ICU patient risk early warning monitoring method and system based on time series data analysis

Abstract

The application discloses an ICU patient risk early warning monitoring method and system based on time series data analysis, relates to the technical field of medical information processing and artificial intelligence, and takes synchronously collected electroencephalogram time series signals and electrocardiogram time series signals as original inputs, extracts an autonomic nervous regulation signal from the electrocardiogram signals through heart rate variability analysis, carries out individualized multi-scale decomposition on the electroencephalogram signals and the heart rate variability signals by using adaptive wavelet packet decomposition, separates out a nervous frequency band component and an autonomic nervous frequency band component, extracts the instantaneous phase of each frequency band component by using Hilbert transformation, calculates a cross-frequency band phase locking index to quantize the brain-heart cross-system phase synchronization strength, continuously generates brain-heart coupling features along a time axis sliding window, compares the brain-heart coupling features with an established personalized baseline reference mode, identifies a coupling mode collapse sequence feature, generates an early warning signal when the feature is identified, and realizes early warning of delirium risk of ICU patients.

Description

Technical Field

[0001] This invention relates to the fields of medical information processing and artificial intelligence technology, specifically to a method and system for risk warning and monitoring of ICU patients based on time-series data analysis. Background Technology

[0002] Patients admitted to the Intensive Care Unit (ICU) are critically ill and require continuous monitoring of multiple physiological parameters to detect the risk of deterioration in a timely manner. Delirium is a common and severe acute brain dysfunction in ICU patients, clinically manifested as attention deficit, fluctuating level of consciousness, and cognitive impairment. It not only prolongs hospital stays and increases mortality but is also closely related to long-term cognitive impairment. Therefore, accurate and timely risk warning for ICU patients, especially early prediction of delirium, is an important issue that urgently needs to be addressed in the field of intensive care.

[0003] Currently, delirium assessment in the ICU mainly relies on subjective tools such as the ICU Confusion Assessment Scale, with intermittent bedside assessments conducted by medical staff. This approach has significant shortcomings: assessment depends on the patient's verbal cooperation, which is often impossible for ICU patients due to sedation, endotracheal intubation, etc., leading to assessment failure or unreliable results; assessment frequency is limited by human resources, making continuous monitoring difficult; subjective scales can only identify symptoms after they have become significant, missing the early intervention window; in addition, existing physiological monitoring methods mostly analyze EEG or heart rate variability in isolation, lacking dynamic quantification of the cross-system coupling relationship between the central nervous system and the autonomic nervous system, and failing to capture delirium prodromes in the subclinical stage of brain-heart dysfunction.

[0004] In summary, there is an urgent need for a method and system that can extract brain-heart cross-system synchronization features from EEG and ECG time-series data and issue early warnings when these features show early regularity breakdown. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for risk warning and monitoring of ICU patients based on time-series data analysis. This invention uses synchronously acquired EEG and ECG time-series signals as raw inputs. It extracts autonomic nervous regulation signals from the ECG signals through heart rate variability analysis. Adaptive wavelet packet decomposition is used to perform individualized multi-scale decomposition of the EEG and heart rate variability signals, separating neural and autonomic frequency band components. Hilbert transform is used to extract the instantaneous phase of each frequency band component, and the cross-frequency band phase-locking index is calculated to quantify the brain-heart cross-system phase synchronization strength. Brain-heart coupling features are continuously generated along a sliding window on the time axis. By comparing these brain-heart coupling features with a personalized baseline reference pattern established in the patient's delirium-free state, coupling pattern disintegration sequence features are identified. When such features are identified, a warning signal is generated. This transforms the subclinical decompensation process of brain-heart functional coupling into a continuously trackable and quantitatively analyzable dynamic feature, achieving objective and early warning of delirium risk in ICU patients and providing a lead time window for clinical intervention.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a risk warning and monitoring method for ICU patients based on time-series data analysis, the specific steps of which are as follows:

[0007] S100: Simultaneously acquires the patient's EEG and ECG timing signals;

[0008] S200. Perform heart rate variability analysis on the electrocardiogram timing signal to generate a heart rate variability signal as an autonomic nervous system regulation signal.

[0009] S300. Adaptive wavelet packet decomposition is used to perform multi-scale decomposition on the EEG time-series signal and the heart rate variability signal to obtain EEG multi-frequency segment sub-signals and autonomic nerve multi-frequency segment signals. Neural frequency band components are extracted from the EEG multi-frequency segment signals, and autonomic nerve frequency band components are extracted from the autonomic nerve multi-frequency segment signals.

[0010] S400. Perform instantaneous phase extraction on the neural frequency band component and the autonomic nerve frequency band component, and calculate the cross-band phase-locking index between them. Continuously calculate the cross-band phase-locking index along the time axis sliding window to generate brain-heart coupling features.

[0011] S500: Compare the brain-heart coupling features with the personalized baseline reference pattern, identify coupling pattern disintegration sequence features, and generate an early warning signal when the coupling pattern disintegration sequence features are identified.

[0012] Furthermore, in S200:

[0013] R-wave detection is performed on the electrocardiogram timing signal, and the time position corresponding to the R-wave peak in each cardiac cycle is marked to obtain the original RR interval sequence. The original RR interval sequence is composed of the time intervals between adjacent R-wave peaks arranged sequentially.

