5, θ wave 0.2, α wave 0.15, γ wave 0.25; all five band values ​​are uniformly substituted into the weighted modified Shannon entropy formula to complete the joint calculation, and any θ, α, or γ bands cannot be removed individually.

[0039] To quantitatively verify the core gain of the theta, alpha, and gamma EEG bands in entropy feature calculation, a controlled experiment was designed using an intracranial pressure temporal trend prediction task as the model performance evaluation standard. Intracranial pressure is a core physiological indicator for assessing brain state. A regression prediction model was constructed based on EEG entropy features, and its performance could be evaluated using the coefficient of determination. It can intuitively represent the degree of fit between EEG entropy features and the actual trend of intracranial pressure changes, and at the same time quantify the long-term stability of the features with baseline drift error.

[0040] The control group retained only the δ and β main bands to construct a weighted Shannon entropy feature input prediction model. The model's determination coefficient for predicting intracranial pressure trends was... The coefficient of variation was only 0.68, indicating a large fitting bias and significant computational error due to EEG baseline drift. The experimental group fully incorporated the three auxiliary bands (θ, α, and γ), and after calculating the corrected Shannon entropy using a five-band joint weighting method, it was input into the same prediction model. The baseline drift error was reduced by 42% simultaneously, with the baseline value increased to 0.85.

[0041] The above comparison results fully demonstrate that the θ, α, and γ bands carry weak EEG rhythm information that cannot be covered by the δ and β main bands. They play a key role in baseline correction and weak abnormal signal identification in the weighted entropy calculation process. The absence of any auxiliary band will significantly weaken the ability of entropy features to represent changes in the brain's physiological state, thus possessing irreplaceable technical benefits.

[0042] Step S106 uses a time-series change rate calculation unit to calculate the time-series change rate of the weighted spectral entropy. The time-series change rate calculation unit employs an algorithm framework combining a variable sliding window and a time decay coefficient. The window duration of the variable sliding window is set to 1-5 seconds; in practice, the window duration is adaptively adjusted according to signal stability. Signal stability is quantitatively assessed by calculating the ratio of the standard deviation to the mean of the EEG signal within the sliding window. When this ratio is lower than a preset stability threshold, the signal is considered to be in a stable phase, and the system automatically uses a long window of 3-5 seconds to improve statistical stability; when this ratio is higher than a preset mutation threshold, the signal is considered to be in a mutation phase, and the system automatically switches to a short window of 1-2 seconds to improve response speed. The time decay coefficient ranges from 0.1 to 0.3. The introduction of the time decay coefficient gives greater weight to recent entropy changes in the calculation, thereby improving the response speed of the time-series change rate to acute changes in intracranial pressure. (Spectral entropy change rate) The calculation formula is: ,in The entropy value at the current moment. Let be the entropy value in the first t seconds, and τ be the duration of the sliding window in seconds. This formula essentially represents the average rate of change of entropy over the sliding window's time span; its sign and magnitude directly reflect the evolving trend of EEG spectral complexity. When the value is positive and continues to increase, it indicates that the complexity of the EEG signal is increasing, which may correspond to a decrease in intracranial pressure; when When the value is negative and continues to decrease, it indicates that the complexity of the EEG signal is decreasing, which may correspond to the process of increasing intracranial pressure.

[0043] Step S107 calculates the frequency band power ratio of the β wave to the δ wave using the power ratio calculation unit. The β wave frequency band is defined as 13~30 Hz, and the δ wave frequency band is defined as 0.5~4 Hz. The calculation formula is as follows: ,in For beta band power, The power of the delta wave band is represented by the β / δ power ratio, a sensitive indicator reflecting the balance between the metabolic state of the cerebral cortex and the inhibitory state of the subcortical structures. Under normal physiological conditions, cerebral cortex activity is dominant, with relatively high β-wave power and relatively low δ-wave power, resulting in a high β / δ power ratio. When increased intracranial pressure leads to insufficient cerebral perfusion, ischemic changes occur in brain tissue, manifested as a decrease in β-wave power and an increase in δ-wave power, leading to a decrease in the β / δ power ratio. By continuously monitoring the dynamic trend of the β / δ power ratio, the influence of intracranial pressure changes on the distribution of the EEG power spectrum can be effectively captured.

