Artificial intelligence-based nervous system data processing method and device and storage medium

Abstract

The invention relates to the technical field of neural signal processing, in particular to a nervous system data processing method and device based on artificial intelligence and a storage medium. The method comprises the following steps: carrying out preprocessing and frequency band decomposition on nervous system data to generate a frequency band data matrix; constructing a phase analysis model based on the frequency band data matrix to generate a functional connection matrix; performing network topology feature extraction on the function connection matrix to obtain a data feature vector; constructing a dynamic evolution model based on the data feature vector to synchronize the stability index; and carrying out multi-layer criterion fusion and abnormity early warning on the synchronization stability index. According to the invention, high-specificity early warning of the abnormal nerve function state is realized.

Description

Technical Field

[0001] This invention relates to the field of neural signal processing technology, and in particular to a method, apparatus and storage medium for neural system data processing based on artificial intelligence. Background Technology

[0002] In recent years, the rapid development of neuroscience and brain-computer interface technology has made it possible to acquire neural system signals with multi-channel, high sampling rates. The massive amounts of neurophysiological data provide an unprecedented foundation for understanding brain functional networks. Computational methods such as functional connectivity analysis and network topology modeling have gradually become important tools for studying brain information processing and disease mechanisms. The introduction of artificial intelligence and machine learning technologies has provided new avenues for extracting dynamic patterns and abnormal features from complex neural data, driving the development of neural function monitoring towards intelligence, personalization, and early warning.

[0003] Existing neural signal processing methods are often based on static or single-band analysis, which makes it difficult to fully characterize the dynamic and coordinated characteristics of brain functional connections; network feature extraction often relies on fixed thresholds or general templates, lacks individual adaptability, and is easily affected by noise and baseline fluctuations; anomaly detection often uses a single criterion or simple threshold, resulting in insufficient specificity of early warning and a high false alarm rate; in addition, traditional methods have limited ability to identify spatiotemporally persistent anomalies, making it difficult to achieve reliable multi-dimensional anomaly fusion judgment in the early stages. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, and storage medium for processing neural system data based on artificial intelligence, so as to solve at least one of the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An artificial intelligence-based method for processing neural system data includes: The nervous system data is preprocessed and decomposed into frequency bands to generate a frequency band data matrix; A phase analysis model is constructed based on the frequency band data matrix to generate a functional connectivity matrix; Network topology features are extracted from the functional connectivity matrix to obtain data feature vectors; A dynamic evolution model is constructed based on data feature vectors, with a synchronous stability index; Multi-level criteria fusion and anomaly early warning are applied to the synchronization stability index.

[0006] Furthermore, the continuously acquired data stream is divided into non-overlapping segments with a fixed duration of 30 seconds to obtain multiple data segments. Each data segment constitutes a data matrix X with dimensions [C, N], where C is the number of channels and N is the number of time points. The preprocessed neural system data of each channel is decomposed into K preset physiological frequency bands, so that each data segment generates K frequency band data matrices, all with dimensions [C, N], where K represents the number of preset frequency bands.

[0007] Furthermore, Hilbert transform is applied to each frequency band data matrix to calculate the instantaneous phase of each channel and each time point. The phase difference sequence between any two channels in the frequency band data matrix corresponding to each preset physiological frequency band is analyzed, and the phase lock value within the data segment of the two channels is analyzed. All channel pairs in the frequency band data matrix corresponding to each preset physiological frequency band are traversed, and the phase lock value between each pair is calculated to obtain a symmetric C×C matrix with diagonal elements of 0, which is used as the functional connection matrix.

[0008] Furthermore, the functional connectivity matrix is ​​thresholded by setting elements with phase-locked values ​​greater than or equal to the connectivity threshold to 1 and elements with phase-locked values ​​less than the connectivity threshold to 0, thereby generating a binary adjacency matrix. Based on the binary adjacency matrix, network topology features are extracted from the functional connectivity matrix to obtain the average connectivity strength, clustering coefficient, feature path length, and small-world scalar. For a data segment, a total of 4×K network topology features are extracted across K frequency bands to form a data feature vector.