[0014] The original RR interval sequence is subjected to ectopic beat identification and removal processing. Abnormal RR intervals that do not conform to the normal sinus rhythm are removed from the sequence, and linear interpolation is performed to fill in the removed positions to obtain the corrected RR interval sequence.

[0015] The modified RR interval sequence is interpolated using a cubic spline interpolation method to convert the non-uniformly sampled RR interval sequence into a uniformly sampled continuous signal. The resampling frequency is set to 4Hz to generate the heart rate variability signal.

[0016] Furthermore, in S300, the adaptive wavelet packet decomposition process is as follows:

[0017] Based on the preset initial wavelet packet decomposition mother wavelet type and maximum decomposition level, construct a complete wavelet packet decomposition tree;

[0018] For each node in the complete wavelet packet decomposition tree, calculate the Shannon entropy value of the wavelet packet coefficient sequence corresponding to the node, where the Shannon entropy value is: ,in, The Shannon entropy value of the node signal. The wavelet packet decomposition coefficients of the nodes, The total number of coefficients for the nodes. For the first The energy normalization probability of each coefficient;

[0019] Starting from the bottom-level leaf nodes of the complete wavelet packet decomposition tree, the process traverses upwards layer by layer. For any parent node, it compares the sum of the entropy values ​​of the two child nodes when retaining them with the entropy value of the parent node when merging the two child nodes and using the parent node as a leaf node. If the sum of the entropy values ​​of the two child nodes is greater than the entropy value of the parent node, the two child nodes are pruned, and the parent node is retained as a new leaf node. If the sum of the entropy values ​​of the two child nodes is less than or equal to the entropy value of the parent node, the two child nodes are retained. The pruning process is recursively executed from bottom to top until the root node, forming the optimal wavelet packet decomposition tree structure that minimizes the total entropy of all leaf nodes in the entire tree.

[0020] Based on the optimal wavelet packet decomposition tree structure, the EEG time-series signal is decomposed to obtain EEG multi-frequency segment sub-signals corresponding to each leaf node of the optimal wavelet packet decomposition tree structure. The EEG multi-frequency segment sub-signals are divided according to the center frequency and bandwidth, respectively corresponding to multiple frequency band sub-signals in the δ band, θ band, α band, β band and γ band.

[0021] The heart rate variability signal is decomposed to obtain the autonomic neural multi-frequency segment sub-signals corresponding to each leaf node of the optimal wavelet packet decomposition tree structure. The autonomic neural multi-frequency segment sub-signals are divided according to the center frequency and bandwidth, respectively corresponding to multiple frequency segment sub-signals in the extremely low frequency band, low frequency band and high frequency band.

[0022] Furthermore, in S300, sub-signals whose frequency range falls within the preset neural activity frequency band range are extracted from the multi-band sub-signals of the EEG as neural frequency band components. The neural frequency band components include theta band sub-signals and alpha band sub-signals. The frequency range of the theta band sub-signals is 4Hz-8Hz, and the frequency range of the alpha band sub-signals is 8Hz-13Hz.

[0023] From the multi-frequency sub-signals of the autonomic nervous system, sub-signals whose frequency range falls within the preset autonomic nervous system modulation frequency band range are extracted as the autonomic nervous system frequency band components. The autonomic nervous system frequency band components include low-frequency sub-signals and high-frequency sub-signals. The frequency range of the low-frequency sub-signals is 0.04Hz-0.15Hz, and the frequency range of the high-frequency sub-signals is 0.15Hz-0.4Hz.

[0024] Furthermore, in S400, the instantaneous phase extraction process is as follows:

[0025] Hilbert transforms are applied to the neural frequency band component and the autonomic neural frequency band component respectively to construct analytical signals, and then the instantaneous phase sequences of the neural frequency band component and the autonomic neural frequency band component are extracted.

[0026] Based on the instantaneous phase sequence, according to Calculate the cross-band phase-locked index ,in, For cross-band phase-locked loop index, For the neural frequency band component at time... The instantaneous phase, For the autonomic neural frequency band components at time 10:00 The instantaneous phase, This is the preset number of sampling points contained within the sliding calculation window. The imaginary unit;

[0027] The sliding calculation window is slid along the time axis with a preset step size, and the cross-frequency band phase-locked index of each window is continuously calculated to obtain the phase-locked index sequence over time, which is used as the brain-heart coupling feature.

[0028] Furthermore, the cross-frequency band phase-locked index is calculated. At the same time, multiple sets of coupling pairs are constructed, including:

[0029] The theta band sub-signal in the neural frequency band component and the low-frequency band sub-signal in the autonomic nerve frequency band component are used as the first coupling pair to calculate the first cross-band phase-locking index.

[0030] The α-band sub-signal in the neural frequency band component and the high-frequency sub-signal in the autonomic nerve frequency band component are used as the second coupling pair to calculate the second cross-band phase-locking index.

[0031] The first cross-band phase-locked index and the second cross-band phase-locked index are together used as components of the brain-heart coupling feature.