[0044] Step S108 involves the feature fusion module 500 coupling the spectral entropy change rate and the β / δ power ratio using an adaptive weight fusion method to construct a comprehensive brain function state index S. The adaptive weight fusion unit uses principal component analysis or linear discriminant analysis to fuse the spectral entropy change rate and the β / δ power ratio, automatically determining the optimal weight coefficients based on the discriminative power of the two features in the training dataset to identify intracranial pressure trends. Specifically, the formula for calculating the comprehensive brain function state index S is: ,in The rate of change of spectral entropy. The ratio of beta wave power to delta wave power. and The adaptive weighting coefficients satisfy the following conditions: The constraints. The value range is 0.4 to 0.7. The value range is 0.3 to 0.6. In practical applications, after analyzing a large amount of labeled data using principal component analysis, it was found that the rate of change of spectral entropy usually contributes more to distinguishing intracranial pressure trends. It can be set to 0.7. When set to 0.3, the formula for calculating the Comprehensive Brain Functional State Index S becomes more specific. . Figure 3 A complete flowchart of the calculation process for the comprehensive brain functional state index S is provided, clearly demonstrating the calculation from the rate of change of spectral entropy. and β / δ power ratio The two input features are fused with adaptive weights to form a complete data stream that is the final output comprehensive brain function state index S.

[0045] Step S109 involves intracranial pressure trend prediction performed by the trend prediction module 600. The trend prediction module 600 incorporates a modified long short-term memory network with residual connections. This network receives the comprehensive brain function state index S from the feature fusion module 500 as its core input feature, and simultaneously receives continuous time-weighted spectral entropy data from the spectral entropy calculation module 300 and β / δ power ratio time-series data from the time-series analysis module 400 as joint input features. The network architecture contains 2-4 hidden layers, each containing 32-128 neurons. The residual connection structure effectively alleviates the gradient vanishing problem during deep network training by directly connecting the output of the previous layer across several layers to the subsequent layers, enabling the network to fully learn the highly nonlinear mapping relationship between EEG signal features and intracranial pressure trends. Through network training, multi-feature time-series correlations are mined. The output layer uses a softmax activation function to output the probability distributions of three states: increased intracranial pressure, stable intracranial pressure, and decreased intracranial pressure. When the probability of increased intracranial pressure exceeds a preset first warning threshold, the system output module 700 triggers a yellow warning; when the probability exceeds a higher second warning threshold, a red warning is triggered. The output is a probability distribution of the intracranial pressure change trend, used to assist clinicians in making diagnostic decisions, without directly providing a disease diagnosis or treatment plan.

[0046] This invention provides a non-invasive intracranial pressure trend analysis data processing system based on electroencephalogram (EEG) signal spectral entropy, used to execute the data processing method described in Example 1. Figure 2 As shown, the system includes seven functional modules: an EEG signal acquisition module 100, a signal preprocessing module 200, a spectrum entropy calculation module 300, a time series analysis module 400, a feature fusion module 500, a trend prediction module 600, and an output module 700. The modules are connected sequentially according to the data processing flow to form a complete closed loop from signal acquisition to result output.

[0047] The EEG signal acquisition module 100 serves as the system's data input, responsible for the targeted acquisition of EEG signals from the subject's scalp. This module employs 4-16 channel scalp electrodes, with electrode sampling impedance controlled within the range of 5000 ohms to 50000 ohms to ensure signal quality. The module includes a built-in sampling parameter adjustment unit, allowing dynamic adjustment of the sampling frequency, sampling window length, and window overlap rate according to monitoring requirements. The acquired raw EEG signals are transmitted to the signal preprocessing module 200 in real-time streaming for further processing. The EEG signal acquisition module 100 also features built-in impedance detection, automatically detecting the contact impedance between each channel electrode and the scalp after electrode placement. If the impedance of a channel exceeds a preset range, a prompt is sent to the operator, allowing for timely adjustment of the electrode position to ensure signal quality.

[0048] The signal preprocessing module 200 is connected to the EEG signal acquisition module 100 and incorporates two core processing units: a customized biorthogonal wavelet denoising unit and a physiologically constrained independent component analysis unit. The wavelet denoising unit performs multi-level decomposition and adaptive threshold denoising on the raw EEG signal. The adaptive threshold dynamically adjusts its value based on the real-time statistical characteristics of the EEG signal noise range, adapting to noise levels in different clinical environments. The independent component analysis unit performs blind source separation and removes artifact components. Physiological constraints include EEG signal frequency range constraints and waveform parameter threshold constraints, ensuring that only signal components conforming to the physiological characteristics of EEG are retained. The effective EEG signal after dual preprocessing is transmitted to the spectral entropy calculation module 300 in a standard format.

