A Method and System for Topology Extraction of Fatigue Auditory Targets Based on Multi-Order Micronetworks

Abstract

A method for extracting the topology of fatigued auditory targets based on multi-order micronetworks includes the following steps: collecting EEG data under fatigued and non-fatigue states; obtaining microstate templates and individual global field power curves through spatiotemporal clustering; constructing multiple individualized analysis time windows; using a multi-objective optimized microstate backfitting method to construct a multi-index joint optimization function and obtain a stable microstate label sequence; quantifying first-order microstate topological features; generating a microstate connection matrix based on frequency band conditions and multi-frequency band fusion based on connectivity indices; constructing a sparsely optimized second-order micronetwork of frequency bands and frequency band fusion, and extracting second-order microstate topological features; comparing the first-order and second-order microstate topological features under fatigued and non-fatigue states, and analyzing the specific changes in parameters and network connections. This invention constructs a multi-order analysis framework of "first-order microstate topology - second-order micronetwork," simultaneously improving analysis efficiency and reliability.

Description

Technical Field

[0001] This invention relates to the fields of neuroscience and brain-computer interface technology, and in particular to a method and system for extracting fatigued auditory target topology based on multi-level micronetworks, which is applicable to neural modulation-related research such as auditory attention analysis. Background Technology

[0002] Fatigue is a critical factor affecting operational performance and safety, especially in high-intensity task environments such as underwater exploration and industrial equipment monitoring. These environments are typically accompanied by continuous and dynamically changing background noise and non-stationary, multi-source signals, placing extremely high cognitive load on operators when identifying auditory targets. With sustained cognitive effort or prolonged task execution, fatigue weakens selective auditory attention, reduces decision-making accuracy, and consequently increases safety risks. Therefore, in-depth research into the brain's processing mechanisms of auditory information under fatigue is of great significance for developing scientific fatigue management strategies, optimizing work processes, and ensuring operational safety in high-risk environments.

[0003] Auditory target recognition is a rapid and comprehensive cognitive process involving attention allocation, contextual updating, and executive control. In complex acoustic environments, individuals must quickly extract key information from background sounds with overlapping spectra and unpredictable temporal distributions, primarily relying on the dynamic coordination of brain regions such as the auditory cortex and frontoparietal networks. Furthermore, neuroimaging studies indicate that saliency networks and executive control networks play a moderating role in fatigue states. Currently, systematic and in-depth research on the neurodynamic mechanisms by which fatigue affects auditory target recognition in complex acoustic environments is lacking, and there is an urgent need to improve the ability to capture and quantify the dynamic changes in neural activity under fatigue. Especially in tasks requiring rapid perception and continuous cognitive control coordination, detailed spatiotemporal analysis methods are still lacking for fatigue-induced abnormalities in brain functional connectivity and fluctuations in network efficiency.

[0004] Currently, electroencephalography (EEG) has become an important tool for studying fatigue-related brain function changes due to its high temporal resolution. However, traditional event-related potential (ERP) analysis relies on average waveforms and a limited number of channels, making it difficult to reveal the dynamic and distributed characteristics of brain activity in continuous and complex auditory scenarios. In recent years, event-related microstate technology has shown its application potential in rapid event processing through spatiotemporal variation clustering analysis of EEG signals. However, existing methods are mostly based on fixed time window divisions and single-index backfitting, which are difficult to adapt to individual differences, susceptible to artifact interference, and lack stability. In addition, the functional networks constructed based on these methods often lack frequency-dimensional coupling features, have limited correlation with event-related components, and limit the comprehensive interpretation of the recognition process. Therefore, when performing auditory cognitive tasks in complex acoustic environments under fatigue conditions, it is urgent to design a multi-frequency event-related micronetwork construction method based on adaptive time windows and multi-index optimized backfitting mechanisms. This method can accurately capture and quantify the topological structure of rapid neural activity by integrating the spatiotemporal topology of microstates with network coupling into a neural monitoring framework, providing theoretical support and technical foundation for fatigue recognition, task scheduling optimization, and the development of auditory target identification systems in high-risk scenarios. Summary of the Invention

[0005] The purpose of this invention is to address the deficiencies in the spatiotemporal feature analysis technology for auditory target identification in the existing technology, and to provide a fatigue-state auditory target topology extraction method and system based on multi-order micronetworks.

[0006] The technical solution adopted to achieve the purpose of this invention is:

[0007] A fatigue-state auditory target topology extraction method based on multi-order micronetworks includes the following steps:

[0008] Step 1: Collect EEG data induced by auditory target recognition task in a continuous background noise environment under fatigued and non-fatigue conditions, and simultaneously record the intensity, frequency, occurrence time of auditory stimulation and subjective fatigue scale results.

[0009] Step 2: Preprocess the EEG data recorded in Step 1, calculate the event-related potential data of each channel, and obtain the microstate template and individual global field power curve through spatiotemporal clustering.

[0010] Step 3: Based on the individual global field power curve obtained in Step 2, an adaptive time window selection mechanism is adopted to automatically identify multiple significant jump points using the dynamic change rate of the individual global field power curve. Multiple individualized analysis time windows are constructed with these points as the center to fully cover the micro-state switching dynamics during important components of the event-related potential.

[0011] Step 4: Based on the individualized analysis time window determined in Step 3, a multi-objective optimization microstate backfitting method is adopted to construct an error and multi-index joint optimization function with the joint indices of maximizing spatial correlation, minimizing global field power error, and duration stability. First, the microstate label sequence is initialized using a greedy search method, and then a dynamic programming strategy is used to optimize the high-loss EEG data segments to minimize the error of the joint indices and obtain a stable microstate label sequence. Within this individualized analysis time window, the first-order microstate topological features are quantified, including global field power, global explained variance, occurrence frequency, duration, and transition probability between microstates.

