Electroencephalogram-based cognitive decline risk assessment method and system

Abstract

The present invention relates to the technical field of medical data processing. Disclosed are an electroencephalogram-based cognitive decline risk assessment method and system. In the present invention, a plurality of electroencephalogram representations in electroencephalogram data of a target group and behavioral data of the target group are used to perform training to obtain a first cognitive decline risk assessment model, a second cognitive decline risk assessment model and a third cognitive decline risk assessment model, wherein the first cognitive decline risk assessment model mainly includes spectral information, information included in the second cognitive decline risk assessment model is mainly connection information of a brain network, and the third cognitive decline risk assessment model mainly includes neural spatio-temporal dynamics information. Thus, doctors can be assisted in performing efficient and accurate cognitive decline risk assessment on subjects to be assessed, and patients can be classified on the basis of the cognitive decline risk types of the patients, thereby helping the doctors to quickly formulate rational diagnosis and treatment strategies and improving the working efficiency and quality.

Description

A method and system for assessing cognitive decline risk based on EEG Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to an EEG-based cognitive decline risk assessment method and system. Background Art

[0002] The pathological features of mild cognitive impairment (MCI) are heterogeneous, with many lesions occurring on a continuum between individuals with dementia and those with cognitive integrity. Indeed, even individuals who do not exhibit any cognitive impairment may harbor neuropathological changes, particularly in the elderly. This may explain why a clear signature of the MCI state has not been established in neuropathological studies.

[0003] In recent years, there is no standardized measurement tool or cutoff score for most cognitive impairments encountered in clinical practice, including dementia, Alzheimer's disease, frontotemporal lobar degeneration, dementia with Lewy bodies, and vascular cognitive impairment. Clinical practice typically requires a combination of clinical history, neurological examination, psychiatric examination, and neuropsychological testing to make a final determination of whether a patient has cognitive impairment. Cognitive function measures can be very helpful in making this determination.

[0004] The MMSE is the most widely used cognitive screening test for general cognitive assessment by physicians. However, the MMSE has a ceiling effect, meaning that normal individuals have a limited dynamic performance range. This increases the likelihood that patients with pre-dementia will score within the normal range (24 points and above). Furthermore, the MMSE has poor sensitivity in distinguishing mild cognitive impairment (MCI), resulting in suboptimal cognitive screening.

[0005] Therefore, there is an urgent need for a new method to assess the risk of cognitive decline to assist doctors in measuring patients' cognitive function efficiently and accurately. SUMMARY OF THE INVENTION

[0006] The present invention provides an EEG-based cognitive decline risk assessment method and system to address the defect that doctors use existing technology to perform cognitive screening tests, but the cognitive screening effect is not ideal. Technical Solutions

[0007] The present invention provides a method for constructing a cognitive decline risk assessment model, comprising:

[0008] Obtaining EEG data and behavioral data of a target group, where the target group includes a healthy group and a group of patients at risk of cognitive decline (e.g., patients with impaired brain function, cognitive decline, behavioral disorders, etc.), and the EEG data has corresponding cognitive decline degree labels;

[0009] Preprocess EEG data;

[0010] Based on the preprocessed EEG data, EEG data and spectral features of multiple regular frequency bands of each individual in the target group are obtained, wherein the regular frequency bands include any one of the following or any combination thereof: Delta (1-4 Hz) frequency band, Theta (4-8 Hz) frequency band, Alpha (8-13 Hz) frequency band, Beta (13-30 Hz) frequency band;

[0011] Selecting an individualized alpha frequency band for each individual in the target group, and obtaining the individualized alpha frequency band EEG data and spectral features of each individual in the target group based on the preprocessed EEG data;

[0012] A first cognitive decline risk assessment model is trained based on the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group;

[0013] Tracing the EEG data of multiple regular frequency bands of each individual in the target group;

[0014] Based on the EEG data of multiple conventional frequency bands after tracing back to the source, PLV was used to construct a brain network. The constructed brain network was then correlated with the behavioral data of the target group to train a second cognitive decline risk assessment model.

[0015] The EEG data of multiple conventional frequency bands after tracing back to the source were segmented into networks to obtain multiple brain functional networks, and the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks were obtained. The third cognitive decline risk assessment model was obtained by training based on the self-time correlation window, permutation entropy, and neuronal variability ratio.

[0016] It should be noted that the EEG data is preferably the EEG data of the target group during a resting-state 64-channel EEG test. During the resting-state 64-channel EEG test, the first condition is to test the neurophysiological mechanism of maintaining a low-alert state with eyes closed. The target group is required to sit quietly, maintain a relaxed mental state (i.e., no goal-directed mental activity), and keep their eyes closed for several minutes (e.g., 8-15 minutes). The second condition is to test the neurophysiological mechanism of maintaining a moderate alertness with eyes open (e.g., 8-15 minutes). The target group is required to open their eyes, stare at a target on the wall, and remain focused.

[0017] Under the above-mentioned resting-state conditions, the experimenter's instructions to the subject must be very precise and consistent. These instructions may be difficult to implement for patients with impaired brain function, cognitive decline, and / or behavioral disorders (for example, patients with dementia caused by Alzheimer's disease). Therefore, the experimenter should pay attention to the subject's behavioral state during the rsEEG (resting-state EEG) recording process, take notes, and instruct the subject to delete rsEEG recording periods characterized by drowsiness and alertness (large offsets and sleep waves in the data) to obtain more accurate EEG data for the target group.

[0018] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, preprocessing of EEG data includes any one of the following items or any combination thereof: downsampling (for example, downsampling to 500 Hz), removing bad segments, interpolating bad leads, filtering, and using ICA to remove artifact interference (artifacts such as electrooculography, electromyography, and electrocardiography).

[0019] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the spectral characteristics of multiple regular frequency bands include the relative energy of the regular frequency bands and the power spectral density of the regular frequency bands. The EEG data and spectral characteristics of multiple regular frequency bands of each individual in the target group are obtained based on the preprocessed EEG data, including:

[0020] According to the EEG data of each regular frequency band, the sum of the squares of the amplitudes of each regular frequency band is obtained as the energy of each regular frequency band;

[0021] According to the energy of each regular frequency band, a ratio of the energy of each regular frequency band to the sum of the energies of all regular frequency bands is obtained as the relative energy of each regular frequency band, and a relative energy ranking of the plurality of regular frequency bands is obtained;

[0022] According to the EEG data of each regular frequency band, the power spectrum density of each regular frequency band is obtained by fast Fourier transform, and the ratio of the power spectrum density between multiple regular frequency bands is obtained.

[0023] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the expression of power spectral density is:

[0024]

[0025] In the expression of power spectrum density, PSD represents power spectrum density, N represents the number of sampling points of EEG signal, represents the result of discrete Fourier transform, that is , Represents the EEG signals of each regular frequency band, Indicates the sampling frequency.

