Epilepsy clinical nursing early warning method and system

Abstract

The invention provides an epilepsy clinical nursing early warning method and system, and relates to the technical field of medical information processing, the epilepsy clinical nursing early warning method comprises the following steps: obtaining multi-modal physiological signal data of an epilepsy patient, the multi-modal physiological signal data comprising electroencephalogram signal data and limb movement signal data; performing principal component analysis processing on the electroencephalogram signal data to obtain dimensionality-reduced electroencephalogram characteristic data; performing time-frequency decomposition on the dimensionality-reduced electroencephalogram characteristic data to obtain energy distribution of a plurality of frequency bands; calculating an information entropy value based on the energy distribution of the plurality of frequency bands; according to the information entropy and the limb movement signal data, the current state of the epileptic is judged through a decision tree model; determining a corresponding hidden danger level according to the current state and generating early warning information; according to the method, electroencephalogram artifacts are eliminated through principal component analysis, the state of the threatened period is recognized before epileptic seizure, and the problems of early warning lag and unstable response in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, specifically to a clinical nursing early warning method and system for epilepsy. Background Technology

[0002] Current clinical nursing safety hazard analysis in neurology mainly relies on manual experience-based screening, where nursing staff identify potential risks and take preventative measures based on past experience. This approach suffers from unstable analysis results, incomplete coverage, and low response efficiency. In recent years, intelligent analysis methods based on machine learning and signal processing have been gradually applied in this field. Typical technical approaches include: using principal component analysis to reduce the dimensionality of pathological indicators, extracting key features, and then inputting them into a classification model for disease diagnosis; using information entropy to quantify and encode feature variables, thereby constructing a decision tree for disease classification; and matching the classification results with corresponding nursing safety hazard standards to determine whether the patient's condition meets the standards and issue alerts.

[0003] However, the aforementioned technical solutions are mainly based on general analysis of static biochemical indicators such as blood glucose and blood pressure, with sampling frequencies typically on the order of minutes or even hours. This is relatively effective for monitoring the state of long-term conditions such as cerebral hemorrhage. But epileptic seizures are transient electrophysiological abnormalities; the transition from the prodromal phase to the tonic-clonic seizure phase may only take tens of seconds. The low sampling frequency of static indicators cannot capture the millisecond-level prodromal features of seizures, resulting in a serious lag in early warning.

[0004] For the detection of epileptic seizures, electroencephalography (EEG) analysis technology has become an effective means due to its high temporal resolution and long-term monitoring capabilities. Existing EEG analysis methods can be divided into two categories: linear and nonlinear methods. While time-domain analysis in linear methods contains complete information, it lacks objectivity and has significant errors. Frequency-domain analysis requires stationary signals, but EEG signals are inherently nonlinear and non-stationary. Time-frequency domain analysis methods, such as wavelet transform, have improved upon this but still have limitations. From the perspective of nonlinear dynamics, entropy-based methods have significant advantages in biological signal processing. Sample entropy, as an improved algorithm for approximate entropy, has the characteristics of being independent of data length and insensitive to data loss. Multidimensional sample entropy, by combining multi-channel EEG signals, can more comprehensively reflect the overall changes in the brain before and after an epileptic seizure.

[0005] However, existing entropy-based epilepsy detection schemes still have the following shortcomings: First, multidimensional sample entropy calculation involves a large number of point-by-point comparisons, and the computational efficiency is difficult to meet the needs of real-time monitoring; Second, epilepsy patients often experience severe electromyography and electrooculography artifacts due to convulsions and blinking, which are difficult to remove using conventional preprocessing methods; Third, existing schemes mostly use deep learning classifiers for binary classification, which cannot distinguish between an impending seizure and an ongoing seizure, and lack graded early warning capabilities; Fourth, the system output is limited to prediction results and alarm signals, and an automated closed loop from monitoring abnormalities to environmental intervention has not been established, making it impossible to achieve proactive protection before a seizure.