[0009] Furthermore, the sliding window size is set to 5 seconds and the sliding step size is 2.5 seconds. The data segment is truncated into multiple short time windows, and short-time feature analysis is performed on the short time window data to obtain short-time feature vectors. For the 4×K network topology features involved in the short-time feature vectors, a time series of length M is generated for each network topology feature according to the truncating order of the sliding window, and this is used as the feature time series. Set up a fluctuation observation window containing 10 short-term windows of data. For each short-term eigenvector, calculate its mean and standard deviation within the fluctuation observation window, and use the ratio of the calculated standard deviation to the mean as the local coefficient of variation.

[0010] Furthermore, the maximum value of the local coefficient of variation within 1 hour is extracted as the historical maximum value of variation, and the minimum value of the local coefficient of variation within 1 hour is extracted as the historical minimum value of variation. The synchronization stability index is then analyzed based on the historical maximum value of variation, the historical minimum value of variation, and the local coefficient of variation.

[0011] Furthermore, during a clearly defined health monitoring period for the user, at least 20 baseline data segments of varying lengths are collected to establish the user's normal characteristic range. While establishing the user's normal characteristic range, the data feature vector of each baseline data segment is analyzed, and the average value μ of each feature is calculated. base and standard deviation σ base The normal feature range for each feature is defined as [μ]. base -2×σ base ,μ base +2×σ base ].

[0012] Furthermore, each feature value in the data feature vector is examined, and feature values ​​that do not belong to their corresponding normal feature range are regarded as abnormal feature values. When the number of abnormal feature values ​​is greater than or equal to 3 and the abnormal feature values ​​come from at least 3 different preset physiological frequency bands, a prompt for multi-frequency band feature joint synchronization deviation is triggered, and an abnormal warning is issued. The number of dimensions with a synchronization stability index less than 0.3 is counted and considered as the number of abnormal dimensions. When the number of abnormal dimensions is greater than or equal to (2×K), a warning is triggered indicating that the synchronization stability of multiple dimensions has been lost and an anomaly warning is issued. When the combined synchronization deviation of multi-band features and the loss of multi-dimensional synchronization stability trigger a prompt within 30 seconds, the functional connection matrix corresponding to the data of the two short time windows before and after the prompt is extracted and used as the abnormal connection matrix. The relative change of each channel pair in the abnormal connection matrix with its corresponding baseline mean is calculated. The relative changes are sorted in descending order, and the channel pairs corresponding to the first 10% and the last 10% of the relative changes are taken as abnormal channels. The abnormal channels are projected onto the two-dimensional scalp plane, and the spatial density clustering algorithm is applied to analyze the distribution of abnormal channels. The neighborhood radius is set to 1.5 times the average distance between electrodes, and the minimum number of points required to form a cluster is 3. When there is more than or equal to 1 cluster, and the number of channels contained in the largest cluster accounts for more than 40% of the number of abnormal channels, it is determined to have spatial clustering. When the cluster with spatial clustering continues to exist for 3 consecutive short time windows, a prompt for spatiotemporally persistent anomalies is triggered, and the highest level of anomaly warning is issued.

[0013] On the other hand, the present invention also provides an artificial intelligence-based neural system data processing device, comprising: The data decomposition unit is used to preprocess and decompose the neural system data to generate a frequency band data matrix; The phase analysis unit is used to construct a phase analysis model based on the frequency band data matrix to generate a functional connectivity matrix; The feature extraction unit is used to extract network topology features from the functional connectivity matrix to obtain data feature vectors; Evolutionary analysis unit, used to construct dynamic evolutionary models based on data feature vectors, to synchronize stability indices; Anomaly warning unit is used to perform multi-level criterion fusion and anomaly warning for the synchronization stability index.

[0014] On the other hand, the present invention also provides a storage medium characterized in that it stores instructions that, when run on a computer, cause the computer to execute the artificial intelligence-based neural system data processing method as described in any of the preceding claims.

[0015] The beneficial effects of this invention are as follows: By integrating multi-band phase analysis, network topology feature extraction, and dynamic evolution modeling, a high-dimensional characterization and dynamic monitoring of neural functional connectivity is achieved. This method possesses a multi-layer criterion fusion mechanism, enabling anomaly judgment from multiple dimensions such as feature deviation, instability loss, and spatiotemporal clustering. This significantly improves the specificity of identifying abnormal neural functional states and the timeliness of early warning. The system supports individualized baseline modeling, effectively overcoming individual differences and environmental interference. It is suitable for long-term health monitoring and early warning of neurological dysfunction, possessing high clinical application value and promising prospects for wider application. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the artificial intelligence-based neural system data processing method in this embodiment.