[0049] The spectrum entropy calculation module 300 is connected to the signal preprocessing module 200 and includes a built-in frequency band power calculation unit and a weighted spectrum entropy calculation unit. The frequency band power calculation unit divides the effective EEG signal into frequency bands and calculates the power of each band. The five fixed frequency bands are 0.5–4 Hz, 4–8 Hz, 8–13 Hz, 13–30 Hz, and 30–100 Hz, corresponding to delta waves, theta waves, alpha waves, beta waves, and gamma waves, respectively. The weighted spectrum entropy calculation unit stores pre-calibrated frequency band weighting coefficients, calculates the weighted spectrum entropy according to the frequency band weighted modified Shannon entropy algorithm, and transmits the calculation result to the time series analysis module 400. When calculating the frequency band power, the frequency band power calculation unit uses the Welch power spectrum estimation method, dividing the long data segment into several overlapping sub-segments, calculating the Fourier transform for each sub-segment, and averaging the results to obtain a smoother and more reliable power spectrum estimation result.

[0050] The timing analysis module 400 incorporates a timing change rate calculation unit and a power ratio calculation unit. The timing change rate calculation unit uses a variable sliding window combined with a time decay coefficient to calculate the timing change rate of the weighted spectral entropy. The window duration of the variable sliding window is 1-5 seconds and adaptively adjusts according to signal stability. The power ratio calculation unit calculates the frequency band power ratio of the β wave and the δ wave. The timing analysis module 400 displays the timing change rate of the spectral entropy. β / δ power ratio The calculation results, such as the weighted spectral entropy of the continuous time series, are transmitted to the feature fusion module 500 and the trend prediction module 600.

[0051] The feature fusion module 500 is connected to the time series analysis module 400 and includes a built-in adaptive weight fusion unit. This unit couples the temporal change rate of spectral entropy with the frequency band power ratio using an adaptive weight fusion method, according to the formula... The comprehensive brain function state index S is calculated. In each calculation, the adaptive weight fusion unit first standardizes the input features to eliminate dimensional differences, and then performs a weighted summation based on a preset feature fusion strategy or weight parameters learned online to obtain the time series of the comprehensive brain function state index S. The feature fusion module 500 transmits the comprehensive index and related feature parameters to the trend prediction module 600.

[0052] The trend prediction module 600 incorporates a Long Short-Term Memory (LSTM) network with residual connections. This LTM network contains 2-4 hidden layers, each with 32-128 neurons. The network is pre-trained with training data including a large number of EEG signal samples synchronously labeled using gold-standard invasive intracranial pressure monitoring equipment. The network receives input features such as the comprehensive brain function state index S, weighted spectral entropy time-series data, and β / δ power ratio time-series data. It utilizes the unique gating mechanism of the LTM network to uncover multi-feature temporal correlations. The introduction of residual connections allows the network to build deeper layers to learn more complex feature representations while ensuring effective gradient propagation during training. The network outputs a probability distribution of intracranial pressure change trends and abnormal warning signals, transmitting the processing results to the output module 700.

[0053] The output module 700 connects to the trend prediction module 600 and uses a visual display screen to show the real-time dynamic trend curve of intracranial pressure. The displayed content includes the intracranial pressure change trend curve, brain metabolic status data, frequency band power ratio time series graph, and comprehensive brain function status index time series graph. The output module 700 is equipped with a graded early warning trigger unit, which sets graded early warning thresholds based on the probability distribution of intracranial pressure change trends. When an early intracranial pressure abnormality is detected, the output module 700 actively triggers a graded early warning prompt, with a yellow warning corresponding to a moderate risk state and a red warning corresponding to a high risk state. The warning information includes key information such as the abnormality type, the current trend probability distribution, and the recommended level of attention, enabling medical staff to grasp the subject's intracranial pressure dynamic status and make clinical decisions as soon as possible.