[0006] Chinese patent document CN111870241B discloses a method for detecting epileptic seizure signals based on optimized multidimensional sample entropy. By constructing multidimensional vectors and calculating optimized multidimensional sample entropy as features, it combines Bi-LSTM neural networks to predict the EEG entropy value for the next 5 minutes and performs binary classification of the seizure phase and normal phase. This method improves computational efficiency and prediction accuracy. However, it still has problems such as using only a single EEG modality data, lacking a mechanism for processing epilepsy-specific artifacts, being unable to achieve graded early warning of the prodromal phase and the seizure phase, and not establishing a closed loop of environmental intervention.

[0007] Chinese patent document CN113314201B discloses a method and system for analyzing safety hazards in clinical nursing in neurology. It uses principal component analysis to reduce the dimensionality of pathological indicators, constructs a decision tree based on information entropy to classify diseases, and matches nursing safety hazard standards with the classification results to issue reminders. This has the technical effect of improving the efficiency and comprehensiveness of analysis. However, it still has some problems, such as the inability to capture millisecond-level electrophysiological abnormalities in epilepsy based on static biochemical indicators, the fact that principal component analysis is only used for dimensionality reduction and does not achieve artifact removal, and the response mechanism is limited to reminders and lacks the ability to actively intervene in the environment. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for early warning of epilepsy clinical care that can effectively eliminate EEG artifacts, capture epileptic seizure precursors in real time, and achieve an automated closed-loop response from monitoring to environmental intervention.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A clinical nursing early warning method for epilepsy includes the following steps: S1: Acquire multimodal physiological signal data of epilepsy patients, including electroencephalogram (EEG) signal data and limb movement signal data; S2: Perform principal component analysis on the EEG signal data to obtain dimensionality-reduced EEG feature data; S3: Perform time-frequency decomposition on the reduced-dimensional EEG feature data to obtain the energy distribution of multiple frequency bands; S4: Calculate the information entropy value based on the energy distribution of the multiple frequency bands; S5: Based on the information entropy value and the limb movement signal data, determine the current state of the epileptic patient using a decision tree model; S6: Determine the corresponding hazard level based on the current status and generate corresponding early warning information.

[0010] Furthermore, the multimodal physiological signal data also includes environmental parameter data, which includes at least one of light intensity data and sound intensity data.

[0011] Further: In step S2, the process of performing principal component analysis on the EEG signal data includes: S21: The EEG signal data is centered to obtain a centered EEG signal matrix; S22: Calculate the covariance matrix of the centered EEG signal matrix; S23: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; S24: Identify artifact components based on the spectral characteristics of the feature vector; S25: After removing the artifact components, the EEG signal is reconstructed to obtain the dimensionality-reduced EEG feature data.

[0012] Furthermore, the artifact components include electrooculography (EOG) artifact components and electromyography (EMG) artifact components.

[0013] Further: In S3, wavelet packet transform is used to perform time-frequency decomposition on the dimensionality-reduced EEG feature data, and the multiple frequency bands include Delta band, Theta band, Alpha band, Beta band and Gamma band.

[0014] Further: In step S4, the process of calculating the information entropy value based on the energy distribution of the multiple frequency bands includes: S41: Calculate the energy value for each frequency band; S42: Calculate the relative energy percentage of each frequency band based on the energy value of each frequency band, and use it as a probability distribution; S43: Calculate the Shannon entropy based on the probability distribution to obtain the information entropy value.

[0015] Furthermore, the decision tree model includes a root node, a second-layer node, and a third-layer node. The decision feature of the root node is the rate of change of EEG information entropy, the decision feature of the second-layer node is the limb movement impact feature, and the decision feature of the third-layer node is vital sign parameters.

[0016] Furthermore: the current state includes one of the following: stable period, prodromal period, ictal period, and postictal period; The hazard levels include Level 1 hazards and Level 2 hazards. The Level 1 hazard corresponds to the current state when it is in the prodromal stage, and the Level 2 hazard corresponds to the current state when it is in the acute stage. S6 further includes: When the hazard level is Level 1, the environmental control equipment is used to adjust the environmental parameters. When the hazard level is Level II, the control nursing equipment will perform protective operations and activate the alarm device.