[0018] Figure 2 This is a flowchart of the method for constructing the phase analysis model in this embodiment.

[0019] Figure 3 This is a flowchart of the method for constructing the dynamic evolution model in this embodiment.

[0020] Figure 4 This is a schematic diagram of the structure of the artificial intelligence-based neural system data processing device in this embodiment. Detailed Implementation

[0021] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a further detailed account of the artificial intelligence-based neural system data processing method, apparatus, and storage medium disclosed in this invention. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated; they can be combined to achieve better technical effects. In the accompanying drawings of the following embodiments, the same reference numerals in each drawing represent the same features or components, which can be applied to different embodiments. Therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0022] It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are merely for illustrative purposes and to aid those skilled in the art in understanding and reading the invention. They are not intended to limit the conditions under which the invention can be implemented. Any modifications to the structure, changes in proportions, or adjustments to size, provided they do not affect the effectiveness or purpose of the invention, should fall within the scope of the technical content disclosed in the invention. The scope of the preferred embodiments of the present invention includes other implementations, wherein functions may be performed not in the order stated or discussed, including substantially simultaneously or in reverse order, depending on the functions involved. This should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0023] Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0024] In the description of the embodiments of this application, " / " means "or", and "and / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" means: A and B exist alone, B exists alone, and A and B exist simultaneously. In the description of the embodiments of this application, "multiple" refers to two or more embodiments.

[0025] Please see Figure 1 As shown, this is an artificial intelligence-based neural system data processing method in this embodiment, including: Step S1 involves preprocessing and decomposing the neural system data to generate a frequency band data matrix.

[0026] Specifically, in step S1 of this embodiment, the nervous system data is the user's multi-channel scalp EEG signal with a sampling frequency of 1000Hz. The continuously acquired data stream is divided into non-overlapping segments for a fixed duration of 30 seconds to obtain multiple data segments. Each data segment constitutes a data matrix X with dimensions [C,N], where C is the number of channels and N is the number of time points. The number of time points is equal to the product of the segmentation duration and the sampling frequency.

[0027] Specifically, in step S1 of this embodiment, the data matrix is ​​standardized and preprocessed, the time series of each channel is linearly fitted, and the fitted curve is subtracted to eliminate slow baseline drift. A zero-phase digital filter is used to bandpass filter the signal of each channel to retain the frequency band closely related to neural activity.

[0028] Specifically, in step S1 of this embodiment, the preprocessed neural system data of each channel is decomposed into K preset physiological frequency bands, so that each data segment generates K frequency band data matrices, all of which have dimensions [C,N], where K represents the number of preset frequency bands.

[0029] Specifically, in this embodiment, the number of preset frequency bands is set to 5, and the corresponding preset physiological frequency bands are: 0.5-4Hz, 4-8 Hz, 8-13 Hz, 13-30Hz, and 30-45Hz.

[0030] Specifically, in step S1 of this embodiment, by standardizing, de-drifting, and bandpass filtering the original neural signal, the interference of low-frequency noise and baseline drift on subsequent analysis is effectively eliminated. Furthermore, the signal is decomposed into multiple physiological frequency bands, preserving the frequency domain information related to different neural activity patterns. This lays the data foundation for subsequent multi-band phase analysis and functional connectivity modeling. This processing method enhances the ability to analyze the frequency domain characteristics of neural signals and improves the granularity of the representation of EEG activity under different functional states.

[0031] Please continue reading. Figure 1 As shown, the artificial intelligence-based neural system data processing method further includes: Step S2: Construct a phase analysis model based on the frequency band data matrix to generate a functional connectivity matrix.

[0032] Please see Figure 2 As shown, this is a method for constructing a phase analysis model, including: Step S21: Extract the instantaneous phase from the frequency band data matrix to obtain the instantaneous phase.

[0033] Specifically, in step S21 of this embodiment, the instantaneous phase of each channel and each time point is calculated by applying Hilbert transform to each frequency band data matrix.

[0034] Specifically, in step S21 of this embodiment, when calculating the instantaneous phase of each channel and each time point, for example, performing a Hilbert transform on the frequency band data matrix of channel c to obtain its analytic signal z. c If (t), then the instantaneous phase of the channel at time t is: φ c (t)=arg(z c (t)), where t represents the time point number.

[0035] Please continue reading. Figure 2 As shown, the method for constructing the phase analysis model further includes: Step S22: Analyze the phase lock value based on the instantaneous phase at each time point.

