An electroencephalogram signal processing method and device based on a brain function complex network and electronic equipment

Abstract

The application provides a brain function complex network-based electroencephalogram signal processing method and device and electronic equipment, and relates to the technical field of electroencephalogram signal processing. The method steps include obtaining electroencephalogram data of a sub-brain concussion group and a healthy control group, and preprocessing the electroencephalogram data. The preprocessed electroencephalogram data is separated according to attention subnetworks, and the phase lock values of the electroencephalogram data under different frequency bands are calculated respectively to generate a subnetwork functional connection matrix. Based on a head model and a brain atlas, the electroencephalogram data is mapped to a source space through a beamforming algorithm to generate a whole-brain functional connection matrix. Global indicators are extracted through the whole-brain functional connection matrix, and node indicators are extracted through the subnetwork functional connection matrix. A classification network model is established, the global indicators and the node indicators are fused and input into the trained classification network model, and the classification results of the sub-brain concussion group and the healthy control group are output. Compared with the traditional method, the accuracy and generalization ability of the classification network model are greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroencephalogram signal processing, and in particular to an electroencephalogram signal processing method and device based on brain function complex network and electronic equipment. BACKGROUND

[0002] Subconcussion refers to head impact without known or clinically diagnosed concussion, which can also occur when the body or torso is rapidly accelerated and decelerated, especially when the brain is free to move within the cranial cavity, resulting in a "drift" phenomenon. The main impact of subconcussion comes from its repeated occurrence, and cumulative exposure can lead to harmful consequences. More and more evidence shows that repetitive subconcussion is common in contact sports and explosive impacts, and can trigger a series of subacute and chronic complications.

[0003] However, in the prior art, the research on the neural mechanism of subconcussion mainly relies on clinical symptom observation, neuropsychological scale evaluation and conventional electroencephalogram analysis, generally using single modal characteristics, which is difficult to effectively integrate global topological properties and node-level functional characteristics, resulting in insufficient accuracy and generalization ability of the neural mechanism research on subconcussion. SUMMARY

[0004] The present application aims to provide an electroencephalogram signal processing method and device based on brain function complex network to solve the problem of insufficient accuracy and generalization ability of the neural mechanism research on subconcussion in the prior art mentioned in the background.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme: an electroencephalogram signal processing method based on brain function complex network, comprising the steps of: obtaining electroencephalogram data of subconcussion group and healthy control group, and preprocessing the electroencephalogram data; separating the preprocessed electroencephalogram data according to attention subnetwork, and calculating the phase lock value of the electroencephalogram data under different frequency bands respectively to generate subnetwork functional connection matrix; mapping the electroencephalogram data to source space based on head model and brain atlas through beamforming algorithm to generate whole brain functional connection matrix; extracting global indicators through the whole brain functional connection matrix and node indicators through the subnetwork functional connection matrix; establishing a classification network model, fusing the global indicators and the node indicators through graph theory and inputting them into the trained classification network model to output the classification results of the subconcussion group and the healthy control group.

[0006] Optionally, the step of obtaining the EEG data of the sub-concussion group and the healthy control group specifically comprises: aligning the physical positions of the EEG electrodes with a standard brain region template for channel positioning, and obtaining the EEG data of the sub-concussion group and the healthy control group; performing 0.5Hz-30Hz band-pass filtering and 50Hz notch filtering, downsampling, whole-brain average reference conversion, segmentation and baseline correction, bad segment deletion, removal of electrooculogram and electromyogram components, and removal of excessively large amplitude on the collected EEG data.

[0007] Optionally, the step of segmentation specifically comprises: under different stimulus cue conditions, taking the stimulus prompt as the origin to intercept a time period of 500ms before and 500ms after the stimulus; and under different stimulus target conditions, taking the target stimulus as the origin to intercept a time period of 1000ms before and 1000ms after the stimulus; and the step of baseline correction specifically comprises: under different stimulus cue conditions, performing baseline correction on 500ms before the stimulus; and under different stimulus target conditions, performing baseline correction on 1000ms before and 500ms before the stimulus.

[0008] Optionally, the step of separating the preprocessed EEG data according to the attention sub-networks specifically comprises: separating the preprocessed EEG data according to the vigilance network, the orientation network and the executive control network.