[0017] Furthermore: S1-S6 are executed continuously using a sliding time window, the duration of which is 1 second to 10 seconds.

[0018] Furthermore: an epilepsy clinical nursing early warning system, implementing the above-mentioned epilepsy clinical nursing early warning method, includes: The data acquisition module is used to acquire multimodal physiological signal data of epilepsy patients, including electroencephalogram (EEG) signal data and limb movement signal data. The signal processing module is used to perform principal component analysis on the EEG signal data to obtain dimensionality-reduced EEG feature data. The time-frequency analysis module is used to perform time-frequency decomposition on the dimensionality-reduced EEG feature data to obtain the energy distribution of multiple frequency bands; The entropy calculation module is used to calculate the information entropy value based on the energy distribution of the multiple frequency bands; The state determination module is used to determine the current state of the epileptic patient based on the information entropy value and the limb movement signal data through a decision tree model; The early warning module is used to determine the corresponding hazard level based on the current status and generate corresponding early warning information; The data acquisition module includes an EEG acquisition unit and a motion sensing unit. The EEG acquisition unit is used to acquire multi-channel EEG signals, and the motion sensing unit includes an accelerometer and a gyroscope.

[0019] Compared with the prior art, the present invention has the following advantages: I. This invention extends principal component analysis from a simple dimensionality reduction tool to a signal purification tool with artifact recognition and removal functions. By analyzing the spectral characteristics of each principal component, it identifies and removes electrooculography (EOG) and electromyography (EMG) artifacts, solving the problem of severe interference with EEG signals caused by blinking and muscle twitching in epilepsy patients. This improves the accuracy of subsequent feature extraction and reduces the false alarm rate caused by artifacts.

[0020] Second, this invention uses wavelet packet transform to decompose the EEG signal into time and frequency and calculate the Shannon entropy of the energy distribution of each frequency band. It can quantify the process of the EEG signal changing from a disordered state to a synchronous rhythmic state during an epileptic seizure. Compared with the minute-level sampling frequency of static biochemical indicators, it realizes the real-time capture of millisecond-level electrophysiological abnormalities, enabling the system to identify the prodromal state before the onset of obvious symptoms such as limb convulsions in epileptic patients, thus providing nursing staff with an early warning window.

[0021] Third, this invention establishes an automated closed-loop response mechanism from monitoring abnormalities to environmental intervention. It performs graded early warning based on the decision tree classification results. During the prodromal period, the lights are automatically dimmed to reduce photosensitive stimulation. During the attack period, the bed rails are raised and an audible and visual alarm is triggered and the attack timer is started. This reduces the reliance on manual operation and solves the problems of delayed nursing response and unstable reliance on manpower in the prior art. Attached Figure Description

[0022] Figure 1 A flowchart of a clinical nursing early warning method for epilepsy provided by the present invention; Figure 2 This is a schematic diagram of the structure of an epilepsy clinical nursing early warning system provided by the present invention. Detailed Implementation

[0023] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 like Figure 1 As shown: This invention provides a clinical nursing early warning method for epilepsy, comprising the following steps: S1, acquire multimodal physiological signal data of epilepsy patients, including electroencephalogram (EEG) signal data and limb movement signal data; S2, principal component analysis is performed on the EEG signal data to obtain the dimensionality-reduced EEG feature data; S3, perform time-frequency decomposition on the dimensionality-reduced EEG feature data to obtain the energy distribution of multiple frequency bands; S4, calculates the information entropy value based on the energy distribution of multiple frequency bands; S5. Based on the information entropy value and limb movement signal data, determine the current state of the epilepsy patient through a decision tree model; S6. Determine the corresponding risk level based on the current state and generate corresponding early warning information.