[0032] Furthermore, in S500, the step of identifying the coupling mode disintegration sequence features is as follows:

[0033] In the initial stage after the patient is admitted to the ICU, when the patient is confirmed to be delirious through clinical assessment, EEG and ECG time-series signals of a preset duration are collected, and the initial brain-heart coupling characteristics in the delirious state are obtained according to S100-S400.

[0034] The cross-band phase-locking index values ​​of all time windows in the initial brain-heart coupling characteristics are statistically calculated to obtain the mean phase-locking index μ and the standard deviation of the phase-locking index σ. The mean phase-locking index μ and the standard deviation of the phase-locking index σ are used as the patient's personalized baseline reference pattern. The personalized baseline reference pattern reflects the individualized characteristic level of brain-heart cross-system phase synchronization of the patient in a normal delirium-free state.

[0035] After entering the real-time monitoring phase, the cross-band phase-locked index value is calculated for each current sliding time window. Execute the judgment logic:

[0036] Judgment condition one: ,in This is a preset multiplier. The value range is 1.5-3.0;

[0037] Judgment condition two: Continuous before the current sliding time window Within each sliding time window, the cross-frequency band phase-locked index value corresponding to each sliding time window satisfies the first judgment condition. The preset threshold for the number of continuous windows. The value range is 4-10;

[0038] Judgment condition three: Continuous The cross-frequency band phase-locked index value corresponding to each sliding time window It satisfies the monotonically non-increasing relation, that is... And there exists at least one pair of adjacent windows that satisfy This indicates that the phase-locked index is in the continuous Within each sliding time window, it exhibits a non-plateau-like, continuous downward trend;

[0039] Only when judgment condition one, judgment condition two and judgment condition three are simultaneously satisfied is it determined that the coupling mode disintegration sequence feature has been detected. The coupling mode disintegration sequence feature indicates that the cross-system phase synchronization between the patient's central nervous system and autonomic nervous system is in progress, and the brain-heart functional coupling has entered a pre-decompensation state.

[0040] Furthermore, in S500, the generated warning signal content includes:

[0041] The warning type identifier indicates that this warning is an early warning for delirium;

[0042] Current value of the cross-band phase-locked loop exponent for the current coupling pair;

[0043] The cross-frequency band phase-locked index Duration;

[0044] The warning signal is sent to the central monitoring station display interface through the hospital information system interface, where the warning information is displayed graphically, and simultaneously pushed to the mobile nursing terminal via wireless network.

[0045] On the other hand, an ICU patient risk warning and monitoring system based on time-series data analysis includes:

[0046] The signal acquisition module is used to simultaneously acquire the patient's EEG and ECG timing signals.

[0047] A heart rate variability analysis module, connected to the signal acquisition module, is used to perform heart rate variability analysis on the electrocardiogram timing signal and generate a heart rate variability signal as an autonomic nervous system regulation signal.

[0048] An adaptive wavelet packet decomposition and frequency band extraction module is connected to the signal acquisition module and the heart rate variability analysis module, respectively, and is used to perform adaptive wavelet packet decomposition on the EEG time-series signal and the heart rate variability signal to obtain EEG multi-frequency segment sub-signals and autonomic nerve multi-frequency segment sub-signals, and extract neural frequency band components and autonomic nerve frequency band components.

[0049] The cross-band phase synchronization analysis module, connected to the adaptive wavelet packet decomposition and frequency band extraction module, is used to perform instantaneous phase extraction on the neural frequency band component and the autonomic neural frequency band component, calculate the cross-band phase locking index, and generate brain-heart coupling features by sliding a window along the time axis.

[0050] The coupling mode breakdown identification and early warning module is connected to the cross-band phase synchronization analysis module. It is used to compare the brain-heart coupling characteristics with the personalized baseline reference pattern, identify coupling mode breakdown sequence characteristics, and generate an early warning signal when the characteristics are identified.

[0051] Compared with existing technologies, this ICU patient risk warning and monitoring method based on time-series data analysis has the following beneficial effects:

[0052] This invention employs an adaptive wavelet packet decomposition algorithm, using information entropy minimization as the cost function to search for the optimal wavelet packet decomposition tree structure from bottom to top. This allows the frequency band division of EEG temporal signals and heart rate variability signals to be automatically determined based on the signal's own characteristics, rather than using fixed frequency band segmentation. This adaptive mechanism can accurately separate individualized neural and autonomic frequency band components, avoiding signal distortion and frequency band aliasing problems caused by individual differences due to fixed frequency band segmentation. The adaptive decomposition can generate an optimal frequency band segmentation scheme that matches the physiological characteristics of each patient, thereby ensuring the purity and reliability of the frequency band components on which subsequent cross-band phase synchronization analysis is based. This guarantees the accuracy and repeatability of brain-heart coupling feature extraction, providing a solid and individualized signal processing foundation for early warning of delirium.

[0053] This invention applies the cross-band phase-locked index to the quantitative assessment of brain-heart cross-system coupling state. By calculating the synchronization strength between the instantaneous phase of the neural frequency band component and the instantaneous phase of the autonomic frequency band component, it quantifies the phase coordination relationship between the central nervous system and the autonomic nervous system, reveals the dynamic functional connection between cortical neural oscillations and autonomic rhythms, provides a clear neurophysiological mechanism explanation for delirium prediction, and elevates the brain-heart coupling state from qualitative judgment to continuous tracking quantitative monitoring, providing objective dynamic characteristics for subsequent pattern recognition and early warning decision-making.