[0009] Optionally, the step of calculating the phase lock value of the EEG data under different frequency bands specifically comprises: calculating the phase lock value of the EEG data under the δ(1Hz-4Hz), θ(4Hz-8Hz), α(8Hz-13Hz) and β(13Hz-30Hz) frequency bands, respectively, and the calculation formula is: wherein, PLV is the phase lock value, N is the number of time points, e is the base number of natural logarithm, i is the imaginary unit, and ΔΦ t is the phase difference of each time point.

[0010] Optionally, the step of mapping the EEG data to the source space based on the head model and the brain atlas through the beamforming algorithm specifically comprises: constructing a forward model based on the BEM head model constructed in the MNI standard space and the AAL-116 brain atlas, combining the EEG signal position information of the EEG electrodes, adopting the beamforming algorithm to preliminarily position the EEG signal, quantifying the functional connection strength between different brain regions according to the positioning result, and aggregating the EEG signal to the 116 standard brain regions defined by the AAL-116 atlas through the nearest neighbor interpolation method.

[0011] Optionally, the global index includes the clustering coefficient, the network average shortest path length, the global efficiency and the local efficiency; and the node index includes the node degree, the betweenness centrality, the node efficiency, the node clustering coefficient and the node local efficiency.

[0012] Optionally, the step of fusing the global index and the node index through graph theory and inputting into the trained classification network model specifically comprises: combining the global index and the node index of all wave bands to obtain a target feature vector, and inputting the target feature vector into the classification network model.

[0013] In another aspect, the application also provides an electroencephalogram signal processing device based on brain function complex network, comprising: an acquisition module, which acquires electroencephalogram data of a sub-brain concussion group and a healthy control group, and pre-processes the electroencephalogram data; a small matrix generation module, which separates the pre-processed electroencephalogram data according to attention sub-networks, and respectively calculates the phase lock values of the electroencephalogram data under different frequency bands to generate sub-network functional connection matrices; a large matrix generation module, which maps the electroencephalogram data to source space based on a head model and a brain atlas through a beamforming algorithm to generate a whole brain functional connection matrix; a feature extraction module, which extracts global indexes through the whole brain functional connection matrix and extracts node indexes through the sub-network functional connection matrices; and a classification module, which establishes a classification network model, fuses the global indexes and the node indexes through graph theory, and inputs them into the trained classification network model to output classification results of the sub-brain concussion group and the healthy control group.

[0014] In another aspect, the application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the electroencephalogram signal processing method based on brain function complex network when executing the computer program.

[0015] Compared with the prior art, the application has the following beneficial effects:

[0016] By separating the electroencephalogram data of the three attention sub-networks of alertness network, directional network and executive control network, and combining the analysis of the delta / theta / alpha / beta multi-frequency bands, multi-dimensional analysis of sub-brain concussion attention network damage is realized; compared with the traditional single-frequency band or single network analysis method, the abnormal neural oscillation mode caused by repetitive head impact can be more comprehensively captured, thereby providing a new perspective for revealing the neurophysiological mechanism of sub-brain concussion.

[0017] The beamforming algorithm is combined with the BEM head model and the AAL-116 brain atlas to accurately map the scalp electrophysiological signals to the source space, and the spatial resolution reaches the level of 116 brain regions, which can accurately locate the brain regions and effectively overcome the interference of volume conduction effect.

[0018] The global index and the node index are fused through graph theory, a multi-scale network feature set is constructed, compared with single index analysis, the small world property change of the whole network can be described, and key brain areas can be located, the neural network features of sub-brain concussion can be comprehensively analyzed, and the robustness of the model classification is improved.

[0019] By fusing multi-dimensional graph theory features through a classification network model, a“electrophysiological signal-graph theory network-machine learning”trinity analysis framework is established, a new paradigm is provided for the neural marker research of mild brain injury, the research of brain concussion is promoted from the behavior to the neural network mechanism level, the automatic classification of sub-brain concussion is realized, and compared with the traditional way, the accuracy and generalization ability of the classification network model are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is a schematic diagram of the method steps of the present application.

[0021] Figure 2 It is a schematic diagram of the device structure of the present application.

[0022] In the figure: 10 - acquisition module, 20 - small matrix generation module, 30 - large matrix generation module, 40 - feature extraction module, 50 - classification module. DETAILED DESCRIPTION

[0023] The scheme of the present application will be clearly and completely explained by combining the drawings in the embodiments of the present application, obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments.