[0026] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the individualized alpha band of each individual in the target group is selected, and the individualized alpha band EEG data and spectral characteristics of each individual in the target group are obtained based on the preprocessed EEG data, including:

[0027] According to the pre-processed EEG data, broadband filtered data and narrowband filtered data are obtained, and the broadband filtered data and narrowband filtered data are non-time filtered data;

[0028] creating a first channel covariance matrix based on the broadband filtered data;

[0029] For each frequency band within a preset range (e.g., 4-13 Hz), a second channel covariance matrix is ​​created based on the narrowband filtered data, and eigendecomposition is performed on the first channel covariance matrix and the second channel covariance matrix to obtain a separating eigenvector for each frequency band, wherein the separating eigenvector can separate the first channel covariance matrix and the second channel covariance matrix to the greatest extent while suppressing data features represented in the first channel covariance matrix and the second channel covariance matrix;

[0030] According to the separated eigenvectors of each frequency band, a eigenvector similarity matrix is ​​constructed, and cluster analysis is performed on the eigenvector similarity matrix to obtain clusters, wherein the clusters represent the frequency band range and the edges of the clusters represent the frequency band boundaries;

[0031] Based on the eigenvector similarity matrix, a first frequency band range was selected in the Alpha (8-13HZ) frequency band, the number of clusters was recorded and the average frequency of the clusters was obtained. A second frequency band range was selected in the Theta (4-8HZ) frequency band, the number of clusters in the second frequency band was recorded and the average frequency of the clusters was obtained. The coupling degree between the average frequency of the clusters in the first frequency band and the average frequency of the clusters in the second frequency band was obtained.

[0032] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the first cognitive decline risk assessment model is trained based on the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group, including:

[0033] A first cognitive decline risk assessment model is trained based on the relative energy and power spectral density ratio of multiple conventional frequency bands, the number of clusters in the first frequency band range, the average frequency of the clusters, and the degree of coupling between the average frequency of the clusters in the first frequency band range and the average frequency of the clusters in the second frequency band range.

[0034] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, tracing the EEG data of multiple conventional frequency bands of each individual in the target group includes:

[0035] A tracing algorithm was used to trace the EEG data of multiple regular frequency bands of each individual in the target group, and the brain regions were segmented by combining the ALL template, and the 64-lead EEG data were converted into EEG data of 116 brain regions.

[0036] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, a brain network is constructed using PLV based on EEG data of multiple conventional frequency bands after tracing, and correlation analysis is performed between the constructed brain network and behavioral data of a target group to train a second cognitive decline risk assessment model, including:

[0037] The phase difference time series distribution of EEG signals between brain regions was evaluated by PLV expression to construct brain networks;

[0038] The constructed brain network was correlated with the behavioral data of the target group to obtain statistically significant network loops. The statistically significant network loops were divided into positively correlated loops and negatively correlated loops according to frequency bands. Each network loop was normalized and regression analysis was performed on the behavioral data to train a second cognitive decline risk assessment model.

[0039] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the PLV expression is:

[0040]

[0041] In PLV expression, The time series of the phase difference of EEG signals between brain regions is The distribution of express and The phase difference, and Represents the EEG signals of different brain regions, N represents The length (i.e. number of points) of .

[0042] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the behavioral data of the target group is the MoCa (Montreal Cognitive Assessment) scale characteristics of the target group. The constructed brain network is subjected to correlation analysis with the behavioral data of the target group to obtain network loops with statistical differences, and the network loops with statistical differences are divided into positive correlation loops and negative correlation loops according to frequency bands, including:

[0043] Conduct correlation analysis between the constructed brain network and the behavioral data of the target group, and select network loops with statistical parameters less than a preset threshold (e.g., p less than 0.05) as network loops with statistical differences;

[0044] According to the frequency band, the network loops with a Pearson correlation greater than 0 with the behavioral data were considered as positively correlated loops, and the network loops with a Pearson correlation less than 0 with the behavioral data were considered as negatively correlated loops.

[0045] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the brain functional network includes any one of the following or any combination thereof: limbic network, visual network, somatosensory network, frontoparietal network, dorsal attention network, ventral attention network, and default network. The EEG data of multiple conventional frequency bands after tracing are subjected to network segmentation to obtain multiple brain functional networks, and the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks are obtained. A third cognitive decline risk assessment model is obtained by training based on the self-time correlation window, permutation entropy, and neuronal variability ratio, including:

[0046] The self-time correlation window of each brain functional network is obtained by the calculation formula of the self-time correlation window;

[0047] The permutation entropy of each brain functional network is obtained by the permutation entropy calculation formula;

[0048] By calculating the standard deviation of the EEG signals of each brain functional network, the neuronal variability ratio of each brain functional network was obtained.

[0049] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the calculation formula of the self-time correlation window is:

[0050]

[0051] In the calculation formula of the self-time correlation window, ACW represents the self-time correlation window. represents the result of continuous Fourier transform, that is , t represents time, x represents the EEG signals of different brain functional networks, Indicates frequency.

[0052] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the calculation formula of the permutation entropy is:

[0053]

[0054] In the calculation formula of permutation entropy, represents the permutation entropy, and x represents the EEG signal of each brain functional network Indicates the probability of the xth sample point appearing in the EEG signal.

[0055] The present invention also provides a method for assessing cognitive decline risk, comprising:

[0056] Receiving EEG data of a subject, wherein the subject is a healthy person, a patient suspected of being at risk of cognitive decline, or a patient at risk of cognitive decline;

[0057] According to the EEG data of the subject, the first cognitive decline risk of the subject is predicted by the first cognitive decline risk assessment model. When the first cognitive decline risk of the subject is higher than or equal to the preset high-risk threshold, the prediction result of the first cognitive decline risk of the subject is output as high risk. When the first cognitive decline risk of the subject is lower than the preset high-risk threshold and equal to or higher than the preset medium-risk threshold, the prediction result of the first cognitive decline risk of the subject is output as medium risk. When the first cognitive decline risk of the subject is lower than the preset medium-risk threshold, the second cognitive decline risk of the subject is predicted by the second cognitive decline risk assessment model. When the second cognitive decline risk of the subject is When the risk is greater than or equal to the preset low-risk threshold, the prediction result that the second cognitive decline risk of the subject is low risk is output; when the second cognitive decline risk of the subject is less than the preset low-risk threshold, the EEG data of the subject is clustered through the third cognitive decline risk assessment model; when the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is less than or equal to the preset distance threshold, the prediction result that the third cognitive decline risk of the subject is low risk is output; when the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is greater than the preset distance threshold, the prediction result that the subject has no cognitive decline risk is output.

[0058] The present invention also provides a cognitive decline risk assessment system, comprising:

[0059] A data receiving module is used to receive EEG data of a subject, wherein the subject is a healthy person, a patient suspected of having a risk of cognitive decline, or a patient at risk of cognitive decline;

[0060] The evaluation module is used to: predict the first cognitive decline risk of the subject to be tested through the first cognitive decline risk evaluation model according to the EEG data of the subject to be tested; when the first cognitive decline risk of the subject to be tested is higher than or equal to the preset high-risk threshold, output the prediction result that the first cognitive decline risk of the subject to be tested is high risk; when the first cognitive decline risk of the subject to be tested is lower than the preset high-risk threshold and equal to or higher than the preset medium-risk threshold, output the prediction result that the first cognitive decline risk of the subject to be tested is medium risk; when the first cognitive decline risk of the subject to be tested is lower than the preset medium-risk threshold, predict the second cognitive decline risk of the subject to be tested through the second cognitive decline risk evaluation model; when the second cognitive decline risk of the subject to be tested is lower than the preset medium-risk threshold, output the prediction result that the first cognitive decline risk of the subject to be tested is medium risk; When the risk of cognitive decline is greater than or equal to the preset low-risk threshold, the prediction result that the second cognitive decline risk of the subject is low risk is output; when the second cognitive decline risk of the subject is less than the preset low-risk threshold, the EEG data of the subject is clustered through the third cognitive decline risk assessment model; when the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is less than or equal to the preset distance threshold, the prediction result that the third cognitive decline risk of the subject is low risk is output; when the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is greater than the preset distance threshold, the prediction result that the subject has no cognitive decline risk is output.