[0036] Specifically, in step S22 of this embodiment, the phase difference sequence Δφ between any two channels in the frequency band data matrix corresponding to each preset physiological frequency band is analyzed. ci,cj (t), Δφ ci,cj (t)=φ ci (t)-φ cj (t), where ci and cj represent any two channels in the frequency band data matrix, and the phase lock value within the data segment of the above two channels is analyzed. The expression for the phase lock value is: Among them, PLV ci,cj This indicates the phase lock value.

[0037] Please continue reading. Figure 2 As shown, the method for constructing the phase analysis model further includes: Step S23: Construct a functional connection matrix based on the phase-locked value.

[0038] Specifically, in step S23 of this embodiment, all channel pairs in the frequency band data matrix corresponding to each preset physiological frequency band are traversed, and the phase lock value between each pair is calculated to obtain a symmetrical C×C matrix with diagonal elements of 0, which is then used as the functional connection matrix. The number of channel pairs is C×(C-1) / 2.

[0039] Specifically, in step S2 of this embodiment, the instantaneous phase of each channel is extracted by Hilbert transform, and the phase lock value between channels is calculated to construct a functional connectivity matrix that reflects the degree of synchronization between different brain regions. This method overcomes the limitation of traditional amplitude correlation analysis in making insufficient use of phase information, and can more sensitively capture the coordinated changes between neural oscillations, providing a quantitative basis for subsequent network topology analysis that reflects the strength of brain functional coupling.

[0040] Please continue reading. Figure 1 As shown, the artificial intelligence-based neural system data processing method further includes: Step S3: Extract network topology features from the functional connection matrix to obtain data feature vectors.

[0041] Specifically, in step S3 of this embodiment, the functional connection matrix is ​​thresholded by setting elements in the functional connection matrix whose phase lock value is greater than or equal to the connection threshold to 1, and elements in the functional connection matrix whose phase lock value is less than the connection threshold to 0, so as to generate a binary adjacency matrix.

[0042] Specifically, in this embodiment, the connection threshold is determined based on the values ​​of each element in the functional connection matrix. The elements in the functional connection matrix are sorted in descending order, and the value corresponding to the 20th element is taken as the connection threshold.

[0043] Specifically, in step S3 of this embodiment, network topology features are extracted from the functional connectivity matrix based on the binary adjacency matrix to obtain the average connectivity strength, clustering coefficient, feature path length, and small-world scalar. For a data segment, a total of 4×K network topology features are extracted across K frequency bands to form a data feature vector.

[0044] Specifically, in step S3 of this embodiment, the average connection strength is the average value of the phase-locked values ​​of the functional connection matrices corresponding to all elements that are 1 in the binary adjacency matrix.

[0045] Specifically, in step S3 of this embodiment, the expression for the clustering coefficient is: D ci =2×E ci / [K ci ×(K ci -1)], where D ci K represents the clustering coefficient. ci E represents the number of connections for channel ci. ci K represents the channel ci ci The actual number of edges between each neighbor.

[0046] Specifically, in step S3 of this embodiment, the feature path length is the average of the shortest paths between all channel pairs in the functional connection matrix.

[0047] Specifically, in step S3 of this embodiment, the small-world scalar is a parameter that reflects whether the functional connectivity matrix has small-world properties. The small-world properties are high clustering coefficients and short feature path lengths. When the small-world scalar is greater than 1, it indicates that the functional connectivity matrix has small-world properties. The analysis method for the small-world scalar is as follows: generate 100 random networks with the same number of channels and edges as the functional connectivity matrix, compare the functional connectivity matrix with the generated random networks, and calculate the average clustering coefficient and average path length between the functional connectivity matrix and the random networks. Small-world scalar = (average clustering coefficient / clustering coefficient) / (average path length / feature path length).

[0048] Specifically, in step S3 of this embodiment, the functional connectivity matrix is ​​converted into a binary adjacency matrix through thresholding, and then multiple network topology indicators such as average connectivity strength, clustering coefficient, feature path length, and small-world scalar are extracted. These features characterize the integration and separation characteristics of the brain's functional network from both global and local perspectives, forming a high-dimensional feature vector, which provides rich network state description information for subsequent dynamic evolution analysis and anomaly detection.