[0012] Step 5: Based on the stable microstate label sequence obtained in Step 4, perform band decomposition on the event-related potential signal corresponding to each microstate segment within the window, use the Morlet wavelet function to improve the time-frequency resolution, and generate a microstate connection matrix based on the connectivity index, which includes frequency segmentation conditions and multi-band fusion.

[0013] Step 6: Based on the microstate connection matrix from Step 5, construct a second-order micronetwork with sparse optimization and frequency band fusion, wherein the optimal sparsity is determined based on structural information entropy; then perform binary processing on the second-order micronetwork and extract the second-order microstate topological features to achieve a quantitative representation of the neural dynamics structure. The second-order microstate topological features include clustering coefficients, feature path lengths, global efficiency, and local efficiency.

[0014] Step 7: Compare the first-order microstate topology features and second-order microstate topology features under fatigue and non-fatigue conditions, and analyze the specific changes in parameters and network connections.

[0015] In the above technical solution, the calculation method for the individualized analysis time window in step 3 is as follows:

[0016] W λ =[t λ -δ,t λ +δ]

[0017] △GFP(t)=|GFP′(t+1)-GFP′(t)|>2·σ △

[0018]

[0019]

[0020] W λ For t λ The corresponding individualized analysis time window, t λ ∈{t1,t2,…,t K} represents the significant transition point after screening, δ represents one side of the fixed buffer, and ΔGFP(t) represents the local mutation amount at time t; σ △ Let be the standard deviation of all ΔGFP(t); GFP(t) be the global field power curve; GFP′(t) be the first derivative of the global field power curve; and GFP″(t) be the second derivative of the global field power curve.

[0021] In the above technical solution, the multi-index joint optimization function in step 4 is as follows:

[0022] L=w1L COR +w2L GFP +w3L Dur

[0023] Where L is the joint optimization function, L COR For spatial correlation error, L GFP For global field power matching error, L Dur For the duration of stability error, w q The entropy weights for each indicator are q = 1, 2, or 3; w q The calculation method is as follows:

[0024]

[0025] Where x qn Let g be the value of the q-th index on the n-th candidate solution. If there are r candidate solutions, then n = 1, ..., r; qn e represents the relative weight of the q-th index on the n-th candidate solution. q Let q be the information entropy of the q-th indicator.

[0026] In the above technical solution, the calculation formulas for each indicator are as follows:

[0027]

[0028] Where M(t) is the event-related spatial electrode potential distribution topographic map at time t, S(t) is the microstate label index corresponding to time t, and T S(t) Let S(t) be the microstate template, and COR(M(t),T) be the microstate template. S(t) ) represents the template matching correlation calculated based on Pearson correlation, and GFP(t) represents the spatial potential distribution intensity of the raw EEG at time t. The mean global field power at all time points labeled S(t) is given. Let be the variance of the duration of microstates k = 1, 2, ..., N, and N be the total number of microstate templates.

[0029] In the above technical solution, COR(M(t),T S(t) The calculation formula for ) is as follows:

[0030]

[0031] Where M(t)=[M1(t),…,M Q [(t)] is the topographic map at time t, i.e. the event-related potential vector of the whole brain electrodes; T is the mean of this topographic map. S(t) =[T S(t),1 ,…,T S(t),Q [] represents the template of the microstate S(t), and Q is the number of electrodes. Let i be the mean of the template corresponding to S(t), and i and j be the channel indices.

[0032] In the above technical solution, step 4, the dynamic programming strategy includes a local cost function with added penalty terms and a global cost function, the algorithms of which are as follows:

[0033]

[0034] C(τ)=min a<τ [C(a)+Cost(a+1,τ,T k )]

[0035] Where Cost(a+1,τ,T) k ) represents the local cost generated after assigning a microstate label k to a continuous EEG data segment from time a+1 to τ; b represents the time points from a+1 to τ, where a represents the end time of the previous fitted segment and τ represents the end time of the current fitted segment, 0≤a<τ; M(b) is the event-related potential topography at time b, represented as a voltage space vector [V1(b),…,V] containing all channels. N (b)];T k For the template of microstate k; COR(M(b),T k ) for M(b) and T k The Pearson correlation; GFP(b) is the global field power of b; For microstate T k The average global field power at the corresponding time point; τ-a+1 is the duration of the current fitted segment in frames; μ k Let be the target average duration of frames for microstate k; α be the duration penalty factor; C(a) be the minimum cumulative cost from the starting point to a; and C(τ) be the minimum cumulative cost from the starting point to τ. Dynamic programming obtains the optimal microstate label sequence by minimizing these values.

[0036] In the above technical solution, the method for calculating the micro-state connection matrix in step 5 is as follows:

[0037]

[0038] in Let f be the microstate connection matrix corresponding to channel i and channel j in the frequency band f∈{theta,alpha,beta,gamma}. The elements of the matrix are the phase lag exponents of the inter-lead signals. Let m be the instantaneous phase of channel i in frequency band f, where m is the local sampling time in the current microstate segment, P is the total number of sample points in the segment, and sign[·] determines the positive or negative direction of the phase difference.

[0039] In the above technical solution, the calculation of the multi-frequency fusion micro-state connection matrix in step 5 is as follows:

[0040]

[0041] in For the multi-frequency fusion microstate connection matrix, i and j are electrode indices, η f The weighting coefficients for frequency band f are the sum of the weights of all frequency bands, ∑η. f =1.

[0042] In the above technical solution, in step 7, a second-order micronetwork is constructed based on the structural information entropy mechanism. The structural information entropy is calculated as follows:

[0043]

[0044] Where H ρ H represents the network structure information entropy with sparsity ρ, measuring the complexity of the connection distribution in the network structure. ρ The larger the value, the more balanced the network connection; p i p is the degree-normalized weight of node i. i ∈[0,1], d i Let i be the degree of node i. It represents the sum of the degrees of all nodes in the network.