[0024] It should be noted that the terms“first”,“second” and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms“include” and“have” and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0025] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" and "comprising" and the like, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0026] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" and "comprising" and the like, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0027] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" and "comprising" and the like, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0028] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" and "comprising" and the like, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0029] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" and "comprising" and the like, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. Figure 1 The method comprises the following steps: acquiring electroencephalogram data of a sub-brain concussion group and a healthy control group, and preprocessing the electroencephalogram data; separating the preprocessed electroencephalogram data according to attention sub-networks, and respectively calculating phase locking values of the electroencephalogram data in different frequency bands to generate sub-network functional connection matrices; mapping the electroencephalogram data to a source space based on a head model and a brain atlas through a beamforming algorithm to generate a whole-brain functional connection matrix; extracting global indexes through the whole-brain functional connection matrix and node indexes through the sub-network functional connection matrices; establishing a classification network model, fusing the global indexes and the node indexes through graph theory, and inputting the global indexes and the node indexes into the trained classification network model to output classification results of the sub-brain concussion group and the healthy control group.

[0030] Specifically, the present application realizes multi-dimensional analysis of sub-brain concussion attention network damage by separating the EEG data of the alertness network, the directed network and the executive control network three attention sub-networks, and combining delta / theta / alpha / beta multi-band analysis. Compared with the traditional single frequency band or single network analysis method, it can more comprehensively capture the abnormal neural oscillation mode caused by repetitive head impact, and provide a new perspective for revealing the neurophysiological mechanism of sub-brain concussion. The beamforming algorithm combined with the BEM head model and the AAL-116 brain atlas is used to accurately map the scalp electrophysiological signals to the source space, and the spatial resolution reaches 116 brain region level, which can accurately locate the brain region and effectively overcome the interference of volume conduction effect. By integrating global indicators and node indicators through graph theory, a multi-scale network feature set is constructed, which can not only describe the small world property change of the whole network, but also locate the key brain region, and can comprehensively analyze the neural network characteristics of sub-brain concussion, and improve the robustness of the model classification. By integrating multi-dimensional graph theory features through the classification network model, a "electrophysiological signal-graph theory network-machine learning" three-in-one analysis framework is established, which provides a new paradigm for the research of neural markers of mild brain injury, promotes the research of brain concussion from the behavior to the neural network mechanism level, realizes the automatic classification of sub-brain concussion, and greatly improves the accuracy and generalization ability of the classification network model compared with the traditional method.

[0031] In some embodiments, the step of obtaining EEG data of the sub-brain concussion group and the healthy control group specifically includes: aligning the physical position of the EEG electrode with the standard brain region template for channel positioning, and obtaining the EEG data of the sub-brain concussion group and the healthy control group; performing 0.5Hz-30Hz band-pass filtering and 50Hz notch filtering, downsampling, whole brain average reference conversion, segmentation and baseline correction, bad segment deletion, removing electrooculogram and electromyogram components, and removing large amplitude on the collected EEG data.

[0032] Specifically, the application covers the main regions of the cerebral cortex with 32 electrodes, including frontal, parietal, temporal and occipital positions, aligns the physical position of the EEG electrode with the standard brain region template for channel positioning, ensures the accurate matching of the physical position of each electrode with the standard brain region template, and provides accurate data basis for subsequent source positioning and functional connection calculation. The 0.5Hz-30Hz band-pass filter can retain the brain rhythm related to cognitive activity, filter out low-frequency drift and high-frequency noise, and the 50Hz notch filter can eliminate power interference and improve signal purity. By performing 0.5Hz-30Hz band-pass filtering and 50Hz notch filtering on the collected EEG data, the signal-to-noise ratio can be improved, and the pollution of artifacts to the functional connection index can be avoided. The sampling rate is reduced from the original value to 250Hz by downsampling, reducing the data volume and reducing the computational complexity, while retaining the key frequency band information. The re-reference step changes the reference electrode from a single point to a whole brain average reference, eliminating the influence of the reference electrode position on the signal distribution, making the potential difference between brain regions more truly reflect neural activity, and avoiding false functional connections or network topology deviations caused by improper selection of the reference electrode. The segmentation step divides the EEG data into independent time periods under different stimulus cues cue conditions or different stimulus targets target conditions, for example, under different stimulus cues cue conditions, taking the appearance of the stimulus cue (no prompt, double prompt, central prompt and spatial prompt) as the origin, and the analysis time period is 500ms before the stimulus appears to 500ms after the stimulus appears; under different stimulus targets target conditions, taking the appearance of the stimulus target (consistent / inconsistent) as the origin, and the analysis time period is 1000ms before the stimulus appears to 1000ms after the stimulus appears, which facilitates direct comparison of network differences under different conditions. The baseline correction step eliminates the interference of the resting state potential on the task-related signal by subtracting the average potential of the baseline period before the stimulus, ensuring that the functional connection index reflects the dynamic changes related to the task, and avoiding the false increase or decrease of the phase lock value and other indicators caused by baseline drift. The bad segment deletion step deletes the trials containing artifacts (such as eye blinking, electromyogram, electrode falling off), ensuring the data quality. The ocular and electromyographic component removal step uses independent component analysis to separate and remove ocular, electromyographic and other non-brain-derived components, which can greatly improve the purity of the signal. The excessive amplitude removal step deletes the trials with abnormal amplitude with ±70μV as the standard, excluding extreme noise. Through the preprocessing of the EEG data, high-quality data input can be provided for subsequent tasks.