[0049] Please continue reading. Figure 1 As shown, the artificial intelligence-based neural system data processing method further includes: Step S4: Construct a dynamic evolution model based on the data feature vectors to synchronize the stability index.

[0050] Please see Figure 3 As shown, this is a method for constructing a dynamic evolution model, including: Step S41: Construct a sliding window to generate feature time series.

[0051] Specifically, in step S41 of this embodiment, the window size of the sliding window is set to 5 seconds and the sliding step is 2.5 seconds. The data segment is truncated into multiple short-time window data, and short-time feature analysis is performed on the short-time window data to obtain short-time feature vectors. For the 4×K network topology features involved in the short-time feature vectors, a time series of length M is generated for each network topology feature according to the truncating order of the sliding window, and it is used as the feature time series. M = (segmentation duration - window size of the sliding window) / sliding step + 1.

[0052] Specifically, in step S41 of this embodiment, performing short-term feature analysis on the short-term window data involves performing the analysis of the short-term window data in steps S1-S3 above to generate a short-term feature vector.

[0053] Please continue reading. Figure 3 As shown, the method for constructing the dynamic evolution model further includes: Step S42: Determine the recent volatility of the characteristic time series in order to determine the local coefficient of variation.

[0054] Specifically, in step S42 of this embodiment, a fluctuation observation window containing 10 short-term window data is set. For each short-term feature vector, its mean and standard deviation within the fluctuation observation window are calculated, and the ratio of the calculated standard deviation to the mean is used as the local coefficient of variation.

[0055] Please continue reading. Figure 3 As shown, the method for constructing the dynamic evolution model further includes: Step S43: Analyze the synchronization stability index based on the local coefficient of variation.

[0056] Specifically, in step S43 of this embodiment, the maximum value of the local coefficient of variation within 1 hour is extracted as the historical maximum value of variation, and the minimum value of the local coefficient of variation within 1 hour is extracted as the historical minimum value of variation. The synchronization stability index is analyzed based on the historical maximum value of variation, the historical minimum value of variation, and the local coefficient of variation. The synchronization stability index is 1 - (local coefficient of variation - historical minimum value of variation) / (historical maximum value of variation - local coefficient of variation).

[0057] Specifically, in step S4 of this embodiment, a sliding window mechanism is introduced to convert long-term signals into short-term feature sequences. By calculating local coefficients of variation and historical extreme values, a synchronous stability index reflecting the fluctuation of each feature over time is constructed. This model can dynamically track the time-varying characteristics of the brain's functional network, identify patterns of instability loss or abnormal fluctuations, and enhance the ability to monitor transient or progressive changes in neural function.

[0058] Please continue reading. Figure 1 As shown, the artificial intelligence-based neural system data processing method further includes: Step S5 involves multi-level criterion fusion and anomaly warning for the synchronization stability index.

[0059] Specifically, in step S5 of this embodiment, during a clearly defined health monitoring period for the user, at least 20 baseline data segments of a specified length are collected to establish the user's normal characteristic range. While establishing the user's normal characteristic range, the data feature vector of each baseline data segment is analyzed, and the average value μ of each feature is calculated. base and standard deviation σ base The normal feature range for each feature is defined as [μ]. base -2×σ base ,μ base +2×σ base ].

[0060] Specifically, in step S5 of this embodiment, each feature value in the data feature vector is checked, and feature values ​​in the data feature vector that do not belong to their corresponding normal feature range are taken as abnormal feature values. When the number of abnormal feature values ​​is greater than or equal to 3 and the abnormal feature values ​​come from at least 3 different preset physiological frequency bands, a prompt for multi-frequency band feature joint synchronization deviation is triggered, and an abnormal warning is issued.

[0061] Specifically, in step S5 of this embodiment, the number of dimensions with a synchronization stability index less than 0.3 is counted and used as the number of abnormal dimensions. When the number of abnormal dimensions is greater than or equal to (2×K), a prompt is triggered indicating the loss of multi-dimensional synchronization stability coordination, and an abnormal warning is issued.