[0025] Specifically, in step S1, EEG signal data is acquired through a multi-channel EEG acquisition device, collecting real-time voltage value sequences from key leads such as the prefrontal and temporal lobes; limb movement signal data is acquired through an inertial measurement unit worn on the wrist or ankle, including triaxial acceleration data. Step S2 uses principal component analysis to process the high-dimensional EEG signal, identifying and removing artifact components while reducing dimensionality. Step S3 uses wavelet packet transform to decompose the EEG signal into different frequency bands. Step S4 calculates the Shannon entropy of the energy distribution of each frequency band; during an epileptic seizure, the EEG signal tends to synchronize, and the entropy value decreases significantly. Step S5 uses the entropy change rate as the core criterion, combined with movement characteristics, to classify the state. Step S6 triggers a graded early warning based on the classification results, achieving closed-loop control from monitoring to intervention.

[0026] In one specific embodiment of this example, the multimodal physiological signal data further includes environmental parameter data, which includes at least one of light intensity data and sound intensity data. Environmental parameters are used to assess whether there are triggering factors in the care environment. Photosensitive epilepsy patients are sensitive to strong light stimulation, and collecting ambient light intensity data helps to adjust environmental conditions in a timely manner.

[0027] In one specific embodiment of this example, the process of performing principal component analysis on the EEG signal data in step S2 includes: S21, The EEG signal data is centered to obtain a centered EEG signal matrix; S22, calculate the covariance matrix of the centered EEG signal matrix; S23, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; S24, Identify artifact components based on the spectral characteristics of the feature vector; S25, after removing artifact components, the EEG signal is reconstructed to obtain the dimensionality-reduced EEG feature data.

[0028] Centralization processing involves subtracting the mean from each channel signal. The covariance matrix reflects the correlation between the signals of each channel. After eigenvalue decomposition, the source of each principal component is determined based on its spectral characteristics. Low-frequency, high-amplitude components correspond to electrooculogram artifacts, while high-frequency, noisy components correspond to electromyogram artifacts. After removing these artifact components, the signal is reconstructed to obtain pure EEG principal components.

[0029] In one specific embodiment of this example, the artifact components include electrooculography (EOG) artifact components and electromyography (EMG) artifact components. During monitoring, epileptic patients often experience EOG interference due to blinking and EMG interference due to muscle twitching. The spectral characteristics of these two types of artifacts differ from normal brain activity signals and can be identified and eliminated by analyzing the spectrum of the feature vectors.

[0030] In one specific embodiment of this example, wavelet packet transform is used to perform time-frequency decomposition on the dimensionality-reduced EEG feature data in step S3. Multiple frequency bands are included, such as the Delta band, Theta band, Alpha band, Beta band, and Gamma band. Wavelet packet transform can decompose non-stationary EEG signals into these five physiologically meaningful frequency bands. During epileptic seizures, energy in specific frequency bands will exhibit abnormal concentration.

[0031] In one specific embodiment of this example, the process of calculating the information entropy value based on the energy distribution of multiple frequency bands in step S4 includes: S41, calculate the energy value for each frequency band; S42, calculate the relative energy percentage of each frequency band based on the energy value of each frequency band, as a probability distribution; S43. Calculate the Shannon entropy based on the probability distribution to obtain the information entropy value.

[0032] Specifically, the energy value of each frequency band is equal to the sum of the squares of the wavelet coefficients of that band, the relative energy proportion is equal to the energy of that band divided by the total energy of all frequency bands, and the Shannon entropy is equal to the negative of the sum of the products of the relative energy proportions of each frequency band and their logarithms. Under normal conditions, EEG signals are chaotic and energy is dispersed across various frequency bands, resulting in a high entropy value; during an epileptic seizure, neurons fire synchronously, energy is concentrated in a specific frequency band, and the entropy value is significantly reduced.

[0033] In one specific implementation of this embodiment, the decision tree model includes a root node, second-level nodes, and third-level nodes. The decision feature of the root node is the rate of change of EEG entropy, the decision feature of the second-level nodes is the limb movement impact feature, and the decision feature of the third-level nodes is vital sign parameters. The root node first determines whether the entropy value has abruptly changed; if the entropy value drops sharply, it enters the abnormal branch. The second-level nodes determine whether there is limb convulsion, and the movement impact feature is represented by the accelerometer value. The third-level nodes further subdivide the state by combining vital signs such as blood oxygen and heart rate.