[0033] In some embodiments, the segmenting step specifically comprises: under different stimulus cue conditions, taking the stimulus cue as the origin, and intercepting a time period of 500 ms before and 500 ms after the stimulus; under different stimulus target conditions, taking the stimulus target as the origin, and intercepting a time period of 1000 ms before and 1000 ms after the stimulus; and the baseline correction step specifically comprises: under different stimulus cue conditions, baseline correction is performed on 500 ms before the stimulus; and under different stimulus target conditions, baseline correction is performed on 1000 ms before and 500 ms before the stimulus.

[0034] Specifically, under different stimulus cue conditions or under different stimulus target conditions, the electroencephalogram data is divided into independent time periods, for example, under different stimulus cue conditions, taking the appearance of the stimulus cue (no cue, double cue, central cue, and spatial cue) as the origin, and the analysis time period is 500 ms before the appearance of the stimulus and 500 ms after the appearance of the stimulus; under different stimulus target conditions, taking the appearance of the stimulus target (consistent / inconsistent) as the origin, and the analysis time period is 1000 ms before the appearance of the stimulus and 1000 ms after the appearance of the stimulus, which facilitates direct comparison of network differences under different conditions. The baseline correction step eliminates the interference of the resting state potential on the task-related signal by subtracting the average potential of the baseline period before the stimulus, ensures that the functional connection index reflects the dynamic changes related to the task, and avoids false increases or decreases in phase-locked values and other indicators caused by baseline drift.

[0035] In some embodiments, the step of separating the preprocessed electroencephalogram data according to attention sub-networks specifically comprises: separating the preprocessed electroencephalogram data according to the vigilance network, the orientation network, and the executive control network.

[0036] Specifically, the preprocessed electroencephalogram data is separated according to the vigilance network, the orientation network, and the executive control network, which decomposes complex cognitive functions into quantifiable and analyzable neural network units, not only improves the scientific rigor of sub-brain concussion research, but also provides clear neurobiological targets for subsequent research, skillfully links electrophysiological signals, cognitive functions, and neural damage together, and can accurately match electroencephalogram data with cognitive function dimensions to enhance sensitivity.

[0037] In some embodiments, the step of calculating the phase-locked value of the electroencephalogram data under different frequency bands specifically comprises: calculating the phase-locked value of the electroencephalogram data under the delta (1 Hz-4 Hz), theta (4 Hz-8 Hz), alpha (8 Hz-13 Hz), and beta (13 Hz-30 Hz) frequency bands, respectively, and the calculation formula is: In the formula, PLV is the phase-locked value, N is the number of time points, e is the base of the natural logarithm, i is the imaginary unit, and ΔΦ t is the phase difference of each time point.

[0038] It needs to be understood that different frequency band neural oscillations are closely related to specific cognitive functions and pathological states. By calculating the PLV in different frequency bands, the damage mode of sub-brain oscillation to different neural networks can be analyzed, wherein, δ (1Hz-4Hz) is related to slow wave activity, sleep regulation and compensation mechanism after brain injury, and abnormal increase may reflect chronic fatigue or white matter damage; θ (4Hz-8Hz) is related to attention, working memory and hippocampus-frontal pathway activity, and the decrease of directional network θ synchronization may indicate spatial attention deficit; α (8Hz-13Hz) is related to cortical inhibition, sensory gating and resting state activity of the default mode network, and the lack of α inhibition of the alert network may reflect decreased arousal; β (13Hz-30Hz) is related to motor control, high-level cognitive control and basal ganglia-cortical loop activity, and the abnormal β synchronization of the executive control network may indicate impaired conflict monitoring function.