[0054] This invention deconstructs sequence features through coupling mode, transforming the traditional single-moment threshold judgment into a sequential identification of the trend pattern of the phase-locked index across frequency bands. This sequential detection mechanism effectively suppresses false positive alarms caused by transient phase-locked index decline due to changes in patient position, brief physiological fluctuations, or signal noise, significantly improving the specificity of early warning. At the same time, this sequential detection logic can capture abnormalities in the early stage when the phase-locked index begins to decline continuously, enabling early prediction of delirium clinical symptoms.

[0055] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0057] Figure 1 A flowchart illustrating the steps of a risk warning and monitoring method for ICU patients based on time-series data analysis;

[0058] Figure 2 This is a data flow diagram of an ICU patient risk warning and monitoring system based on time-series data analysis.

[0059] Figure 3 This is a flowchart of the coupling mode disintegration sequence feature identification and early warning decision-making process in an embodiment of the present invention. Detailed Implementation

[0060] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0061] In the description of this invention, the term "signal" refers to an electrophysiological quantity that varies over time, the term "coupling" refers to the interaction between two signals or systems in the time or frequency domain, and the term "decomposition" refers to the process of breaking down an original signal into multiple sub-signals with different frequency components through mathematical transformation. Those skilled in the art will understand that the above terms have clear physical meanings and mathematical definitions in this invention and should not be construed as limiting the scope of protection of this invention. Furthermore, the technical features involved in the following embodiments can be combined with each other to form new implementation methods, provided that they do not conflict with each other.

[0062] The present invention provides a risk warning and monitoring method for ICU patients based on time-series data analysis, such as... Figure 1 As shown, the steps of this method are as follows:

[0063] S100: Simultaneously acquires the patient's EEG and ECG timing signals;

[0064] S200. Perform heart rate variability analysis on the electrocardiogram timing signal to generate a heart rate variability signal as an autonomic nervous system regulation signal.

[0065] S300. Adaptive wavelet packet decomposition is used to perform multi-scale decomposition on the EEG time-series signal and the heart rate variability signal to obtain EEG multi-frequency segment sub-signals and autonomic nerve multi-frequency segment signals. Neural frequency band components are extracted from the EEG multi-frequency segment signals, and autonomic nerve frequency band components are extracted from the autonomic nerve multi-frequency segment signals.

[0066] S400. Perform instantaneous phase extraction on the neural frequency band component and the autonomic nerve frequency band component, and calculate the cross-band phase-locking index between them. Continuously calculate the cross-band phase-locking index along the time axis sliding window to generate brain-heart coupling features.

[0067] S500: Compare the brain-heart coupling features with the personalized baseline reference pattern, identify coupling pattern disintegration sequence features, and generate an early warning signal when the coupling pattern disintegration sequence features are identified.

[0068] This invention enables early and objective warning of delirium risk by performing in-depth analysis and feature mining on the synchronously collected EEG and ECG time-series signals of ICU patients. The method constructs a complete processing link from signal acquisition, preprocessing, adaptive decomposition, feature extraction to pattern recognition, and transforms the brain-heart functional coupling state and its dynamic evolution process into quantifiable and traceable computational indicators, providing a new auxiliary decision-making tool for clinical practice.

[0069] On the other hand, in a preferred embodiment of the present invention, the ICU patient risk warning and monitoring system based on time-series data analysis includes: a signal acquisition module, a heart rate variability analysis module, an adaptive wavelet packet decomposition and frequency band extraction module, a cross-frequency band phase synchronization analysis module, and a coupling mode disintegration identification and warning module, wherein:

[0070] The signal acquisition module is used to simultaneously acquire the patient's EEG and ECG timing signals. The EEG timing signals are acquired using a multi-channel EEG machine via Ag / AgCl electrodes placed on the patient's head, with a sampling frequency set at no less than 256Hz to fully capture neurophysiological signals with an upper frequency limit below 100Hz. The ECG timing signals are acquired using standard three-lead or five-lead ECG monitoring electrodes attached to the corresponding location on the patient's chest, with a sampling frequency set at no less than 500Hz to accurately capture R-wave peaks and provide sufficient temporal resolution for subsequent RR interval extraction. All electrodes in contact with the patient's body are medical-grade low-impedance electrodes, and a hardware filter is installed at the acquisition front end to initially suppress power frequency interference and electromyographic noise.

[0071] The heart rate variability analysis module is connected to the signal acquisition module and is used to perform heart rate variability analysis on the electrocardiogram timing signal to generate a heart rate variability signal as an autonomic nervous system regulation signal.

[0072] The adaptive wavelet packet decomposition and frequency band extraction module is connected to the signal acquisition module and the heart rate variability analysis module, respectively, and is used to perform adaptive wavelet packet decomposition on the EEG time series signal and the heart rate variability signal to obtain EEG multi-frequency segment sub-signals and autonomic nerve multi-frequency segment sub-signals, and extract the neural frequency band components and the autonomic nerve frequency band components.