[0061] The present invention also provides an electronic device, comprising a processor and a memory storing a computer program, wherein when the processor executes the computer program, the method for constructing any of the above-mentioned cognitive decline risk assessment models is implemented.

[0062] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned methods for constructing a cognitive decline risk assessment model.

[0063] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute any of the above-mentioned methods for constructing a cognitive decline risk assessment model. Beneficial effects

[0064] The present invention provides an EEG-based cognitive decline risk assessment method and system, which utilizes multiple EEG representations in the EEG data of the target group and behavioral data training to obtain a first cognitive decline risk assessment model, a second cognitive decline risk assessment model, and a third cognitive decline risk assessment model.

[0065] The first cognitive decline risk assessment model primarily includes spectral information, which is relatively stable and not easily altered by treatment. This information is primarily found in patients with mild cognitive impairment and dementia. It can be used to define medium and high risk of cognitive decline, representing a stable brain function state and serving as a priority diagnostic criterion. The second cognitive decline risk assessment model primarily includes brain network connectivity information, which can be restored with treatment and is highly variable. This model primarily includes patients with subjective cognitive decline and mild cognitive impairment, and can be used to define low and medium risk of cognitive decline, representing neuroplasticity and serving as a secondary diagnostic criterion. The third cognitive decline risk assessment model primarily includes neural spatiotemporal dynamics information. This feature appears in the early stages of cognitive decline, namely the subjective cognitive decline stage, and can be used to define low risk of cognitive decline. It is primarily used for early risk warning. The first, second, and third cognitive decline risk assessment models can represent the progression of cognitive decline, from the early, recoverable third and second cognitive decline risk assessment models to the stable, difficult-to-recover first cognitive decline risk assessment model.

[0066] The first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model can assist doctors in conducting efficient and accurate cognitive decline risk assessments on subjects, and can classify patients according to their cognitive decline risk types, helping doctors to quickly formulate reasonable diagnosis and treatment strategies and improve work efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0068] FIG1 is a flow chart showing one of the methods for constructing a cognitive decline risk assessment model provided by the present invention.

[0069] Figure 2 is a second flow chart of a method for constructing a cognitive decline risk assessment model provided by the present invention, wherein Model 1, Model 2, and Model 3 correspond to the first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model, respectively.

[0070] Figure 3 is a flow chart of cognitive screening using a cognitive decline risk assessment model provided by the present invention, wherein Model 1, Model 2, and Model 3 correspond to the first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model, respectively.

[0071] FIG4 is a schematic structural diagram of a system for constructing a cognitive decline risk assessment model provided by the present invention.

[0072] FIG5 is a schematic structural diagram of an electronic device provided by the present invention. Modes for Carrying Out the Invention

[0073] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the embodiments described are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0074] The following describes the EEG-based cognitive decline risk assessment method and system provided by the present invention with reference to FIG1 to FIG5 .

[0075] Figures 1 and 2 are flow charts of a method for constructing a cognitive decline risk assessment model provided by the present invention. Referring to Figure 1 , a method for constructing a cognitive decline risk assessment model provided by the present invention may include:

[0076] Step S110: Acquire EEG data and behavioral data of a target group, wherein the target group includes a healthy group and a patient group at risk of cognitive decline (e.g., a patient group with impaired brain function, cognitive decline, behavioral disorders, etc.), and the EEG data has a corresponding cognitive decline degree label;

[0077] Step S120: Preprocess the EEG data. In one embodiment, the preprocessing may include any one of the following or any combination thereof: downsampling (e.g., downsampling to 500 Hz), removing bad segments, interpolating bad leads, filtering, and using ICA to remove artifacts (e.g., electrooculography, electromyography, and electrocardiography). Various preprocessing functions of the EEG data may be used to implement these functions.

[0078] Step S130: Obtain EEG data and spectral features of multiple conventional frequency bands for each individual in the target group based on the preprocessed EEG data, wherein the conventional frequency bands include any one of the following or any combination thereof: Delta (1-4 Hz) frequency band, Theta (4-8 Hz) frequency band, Alpha (8-13 Hz) frequency band, and Beta (13-30 Hz) frequency band;

[0079] Step S140: selecting an individualized alpha band for each individual in the target group, and obtaining the individualized alpha band EEG data and spectral features of each individual in the target group based on the preprocessed EEG data;

[0080] Step S150: training a first cognitive decline risk assessment model based on the spectral characteristics of multiple regular frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group;

[0081] Step S160: tracing the EEG data of multiple regular frequency bands of each individual in the target group;

[0082] Step S170: constructing a brain network using PLV based on the EEG data of multiple conventional frequency bands after tracing, and performing correlation analysis between the constructed brain network and the behavioral data of the target group to train a second cognitive decline risk assessment model;

[0083] Step S180: Perform network segmentation on the EEG data of multiple regular frequency bands after tracing back to obtain multiple brain functional networks, and obtain the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks; and obtain a third cognitive decline risk assessment model based on the self-time correlation window, permutation entropy, and neuronal variability ratio training.

[0084] In one embodiment, the EEG data is preferably EEG data of the target group during a resting-state 64-channel EEG test.

[0085] The following will describe in detail the process of the target group undergoing a resting-state 64-channel EEG test.

[0086] (1) Preliminary assessment of the subject's condition before EEG recording

[0087] On the days leading up to the rsEEG rhythm recording, instruct the subjects to have a regular sleep schedule the night before the recording. Subjects should also be informed to refrain from consuming psychoactive substances and medications (i.e., food and beverages including nicotine, caffeine, alcohol, and any other form of stimulants) on the morning of the experiment.

[0088] Subjects may take psychoactive medications (i.e., benzodiazepines, antidepressants, etc.) as normal the day before the EEG recording, but not on the morning of the recording (this behavioral decision should obviously be agreed upon after appropriate clinical consultation).

[0089] In this setting, residual effects of benzodiazepines or other psychoactive drugs should be anticipated. For example, residual effects on rsEEG activity may occur the following day. Furthermore, residual drug agents may lead to an increase in the rsEEG beta rhythm and other effects the following day. Furthermore, patients dependent on certain medications may experience altered psychophysiological states due to delays in taking their medication that day, leading to personal effects such as anxiety or early sleepiness. These factors should be addressed in the general subject assessment prior to rsEEG recording.

[0090] The best time to record rsEEG rhythms is in the morning after a satisfying light breakfast.

[0091] A brief interview was conducted with the subjects to confirm the sleep quality of the standard subjects the night before the recording and the above conditions.