[0039] Specifically, by calculating the phase locking value in different frequency bands, more detailed feature input can be provided for subsequent graph theory feature fusion and classification network model, the robustness of the classification network model is enhanced, and multi-dimensional and interpretable features are provided for the model.

[0040] In some embodiments, the step of mapping the electroencephalogram data to the source space based on the head model and the brain atlas specifically comprises: constructing a forward model based on the BEM head model and the AAL-116 brain atlas constructed in the MNI standard space, combining the electroencephalogram signal position information of the EEG electrodes, using a beamforming algorithm to preliminarily locate the electroencephalogram signal, using the phase locking value to quantify the functional connection strength between different brain regions according to the positioning result, and using the nearest neighbor interpolation method to aggregate the electroencephalogram signal to 116 standard brain regions defined by the AAL-116 atlas.

[0041] Specifically, the head model is a three-dimensional head structure model in neuroimaging research, which is a mathematical model for describing the electromagnetic signal propagation path of the head, connecting the signals recorded on the scalp with the brain source activity. By combining the anatomic structure information of the head, such as the conductive properties of the scalp, skull, cerebrospinal fluid and brain tissue, the head model can solve the forward problem, i.e. predicting the scalp signal distribution from the brain EEG signal, and solve the inverse problem, i.e. calculating the brain EEG signal from the scalp signal. The Boundary Element Model (BEM) volume conductor model based on the MNI standard space is adopted in the present application, which can accurately describe the conductive properties of different tissues of the head, providing a reliable basis for source analysis and functional connectivity construction. The brain atlas is used to identify and locate the spatial position of different brain regions and their relationship, which includes the anatomical structure, functional characteristics and the connection relationship between regions of the brain. The AAL atlas divides the brain into 116 regions, each of which has its unique number, and the AAL-116 brain atlas is used in the functional connectivity part of the present application, combined with the EEG signal position information of the EEG electrode to construct a forward model, and a beamforming algorithm is used to preliminarily locate the EEG signal, and according to the positioning result, the functional connection strength between different brain regions is quantified by using the phase-locked value, and the EEG signal is aggregated to the 116 standard brain regions defined by the AAL-116 atlas by using the nearest neighbor interpolation method, and the original EEG signal is converted from the electrode space to the anatomical space, solving the spatial ambiguity problem of the signal source, and providing a key basis for quantifying the functional connection between brain regions and revealing the neural mechanism of sub-brain concussion.

[0042] In some embodiments, the global indicators include the clustering coefficient, the network average shortest path length, the global efficiency, the local efficiency; the node indicators include the node degree, the betweenness centrality, the node efficiency, the node clustering coefficient, the node local efficiency.

[0043] Specifically, the clustering coefficient reflects the density of connections between the adjacent regions of a node, measures the ability of the whole network to transmit information in a short distance, and if the clustering coefficient of the subcerebral concussion group is significantly reduced, it may reflect the impaired ability of local information integration of attention networks, such as alertness network and directional network. The average shortest path length of the network describes the average value of the shortest path between any two nodes in the network, measures the global information transmission efficiency, and if the average shortest path length of the network of the subcerebral concussion group is significantly increased, it may indicate that the global information integration ability of the brain is decreased, resulting in impaired attention and executive function. The global efficiency describes the average value of the reciprocal of the shortest path length between any two nodes in the network, measures the ability of the whole network to transmit information in a long distance, and if the global efficiency of the subcerebral concussion group is significantly reduced, it may reveal the impaired global information transmission ability of the brain network. The local efficiency (Eloc) reflects the efficiency of the neighborhood subnetwork of a node after the node is removed, measures the local information integration ability between brain regions and the functional modular characteristics between regions, and if the local efficiency of the subcerebral concussion group is significantly reduced, it may indicate that the local functional modularization of attention networks, such as directional and executive control networks, is abnormal, resulting in decreased cognitive flexibility.