[0073] The cross-band phase synchronization analysis module is connected to the adaptive wavelet packet decomposition and frequency band extraction module. It is used to perform instantaneous phase extraction on the neural frequency band component and the autonomic neural frequency band component, calculate the cross-band phase locking index, and generate brain-heart coupling features by sliding a window along the time axis.

[0074] The coupling mode breakdown identification and early warning module is connected to the cross-band phase synchronization analysis module. It is used to compare the brain-heart coupling characteristics with the personalized baseline reference pattern, identify coupling mode breakdown sequence characteristics, and generate an early warning signal when identified.

[0075] In a preferred embodiment of the present invention, the specific implementation process of steps S100 and S200 is as follows:

[0076] After the system starts, it synchronously acquires the patient's EEG and ECG timing signals through the signal acquisition module, and records the acquired EEG timing signals as follows: ECG timing signals are denoted as ,in, Indicates the discrete sampling time;

[0077] The original signals were preprocessed. The EEG signals were filtered using a bandpass filter of 0.5Hz-70Hz to remove EEG and EMG artifacts. The ECG signals were filtered using a bandpass filter of 0.5Hz-40Hz to highlight the QRS complex.

[0078] Heart rate variability analysis was performed on the electrocardiogram (ECG) time-series signal, and the preprocessed ECG signal was analyzed. R-wave peak detection was performed, and the time and location of the R-wave peak in each cardiac cycle were recorded to form an R-wave position sequence. The time interval between two adjacent R-wave peaks is calculated to obtain the original RR interval sequence. ,in The original RR interval sequence contains abnormal RR intervals caused by premature beats and conduction blocks. This embodiment uses a combination of median filtering and thresholding to identify ectopic beats. The ratio of the current RR interval to the median of the two preceding and following RR intervals is calculated. If this ratio exceeds a preset threshold range, the RR interval is considered an abnormal value. If three or more consecutive RR intervals are marked as abnormal, it is determined to be a non-sinus rhythm segment, and all data in this segment is removed. For the locations of the removed abnormal RR intervals, linear interpolation is used to fill in the gaps. If the first... RR interval If a period is removed, its adjacent non-removed RR interval is used. and The average value is used instead, and after traversing the entire sequence, the corrected RR interval sequence is obtained. ;

[0079] The corrected RR interval sequence is non-uniformly distributed on the time axis. In this embodiment, a cubic spline interpolation method is used to convert the corrected RR interval sequence into a uniformly sampled continuous signal. A cubic spline interpolation function is constructed with the time of occurrence of each RR interval as the x-axis and the value of that RR interval as the y-axis. The interpolation function is then sampled at equal intervals on the time axis at a resampling frequency of 4Hz, thereby generating a uniformly sampled heart rate variability signal. This resampling frequency can fully preserve the effective information of the heart rate variability signal in the 0.003Hz-0.4Hz frequency band. As an autonomic nervous system regulatory signal, it is used to characterize the modulation activity of the sympathetic and vagus nerves on the sinoatrial node of the heart.

[0080] In a preferred embodiment of the present invention, the adaptive wavelet packet decomposition process in step S300 specifically includes:

[0081] Construct a complete wavelet packet decomposition tree, pre-setting the mother wavelet type and maximum decomposition level. In this embodiment, the mother wavelet uses the Daubechies series wavelet, specifically db8 or db10, and the maximum decomposition level is set accordingly. The frequency of the signal and the lowest frequency band to be extracted are determined based on the sampling frequency of the signal. For EEG signals with a sampling frequency of 256Hz, the following settings are used: The width of the lowest sub-band is Based on the preset mother wavelet and maximum decomposition level Construct a wavelet packet decomposition tree with a full binary tree structure, where the root node corresponds to the original signal, and each parent node is decomposed into low-frequency approximation child nodes and high-frequency detail child nodes;

[0082] For each node in the complete wavelet packet decomposition tree, calculate the Shannon entropy value of its corresponding wavelet packet coefficient sequence. For a given node, its wavelet packet decomposition coefficient sequence is: ,in, Given the total number of coefficients for this node, calculate the energy normalization probability for each coefficient: This probability reflects the first The proportion of each coefficient in the total energy of a node, and the entropy value of that node. ,like Then define Entropy The larger the entropy value, the more dispersed the energy distribution of the signal and the higher its complexity; the smaller the entropy value, the more concentrated the energy of the signal and the simpler its structure.

[0083] From the first complete tree Starting from the first layer, traverse upwards layer by layer. For any non-leaf parent node, execute the decision logic to obtain the left and right child nodes of that parent node, and calculate the total entropy value when retaining the child nodes. ,in, and These are the entropy values ​​of the left and right child nodes, respectively. Additionally, the entropy value is calculated when merging child nodes and making the parent node a leaf node. ,Compare and The size relationship, if This indicates that decomposing the signal into two child nodes increases the overall information complexity. In this case, these two child nodes should be removed, and the parent node should be retained as a new leaf node. If If the decomposition to child nodes reduces information complexity, then these two child nodes are retained. This comparison process is executed recursively from bottom to top until the root node. The final tree structure is the optimal wavelet packet decomposition tree structure that minimizes the sum of the entropy of all leaf nodes in the entire tree.