[0092] (2) EEG recording environment requirements

[0093] A quiet and dimly lit room.

[0094] Noise levels inside the room should be negligible.

[0095] The subject should lie in a comfortable semi-reclined armchair or bed.

[0096] The wall in front of him / her should be painted with a uniform light-colored (e.g., white, light yellow, or green) target, with only one central fixation target at his / her eye level.

[0097] (3) EEG recording

[0098] The subjects were asked to wear a 64-lead electrode cap and apply conductive paste to each lead electrode to ensure that the impedance between the skin and the electrode was less than 50k ohms. The sampling rate was set to 5000 Hz.

[0099] The first condition tests the neurophysiological mechanisms underlying the maintenance of a low-alert state with eyes closed, requiring the subject to sit quietly, maintain a relaxed mental state (i.e., no goal-directed mental activity), and keep their eyes closed for several minutes (e.g., 8–15 minutes).

[0100] The second condition is to test the neurophysiological mechanism of maintaining moderate alertness with eyes open (e.g., 8-15 minutes), requiring the target group to open their eyes, look at a target on the wall and maintain concentration.

[0101] Under the above-mentioned resting-state conditions, the experimenter's instructions to the subject must be very precise and consistent. These instructions may be difficult to implement for patients with impaired brain function, cognitive decline, and / or behavioral disorders (for example, patients with dementia caused by Alzheimer's disease). Therefore, the experimenter should pay attention to the subject's behavioral state during the rsEEG (resting-state EEG) recording process, take notes, and instruct the subject to delete rsEEG recording periods characterized by drowsiness and alertness (large offsets and sleep waves in the data) to obtain more accurate EEG data for the target group.

[0102] EEG data is used to assess and analyze the risk of cognitive decline. Due to its high temporal resolution, simple equipment and ease of use, EEG can not only objectively reflect the state of the brain, but also has simpler implementation conditions than MRI, making it easier to promote in clinical environments.

[0103] In one embodiment, step S130 may filter the preprocessed EEG data to obtain EEG data of a regular frequency band (fixed frequency band). The spectral characteristics of the regular frequency band may include the relative energy of the regular frequency band and the power spectral density of the regular frequency band.

[0104] In one embodiment, step S130 may include:

[0105] According to the EEG data of each regular frequency band, the sum of the squares of the amplitudes of each regular frequency band is obtained as the energy of each regular frequency band;

[0106] According to the energy of each regular frequency band, the ratio of the energy of each regular frequency band to the sum of the energies of all regular frequency bands is obtained as the relative energy of each regular frequency band (i.e., the relative power in FIG2 ), and the relative energy ranking of the multiple regular frequency bands is obtained;

[0107] According to the EEG data of each regular frequency band, the power spectrum density of each regular frequency band is obtained by fast Fourier transform, and the ratio of the power spectrum density between multiple regular frequency bands is obtained (i.e., the frequency band power ratio in Figure 2).

[0108] Among them, Power Spectral Density (PSD) is a characteristic that describes the power distribution carried by a signal or system in each frequency band. For discrete signals, its power spectral density can be calculated using the Fast Fourier Transform (FFT). The expression for power spectral density can be:

[0109]

[0110] In the expression of power spectrum density, PSD represents power spectrum density, N represents the number of sampling points of EEG signal, represents the result of discrete Fourier transform, that is , Represents the EEG signals of each regular frequency band, Indicates the sampling frequency.

[0111] Some existing studies have examined fixed frequency boundaries, but fixed boundaries prevent more nuanced and potentially informative analysis of how these boundaries vary across groups (e.g., patient populations or genetic backgrounds) or other individual factors (e.g., age, personality, task performance, etc.). Therefore, this example combines conventional frequency bands with personalized alpha bands for cognitive decline risk assessment analysis.

[0112] In one embodiment, step S140 may use a gedBounds method to extract the individualized alpha band. The gedBounds method can be divided into three stages. Step S140 may include:

[0113] According to the pre-processed EEG data, broadband filtered data and narrowband filtered data are obtained, and the broadband filtered data and narrowband filtered data are non-time filtered data;

[0114] Create the first channel covariance matrix based on the broadband filtered data, denoted as matrix R, where matrix R can be obtained by calculating the covariance of 64-channel broadband data (i.e., preprocessed EEG data);

[0115] For each frequency band within a preset range (e.g., 4-13 Hz), a second channel covariance matrix is ​​created based on the narrowband filtered data, denoted as matrix S (matrix S can be obtained by calculating the covariance of narrowband data (i.e., broadband data is filtered to obtain data of a specific frequency band; this embodiment mainly examines the 4-13 Hz frequency band)), and eigendecomposition is performed on the first channel covariance matrix and the second channel covariance matrix, and a generalized eigendecomposition of matrix S and matrix R is calculated (this decomposition identifies a spatial filter (a set of weights for all channels) to separate matrix S and matrix R to the greatest extent possible while suppressing data features represented in the two matrices), obtaining a separation eigenvector for each frequency band (i.e., storing the eigenvector with the greatest separation), wherein the separation eigenvector can separate the first channel covariance matrix and the second channel covariance matrix to the greatest extent while suppressing data features represented in the first channel covariance matrix and the second channel covariance matrix;

[0116] constructing a feature vector similarity matrix based on the separated feature vectors of each frequency band, and performing cluster analysis on the feature vector similarity matrix to obtain clusters (i.e., calculating the pairwise square correlations between the separated feature vectors from all frequencies within a preset range and storing them in the feature vector similarity matrix, and then performing cluster analysis on the feature vector similarity matrix to determine highly similar "blocks" on the diagonal), wherein the clusters represent the frequency band ranges and the edges of the clusters represent the frequency band boundaries;

[0117] Based on the eigenvector similarity matrix, a first frequency band range was selected in the Alpha (8-13 Hz) frequency band, the number of clusters (i.e., frequency band principal components) in the first frequency band was recorded, and the average frequency of the clusters was obtained. A second frequency band range was selected in the Theta (4-8 Hz) frequency band, the number of clusters in the second frequency band was recorded, and the average frequency of the clusters was obtained. The coupling degree between the average frequency of the clusters in the first frequency band and the average frequency of the clusters in the second frequency band was also obtained. The coupling degree was defined as: 2*theta frequency -alpha frequency.

[0118] In one embodiment, step S150 can train a first cognitive decline risk assessment model based on the relative energy and power spectral density ratios of multiple regular frequency bands, the number of clusters in the first frequency band, the average frequency of the clusters, and the degree of coupling between the average frequency of the clusters in the first frequency band and the average frequency of the clusters in the second frequency band. That is, the first cognitive decline risk assessment model learns the relationship between spectral EEG data and different cognitive decline risk levels.

[0119] In one embodiment, step S160 can use a tracing algorithm to trace the EEG data of multiple regular frequency bands of each individual in the target group, combine the ALL template to segment the brain regions, and convert the 64-lead EEG data into EEG data of 116 brain regions.