[0054] In actual clinical applications, the data processing method and system provided in this embodiment operate as follows: After admitting patients with traumatic brain injury to the intensive care unit, medical staff place 4-16 channel scalp electrodes on the patient's scalp according to the standard lead protocol of the International Society for Electroencephalography (ESE). Monitoring is initiated after the electrode sampling impedance is adjusted to below 50,000 ohms. The EEG signal acquisition module 100 continuously acquires scalp EEG signals at a sampling frequency of 250 Hz, a window length of 3 seconds, and a window overlap rate of 55%. The raw signal data stream is transmitted in real time to the signal preprocessing module 200. The biorthogonal wavelet denoising unit in the signal preprocessing module 200 performs six-level wavelet decomposition on the signal. The adaptive threshold algorithm dynamically adjusts the threshold parameters based on real-time noise statistical characteristics, effectively suppressing 50 Hz power frequency interference and high-frequency electromyography artifacts. The independent component analysis unit with physiological constraints performs blind source separation after denoising, eliminating eye movement artifacts and ECG artifacts through frequency range constraints and waveform parameter threshold constraints, obtaining a clean and effective EEG signal. The spectral entropy calculation module 300 divides the effective EEG signal into five fixed frequency bands and calculates the power of each band. After normalizing the power, it calculates the weighted spectral entropy according to preset frequency band weighting coefficients. The temporal analysis module 400 uses a sliding window with adaptively varying window duration to calculate the temporal change rate of the spectral entropy, and simultaneously calculates the power ratio of the β wave to the δ wave. The feature fusion module 500 uses an adaptive weighted coupling method to fuse the spectral entropy change rate and the β / δ power ratio into a comprehensive brain function state index. The long short-term memory network with residual connections in the trend prediction module 600 outputs a probability distribution of intracranial pressure change trends based on the comprehensive brain function state index and temporal features. When the probability distribution shows that the probability of an upward trend in intracranial pressure exceeds the warning threshold, the output module 700 presents a warning message on the display screen and triggers an audible and visual alarm to alert medical staff to changes in the patient's condition. The entire monitoring process can continue for 24 hours or even longer, providing continuous, stable, and non-invasive trend reference data for clinical intracranial pressure management.

[0055] This embodiment achieves safe and non-invasive monitoring of intracranial pressure trends through scalp EEG acquisition. A dual preprocessing technique combining biorthogonal wavelet transform and physiologically constrained independent component analysis ensures high robustness of signal processing. Frequency-weighted modified Shannon entropy is used as the core quantitative indicator to establish a clear physiological correlation with intracranial pressure changes. A combination of variable sliding window and adaptive weight fusion enhances the dynamic capture capability of gradual changes in intracranial pressure. Finally, a long short-term memory network with residual connectivity is used to achieve intelligent prediction and early grading of intracranial pressure trends, providing a safe, reliable, accurate, and intelligent non-invasive monitoring solution for clinical intracranial pressure management.

[0056] In the specific implementation of this invention, the specific values ​​of the frequency band weighting coefficients can be personalized based on clinical validation data. The optimal weighting coefficients may differ for different diseases or individuals. The system supports adaptive optimization of the weighting coefficients through machine learning algorithms to improve assessment accuracy. The threshold for signal stability can also be appropriately adjusted according to the noise level of the monitoring environment. In monitoring environments with strong noise interference, the stability judgment criteria can be appropriately relaxed to avoid frequent switching of window length. Before deployment, the long short-term memory network in the trend prediction module needs to be trained offline using sufficient labeled EEG data. The training data should cover samples of different ages, different diseases, and different intracranial pressure change trends to ensure the generalization ability of the model. In actual use, the system also supports joint analysis of multi-lead signals. When the signal quality of multiple scalp electrode channels meets the requirements, multi-lead signals can be processed in parallel and the results fused to further improve the reliability of the assessment.