[0045] Another aspect of the present invention includes a system for performing the fatigue-state auditory target topology extraction method, the system comprising an experimental paradigm adjustment module, a data recording and preprocessing module, a first-order microstate topology extraction module, and a second-order micronetwork topology extraction module, wherein:

[0046] The experimental paradigm adjustment module includes an experimental procedure display unit, an auditory target stimulus design unit, a background noise adjustment unit, and a KSS scale design unit.

[0047] The data recording and preprocessing module includes an EEG data recording unit, an EEG preprocessing unit, a KSS scale scoring recording unit, and a stimulation parameter recording unit.

[0048] The first-order microstate topology extraction module includes a trial window length selection unit, an adaptive time window adjustment unit, a multi-index optimization and inverse simulation unit, a first-order topology quantization unit, and a first-order topology visualization unit.

[0049] The second-order micro-network topology extraction module includes a micro-state connection matrix unit, a sparsification unit, a second-order topology quantization unit, and a second-order topology visualization unit.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1. Construction of an experimental environment with high ecological validity: This invention constructs an experimental paradigm of continuous dynamic background noise superimposed with multi-source target stimuli, and combines it with headphones to construct a virtual surround sound field to simulate the acoustic interference characteristics in actual work scenarios; simultaneously, it links the KSS scale to achieve quantitative definition of fatigue state, so that the simulated environment is highly matched with the core features of real auditory target identification tasks in terms of acoustic parameters, task mode and state monitoring dimensions, providing a data foundation with higher ecological validity for subsequent neural mechanism analysis.

[0052] 2. Improving the accuracy and stability of microstate sequence extraction: Addressing the problem that fixed time windows in existing microstate analysis are difficult to adapt to individual EEG dynamics, this invention employs an adaptive time window selection mechanism. Based on significant jump points in the individual's global field power curve, an analysis window is dynamically constructed, accurately covering the microstate switching process of key components of event-related potentials. Simultaneously, through a multi-index optimization inverse fitting method (integrating spatial correlation, global field power error, and temporal stability indicators), combined with a dynamic programming strategy to optimize the label sequence, artifact interference is effectively reduced, significantly improving the stability and consistency of the microstate label sequence, overcoming the result fluctuation problem caused by traditional single-index inverse fitting.

[0053] 3. Deep Topological Analysis of Multi-Order Micronetworks: This invention innovatively constructs a multi-order analysis framework of "first-order microstate topology - second-order micronetwork," breaking through the limitations of single-dimensional analysis. The first-order topology characterizes the dynamic switching of EEG spatial patterns by quantifying parameters such as global field power and transition probability. The second-order network first constructs a connection matrix through frequency band analysis and multi-frequency fusion, and then achieves sparsity optimization based on structural information entropy. Combined with parameters such as clustering coefficient and global efficiency, it reveals the hierarchical association of cross-frequency band neural connections. This framework not only achieves comprehensive capture and network hierarchy quantification of neural connection patterns from theta to gamma frequency bands, filling the gap in frequency coupling and deep topological analysis, but also improves analysis efficiency and reliability simultaneously by eliminating redundant connections and retaining core features, ultimately achieving comprehensive quantification of neurodynamic structures.

[0054] 4. Highly efficient integrated automated analysis system: This invention integrates experimental paradigm adjustment, data recording preprocessing, multi-level topology extraction and visualization modules to achieve seamless integration from experimental design and signal processing to feature analysis; it supports flexible adjustment of parameters such as window length selection and sparsity setting, reduces human intervention errors, improves the efficiency and consistency of fatigue-state auditory target topology extraction, and provides a standardized analysis tool for auditory cognitive fatigue research. Attached Figure Description

[0055] Figure 1 This is a flowchart of a fatigue-based auditory target topology extraction method based on multi-order micronetworks.

[0056] Figure 2 This is a schematic diagram of a paradigm for auditory target identification in an environment with background noise interference under fatigue conditions.

[0057] Figure 3 A schematic diagram illustrating the construction of a first-order microstate topology sequence based on adaptive window selection and multi-index inverse optimization.

[0058] Figure 4 A schematic diagram of the microstate connection matrix construction for frequency division and multi-frequency fusion.

[0059] Figure 5 This is a schematic diagram of the construction of the second-order micro-network topology after sparsification based on the structural information entropy mechanism.

[0060] Figure 6 This is a schematic diagram of a fatigue-based auditory target topology extraction system based on a multi-order micronetwork. Detailed Implementation

[0061] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0062] Example 1

[0063] like Figure 1 As shown, a fatigue-state auditory target topology extraction method based on multi-order micronetworks includes the following steps:

[0064] Step 1: Collect EEG data induced by auditory target recognition task in a continuous background noise environment under fatigued and non-fatigue conditions, and simultaneously record the intensity, frequency, occurrence time of auditory stimuli and the results of the subjective fatigue scale.

[0065] Step 2: Preprocess the EEG data recorded in Step 1, calculate the event-related potential data of each channel, and obtain the microstate template and individual global field power curve through spatiotemporal clustering.

[0066] Step 3: Based on the individual global field power curve obtained in Step 2, an adaptive time window selection mechanism is adopted to automatically identify multiple significant jump points using the dynamic change rate of the individual global field power curve. Multiple individualized analysis time windows are constructed with these points as the center to fully cover the micro-state switching dynamics during important components of the event-related potential.

[0067] Step 4: Based on the individualized analysis time window determined in Step 3, a multi-objective optimization microstate backfitting method is used to construct an error and optimization function with spatial correlation maximization, global field power error minimization, and temporal stability as joint indicators. First, the microstate label sequence is initialized using a greedy search approach. Then, a dynamic programming strategy is used to optimize high-loss segments, minimizing the error of the joint indicators to obtain a stable microstate label sequence. Within this individualized analysis time window, the first-order microstate topological features are quantified, including global field power, global explained variance, occurrence frequency, duration, and transition probability between microstates.