[0044] Specifically, the node degree represents the number of edges or neighbor nodes directly connected to a node, measures the direct connectivity of a node in the network, and if the node degree of a specific brain region, such as the frontal lobe or the parietal lobe, in the subcerebral concussion group is significantly reduced, it may indicate that the direct information input / output ability of these regions is impaired, resulting in abnormal attention or executive function. The betweenness centrality examines the contribution of each node to the shortest path between all other pairs of points, measures the importance of a node as a hub of information flow in the network, and if the betweenness centrality of the default network, such as the precuneus, in the subcerebral concussion group is significantly reduced, it may indicate that its ability to coordinate the default mode and task-related networks is impaired, resulting in decreased cognitive flexibility. The node efficiency reflects the efficiency of information transfer between a node and other nodes, measures the information transfer ability of a single node in the network, and is a manifestation of global efficiency at the node level. If the node efficiency of the sensorimotor area, such as the precentral gyrus, in the subcerebral concussion group is significantly reduced, it may be related to the decreased motor control ability. The node clustering coefficient reflects whether the direct neighbors of a node are connected to each other, measures the density of connections in the neighborhood of a single node, and if the node clustering coefficient of the temporal lobe in the subcerebral concussion group is significantly reduced, it may indicate that the local network efficiency related to speech or memory processing is decreased. The node local efficiency reflects the ability of other nodes in the neighborhood to maintain information flow when a node is removed, measures the information processing efficiency of a single node in its neighborhood network, and if the node local efficiency of the anterior cingulate gyrus in the subcerebral concussion group is significantly reduced, it may indicate that the stability of its emotion regulation or conflict monitoring network is impaired.

[0045] In some embodiments, the step of fusing the global indicators and the node indicators by graph theory and inputting into the trained classification network model specifically comprises: combining the global indicators and the node indicators of all wave bands to obtain a target feature vector, and inputting the target feature vector into the classification network model.

[0046] Specifically, the global indicators and the node indicators are fused by graph theory, the global indicators provide a whole network perspective, the node indicators supplement local details, and the combination of the two reveals the complexity of the network. In graph theory, a complex network can be expressed as a graph, and the brain network is abstracted as a graph composed of nodes and edges, i.e., brain regions and functional connections. Through graph theory, a complex brain system can be abstracted into a simple geometric representation, i.e., many nodes and the relationship between the nodes, and the global indicators and the node indicators are ingeniously fused together to provide accurate and multi-dimensional feature input for subsequent model classification tasks.

[0047] On the other hand, referring to Figure 2 The application also provides a brain function complex network-based electroencephalogram signal processing device, which comprises: an acquisition module 10 that acquires electroencephalogram data of a sub-brain concussion group and a healthy control group and pre-processes the electroencephalogram data; a small matrix generation module 20 that separates the pre-processed electroencephalogram data according to attention sub-networks and respectively calculates the phase lock values of the electroencephalogram data under different frequency bands to generate sub-network functional connection matrices; a large matrix generation module 30 that maps the electroencephalogram data to a source space based on a head model and a brain atlas through a beamforming algorithm to generate a whole brain functional connection matrix; a feature extraction module 40 that extracts global indicators from the whole brain functional connection matrix and extracts node indicators from the sub-network functional connection matrices; and a classification module 50 that establishes a classification network model, fuses the global indicators and the node indicators and inputs them into the trained classification network model to output a classification result of the sub-brain concussion group and the healthy control group.

[0048] On the other hand, the application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the brain function complex network-based electroencephalogram signal processing method when executing the computer program.

[0049] If the integrated unit is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0050] A person of ordinary skill in the art can understand that all or part of the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of the method. In the embodiments of the present application, any reference to memory, storage, database or other medium can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0051] The above is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent transformation or direct or indirect application in the related technical field based on the content of the present application specification and drawings is also included in the patent protection scope of the present application.

Claims

1. A method for processing electroencephalogram (EEG) signals based on brain function complex network, characterized by the following steps The method comprises the following steps: Obtaining electroencephalogram (EEG) data of a sub-concussion group and a healthy control group, and preprocessing the EEG data; Separating the preprocessed EEG data according to attention sub-networks, and calculating phase locking values (PLVs) of the EEG data in different frequency bands respectively to generate sub-network functional connectivity matrices; Mapping the EEG data to a source space based on a head model and a brain atlas by a beamforming algorithm to generate a whole-brain functional connectivity matrix; Extracting global indicators from the whole-brain functional connectivity matrix and node indicators from the sub-network functional connectivity matrices; Establishing a classification network model, fusing the global indicators and the node indicators by graph theory, and inputting the indicators into the trained classification network model to output classification results of the sub-concussion group and the healthy control group.