[0034] In one specific embodiment of this example, the current state includes one of the following: stable period, prodromal period, attack period, and post-attack period; the hazard level includes Level 1 hazard and Level 2 hazard, with Level 1 hazard corresponding to the prodromal period and Level 2 hazard corresponding to the attack period; step S6 further includes: when the hazard level is Level 1 hazard, controlling the environmental adjustment device to adjust environmental parameters; when the hazard level is Level 2 hazard, controlling the nursing equipment to perform protective operations and activate the alarm device. Level 1 hazard response includes sending a warning to nursing staff and dimming the ambient lights to reduce photosensitive stimulation; Level 2 hazard response includes triggering an audible and visual alarm, raising the bed rails to prevent falls, and starting an attack timer to monitor attack duration.

[0035] In one specific embodiment of this example, steps S1 to S6 are executed continuously using a sliding time window, with the time window duration ranging from 1 to 10 seconds. This sliding window mechanism enables the system to continuously monitor the dynamic changes in EEG entropy values, achieving continuous streaming computation and thus timely capturing premonitory symptoms.

[0036] Example 2 like Figure 2 As shown, the present invention also provides an epilepsy clinical nursing early warning system for implementing the above method, comprising: The data acquisition module is used to acquire multimodal physiological signal data from epilepsy patients; The signal processing module is used to perform principal component analysis on EEG signal data. The time-frequency analysis module is used to perform time-frequency decomposition on the dimensionality-reduced EEG feature data; The entropy calculation module is used to calculate information entropy values ​​based on the energy distribution of multiple frequency bands; The status determination module is used to determine the current status of epilepsy patients through a decision tree model. The early warning module is used to determine the level of potential hazards based on the current status and generate early warning information.

[0037] The data acquisition module includes an EEG acquisition unit and a motion sensing unit. The EEG acquisition unit acquires multi-channel EEG signals, and the motion sensing unit includes an accelerometer and a gyroscope to acquire acceleration and angular velocity data of limb movements.

[0038] Example 3 The core algorithm formula involved in this invention will be described in detail below through a specific embodiment.

[0039] In this embodiment, an 8-channel EEG acquisition device is used to monitor epilepsy patients. The sampling frequency is 256Hz, and each time window contains 1280 sampling points, i.e., a 5-second time window.

[0040] In the principal component analysis (PCA) stage, the observation matrix is ​​first constructed. Let the number of EEG signal channels be m, and the number of sampling points per channel be n. Then the observation matrix X is an m-row, n-column real matrix, as shown in the formula:

[0041] Where X is the observation matrix, Let m be a real matrix space with m rows and n columns, where m is the number of channels of the EEG signal and n is the number of sampling points for each channel.

[0042] In this embodiment, m equals 8, n equals 1280, and the dimension of the observation matrix X is 8 rows and 1280 columns. The element in the i-th row and t-th column of matrix X represents the voltage value of the i-th channel at the t-th sampling time.

[0043] The observation matrix is ​​centered. Let the original signal of the i-th channel be x. i(t) , where t represents the sampling time, and the value of t ranges from 1 to n. Calculate the mean of all sampling points in the i-th channel, denoted as E. [xi] E [xi] The value is equal to the sum of the voltage values ​​of all sampling points in the i-th channel divided by the number of sampling points n. The centered signal is denoted as... i(t) , i(t) =x i(t) -E [xi] After performing the above centralization operation on all m channels, a centralized matrix is ​​obtained. .

[0044] Calculate the covariance matrix. The covariance matrix is ​​denoted as C. x The formula is:

[0045] Among them, C x Here is the covariance matrix; n is the number of sampling points; T It is the transpose of the centered matrix.

[0046] Covariance matrix C x It is an m x m square matrix; in this embodiment, it is 8 x 8 columns. x The element in the i-th row and j-th column represents the covariance between the i-th channel and the j-th channel signal.