[0068] Step 5: Based on the stable microstate label sequence obtained in Step 4, perform band decomposition on the event-related potential signal corresponding to each microstate segment within the window, use the Morlet wavelet function to improve the time-frequency resolution, and generate a microstate connection matrix with frequency segment conditions and a multi-frequency fusion microstate connection matrix based on the connectivity index.

[0069] Step 6: Based on the microstate connection matrix and multi-frequency fusion microstate connection matrix from Step 5, construct a second-order micronetwork with sparse optimization and frequency band fusion, wherein the optimal sparsity is determined based on structural information entropy; then perform binary processing on the second-order micronetwork and extract the second-order microstate topological features to achieve a quantitative representation of the neural dynamics structure. The second-order microstate topological features include clustering coefficients, feature path lengths, global efficiency, and local efficiency.

[0070] Step 7: Compare the topological features of first-order and second-order microstates under fatigue and non-fatigue states, analyze the specific changes in parameters and network connections, reveal the impact of fatigue on the neural dynamics of auditory target recognition, and discover that the changes in network parameters related to specific microstates reflect the compensatory redistribution of neural resources and the existence of a cognitive regulation mechanism of "mid-term compensation - late-term impairment".

[0071] Example 2

[0072] A fatigue-state auditory target topology extraction method based on multi-order micronetworks includes the following steps:

[0073] Step 1: Auditory target recognition task paradigm in continuous background noise interference environment under fatigued and non-fatigue states, including:

[0074] Step 1.1, constructing a noise interference and target stimulus system: Continuous underwater noise is used as the background interference signal, superimposed with a target stimulus sequence; the target stimulus includes four types of equally probable sound signals (including two types of marine animal calls and two types of ship radiated noise, where the marine animal calls are sourced from the Watkins Marine Mammal Sound Database and the ship radiated noise samples are sourced from the ShipsEar database). Adobe Audition 2022 is used to mix the background noise and target stimulus. All stimuli and background noise are normalized to -25dB sound pressure level and 48kHz sampling rate. Each type of stimulus is presented 45 times in a pseudo-random order, with a single stimulus duration of 1s and a stimulus interval of 2s.

[0075] Step 1.2: Design the task execution process under two states: Participants received auditory stimuli through closed-back headphones (Sennheiser HD280PRO), maintaining focus while looking at the white cross displayed on the monitor in front of them; they quickly responded to all target stimuli by pressing buttons ("u", "i", "o", and "p" keys correspond to the four types of stimuli). The task was conducted twice: the first time in a non-fatigue state (KSS score < 4 before the experiment), and the second time in a fatigue state (after successful induction of fatigue in the AX-continuous attention task, i.e., KSS button scores ≥ 8 for two consecutive times). The duration of each underwater sound task was 12 minutes. The KSS scale appeared every 15 minutes in the AX-continuous attention task. The specific experimental paradigm is as follows: Figure 2 As shown.

[0076] Step 2: Acquisition and preprocessing of EEG signals evoked by auditory targets in fatigued and non-fatigue states, as well as acquisition of micro-state templates and individual global field power curves:

[0077] Step 2.1: Acquisition of EEG signals induced by auditory targets in fatigued and non-fatigue states.

[0078] A 64-channel NeuroScan EEG acquisition system (Compumedics, Australia) was used, with electrodes arranged according to the international 10-20 system. The Ref electrode was used as the reference electrode, and the EEG signals were amplified by the NeuroScanSynAmps2 amplifier. Before acquisition, all electrode impedances were ensured to be below 20kΩ, and the signal sampling rate was set to 1000Hz. EEG signal acquisition was synchronized with the stimulus presentation of the auditory target recognition task (the stimulus was presented via MATLAB R2019a combined with Psychtoolbox-3). The EEG signals induced by the participants' responses to underwater target stimuli (marine animal calls, ship radiated noise, etc.) were recorded in both the non-fatigue state (first auditory target recognition task, corresponding to a KSS score <4) and the fatigue state (corresponding to two consecutive KSS scores ≥8), capturing the dynamic neural electrical activity induced by the stimulus.

[0079] Step 2.2, EEG data preprocessing.

[0080] First, complete labels for the stimulation sequences were added. Then, bandpass filters with a 1Hz low-pass and an 80Hz high-pass were applied to attenuate low-frequency drift and power frequency noise. Next, useless channels were identified and marked, and data completion was performed using spherical interpolation. The reference electrode was recalibrated to the average value of all electrodes to improve the signal-to-noise ratio. Ocular artifacts such as eye movement and blinking were separated and removed through independent component analysis. The data was segmented according to the stimulation start time, with each segment covering 200ms before stimulation to 1000ms after stimulation, and baseline correction was performed using the 200ms before stimulation as the baseline. Finally, abnormal data segments with a maximum-to-minimum voltage difference exceeding 100μV within each segment were removed to ensure the reliability of subsequent data analysis.

[0081] Step 2.3: Spatiotemporal clustering generates micro-state templates and individual global field power curves.

[0082] Based on the preprocessed EEG data from step 2.2, event-related potential data for each channel are calculated, and microstate templates and individual global field power curves are obtained through spatiotemporal clustering.

[0083] Step 3: Design an adaptive time window selection mechanism to construct an individualized analysis time window.

[0084] Based on the individual global field power curve obtained in step 2, an adaptive time window selection mechanism is adopted to automatically identify multiple significant jump points using the dynamic change rate of the individual global field power curve. Multiple individualized analysis time windows are constructed with these points as the center to comprehensively cover the micro-state switching dynamics during important components of the event-related potential.