[0062] Specifically, in step S5 of this embodiment, when the multi-band feature joint synchronization deviation and multi-dimensional synchronization stability collaborative loss are successively triggered within 30 seconds, the functional connection matrix corresponding to the data of the two short time windows before and after the triggering prompt time is extracted and used as the abnormal connection matrix. The relative change of each channel pair in the abnormal connection matrix with its corresponding baseline mean is calculated. The relative change is sorted in descending order, and the channel pairs corresponding to the first 10% and the last 10% of the relative change are taken as abnormal channels. The abnormal channels are projected onto the two-dimensional scalp plane, and the spatial density clustering algorithm is applied to analyze the distribution of abnormal channels. The neighborhood radius is set to 1.5 times the average distance between electrodes, and the minimum number of points required to form a cluster is 3. When there is more than or equal to 1 cluster, and the number of channels contained in the largest cluster accounts for more than 40% of the number of abnormal channels, it is determined that there is spatial clustering. When the cluster with spatial clustering continues to exist for 3 consecutive short time windows, the prompt of spatiotemporal persistence is triggered, and the highest level of abnormal warning is issued.

[0063] Specifically, in step S5 of this embodiment, by establishing an individualized normal feature range and combining multi-level criteria such as multi-band feature deviation, multi-dimensional stability loss, and spatiotemporal aggregation of abnormal channels, a hierarchical early warning mechanism is realized. This fusion strategy effectively reduces the false alarm rate and improves the specificity of identifying real abnormal events. It is especially suitable for early and reliable early warning of spatiotemporally persistent neurological functional abnormalities.

[0064] Please see Figure 4 As shown, this is an artificial intelligence-based neural system data processing device according to this embodiment, including: The data decomposition unit is used to preprocess and decompose the neural system data to generate a frequency band data matrix; The phase analysis unit is used to construct a phase analysis model based on the frequency band data matrix to generate a functional connectivity matrix; The feature extraction unit is used to extract network topology features from the functional connectivity matrix to obtain data feature vectors; Evolutionary analysis unit, used to construct dynamic evolutionary models based on data feature vectors, to synchronize stability indices; Anomaly warning unit is used to perform multi-level criterion fusion and anomaly warning for the synchronization stability index.

[0065] In the above description, the disclosure of this invention is not intended to limit itself to these aspects. Rather, within the scope of the objectives of this disclosure, components can be selectively and operationally combined in any number. Furthermore, terms such as “comprising,” “encompassing,” and “having” should be interpreted by default as inclusive or open-ended, rather than exclusive or closed, unless explicitly defined as such. All technical, scientific, or other terms are to be understood by those skilled in the art, unless defined as such. Public terms found in dictionaries should not be interpreted in the context of the relevant technical documents in an overly idealistic or impractical manner, unless explicitly defined as such in this disclosure. Any modifications or alterations made by those skilled in the art based on the foregoing disclosure are within the scope of the claims.