[0084] Based on the optimal wavelet packet decomposition tree structure, the EEG temporal signal is processed. Wavelet packet decomposition is performed. After decomposition, each leaf node of the tree corresponds to an independent EEG multi-band sub-signal. The center frequency and bandwidth of each sub-signal are uniquely determined by the depth and path of the node in the tree. According to the frequency band, they correspond to multiple frequency band sub-signals in the δ band, θ band, α band, β band and γ band respectively. The frequency range of the δ band is 0.5Hz-4Hz, the frequency range of the θ band is 4Hz-8Hz, the frequency range of the α band is 8Hz-13Hz, the frequency range of the β band is 13Hz-30Hz, and the frequency range of the γ band is 30Hz-100Hz.

[0085] The optimal wavelet packet decomposition tree structure is used to analyze the heart rate variability signal. The decomposition yields multi-frequency sub-signals of the autonomic nervous system corresponding to each leaf node of the tree structure. These sub-signals are divided according to their center frequency and bandwidth, corresponding to various frequency bands in the extremely low frequency band, low frequency band, and high frequency band. The frequency range of the extremely low frequency band is... The frequency range of the low-frequency band is The frequency range of the high-frequency band is .

[0086] In a preferred embodiment of the present invention, sub-signals whose frequency range falls within a preset neural activity frequency band range are extracted from the multi-band EEG sub-signals as the neural frequency band component. The neural frequency band component includes theta band sub-signals and alpha band sub-signals. Specifically, the operation involves: traversing all the EEG lobe node sub-signals obtained in the traversal step, calculating the center frequency of each sub-signal, and determining the sub-signals with a center frequency between 4Hz and 8Hz as... Frequency band sub-signals Sub-signals with center frequencies between 8Hz and 13Hz are identified as α-band sub-signals. If a frequency band in the optimal decomposition tree is further subdivided into narrower sub-bands, then these sub-signals that continuously cover that frequency band are superimposed to synthesize a complete frequency band. Frequency band signal or alpha band signal.

[0087] The autonomic nervous system extracts sub-signals whose frequency range falls within a preset autonomic nervous system modulation frequency band from the multi-frequency sub-signals. These sub-signals are used as the autonomic nervous system frequency band components. The autonomic nervous system frequency band components include low-frequency sub-signals and high-frequency sub-signals. Specifically, all obtained autonomic nervous system leaf node sub-signals are traversed, and sub-signals with a center frequency between 0.04Hz and 0.15Hz are identified as LF frequency band sub-signals. Sub-signals with center frequencies between 0.15Hz and 0.4Hz are identified as HF band sub-signals. If a frequency band is decomposed into multiple sub-bands, these sub-signals are synthesized to obtain a complete LF band signal or HF band signal.

[0088] In a preferred embodiment of the present invention, such as Figure 3 As shown, the specific implementation of step S400 is as follows:

[0089] The instantaneous phase of the narrowband signal is extracted using Hilbert transform. In this embodiment, the theta band EEG signal is used. For example, its Hilbert transform is defined as: ,in, Represents Cauchy's principal value integral, by and its Hilbert transformation Constructing an analytic signal: ,in, The imaginary unit, Instantaneous amplitude, This yields the desired instantaneous phase sequence. Similarly, Hilbert transforms are performed on the α-band sub-signal in the neural frequency band component, and the LF and HF sub-signals in the autonomic frequency band component, respectively, to obtain the corresponding instantaneous phase sequences. , and .

[0090] Construct two sets of cross-frequency band coupling pairs, the first coupling pair being... The instantaneous phase difference sequence between the frequency band neural signal and the LF band autonomic nerve signal is as follows: The second coupling pair consists of an α-band neural signal and an HF-band autonomic nerve signal, and their instantaneous phase difference sequence is as follows: For each coupling pair, the phase-locked index is calculated using the sliding time window method. Let the preset length of the sliding calculation window be... The number of sampling points contained in this window is ,but: ,in, The sampling frequency of the signal is given for all sampling points within the window. Calculate the PLV value of this window: , among which, among which, For cross-band phase-locked loop index, For the neural frequency band component at time... The instantaneous phase, For the autonomic neural frequency band components at time 10:00 The instantaneous phase, This is the preset number of sampling points contained within the sliding calculation window. The value of PLV is the imaginary unit, and its range is [0, 1]. When the phase difference between two signals is constant, PLV approaches 1; when the phase difference is randomly distributed, PLV approaches 0.

[0091] Along the time axis with a preset step size Slide the window, continuously calculate the PLV value for each window, and combine the PLV sequences of the first coupling pair. PLV sequence of the second coupling pair Together, they form part of the brain-heart coupling feature.

[0092] In a preferred embodiment of the present invention, in step S500, for each current sliding time window PLV value calculated in the real-time monitoring phase, a judgment logic is executed: In this embodiment, based on... Let's take a coupling pair as an example to illustrate. The judgment logic for coupled pairs is exactly the same;

[0093] Condition 1: Calculate the PLV value of the current window. To determine whether it is below the baseline level, the specific criteria are as follows: ,in, This is a preset multiplier coefficient used to control the sensitivity and specificity of the alarm. In this embodiment, The value range is 1.5-3.0.