[0085] By analyzing the zero-crossing points of the second derivative in the individual global field power curve, mutation locations reflecting the spatial distribution structural changes of EEG signals are identified. The local rate of change of the first derivative (ΔGFP(t)) is used as a sensitive indicator to screen for significant mutation points exceeding twice the standard deviation of the global rate of change. To suppress misidentification of dense jumps caused by high-frequency noise or transient spikes, a minimum jump interval constraint mechanism is introduced during the candidate jump point screening process. If the time interval between any two candidate jump points is less than a preset threshold (e.g., 30ms), only jump points with larger absolute values ​​of local mutations are retained. Multiple individualized analysis time windows are constructed by expanding a fixed-length buffer zone to both sides of the screened significant jump points. The specific calculations are as follows:

[0086] W λ =[t λ -δ,t λ +δ]

[0087] △GFP(t)=|GFP′(t+1)-GFP′(t)|>2·σ△

[0088]

[0089] Among them W λ For t λ The corresponding individualized analysis time window, t λ ∈{t1,t2,…,t K} represents the significant transition point after screening, δ represents one side of the fixed buffer (set δ = 50ms, fixed total window length is 100ms), the window length of 2δ covers an important component of the event-related potential (for example, in this embodiment, the P2 component segment is extended 50ms on both sides of the 3rd significant transition point as the center, and the P3 component segment is extended 50ms on both sides of the 5th significant transition point as the center), ΔGFP(t) is the local mutation amount at time t; σ △ ...

[0090] Step 4: Employ a multi-index joint optimization method to perform backfitting of microstates and quantize the topological properties of first-order microstates within the quantization window. Specifically, this includes:

[0091] Step 4.1: Design a multi-index joint optimization function. The multi-index optimization backfit uses the entropy weight method to dynamically allocate weights for spatial correlation, global field power amplitude, and duration stability, ensuring that the contribution of each index is proportional to its information content. The joint optimization function is expressed as follows:

[0092] L=w1L COR +w2L GFP +w3L Dur

[0093] Where L is the joint optimization function, L COR For spatial correlation error, L GFP For global field power matching error, L Dur This refers to the duration of stability error. q The entropy weights for each indicator, q = 1, 2, or 3, are calculated as follows:

[0094]

[0095] Where x qnLet q be the value of the q-th index (spatial correlation error, global field power matching error, or duration stability error) on the n-th candidate solution. The candidate solution is a combination of different microstate label sequences generated by greedy search and dynamic programming. If there are r candidate solutions, then n = 1, ..., r. g is the sum of the values ​​of the q-th index across all candidate solutions; qn e represents the relative weight of the q-th index on the n-th candidate solution. q Let e ​​be the information entropy of the q-th index, reflecting the degree of dispersion of information of this index among candidate solutions. q The larger the value, the less effective the index is in distinguishing candidate solutions; w q w represents the final weight of the q-th indicator. q The larger the value, the greater the contribution of that indicator. The calculation formulas for each indicator are as follows:

[0096]

[0097] Where M(t) is the event-related spatial electrode potential distribution topographic map at time t, S(t) is the microstate label index corresponding to time t, and T S(t) Let S(t) be the microstate template, and COR(M(t),T) be the microstate template. S(t) ) represents the template matching correlation calculated based on Pearson correlation, and GFP(t) represents the spatial potential distribution intensity of the raw EEG at time t. The mean global field power at all time points labeled S(t) is given. Let be the variance of the duration of the microstates k = 1, 2, ..., N, and N be the total number of microstate templates. COR(M(t), T) S(t) The calculation formula for ) is as follows:

[0098]

[0099] Where M(t)=[M1(t),…,M Q [(t)] is the topographic map at time t, i.e. the event-related potential vector of the whole brain electrodes; T is the mean of this topographic map. S(t) =[T S(t),1 ,…,T S(t),Q ] represents the template of the microstate S(t), and Q is the number of electrodes (e.g., 64 channels). Let i be the mean of the template corresponding to S(t), and i and j be the channel indices.

[0100] Step 4.2: Initialize the microstate label sequence using a greedy search method.

[0101] The label sequence is initialized using a greedy search method, and the micro-state template that best matches the topographic map is selected at each time step, providing a reliable starting point for subsequent dynamic optimization.

[0102] Step 4.3: Optimize high-impairment EEG data segments using a dynamic programming strategy. The dynamic programming strategy includes a local cost function with added penalty terms and a global cost function, the algorithms of which are as follows:

[0103]

[0104] C(τ)=min a<τ [C(a)+Cost(a+1,τ,T k )]

[0105] Where Cost(a+1,τ,T) k ) represents the local cost generated after assigning a microstate label k to a continuous EEG data segment from time a+1 to τ; b represents the time points from a+1 to τ, where a represents the end time of the previous fitted segment and τ represents the end time of the current fitted segment, 0≤a<τ; M(b) is the event-related potential topography at time b, represented as a voltage space vector [V1(b),…,V] containing all channels. N (b)];T k For the template of microstate k; COR(M(b),T k ) for M(b) and T k The Pearson correlation; GFP(b) is the global field power of b; For microstate T k The average global field power at the corresponding time point; τ-a+1 is the duration of the current fitted segment in frames; μ k Let be the target average duration of frames for microstate k; α be the duration penalty factor; C(a) be the minimum cumulative cost from the starting point to a; and C(τ) be the minimum cumulative cost (global cost) from the starting point to τ. Dynamic programming obtains the optimal microstate label sequence by minimizing these values. First-order microstate topological feature extraction through multi-index optimization of backfitting addresses the instability problem of traditional microstate analysis, thus characterizing the "spatiotemporal dynamics" of neural activity.

[0106] During the dynamic programming process, the optimal previous jump time point 'a' is recorded point by point in time. * and its corresponding optimal label number k * This serves as a backtracking pointer. Finally, starting from the end of the time series, we backtrack segment by segment to reconstruct the optimal segment boundaries and their micro-state labels, constructing the optimal label sequence of the complete EEG. For example... Figure 3The diagram shows a first-order microstate topology sequence based on adaptive window selection and multi-index inverse simulation optimization. The adaptive window selection uses the fifth significant transition point as a baseline, extending both sides by 50ms to form an analysis window, accurately matching the P2 component segment of auditory-related potentials in fatigue states. Multi-index inverse simulation optimization introduces quantitative indicators of topological stability, feature recognition, and noise suppression effect, backfitting the initial topology sequence and iteratively optimizing the parameters. The resulting first-order microstate topology sequence more realistically reflects the dynamic topological characteristics of auditory target-related EEG activity in fatigue states.