Claims

1. An artificial intelligence-based neural system data processing method, characterized by, The method comprises the following steps: Preprocessing and frequency band decomposition of nervous system data to generate a frequency band data matrix; Constructing a phase analysis model based on the frequency band data matrix to generate a functional connectivity matrix; Network topology feature extraction on the functional connectivity matrix to obtain a data feature vector; Based on the data feature vector, a dynamic evolution model is constructed to obtain a synchronization stability index; Multi-layer criterion fusion and abnormal early warning on the synchronization stability index. 2.The artificial intelligence-based neural system data processing method of claim 1, wherein The continuously collected data stream is non-overlappingly segmented into multiple data segments with a fixed length of 30 seconds, and each data segment constitutes a data matrix X with dimensions [C, N], where C is the number of channels and N is the number of time points. The preprocessed nervous system data of each channel is decomposed into K preset physiological frequency bands, so that each data segment derives K frequency band data matrices, each with dimensions [C, N], and K represents the number of preset frequency bands. 3.The artificial intelligence-based neural system data processing method of claim 2, wherein, For each frequency band data matrix, the Hilbert transform is used to calculate the instantaneous phase of each channel and each time point. The phase difference sequence between any two channels in the frequency band data matrix corresponding to each preset physiological frequency band is analyzed, and the phase locking value within the data segment of the above two channels is analyzed. The phase locking value between each channel pair in the frequency band data matrix corresponding to each preset physiological frequency band is calculated to obtain a symmetric C×C matrix with diagonal elements of 0, which is used as the functional connectivity matrix. 4.The artificial intelligence-based neural system data processing method of claim 3, wherein, Thresholding is performed on the functional connectivity matrix, and elements in the functional connectivity matrix with a phase locking value greater than or equal to the connection threshold are set to 1, and elements in the functional connectivity matrix with a phase locking value less than the connection threshold are set to 0 to generate a binary adjacency matrix. Network topology features such as average connection strength, clustering coefficient, characteristic path length, and small-world scalar are extracted from the functional connectivity matrix based on the binary adjacency matrix to obtain a data feature vector. 5.The artificial intelligence-based neural system data processing method of claim 4, wherein, The window size of the sliding window is set to 5 seconds, and the sliding step is set to 2.5 seconds. The data segment is cut into multiple short-time window data, and short-time feature analysis is performed on the short-time window data to obtain a short-time feature vector. For the 4×K network topology features involved in the short-time feature vector, a time series with a length of M is generated for each network topology feature in the order of the sliding window, which is used as a feature time series. A fluctuation observation window containing 10 short-time window data is set. For each short-time feature vector, the average value and standard deviation within the fluctuation observation window are calculated, and the ratio of the calculated standard deviation to the average value is used as the local variation coefficient. 6.The artificial intelligence-based neural system data processing method of claim 5, wherein, The maximum value of the local variation coefficient within 1 hour is extracted as the historical variation maximum value, and the minimum value of the local variation coefficient within 1 hour is extracted as the historical variation minimum value. The synchronization stability index is analyzed based on the historical variation maximum value, the historical variation minimum value, and the local variation coefficient. 7.The artificial intelligence-based neural system data processing method of claim 6, wherein, In the explicit health monitoring period of the user, at least 20 baseline data segments with a length of the segmentation duration are collected to establish the normal feature range of the user, when establishing the normal feature range of the user, the data feature vector of each baseline data segment is analyzed, and the average value μ of each feature is calculated base And the standard deviation σ base The normal feature range of each feature is defined as [μ base -2×σ base ,μ base +2×σ base ]. 8.The artificial intelligence-based neural system data processing method of claim 7, wherein, The feature values in the data feature vector are checked, and the feature values in the data feature vector that do not belong to the corresponding normal feature range are taken as abnormal feature values. When the number of abnormal feature values is greater than or equal to 3 and the abnormal feature values come from at least 3 different preset physiological frequency bands, a prompt of multi-band feature association synchronization deviation is triggered, and an abnormal warning is performed. The number of dimensions with a synchronization stability index less than 0.3 is counted and taken as the number of abnormal dimensions. When the number of abnormal dimensions is greater than or equal to (2xK), a prompt of multi-dimensional synchronization stability cooperation loss is triggered, and an abnormal warning is performed. When the multi-band feature association synchronization deviation and the multi-dimensional synchronization stability cooperation loss trigger the prompt in succession within 30 seconds, the functional connectivity matrix corresponding to the 2 short time window data before and after the time when the prompt is triggered is extracted and taken as an abnormal connectivity matrix. The relative change amount of each channel pair in the abnormal connectivity matrix to the baseline mean value corresponding thereto is calculated, the relative change amounts are sorted in descending order, the channel pairs corresponding to the top 10% and bottom 10% relative change amounts are taken as abnormal channels, the abnormal channels are projected onto a two-dimensional scalp plane, a spatial density clustering algorithm is applied to analyze the distribution of the abnormal channels, the neighborhood radius is set to 1.5 times the average distance between electrodes, and the minimum number of points required to form a cluster is 3. When the number of clusters is greater than or equal to 1 and the number of channels contained in the largest cluster accounts for more than 40% of the number of abnormal channels, it is determined that there is spatial aggregation. When the cluster with spatial aggregation exists continuously for 3 consecutive short time windows, an abnormality with spatiotemporal persistence is triggered, and the highest level of abnormal warning is performed.

9. An artificial intelligence-based nervous system data processing apparatus applied to the artificial intelligence-based nervous system data processing method according to any one of claims 1 to 8, characterized by, The method comprises: a data decomposition unit configured to preprocess and frequency band decompose the nervous system data to generate a frequency band data matrix; a phase analysis unit configured to construct a phase analysis model based on the frequency band data matrix to generate a functional connectivity matrix; a feature extraction unit configured to extract network topology features of the functional connectivity matrix to obtain a data feature vector; an evolution analysis unit configured to construct a dynamic evolution model based on the data feature vector to obtain a synchronization stability index; an abnormal warning unit configured to perform multi-layer criterion fusion and abnormal warning on the synchronization stability index.

10. A storage medium, characterized by The computer is caused to execute the artificial intelligence-based nervous system data processing method according to any one of claims 1-8 when the instructions are run on the computer.

Images