2. The brain function complex network-based electroencephalogram signal processing method according to claim 1, characterized in that, The step of obtaining EEG data of a sub-concussion group and a healthy control group and preprocessing the EEG data comprises the following steps: Aligning physical positions of EEG electrodes with a standard brain region template for channel positioning, and obtaining EEG data of a sub-concussion group and a healthy control group; Performing 0.5-30 Hz band-pass filtering and 50 Hz notch filtering, downsampling, whole-brain average reference conversion, segmentation and baseline correction, bad segment deletion, removal of electrooculogram (EOG) and electromyogram (EMG) components, and removal of excessive amplitude values on the collected EEG data.

3. The brain function complex network-based electroencephalogram signal processing method according to claim 2, characterized in that, The segmentation step comprises the following steps: Under different stimulus cue conditions, taking a stimulus prompt as the origin, and intercepting a time period of 500 ms before the stimulus to 500 ms after the stimulus; Under different stimulus target conditions, taking a target stimulus as the origin, and intercepting a time period of 1000 ms before the stimulus to 1000 ms after the stimulus; The baseline correction step comprises the following steps: Under different stimulus cue conditions, performing baseline correction on 500 ms before the stimulus; Under different stimulus target conditions, performing baseline correction on 1000 ms before the stimulus to 500 ms before the stimulus.

4. The brain function complex network-based electroencephalogram signal processing method according to claim 1, characterized in that, The step of separating the preprocessed EEG data according to attention sub-networks comprises the following steps: Separating the preprocessed EEG data according to an alertness network, a directional network, and an executive control network.

5. The brain function complex network-based electroencephalogram signal processing method according to claim 1, characterized in that, The step of calculating PLVs of the EEG data in different frequency bands respectively comprises the following steps: Calculating PLVs of the EEG data in δ (1-4 Hz), θ (4-8 Hz), α (8-13 Hz), and β (13-30 Hz) frequency bands respectively, and the calculation formula is as follows: where PLV is the phase lock value, N is the number of time points, e is the base of the natural logarithm, i is the imaginary unit, and ΔΦ is the phase difference for each time point. t where PLV is the phase lock value, N is the number of time points, e is the base of the natural logarithm, i is the imaginary unit, and ΔΦ is the phase difference for each time point.

6. The brain function complex network-based electroencephalogram signal processing method according to claim 5, characterized in that, The step of mapping the EEG data to a source space based on a head model and a brain atlas by a beamforming algorithm comprises the following steps: Based on a BEM head model and an AAL-116 brain atlas constructed in MNI standard space, a forward model is constructed in combination with EEG signal position information of EEG electrodes, a beamforming algorithm is used to preliminarily locate the EEG signals, the functional connectivity strength between different brain regions is quantified according to the location results, and the EEG signals are aggregated into 116 standard brain regions defined in the AAL-116 atlas by a nearest neighbor interpolation method. 7.The brain function complex network-based electroencephalogram signal processing method according to claim 1, characterized in that, The global indicators include the clustering coefficient, the network average shortest path length, the global efficiency, and the local efficiency; and the node indicators include the node degree, the betweenness centrality, the node efficiency, the node clustering coefficient, and the node local efficiency. 8.The brain function complex network-based electroencephalogram signal processing method according to claim 1, characterized in that, The step of fusing the global indicators and the node indicators by graph theory and inputting into the trained classification network model specifically includes: combining the global indicators and the node indicators of all wave bands to obtain a target feature vector, and inputting the target feature vector into the classification network model.

9. An electroencephalogram signal processing device based on brain function complex network, characterized in that, The method comprises the following steps: an acquisition module, which acquires electroencephalogram (EEG) data of a subconcussion group and a healthy control group, and pre-processes the EEG data; a small matrix generation module, which separates the pre-processed EEG data according to attention subnetworks, and respectively calculates phase locking values of the EEG data under different frequency bands to generate subnetwork functional connection matrices; a large matrix generation module, which maps the EEG data to a source space based on a head model and a brain atlas through a beamforming algorithm to generate a whole-brain functional connection matrix; a feature extraction module, which extracts global indicators from the whole-brain functional connection matrix, and extracts node indicators from the subnetwork functional connection matrices; a classification module, which establishes a classification network model, fuses the global indicators and the node indicators by graph theory, and inputs them into the trained classification network model to output a classification result of the subconcussion group and the healthy control group.

10. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the EEG signal processing method based on a brain function complex network according to any one of claims 1 to 8.

Images