[0120] Specifically, EEG source localization, also known as the inverse EEG problem, involves inferring the estimated location, direction, and intensity of the source of neural activity within the brain based on the potential signals recorded by the headpiece. The inverse EEG problem is a nonlinear optimization problem. Given the computational complexity, this embodiment approximates it to a linear problem, Y = AX, where Y = the actual signal recorded by the headpiece electrodes, X = the source information vector to be spatially localized, and A = the transfer matrix (gain matrix). This is the solution to the forward EEG problem, which can be obtained by constructing a suitable head model. Specifically, the scalp EEG signal = cortical electrical activity × the transfer matrix (conduction effects of cerebrospinal fluid, meninges, skull, scalp, etc.). The specific algorithm uses the MNE algorithm included with the Brainstorm toolkit.

[0121] Finally, the scalp EEG data was mapped to different brain regions using a traceability algorithm. Brain segmentation was performed using the AAL template, an anatomically based automatic labeling atlas used for brain image analysis. The AAL template divides the brain into multiple regions based on its structure and function. Each region corresponds to a specific brain function or anatomical structure, encompassing a total of 116 brain regions. Ultimately, the 64-lead EEG data was converted into EEG data for 116 brain regions.

[0122] In one embodiment, step S170 may include:

[0123] The PLV expression is used to evaluate the time series distribution of the phase difference of EEG signals between brain regions to construct the brain network. The PLV expression is:

[0124]

[0125] In PLV expression, The time series of the phase difference of EEG signals between brain regions is A larger PLV value indicates that the phase difference time series distribution only occupies the unit circle. A very small part, the PLV value range is [0, 1]. If PLV is equal to 1, it means that the phase difference time series distribution is constant in the entire time series range; if PLV is less than 1, it means that the phase difference is evenly distributed in Within the range, express and The phase difference, and Represents the EEG signals of different brain regions, N represents The length (i.e., number of points) of

[0126] The constructed brain network was correlated with the behavioral data of the target group to obtain statistically significant network loops. The statistically significant network loops were divided into positively correlated loops and negatively correlated loops according to frequency bands. Each network loop was normalized and regression analysis was performed on the behavioral data to train a second cognitive decline risk assessment model.

[0127] In one embodiment, the behavioral data of the target group is preferably the MoCa scale (Montreal Cognitive Assessment) characteristics of the target group. The constructed brain network can be subjected to correlation analysis with the behavioral data of the target group, and network loops with statistical parameters less than a preset threshold (for example, p less than 0.05) are selected as network loops with statistical differences; according to the frequency band, network loops with a Pearson correlation greater than 0 with the behavioral data are regarded as positively correlated loops, and network loops with a Pearson correlation less than 0 with the behavioral data are regarded as negatively correlated loops.

[0128] This embodiment uses multidimensional brain network features to directly correlate with behavioral scales to obtain multidimensional brain network loops related to cognition, rather than single or multiple connections, which can increase the stability of the second cognitive decline risk assessment model.

[0129] Based on the results of brain imaging research, the 116 brain regions can be divided into seven networks based on brain function. Each network contains multiple brain regions, referred to as brain function. The brain function network in this embodiment can include any one of the following or any combination thereof: limbic network, visual network, somatosensory network, frontoparietal network, dorsal attention network, ventral attention network, and default network.

[0130] In one embodiment, step S180 may include:

[0131] The auto-temporal correlation window (ACW) of each brain functional network is obtained by the calculation formula of the auto-temporal correlation window, which characterizes the separation and integration functions of each brain functional network;

[0132] The permutation entropy (PE) of each brain functional network is obtained by the permutation entropy calculation formula, which characterizes the functional activation of each brain functional network;

[0133] By calculating the standard deviation of the EEG signals of each brain functional network, the neuronal variability ratio of each brain functional network is obtained, which characterizes the target group's ability to balance external and internal sensations.

[0134] Among them, the autocorrelation window (ACW) is a tool used to measure the degree of correlation between brain neural activity and itself at different time points. The calculation formula of the autocorrelation window is:

[0135]

[0136] In the calculation formula of the time correlation window, ACW represents the auto-time correlation window, represents the result of continuous Fourier transform, that is , t represents time, x represents the EEG signals of different brain functional networks, Indicates frequency.

[0137] The autocorrelation window can be used to study the dynamic changes in brain neural activity and time-related information processing and integration. Specifically, it can be used to measure changes in the delay and temporal correlation of brain neural activity, and to explore the time course and time scale of brain information processing.

[0138] And, the calculation formula of permutation entropy is:

[0139]

[0140] In the calculation formula of permutation entropy, represents the permutation entropy, x represents the EEG signal of each brain functional network, Indicates the probability of the xth sample point appearing in the EEG signal.

[0141] In this embodiment, by analyzing the cognitive decline risk level labels of the original EEG data, the first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model can determine thresholds for differentiating cognitive decline risk levels, such as a high-risk threshold, a medium-risk threshold, and a low-risk threshold. Furthermore, by analyzing the effectiveness and characteristic characterization before and after treatment, the hierarchical significance between the first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model can be determined to determine the hierarchical diagnostic process and the clinical significance of each model.

[0142] A target population of 90 individuals, including an EEG dataset and a behavioral dataset (Moca scale), includes 18 healthy individuals, 7 individuals with subjective cognitive decline, 11 individuals with mild cognitive impairment, and 18 individuals with dementia. Data from each individual after rTMS treatment is also included, for a total of 18 + (7 + 11 + 18) * 2 = 90 individuals. The risk classification thresholds for the first and second cognitive decline risk assessment models are calculated based on the maximum data classification accuracy. The third cognitive decline risk assessment model is specifically designed as follows: EEG data from the healthy group and the patient group at risk of cognitive decline are clustered, and the distance between the EEG data of the individual being tested and the healthy group cluster and the patient group at risk of cognitive decline cluster cluster is calculated, with the individual being assigned to the closest cluster.

[0143] When actually using the first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model constructed by the method for constructing a cognitive decline risk assessment model provided by the present invention, the EEG data of the subject can be obtained first, and then the EEG data of the subject can be input into the first cognitive decline risk assessment model to predict the first cognitive decline risk value of the subject. When the first cognitive decline risk value of the subject is higher than or equal to the preset high-risk threshold, the first cognitive decline risk assessment model outputs a prediction result that the first cognitive decline risk of the subject is high risk. When the first cognitive decline risk value of the subject is less than the preset high-risk threshold and equal to or greater than the preset medium-risk threshold, the first cognitive decline risk assessment model outputs a prediction result that the first cognitive decline risk of the subject is medium risk. When the first cognitive decline risk value of the subject is less than the preset medium-risk threshold, the first cognitive decline risk assessment model continues to output a prediction result that the first cognitive decline risk of the subject is medium risk. The second cognitive decline risk assessment model predicts the second cognitive decline risk value of the subject. When the second cognitive decline risk value of the subject is greater than or equal to the preset low risk threshold, the second cognitive decline risk assessment model outputs a prediction result that the second cognitive decline risk of the subject is low risk. When the second cognitive decline risk value of the subject is less than the preset low risk threshold, the third cognitive decline risk assessment model is used to cluster the EEG data of the subject. When the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is less than or equal to the preset distance threshold, the third cognitive decline risk assessment model outputs a prediction result that the third cognitive decline risk of the subject is low risk. When the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is greater than the preset distance threshold, the third cognitive decline risk assessment model outputs a prediction result that the subject has no cognitive decline risk.