[0107] Step 4.4: Quantify the topological sequence properties of first-order microstates. Parameters for quantifying the topological sequence information of first-order microstates include the global field power, global explained variance, frequency of occurrence, duration, and transition probability between microstates for each microstate within the window. These first-order parameters directly reflect the switching efficiency of EEG spatial patterns. Abnormalities in various parameters under fatigue conditions can quantify the impact of fatigue on auditory attention and can serve as neural biomarkers of fatigue.

[0108] Step 5: Construct the micro-state connection matrix for frequency band conditions and multi-band fusion, specifically including:

[0109] Step 5.1: Perform band decomposition using the Morlet wavelet function. Each micro-state segment corresponds to the original signal and is decomposed into bands (theta: 4-8Hz, alpha: 8-13Hz, beta: 13-30Hz, gamma: 30-50Hz).

[0110] Step 5.2: Calculate the frequency-segmented microstate connectivity matrix. The connectivity index is the phase lag index, which estimates the stability of EEG signal coupling directions in different frequency bands by measuring the consistency of phase difference shift directions between leads, thereby constructing the frequency-segmented microstate connectivity matrix. The construction of the frequency-segmented and multi-frequency fusion connectivity matrix fills the gap in frequency dimension analysis, and the three-dimensional features of "time-space-frequency" comprehensively reflect the frequency dimension characteristics of neural activity; in addition, the connectivity matrix analyzes the specific mechanisms of neural coupling under fatigue state and quantifies abnormal coupling in specific frequency bands. The specific calculation formula is as follows:

[0111]

[0112] in Let f be the microstate connection matrix corresponding to channel i and channel j in the frequency band f∈{theta,alpha,beta,gamma}. The elements of the matrix are the phase lag exponents of the inter-lead signals. Let m be the instantaneous phase of channel i in frequency band f, where m is the local sampling time in the current microstate segment, P is the total number of sample points in the segment, and sign[·] determines the positive or negative direction of the phase difference. Figure 4The micro-state connectivity matrix is ​​composed of frequency division and multi-frequency fusion. Frequency division clearly presents the local connectivity characteristics of fatigue-state auditory-related EEG activities in each frequency band. Multi-frequency fusion dynamically allocates weights based on the contribution of each frequency band feature, integrates information to construct a comprehensive connectivity matrix, and reflects the correlation patterns between different frequency bands from aspects such as the coordinated change trend of connectivity strength, the cross-frequency band association patterns of key nodes, and the covariance characteristics of topology, providing a quantitative basis for network connectivity analysis.

[0113] Step 5.3: Generate the multi-band fusion microstate connection matrix. The calculation of the multi-band fusion microstate connection matrix is ​​as follows:

[0114]

[0115] in For the multi-frequency fusion microstate connection matrix, i and j are electrode indices, η f The weighting coefficients for frequency band f are the sum of the weights of all frequency bands, ∑η. f =1 (If set to equal weight, the weight of each frequency band is 1 / 4).

[0116] Step 6: Construct a second-order micronetwork and extract network topology features. This specifically includes:

[0117] Step 6.1: Construct a second-order micronetwork based on the structural information entropy mechanism. The structural information entropy is calculated as follows:

[0118]

[0119]

[0120] By selecting the optimal sparsity through structural information entropy, interference from redundant network connections is resolved. H ρ H represents the network structure information entropy when the sparsity is ρ (e.g., the sparsity iterates from 0.1 to 0.5 with a step size of 0.01), measuring the complexity of the connection distribution in the network structure. ρ The larger the value, the more balanced the network connection; p i p is the degree-normalized weight of node i, used to quantify the balance of the connection distribution in the undirected network. i ∈[0,1], d i Let i be the degree of node i. The sum of the degrees of all nodes in the network; select the sparsity ρ that maximizes the structural information entropy. * (e.g., 0.3) is the optimal sparsity, at which point the network structure is most balanced and the distribution is most uniform.

[0121] like Figure 5This is a topological diagram of the second-order micronetwork after sparsification based on the structural information entropy mechanism. The structural information entropy mechanism selects key connections that significantly contribute to the topological features of auditory targets in fatigue state and eliminates redundant connections to achieve network sparsification. The sparsified second-order micronetwork topology retains core topological features while reducing network complexity, making the hierarchical associations of EEG networks related to auditory targets in fatigue state clearer.

[0122] Step 6.2: Extract the topological features of the second-order micronetwork. After binary processing of the sparse second-order micronetwork, the parameters of the network topology structure are quantified, including clustering coefficients, feature path lengths, global efficiency, and local efficiency, thus achieving a quantitative characterization of the neural dynamics structure. The extraction of the topological features of the second-order micronetwork reveals the "hierarchical association" of neural connections, integrating multidimensional information of "time window-space-frequency-connection strength," capturing network hierarchical features that cannot be extracted from the first-order topology structure. In addition, the network topology parameters quantify the efficiency fluctuations of the neural network under fatigue state, comprehensively reflecting the "network-level anomalies" of neural activity under fatigue state, providing a more in-depth quantitative basis for the study of fatigue-related neural mechanisms, and also providing a reference for subsequent task scheduling optimization.

[0123] Step 7: Compare the topological features of first-order and second-order microstates under fatigue and non-fatigue states, analyze the specific changes in parameters and network connections, reveal the impact of fatigue on the neural dynamics of auditory target recognition, and discover that the changes in network parameters related to specific microstates reflect the compensatory redistribution of neural resources and the existence of a cognitive regulation mechanism of "mid-term compensation - late-term impairment".