[0094] Condition 2: The consecutive windows preceding the current window The PLV values ​​of all sliding windows satisfy the first judgment condition. This is a preset threshold for the number of continuous windows. In this embodiment, The value range is 4-10.

[0095] Judgment condition 3: For continuous conditions that satisfy condition 2 The sequence of PLV values ​​for each window, denoted as ,in Given the current window value, determine whether the sequence satisfies the monotonically non-increasing relation, i.e.: And there exists at least one pair of adjacent windows. and satisfy: This condition ensures that the PLV value exhibits a non-plateau-like continuous decline within a continuous window.

[0096] The coupling mode disintegration sequence feature is detected if and only if all three conditions are met simultaneously. Physiologically, this feature indicates a progressive loss of cross-system phase synchronization between the patient's central nervous system and autonomic nervous system, and that the brain-heart functional coupling has entered a pre-decompensation state, immediately generating a warning signal.

[0097] In a preferred embodiment of the present invention, after the warning signal is generated, the generated warning signal includes: a warning type identifier, indicating that this warning is an early warning of delirium; trigger source information, indicating the type of coupling pair that triggers the warning; current quantization index, providing the specific value of the cross-band phase-locked index of the current window; deviation degree, calculating and displaying the standard deviation multiple of the current PLV value below the baseline mean; and duration, recording and displaying the PLV value from the first trigger condition to the current time.

[0098] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A risk warning and monitoring method for ICU patients based on time-series data analysis, characterized in that, The steps of this method are as follows: S100: Simultaneously acquires the patient's EEG and ECG timing signals; S200. Perform heart rate variability analysis on the electrocardiogram timing signal to generate a heart rate variability signal as an autonomic nervous system regulation signal. S300. Adaptive wavelet packet decomposition is used to perform multi-scale decomposition on the EEG time-series signal and the heart rate variability signal to obtain EEG multi-frequency segment sub-signals and autonomic nerve multi-frequency segment signals. Neural frequency band components are extracted from the EEG multi-frequency segment signals, and autonomic nerve frequency band components are extracted from the autonomic nerve multi-frequency segment signals. In step S300, sub-signals whose frequency range falls within the preset neural activity frequency band range are extracted from the multi-band sub-signals of the EEG as neural frequency band components. The neural frequency band components include theta band sub-signals and alpha band sub-signals. The frequency range of the theta band sub-signals is 4Hz-8Hz, and the frequency range of the alpha band sub-signals is 8Hz-13Hz. From the multi-band sub-signals of the autonomic nervous system, sub-signals whose frequency range falls within the preset autonomic nervous system modulation frequency band range are extracted as the autonomic nervous system frequency band components. The autonomic nervous system frequency band components include low-frequency sub-signals and high-frequency sub-signals. The frequency range of the low-frequency sub-signals is 0.04Hz-0.15Hz, and the frequency range of the high-frequency sub-signals is 0.15Hz-0.4Hz. S400. Perform instantaneous phase extraction on the neural frequency band component and the autonomic nerve frequency band component, and calculate the cross-band phase-locking index between them. Continuously calculate the cross-band phase-locking index along the time axis sliding window to generate brain-heart coupling features. Calculate the cross-band phase-locked index At the same time, multiple sets of coupling pairs are constructed, including: The theta band sub-signal in the neural frequency band component and the low-frequency band sub-signal in the autonomic nerve frequency band component are used as the first coupling pair to calculate the first cross-band phase-locking index. The α-band sub-signal in the neural frequency band component and the high-frequency sub-signal in the autonomic nerve frequency band component are used as the second coupling pair to calculate the second cross-band phase-locking index. The first cross-band phase-locked index and the second cross-band phase-locked index are together considered as components of the brain-heart coupling feature; S500. The brain-heart coupling features are compared with the personalized baseline reference pattern to identify coupling pattern disintegration sequence features. When the coupling pattern disintegration sequence features are identified, a delirium risk warning signal is generated.

2. The ICU patient risk warning and monitoring method based on time-series data analysis according to claim 1, characterized in that, In S200: R-wave detection is performed on the electrocardiogram timing signal, and the time position corresponding to the R-wave peak in each cardiac cycle is marked to obtain the original RR interval sequence; The corrected RR interval sequence is interpolated using a cubic spline interpolation method to convert the non-uniformly sampled RR interval sequence into a uniformly sampled continuous signal, thereby generating the heart rate variability signal.