[0144] The method for constructing a cognitive decline risk assessment model provided by the present invention utilizes multiple EEG representations in the EEG data of the target group and behavioral data training to obtain a first cognitive decline risk assessment model, a second cognitive decline risk assessment model, and a third cognitive decline risk assessment model.

[0145] The first cognitive decline risk assessment model primarily includes spectral information, which is relatively stable and not easily altered by treatment. This information is primarily found in patients with mild cognitive impairment and dementia. It can be used to define medium and high risk of cognitive decline, representing a stable brain function state and serving as a priority diagnostic criterion. The second cognitive decline risk assessment model primarily includes brain network connectivity information, which can be restored with treatment and is highly variable. This model primarily includes patients with subjective cognitive decline and mild cognitive impairment, and can be used to define low and medium risk of cognitive decline, representing neuroplasticity and serving as a secondary diagnostic criterion. The third cognitive decline risk assessment model primarily includes neural spatiotemporal dynamics information. This feature appears in the early stages of cognitive decline, namely the subjective cognitive decline stage, and can be used to define low risk of cognitive decline. It is primarily used for early risk warning. The first, second, and third cognitive decline risk assessment models can represent the progression of cognitive decline, from the early, recoverable third and second cognitive decline risk assessment models to the stable, difficult-to-recover first cognitive decline risk assessment model.

[0146] The first cognitive decline risk assessment model, the second cognitive decline risk assessment model, and the third cognitive decline risk assessment model can assist doctors in conducting efficient and accurate cognitive decline risk assessments on subjects, and can classify patients according to their cognitive decline risk types, helping doctors to quickly formulate reasonable diagnosis and treatment strategies and improve work efficiency and quality.

[0147] The method for constructing a cognitive decline risk assessment model provided by the present invention also has the following advantages:

[0148] 1. Propose an individualized feature extraction method based on EEG.

[0149] Extract individualized features through adaptive clustering based on EEG. Adaptive algorithms fully account for individual differences, making the extracted features more personalized. Using EEG signals better compensates for the clinical difficulties of MRI.

[0150] 2. Establish a multi-dimensional EEG hierarchical diagnostic model.

[0151] By comparing the data before and after treatment of MCI, the clinical significance of EEG features in different dimensions is captured. According to the intrinsic connection of EEG features in different dimensions, MCI patients are stratified, giving the model results clinical significance and enhancing interpretability.

[0152] 3. Capture brain networks related to cognition.

[0153] By mapping cognitive scales based on multiple EEG features, we capture cognitive-related neural circuits rather than single connections, thereby increasing the generalization and repeatability of the model.

[0154] The following describes a system for constructing a cognitive decline risk assessment model provided by the present invention. The system for constructing a cognitive decline risk assessment model described below and the method for constructing a cognitive decline risk assessment model described above can refer to each other.

[0155] 4 , a system for constructing a cognitive decline risk assessment model provided by the present invention may include:

[0156] A data acquisition module is used to: acquire EEG data and behavioral data of a target group, wherein the target group includes a healthy group and a group of patients at risk of cognitive decline (e.g., patients with impaired brain function, cognitive decline, behavioral disorders, etc.), and the EEG data has corresponding cognitive decline degree labels;

[0157] Preprocessing module, used to: preprocess EEG data;

[0158] A conventional frequency band feature extraction module is used to obtain EEG data and spectral features of multiple conventional frequency bands for each individual in the target group based on the preprocessed EEG data, wherein the conventional frequency bands include any one of the following or any combination thereof: Delta (1-4 Hz) frequency band, Theta (4-8 Hz) frequency band, Alpha (8-13 Hz) frequency band, and Beta (13-30 Hz) frequency band;

[0159] The individualized alpha band feature extraction module is used to: select the individualized alpha band of each individual in the target group, and obtain the individualized alpha band EEG data and spectral features of each individual in the target group based on the preprocessed EEG data;

[0160] A first training module is configured to: train a first cognitive decline risk assessment model based on the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group;

[0161] The tracing module is used to: trace the EEG data of multiple regular frequency bands of each individual in the target group;

[0162] The second training module is used to: construct a brain network using PLV based on the EEG data of multiple conventional frequency bands after tracing, and conduct correlation analysis between the constructed brain network and the behavioral data of the target group to train a second cognitive decline risk assessment model;

[0163] The third training module is used to: perform network segmentation on the EEG data of multiple conventional frequency bands after tracing back to obtain multiple brain functional networks, and obtain the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks, and obtain the third cognitive decline risk assessment model based on the self-time correlation window, permutation entropy, and neuronal variability ratio training.

[0164] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the conventional frequency band feature extraction module may include:

[0165] The first extraction submodule is configured to obtain, based on the EEG data of each regular frequency band, the sum of the squares of the amplitudes of each regular frequency band as the energy of each regular frequency band;

[0166] The second extraction submodule is configured to obtain, based on the energy of each regular frequency band, a ratio of the energy of each regular frequency band to the sum of the energies of all regular frequency bands as the relative energy of each regular frequency band, and obtain a ranking of the relative energies of the plurality of regular frequency bands;

[0167] The third extraction submodule is used to obtain the power spectrum density of each regular frequency band through fast Fourier transform according to the EEG data of each regular frequency band, and obtain the ratio of the power spectrum density between multiple regular frequency bands.

[0168] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the individualized alpha band feature extraction module may include:

[0169] The data processing submodule is used to obtain broadband filtered data and narrowband filtered data according to the preprocessed EEG data, wherein the broadband filtered data and narrowband filtered data are non-time filtered data;

[0170] Creating a submodule for: creating a first channel covariance matrix based on the broadband filtered data;

[0171] a separation submodule, configured to: for each frequency band within a preset range (e.g., 4-13 Hz), create a second channel covariance matrix based on the narrowband filtered data, and perform eigendecomposition on the first channel covariance matrix and the second channel covariance matrix to obtain a separation eigenvector for each frequency band, wherein the separation eigenvector can separate the first channel covariance matrix and the second channel covariance matrix to the greatest extent while suppressing data features represented in the first channel covariance matrix and the second channel covariance matrix;

[0172] A clustering submodule is used to: construct a feature vector similarity matrix based on the separated feature vectors of each frequency band, and perform cluster analysis on the feature vector similarity matrix to obtain clusters, wherein the clusters represent the frequency band range and the edges of the clusters represent the frequency band boundaries;

[0173] The fourth extraction submodule is used to: select a first frequency band range in the Alpha (8-13HZ) frequency band, record the number of clusters in the first frequency band and obtain the average frequency of the clusters, select a second frequency band range in the Theta (4-8HZ) frequency band, record the number of clusters in the second frequency band and obtain the average frequency of the clusters, and obtain the degree of coupling between the average frequency of the clusters in the first frequency band and the average frequency of the clusters in the second frequency band.

[0174] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the first training module may include a first training sub-module, which is used to: train a first cognitive decline risk assessment model based on the relative energy of multiple conventional frequency bands, the power spectral density ratio, the number of clusters in the first frequency band range, the average frequency of the clusters, and the degree of coupling between the average frequency of the clusters in the first frequency band range and the average frequency of the clusters in the second frequency band range.