[0124] In first-order microstate topological feature analysis, the differences between fatigued and non-fatigue states are mainly reflected in the following: In the fatigued state, the global field power of microstates corresponding to the mid-stage recognition phase (such as the P2 component) increases, reflecting enhanced compensatory local neural activity; while the global explained variance, duration, and coverage of microstates corresponding to the late-stage decision-making phase (such as the P3 component) decrease, indicating impaired consistency and stability of neural activity at this stage. Simultaneously, the enhanced activation of some microstates associated with the P3 stage may be a compensatory mechanism. In second-order micronetwork topological feature analysis, the differences are manifested as follows: In fatigue, the global efficiency of the network corresponding to the mid-stage recognition phase increases, the feature path length shortens, and cross-regional information transmission becomes more efficient, reflecting enhanced compensatory integration; in the late-stage decision-making phase, the clustering coefficient and local efficiency of the network decrease, and the local brain region coordination ability and subnetwork processing efficiency decline, reflecting network-level functional decay.

[0125] Example 3

[0126] This embodiment constructs a fatigue-state auditory target topology extraction system based on multi-order micronetworks, based on the analysis steps in Embodiment 1 or Embodiment 2. This system can complete parameter adjustment and experimental data recording for fatigue-state auditory target identification experiments in background noise interference environments, and achieve precise processing throughout the entire process, including EEG signal preprocessing, first-order microstate topology feature extraction and analysis, second-order micronetwork topology feature extraction and analysis, and visualization, thereby improving data analysis efficiency and quality. Figure 6 As shown, the system specifically includes the following components:

[0127] (1) Experimental Paradigm Adjustment Module: This module is used to generate the paradigm flow and adjust the experimental parameters, including the following units: ① Experimental Flow Display Unit: It can display the complete process and key nodes of the experiment in real time, so that the experimenter and the subject can clearly understand the experimental progress; ② Auditory Target Stimulus Design Unit: It can design parameters such as frequency, intensity and duration of target stimuli according to the auditory target identification needs of fatigue state, and supports the random generation of stimulus sequences; ③ Background Noise Adjustment Unit: It can generate different types of background noise signals and can accurately adjust the noise intensity and signal-to-noise ratio to simulate diverse noise interference environments; ④ KSS Scale Design Unit: It has built-in standardized presentation and data acquisition functions of the KSS fatigue scale.

[0128] (2) Data Recording and Preprocessing Module: This module mainly realizes the recording of experimental data, preprocessing of EEG signals, and organization of basic data. It includes: ① EEG data recording unit: compatible with common EEG acquisition devices, synchronously recording the EEG signals of subjects in the fatigued auditory target identification experiment to ensure the completeness and accuracy of data acquisition; ② EEG preprocessing unit: performs rereference, filtering, artifact removal, and segmentation on the acquired EEG signals in real time to improve signal quality; ③ KSS scale scoring recording unit: automatically receives and stores KSS scale scoring data to ensure the validity of the scoring data; ④ Stimulus parameter recording unit: synchronously records auditory target stimulus and background noise parameters to form a parameter log corresponding to the EEG data.

[0129] (3) First-order microstate topology extraction module: This module is used to extract and optimize the topological features of first-order microstates, including the following units: ① Trial window length selection unit: It can select the required epoch window length according to the characteristics of fatigue-state auditory related EEG signals to adapt to the time dimension analysis requirements of experimental data; ② Adaptive time window adjustment unit: By determining the order of significant jump points, it can dynamically adjust the time window by associating fatigue-state auditory related event-related potential components to improve the matching degree between the window and signal features; ③ Multi-index optimization and reverse simulation unit: It supports the selection of optimization function index type, and can configure the optimization function weight (entropy weight / equal weight) and dynamic programming selection to realize the reverse fitting optimization of the topological sequence; ④ First-order topology structure quantification unit: It can calculate the global field power, global explained variance, occurrence frequency, duration and transition probability between microstates and other quantitative parameters to realize the quantitative description of topological features; ⑤ First-order topology visualization unit: It can generate the visualization results of microstate templates and in-window sequences, and intuitively present the dynamic change characteristics of the first-order microstate topology.

[0130] (4) Second-order micro-network topology extraction module: This module is used to complete the construction, optimization and feature extraction of second-order micro-network topology, including the following units: ① Micro-state connection matrix unit: Supports frequency band selection function, and can select frequency band processing or frequency band fusion processing method to construct the connection matrix according to the analysis requirements; ② Sparsification unit: Can apply the structural information entropy mechanism to perform network sparsification processing, and also supports users to independently select sparsity parameters to perform sparsification operation and eliminate redundant connections; ③ Second-order topology structure quantification unit: Can calculate quantitative indicators such as clustering coefficient, feature path length, global efficiency and local efficiency, and realize quantitative analysis of second-order network topology features; ④ Second-order topology visualization unit: Can generate visualization images of frequency micro-state connection matrix, second-order micro-network and binary network, intuitively display the topology structure and connection features of second-order micro-network.