[0047] Perform eigenvalue decomposition on the covariance matrix. The decomposition yields the eigenvalue diagonal matrix Λ and the eigenvector matrix U, satisfying the following formula:

[0048]

[0049] Eigenvalue diagonal matrix The diagonal elements are , Until And arranged in descending order, each eigenvalue λ i For each eigenvector, the i-th column of the eigenvector matrix U is λ. i The corresponding feature vector.

[0050] Artifact components are identified based on the spectral characteristics of the eigenvectors. Spectral analysis is performed on the principal component signal corresponding to each eigenvector. Electrooculography (EOG) artifacts are characterized by low-frequency, large-amplitude components, with the main energy concentrated below 4Hz; electromyography (EMG) artifacts are characterized by high-frequency, chaotic components, with the main energy distributed above 30Hz. The eigenvalues ​​corresponding to the EOG and EMG artifacts are denoted as λartifact. A selection matrix W is constructed, in which the eigenvector columns corresponding to the artifacts are removed. The formula for the purified EEG principal component signal Y is:

[0051] In the time-frequency energy entropy calculation stage, the purified EEG principal component signal Y(t) is subjected to J-level wavelet packet decomposition. In this embodiment, J equals 3, i.e., a 3-level decomposition is performed. The formula is:

[0052] The wavelet coefficients of the k-th frequency band node in the j-th layer after decomposition are denoted as d. j,k (t), where j represents the decomposition level, and the value of j ranges from 1 to J; k represents the index of the frequency band node in that level; and t represents the index of the time sampling point. Three-level wavelet packet decomposition can decompose the signal into 8 frequency band nodes, corresponding to physiological frequency bands such as Delta, Theta, Alpha, Beta, and Gamma bands.

[0053] The specific formula for calculating the energy of each frequency band node is as follows:

[0054] Among them, E j,k Let be the energy of the k-th frequency band node in the j-th layer, j be the layer number of the wavelet packet decomposition, k be the index of the frequency band node in that layer, t be the index of the time sampling point, n be the total number of sampling points, and d be the energy of the k-th frequency band node in the j-th layer. j,k (t) represents the wavelet coefficients of the k-th frequency band node in the j-th layer at time t.

[0055] Calculate the relative energy percentage of each frequency band node.

[0056]

[0057] Where, p j,k E represents the relative energy percentage of the k-th frequency band node in the j-th layer. j,k E represents the energy of the k-th frequency band node in the j-th layer. j,k p represents the total energy of all frequency band nodes in the j-th layer, with a relative energy percentage of p. j,k This forms a probability distribution.

[0058] Calculate the time-frequency energy entropy using the Shannon entropy formula.

[0059]

[0060] Among them, H WPE Let p be the time-frequency energy entropy of the wavelet packet. j,k This represents the relative energy percentage of the k-th frequency band node in the j-th layer.

[0061] In this embodiment, when the patient is in a normal state, the EEG signal energy is distributed across various frequency bands. Assuming the relative energy proportions of the eight frequency band nodes are 0.15, 0.12, 0.13, 0.14, 0.11, 0.12, 0.11, and 0.12 respectively, H is calculated. WPE The entropy value is approximately 2.98, indicating a high entropy. When a patient enters the prodromal phase of epilepsy, the energy in the Theta band is abnormally concentrated. Assuming the relative energy percentages of the eight band nodes are 0.05, 0.60, 0.08, 0.07, 0.05, 0.06, 0.05, and 0.04, the calculated H... WPE The entropy value is approximately 2.08, indicating a significant decrease. The system detected that the entropy value dropped from 2.98 to 2.08, a decrease of about 30%. This decrease exceeds the threshold, and the root node of the decision tree determines that it has entered an abnormal branch, triggering the early warning process.

[0062] As can be seen from the above formula derivation process, principal component analysis achieves artifact removal and dimensionality reduction of EEG signals, wavelet packet decomposition achieves time-frequency analysis of signals, and Shannon entropy achieves quantitative characterization of the complexity of EEG signals. During an epileptic seizure, abnormal synchronous discharge of neurons leads to enhanced rhythmicity of EEG signals, manifested as energy concentration and entropy decrease in specific frequency bands. This invention utilizes this characteristic to achieve early warning of epileptic seizures.