[0175] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the tracing module may include a tracing sub-module, which is used to: use a tracing algorithm to trace the EEG data of multiple regular frequency bands of each individual in the target group, combine the ALL template to segment the brain areas, and convert the 64-lead EEG data into EEG data of 116 brain areas.

[0176] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the second training module may include:

[0177] Construct submodules for: evaluating the time series distribution of phase difference of EEG signals between brain regions through PLV expression to construct brain networks;

[0178] The correlation analysis submodule is used to: perform correlation analysis on the constructed brain network and the behavioral data of the target group to obtain network loops with statistical differences, and divide the network loops with statistical differences into positive correlation loops and negative correlation loops according to frequency bands, and normalize each network loop and perform regression analysis on the behavioral data to train a second cognitive decline risk assessment model.

[0179] Among them, the PLV expression is:

[0180]

[0181] In PLV expression, The time series of the phase difference of EEG signals between brain regions is The distribution of express and The phase difference, and Represents the EEG signals of different brain regions, N represents The length (i.e. number of points) of .

[0182] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the correlation analysis submodule may include:

[0183] The screening submodule is used to: perform correlation analysis between the constructed brain network and the behavioral data of the target group, and select network loops with statistical parameters less than a preset threshold (e.g., p less than 0.05) as network loops with statistical differences;

[0184] The classification submodule is used to: according to the frequency band, regard the network loop whose Pearson correlation with the behavioral data is greater than 0 as a positive correlation loop, and regard the network loop whose Pearson correlation with the behavioral data is less than 0 as a negative correlation loop.

[0185] According to a system for constructing a cognitive decline risk assessment model provided by the present invention, the brain functional network includes any one of the following or any combination thereof: limbic network, visual network, somatosensory network, frontoparietal network, dorsal attention network, ventral attention network, and default network. The third training module may include:

[0186] The self-time correlation window obtaining submodule is used to obtain the self-time correlation window of each brain functional network through the calculation formula of the self-time correlation window;

[0187] The permutation entropy obtaining submodule is used to: obtain the permutation entropy of each brain functional network through the permutation entropy calculation formula;

[0188] The neuron variability ratio obtaining submodule is used to obtain the neuron variability ratio of each brain functional network by calculating the standard deviation of the EEG signal of each brain functional network.

[0189] Among them, the calculation formula of the self-time correlation window is:

[0190]

[0191] In the calculation formula of the self-time correlation window, ACW represents the self-time correlation window. represents the result of continuous Fourier transform, that is , t represents time, x represents the EEG signals of different brain functional networks, Indicates frequency.

[0192] According to a method for constructing a cognitive decline risk assessment model provided by the present invention, the calculation formula of the permutation entropy is:

[0193]

[0194] In the calculation formula of permutation entropy, represents the permutation entropy, x represents the EEG signal of each brain functional network, Indicates the probability of the xth sample point appearing in the EEG signal.

[0195] The present invention also provides a cognitive decline risk assessment system, which may include:

[0196] A data receiving module is used to receive EEG data of a subject, wherein the subject is a healthy person, a patient suspected of having a risk of cognitive decline, or a patient at risk of cognitive decline;

[0197] The evaluation module is used to: predict the first cognitive decline risk value of the subject through the first cognitive decline risk evaluation model according to the EEG data of the subject; when the first cognitive decline risk value of the subject is higher than or equal to the preset high risk threshold, output the prediction result that the first cognitive decline risk of the subject is high risk; when the first cognitive decline risk value of the subject is lower than the preset high risk threshold and equal to or higher than the preset medium risk threshold, output the prediction result that the first cognitive decline risk of the subject is medium risk; when the first cognitive decline risk value of the subject is lower than the preset medium risk threshold, predict the second cognitive decline risk value of the subject through the second cognitive decline risk evaluation model; when the first cognitive decline risk value of the subject is lower than the preset medium risk threshold, output the prediction result that the first cognitive decline risk of the subject is medium risk. When the second cognitive decline risk value is greater than or equal to the preset low-risk threshold, the prediction result that the second cognitive decline risk of the subject is low risk is output; when the second cognitive decline risk value of the subject is less than the preset low-risk threshold, the EEG data of the subject is clustered using the third cognitive decline risk assessment model; when the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is less than or equal to the preset distance threshold, the prediction result that the third cognitive decline risk of the subject is low risk is output; when the distance between the EEG data of the subject and the cluster of patients with cognitive decline risk is greater than the preset distance threshold, the prediction result that the subject has no cognitive decline risk is output.

[0198] FIG5 illustrates a schematic diagram of the physical structure of an electronic device. As shown in FIG5 , the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840. The processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute a method for constructing a cognitive decline risk assessment model, which includes:

[0199] Obtaining EEG data and behavioral data of a target group, wherein the target group includes a healthy group and a patient group at risk of cognitive decline, and the EEG data has corresponding cognitive decline degree labels;

[0200] Preprocess EEG data;

[0201] Based on the preprocessed EEG data, EEG data and spectral features of multiple regular frequency bands of each individual in the target group are obtained, wherein the regular frequency bands include any one of the following or any combination thereof: Delta (1-4 Hz) frequency band, Theta (4-8 Hz) frequency band, Alpha (8-13 Hz) frequency band, Beta (13-30 Hz) frequency band;

[0202] Selecting an individualized alpha frequency band for each individual in the target group, and obtaining the individualized alpha frequency band EEG data and spectral features of each individual in the target group based on the preprocessed EEG data;

[0203] A first cognitive decline risk assessment model is trained based on the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group;

[0204] Tracing the EEG data of multiple regular frequency bands of each individual in the target group;

[0205] Based on the EEG data of multiple conventional frequency bands after tracing back to the source, PLV was used to construct a brain network. The constructed brain network was then correlated with the behavioral data of the target group to train a second cognitive decline risk assessment model.

[0206] The EEG data of multiple conventional frequency bands after tracing back to the source were segmented into networks to obtain multiple brain functional networks, and the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks were obtained. The third cognitive decline risk assessment model was obtained by training based on the self-time correlation window, permutation entropy, and neuronal variability ratio.

[0207] Furthermore, the logic instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0208] On the other hand, the present invention further provides a computer program product, comprising a computer program, which may be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for constructing a cognitive decline risk assessment model provided by the above methods, which includes:

[0209] Obtaining EEG data and behavioral data of a target group, wherein the target group includes a healthy group and a patient group at risk of cognitive decline, and the EEG data has corresponding cognitive decline degree labels;

[0210] Preprocess EEG data;

[0211] Based on the preprocessed EEG data, EEG data and spectral features of multiple regular frequency bands of each individual in the target group are obtained, wherein the regular frequency bands include any one of the following or any combination thereof: Delta (1-4 Hz) frequency band, Theta (4-8 Hz) frequency band, Alpha (8-13 Hz) frequency band, Beta (13-30 Hz) frequency band;

[0212] Selecting an individualized alpha frequency band for each individual in the target group, and obtaining the individualized alpha frequency band EEG data and spectral features of each individual in the target group based on the preprocessed EEG data;

[0213] A first cognitive decline risk assessment model is trained based on the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group;

[0214] Tracing the EEG data of multiple regular frequency bands of each individual in the target group;

[0215] Based on the EEG data of multiple conventional frequency bands after tracing back to the source, PLV was used to construct a brain network. The constructed brain network was then correlated with the behavioral data of the target group to train a second cognitive decline risk assessment model.