[0131] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for topological extraction of fatigued auditory targets based on multi-order micronetworks, characterized in that, Includes the following steps: Step 1: Collect EEG data induced by auditory target recognition task in a continuous background noise environment under fatigued and non-fatigue conditions, and simultaneously record the intensity, frequency, occurrence time of auditory stimulation and subjective fatigue scale results. Step 2: Preprocess the EEG data recorded in Step 1, calculate the event-related potential data of each channel, and obtain the microstate template and individual global field power curve through spatiotemporal clustering. Step 3: Based on the individual global field power curve obtained in Step 2, an adaptive time window selection mechanism is adopted to automatically identify multiple significant jump points using the dynamic change rate of the individual global field power curve. Multiple individualized analysis time windows are constructed with these points as the center to fully cover the micro-state switching dynamics during important components of the event-related potential. Step 4: Based on the individualized analysis time window determined in Step 3, a multi-objective optimization microstate backfitting method is adopted to construct an error and multi-index joint optimization function with the joint indices of maximizing spatial correlation, minimizing global field power error, and duration stability. First, the microstate label sequence is initialized using a greedy search method, and then a dynamic programming strategy is used to optimize the high-loss EEG data segments to minimize the error of the joint indices and obtain a stable microstate label sequence. Within this individualized analysis time window, the first-order microstate topological features are quantified, including global field power, global explained variance, occurrence frequency, duration, and transition probability between microstates. Step 5: Based on the stable microstate label sequence obtained in Step 4, perform band decomposition on the event-related potential signal corresponding to each microstate segment within the window, use the Morlet wavelet function to improve the time-frequency resolution, and generate a microstate connection matrix based on the connectivity index, which includes frequency segmentation conditions and multi-band fusion. Step 6: Based on the microstate connection matrix from Step 5, construct a second-order micronetwork with sparse optimization and frequency band fusion, wherein the optimal sparsity is determined based on structural information entropy; then perform binary processing on the second-order micronetwork and extract the second-order microstate topological features to achieve a quantitative representation of the neural dynamics structure. The second-order microstate topological features include clustering coefficients, feature path lengths, global efficiency, and local efficiency. Step 7: Compare the first-order microstate topology features and second-order microstate topology features under fatigue and non-fatigue conditions, and analyze the specific changes in parameters and network connections.

2. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 1, characterized in that, In step 3, the calculation method for the individualized analysis time window is as follows: for The corresponding individualized analysis time window, These are the significant transition points after screening. For one side of the fixed buffer zone, for The amount of local mutation at time; For all Standard deviation; This is the global field power curve; The first derivative of the global field power curve. This is the second derivative of the global field power curve.

3. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 1, characterized in that, In step 4, the multi-index joint optimization function is as follows: in For joint optimization functions, This is spatial correlation error. For global field power matching error, For the duration of stability error, The entropy weights for each indicator, q =1, 2, or 3; The calculation method is as follows: in For the first The first indicator in the The values ​​that can be taken on the candidate solutions, if they exist If there are 10 candidate solutions, then ; For the first The first indicator in the The relative weights on each candidate solution; For the first Information entropy of each indicator.

4. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 3, characterized in that, The calculation formulas for each indicator are as follows: in for Event-related spatial electrode potential distribution topographic map at a given time. For a moment The corresponding micro-state label index, for The corresponding micro-state template, For template matching correlation calculated based on Pearson correlation, For all those marked as The average global field power at a given time point. microstate Duration variance This represents the total number of microstate templates.

5. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 4, characterized in that, The calculation formula is as follows: in, For a moment The topographic map, i.e., the event-related potential vector of the whole brain electrodes; This is the mean value of the topographic map. Representing microstates template, For the number of electrodes, for The mean of the corresponding template, and For channel indexing.

6. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 1, characterized in that, In step 4, the dynamic programming strategy includes a local cost function with added penalty terms and a global cost function, the algorithms for which are as follows: in For a moment arrive A continuous EEG data segment is assigned a microstate label The resulting local cost; For traversal to The point in time, This indicates the end time of the previous fitted segment. This indicates the end time point of the current fitted segment. ; For a moment The event-related potential topography is represented as a voltage space vector containing all channels. ; It is a microstate; for and Pearson correlation; for The global field power; microstate The average global field power at the corresponding time point; This represents the number of frames the current fitted segment has lasted; microstate The target average sustained frame rate; The duration penalty factor; From the starting point to Minimum cumulative cost; From the starting point to The minimum cumulative cost is obtained by dynamic programming, which minimizes this value to obtain the optimal sequence of microstate labels.

7. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 1, characterized in that, In step 5, the method for calculating the micro-state connection matrix is ​​as follows: in For channel With channel In frequency band The corresponding microstate connection matrix, The elements of the matrix are the phase lag exponents of the inter-lead signals. For channel In frequency band The instantaneous phase, where This represents a local sampling moment within the current microstate segment. This represents the total number of sample points within the segment. Determine the positive or negative direction of the phase difference.

8. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 1, characterized in that, In step 5, the calculation of the micro-state connection matrix of the multi-band fusion is as follows: in For multi-band fusion micro-state connection matrix, and For electrode indexing, frequency band The weighting coefficients, the weights of all frequency bands, and the sum of the weights. .

9. The fatigue-state auditory target topology extraction method based on multi-order micronetworks as described in claim 1, characterized in that, In step 7, a second-order micronetwork is constructed based on the structural information entropy mechanism. The structural information entropy is calculated as follows: in For sparsity is Network structure information entropy measures the complexity of the connection distribution in a network structure. The larger the network size, the more balanced the network connection. For nodes Degree normalized weights, , ; For nodes The degree, It represents the sum of the degrees of all nodes in the network.

10. A system capable of performing the fatigue-state auditory target topology extraction method as described in claim 1, characterized in that, The system includes an experimental paradigm adjustment module, a data recording and preprocessing module, a first-order microstate topology extraction module, and a second-order micronetwork topology extraction module, wherein: The experimental paradigm adjustment module includes an experimental procedure display unit, an auditory target stimulus design unit, a background noise adjustment unit, and a KSS scale design unit. The data recording and preprocessing module includes an EEG data recording unit, an EEG preprocessing unit, a KSS scale scoring recording unit, and a stimulation parameter recording unit; The first-order microstate topology extraction module includes a trial window length selection unit, an adaptive time window adjustment unit, a multi-index optimization and inverse simulation unit, a first-order topology quantization unit, and a first-order topology visualization unit. The second-order micro-network topology extraction module includes a micro-state connection matrix unit, a sparsification unit, a second-order topology quantization unit, and a second-order topology visualization unit.

Images