[0063] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A clinical nursing early warning method for epilepsy, characterized in that, Includes the following steps: S1: Acquire multimodal physiological signal data of epilepsy patients, including electroencephalogram (EEG) signal data and limb movement signal data; S2: Perform principal component analysis on the EEG signal data to obtain dimensionality-reduced EEG feature data; S3: Perform time-frequency decomposition on the reduced-dimensional EEG feature data to obtain the energy distribution of multiple frequency bands; S4: Calculate the information entropy value based on the energy distribution of the multiple frequency bands; S5: Based on the information entropy value and the limb movement signal data, determine the current state of the epileptic patient using a decision tree model; S6: Determine the corresponding hazard level based on the current status and generate corresponding early warning information.

2. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: The multimodal physiological signal data also includes environmental parameter data, which includes at least one of light intensity data and sound intensity data.

3. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: In step S2, the process of performing principal component analysis on the EEG signal data includes: S21: The EEG signal data is centered to obtain a centered EEG signal matrix; S22: Calculate the covariance matrix of the centered EEG signal matrix; S23: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; S24: Identify artifact components based on the spectral characteristics of the feature vector; S25: After removing the artifact components, the EEG signal is reconstructed to obtain the dimensionality-reduced EEG feature data.

4. The method for early warning of epilepsy clinical nursing according to claim 3, characterized in that: The artifact components include electrooculography (EOG) artifact components and electromyography (EMG) artifact components.

5. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: In step S3, wavelet packet transform is used to perform time-frequency decomposition on the dimensionality-reduced EEG feature data. The multiple frequency bands include Delta band, Theta band, Alpha band, Beta band, and Gamma band.

6. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: In step S4, the process of calculating the information entropy value based on the energy distribution of the multiple frequency bands includes: S41: Calculate the energy value for each frequency band; S42: Calculate the relative energy percentage of each frequency band based on the energy value of each frequency band, and use it as a probability distribution; S43: Calculate the Shannon entropy based on the probability distribution to obtain the information entropy value.

7. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: The decision tree model includes a root node, a second-layer node, and a third-layer node. The decision feature of the root node is the rate of change of EEG information entropy, the decision feature of the second-layer node is the limb movement impact feature, and the decision feature of the third-layer node is vital sign parameters.

8. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: The current state includes one of the following: stable phase, prodromal phase, flare-up phase, and post-flare-up phase; The hazard levels include Level 1 hazards and Level 2 hazards. The Level 1 hazard corresponds to the current state when it is in the prodromal stage, and the Level 2 hazard corresponds to the current state when it is in the acute stage. S6 further includes: When the hazard level is Level 1, the environmental control equipment is used to adjust the environmental parameters. When the hazard level is Level II, the control nursing equipment will perform protective operations and activate the alarm device.

9. The method for early warning of epilepsy clinical nursing according to claim 1, characterized in that: S1-S6 are executed continuously using a sliding time window, the duration of which is 1 to 10 seconds.

10. An epilepsy clinical nursing early warning system, implementing the epilepsy clinical nursing early warning method as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire multimodal physiological signal data of epilepsy patients, including electroencephalogram (EEG) signal data and limb movement signal data. The signal processing module is used to perform principal component analysis on the EEG signal data to obtain dimensionality-reduced EEG feature data. The time-frequency analysis module is used to perform time-frequency decomposition on the dimensionality-reduced EEG feature data to obtain the energy distribution of multiple frequency bands; The entropy calculation module is used to calculate the information entropy value based on the energy distribution of the multiple frequency bands; The state determination module is used to determine the current state of the epileptic patient based on the information entropy value and the limb movement signal data through a decision tree model; The early warning module is used to determine the corresponding hazard level based on the current status and generate corresponding early warning information; The data acquisition module includes an EEG acquisition unit and a motion sensing unit. The EEG acquisition unit is used to acquire multi-channel EEG signals, and the motion sensing unit includes an accelerometer and a gyroscope.

Images