3. The ICU patient risk warning and monitoring method based on time-series data analysis according to claim 1, characterized in that, In S300, the adaptive wavelet packet decomposition process is as follows: Based on the preset initial wavelet packet decomposition mother wavelet type and maximum decomposition level, construct a complete wavelet packet decomposition tree; For each node in the complete wavelet packet decomposition tree, calculate the Shannon entropy value of the wavelet packet coefficient sequence corresponding to the node, where the Shannon entropy value is: ,in, The Shannon entropy value of the node signal. The wavelet packet decomposition coefficients of the node are... The total number of coefficients for the nodes. For the first The energy normalization probability of each coefficient; Based on the optimal wavelet packet decomposition tree structure, the EEG time-series signal is decomposed to obtain EEG multi-frequency segment sub-signals corresponding to each leaf node of the optimal wavelet packet decomposition tree structure. The EEG multi-frequency segment sub-signals are divided according to the center frequency and bandwidth, respectively corresponding to multiple frequency band sub-signals in the δ band, θ band, α band, β band and γ band. The heart rate variability signal is decomposed to obtain the autonomic neural multi-frequency segment sub-signals corresponding to each leaf node of the optimal wavelet packet decomposition tree structure. The autonomic neural multi-frequency segment sub-signals are divided according to the center frequency and bandwidth, respectively corresponding to multiple frequency segment sub-signals in the extremely low frequency band, low frequency band and high frequency band.

4. The ICU patient risk warning and monitoring method based on time-series data analysis according to claim 1, characterized in that, In S400, the instantaneous phase extraction process is as follows: Hilbert transforms are applied to the neural frequency band component and the autonomic neural frequency band component respectively to construct analytical signals, and then the instantaneous phase sequences of the neural frequency band component and the autonomic neural frequency band component are extracted. Based on the instantaneous phase sequence, according to Calculate the cross-band phase-locked index ,in, For cross-band phase-locked loop index, For the neural frequency band component at time... The instantaneous phase, For the autonomic neural frequency band components at time 10:00 The instantaneous phase, This is the preset number of sampling points contained within the sliding calculation window. The imaginary unit; The sliding calculation window is slid along the time axis with a preset step size, and the cross-frequency band phase-locked index of each window is continuously calculated to obtain the phase-locked index sequence over time, which is used as the brain-heart coupling feature.

5. The ICU patient risk warning and monitoring method based on time-series data analysis according to claim 1, characterized in that, In S500, the identification steps for coupling mode disintegration sequence features are as follows: In the initial stage after the patient is admitted to the ICU, when the patient is confirmed to be delirious through clinical assessment, EEG and ECG time-series signals of a preset duration are collected, and the initial brain-heart coupling characteristics in the delirious state are obtained according to S100-S400. The phase-locked index values ​​across all time windows in the initial brain-heart coupling characteristics are statistically calculated to obtain the mean phase-locked index μ and the standard deviation of the phase-locked index σ. The mean phase-locked index μ and the standard deviation of the phase-locked index σ are used as the patient's personalized baseline reference pattern. After entering the real-time monitoring phase, the cross-band phase-locked index value is calculated for each current sliding time window. Execute the judgment logic: Judgment condition one: ,in This is a preset multiplier. The value range is 1.5-3.0; Judgment condition two: Continuous before the current sliding time window Within each sliding time window, the cross-frequency band phase-locked index value corresponding to each sliding time window satisfies the first judgment condition. The preset threshold for the number of continuous windows. The value range is 4-10; Judgment condition three: Continuous The cross-frequency band phase-locked index value corresponding to each sliding time window It satisfies the monotonically non-increasing relation, that is... And there exists at least one pair of adjacent windows that satisfy ; The coupling mode disintegration sequence feature is determined to be detected only when all three conditions are met simultaneously.

6. The ICU patient risk warning and monitoring method based on time-series data analysis according to claim 1, characterized in that, In S500, the generated warning signal content includes: The warning type identifier indicates that this warning is an early warning for delirium; Current value of the cross-band phase-locked loop exponent for the current coupling pair; The cross-frequency band phase-locked index Duration; The warning signal is sent to the central monitoring station display interface through the hospital information system interface, where the warning information is displayed graphically, and simultaneously pushed to the mobile nursing terminal via wireless network.

7. An ICU patient risk warning and monitoring system based on time-series data analysis, applicable to the ICU patient risk warning and monitoring method based on time-series data analysis as described in any one of claims 1-6, characterized in that, The system includes: The signal acquisition module is used to simultaneously acquire the patient's EEG and ECG timing signals. A heart rate variability analysis module, connected to the signal acquisition module, is used to perform heart rate variability analysis on the electrocardiogram timing signal and generate a heart rate variability signal as an autonomic nervous system regulation signal. An adaptive wavelet packet decomposition and frequency band extraction module is connected to the signal acquisition module and the heart rate variability analysis module, respectively, and is used to perform adaptive wavelet packet decomposition on the EEG time-series signal and the heart rate variability signal to obtain EEG multi-frequency segment sub-signals and autonomic nerve multi-frequency segment sub-signals, and extract neural frequency band components and autonomic nerve frequency band components. The cross-band phase synchronization analysis module, connected to the adaptive wavelet packet decomposition and frequency band extraction module, is used to perform instantaneous phase extraction on the neural frequency band component and the autonomic neural frequency band component, calculate the cross-band phase locking index, and generate brain-heart coupling features by sliding a window along the time axis. The coupling mode breakdown identification and early warning module is connected to the cross-band phase synchronization analysis module. It is used to compare the brain-heart coupling characteristics with the personalized baseline reference pattern, identify coupling mode breakdown sequence characteristics, and generate an early warning signal when the characteristics are identified.

Images