[0057] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy, characterized in that, Includes the following steps: Step S101: Acquire EEG signals: sampling frequency is 200Hz~300Hz, sampling window length is 2s~4s, and window overlap rate is 50%~60%; Step S102, Dual Preprocessing: The EEG signal is decomposed into multiple layers using a custom wavelet basis function, and signal denoising is completed with an adaptive threshold. Then, blind source separation and artifact removal are performed through independent component analysis with physiological constraints. Step S103, Frequency Band Division and Power Calculation: The preprocessed EEG signal is divided into five fixed frequency bands. The corresponding wavebands and frequency ranges of the five fixed frequency bands are: delta wave 0.5-4Hz, theta wave 4-8Hz, alpha wave 8-13Hz, beta wave 13-30Hz, and gamma wave 30-100Hz. Among them, delta wave and beta wave are the core response wavebands to changes in intracranial pressure, theta wave and alpha wave represent the basic resting brain activity level, and gamma wave captures the characteristics of abnormal high-frequency synchronous discharge of neurons. All five fixed frequency bands are involved in the weighted spectral entropy joint calculation, and the power of each frequency band is calculated to form the frequency band power distribution. S104. Global normalization processing: Normalize the power of each frequency band to obtain the probability distribution of the power proportion of each frequency band. S105. Calculate the weighted spectral entropy: Calculate the weighted spectrum H using the band-weighted modified Shannon entropy. The calculation formula is as follows: ， in, These are frequency band weighting coefficients. The proportion of power in the i-th frequency band to the total power. The total number of frequency bands is given; the time-series change rate of the weighted spectral entropy is calculated using a variable sliding window combined with a time attenuation coefficient; the power ratio between the δ-wave and β-wave frequency bands is calculated based on the aforementioned division. S106. Calculate the time series change rate: The time series change rate of the weighted spectral entropy is calculated using a variable sliding window combined with the time decay coefficient; S107. Calculation of the band power ratio: The formula for calculating the band power ratio of β wave to δ wave is: [Formula omitted] ,in For beta band power, Power in the delta band; S108. Constructing a comprehensive brain function state index S: using an adaptive weight fusion method to couple the spectral entropy change rate. A comprehensive brain functional state index S was constructed using the β / δ power ratio, and the weights were determined by principal component analysis or linear discriminant analysis. S109. Perform intracranial pressure trend prediction: Use a long short-term memory network with residual connections to perform intracranial pressure trend prediction.

2. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 1, characterized in that, The range of values ​​for the weighting coefficients of the frequency bands corresponding to the δ wave and β wave is greater than that of the weighting coefficients of the other frequency bands.

3. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 1, characterized in that, The customized wavelet basis function uses a biorthogonal wavelet basis, and the adaptive threshold is dynamically adjusted according to the noise range of the EEG signal.

4. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 1, characterized in that, In the independent component analysis with physiological constraints, the physiological constraints include frequency range constraints of EEG signals and threshold constraints of waveform parameters.

5. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 1, characterized in that... The variable sliding window has a window duration of 1s to 5s, and the window duration is adaptively adjusted according to the signal stability.

6. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 1, characterized in that, The formula for calculating the comprehensive brain function state index S is as follows: Where S is the comprehensive brain function state index, The rate of change of spectral entropy. The ratio of beta wave power to delta wave power. and The preset weighting coefficients, and .

7. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 7, characterized in that, The rate of change of spectral entropy The calculation formula is: ,in The entropy value at the current moment. Let τ be the entropy value for the first t seconds, and τ be the duration of the sliding window in seconds.

8. The data processing method based on intracranial pressure trend analysis using electroencephalogram (EEG) signal spectral entropy as described in claim 1, characterized in that, The frequency band weighting coefficient The value range is 0.1 to 0.8, where the weighting coefficient for the δ wave ranges from 0.4 to 0.8, and the weighting coefficient for the β wave ranges from 0.3 to 0.

7.

9. A data processing system for non-invasive intracranial pressure trend analysis based on electroencephalogram (EEG) signal spectral entropy, characterized in that, The method for performing intracranial pressure trend analysis data processing based on the spectral entropy of electroencephalogram (EEG) signals according to any one of claims 1 to 9 comprises: an EEG signal acquisition module for acquiring EEG signals from the scalp of a subject; a signal preprocessing module connected to the EEG signal acquisition module for performing wavelet decomposition denoising and independent component analysis artifact separation; a spectral entropy calculation module connected to the signal preprocessing module for calculating multi-band power distribution and weighted spectral entropy; a time-series analysis module for calculating the time-series change rate of spectral entropy and the ratio of band power; a feature fusion module connected to the time-series analysis module for constructing a comprehensive brain function state index through adaptive weights; a trend prediction module with a built-in long short-term memory network with residual connection structure; and an output module for displaying intracranial pressure trend curves and early warning information.

Images