[0216] The EEG data of multiple conventional frequency bands after tracing back to the source were segmented into networks to obtain multiple brain functional networks, and the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks were obtained. The third cognitive decline risk assessment model was obtained by training based on the self-time correlation window, permutation entropy, and neuronal variability ratio.

[0217] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a cognitive decline risk assessment model provided by the above methods, the method comprising:

[0218] Obtaining EEG data and behavioral data of a target group, wherein the target group includes a healthy group and a patient group at risk of cognitive decline, and the EEG data has corresponding cognitive decline degree labels;

[0219] Preprocess EEG data;

[0220] Based on the preprocessed EEG data, EEG data and spectral features of multiple regular frequency bands of each individual in the target group are obtained, wherein the regular frequency bands include any one of the following or any combination thereof: Delta (1-4 Hz) frequency band, Theta (4-8 Hz) frequency band, Alpha (8-13 Hz) frequency band, Beta (13-30 Hz) frequency band;

[0221] Selecting an individualized alpha frequency band for each individual in the target group, and obtaining the individualized alpha frequency band EEG data and spectral features of each individual in the target group based on the preprocessed EEG data;

[0222] A first cognitive decline risk assessment model is trained based on the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band of each individual in the target group;

[0223] Tracing the EEG data of multiple regular frequency bands of each individual in the target group;

[0224] Based on the EEG data of multiple conventional frequency bands after tracing back to the source, PLV was used to construct a brain network. The constructed brain network was then correlated with the behavioral data of the target group to train a second cognitive decline risk assessment model.

[0225] The EEG data of multiple conventional frequency bands after tracing back to the source were segmented into networks to obtain multiple brain functional networks, and the self-time correlation window, permutation entropy, and neuronal variability ratio of each brain functional network in the multiple brain functional networks were obtained. The third cognitive decline risk assessment model was obtained by training based on the self-time correlation window, permutation entropy, and neuronal variability ratio.

[0226] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0227] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0228] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention. CROSS-REFERENCE TO RELATED APPLICATIONS

[0229] This application claims priority to the Chinese patent application (application number 202410096703.2) filed on January 23, 2024, the entire contents of which are incorporated herein by reference. Industrial Applicability

[0230] The present invention utilizes multiple EEG representations and behavioral data from the target group's EEG data to train a first cognitive decline risk assessment model, a second cognitive decline risk assessment model, and a third cognitive decline risk assessment model. The first cognitive decline risk assessment model mainly includes spectral information, the second cognitive decline risk assessment model mainly includes brain network connection information, and the third cognitive decline risk assessment model mainly includes neural spatiotemporal dynamics information. It can assist doctors in conducting efficient and accurate cognitive decline risk assessment on subjects, and can classify patients according to their cognitive decline risk type, which helps doctors quickly formulate reasonable diagnosis and treatment strategies and improve work efficiency and quality.

Claims

1. A method for constructing a risk assessment model for cognitive decline, characterized in that Including: Obtaining electroencephalogram (EEG) data and behavioral data of a target group, where the target group includes a healthy group and a patient group at risk of cognitive decline, and the EEG data has corresponding cognitive decline degree labels; Preprocessing the EEG data; Based on the preprocessed EEG data, obtaining the EEG data and spectral features of multiple conventional frequency bands for each individual in the target group, where the conventional frequency bands include any one or any combination of the following: Delta frequency band, Theta frequency band, Alpha frequency band, Beta frequency band; Selecting the individualized alpha frequency band for each individual in the target group, and based on the preprocessed EEG data, obtaining the EEG data and spectral features of the individualized alpha frequency band for each individual in the target group; Training a first cognitive decline risk assessment model based on the spectral features of multiple conventional frequency bands and the spectral features of the individualized alpha frequency band for each individual in the target group; Tracing the EEG data of multiple conventional frequency bands for each individual in the target group; Based on the traced EEG data of multiple conventional frequency bands, constructing a brain network using the phase locking value (PLV), and performing a correlation analysis between the constructed brain network and the behavioral data of the target group to train a second cognitive decline risk assessment model; Performing network segmentation on the traced EEG data of multiple conventional frequency bands to obtain multiple brain functional networks, and obtaining the self-time correlation window, permutation entropy, and neuron variability ratio of each brain functional network in the multiple brain functional networks, and training a third cognitive decline risk assessment model based on the self-time correlation window, permutation entropy, and neuron variability ratio; 2. The method for constructing a cognitive decline risk assessment model according to claim 1, characterized in that The spectral features of multiple conventional frequency bands include the relative energy of the conventional frequency band and the power spectral density of the conventional frequency band. The obtaining of the EEG data and spectral features of multiple conventional frequency bands for each individual in the target group based on the preprocessed EEG data includes: Based on the EEG data of each conventional frequency band, obtaining the sum of the squared amplitudes of each conventional frequency band as the energy of each conventional frequency band; Based on the energy of each conventional frequency band, obtaining the ratio of the energy of each conventional frequency band to the sum of the energies of all conventional frequency bands as the relative energy of each conventional frequency band, and obtaining the magnitude ranking of the relative energies of multiple conventional frequency bands; Based on the EEG data of each conventional frequency band, obtaining the power spectral density of each conventional frequency band through fast Fourier transform, and obtaining the ratio of the power spectral densities between multiple conventional frequency bands; Preferably, the expression of the power spectral density is: ； In the expression of the power spectral density, PSD represents the power spectral density, and N represents the number of sampling points of the electroencephalogram signal. Indicates the result of the discrete Fourier transform, i.e., ， Represent the electroencephalogram signals of each conventional frequency band, representing the sampling frequency.

3. The method for constructing a cognitive decline risk assessment model according to claim 2, wherein The selecting of the individualized alpha frequency band for each individual in the target group and the obtaining of the EEG data and spectral features of the individualized alpha frequency band for each individual in the target group based on the preprocessed EEG data include: Based on the preprocessed EEG data, obtaining broadband filtered data and narrowband filtered data, where the broadband filtered data and the narrowband filtered data are non-time filtered data; Creating a first channel covariance matrix based on the broadband filtered data; For each frequency band within a preset range, create a second channel covariance matrix based on the narrowband filtered data, and perform eigenvalue decomposition on the first channel covariance matrix and the second channel covariance matrix to obtain the separated eigenvectors for each frequency band. Among them, the separated eigenvectors can maximize the separation between the first channel covariance matrix and the second channel covariance matrix while suppressing the data characteristics represented in the first channel covariance matrix and the second channel covariance matrix; Construct an eigenvector similarity matrix based on the separated eigenvectors for each frequency band, and perform clustering analysis on the eigenvector similarity matrix to obtain clustering clusters. Among them, the clustering clusters represent frequency band ranges, and the edges of the clustering clusters represent frequency band boundaries; Based on the eigenvector similarity matrix, select a first frequency band range in the Alpha (8 - 13HZ) frequency band, record the number of clustering clusters within the first frequency band range and obtain the average frequency of the clustering clusters, select a second frequency band range in the Theta (4 - 8HZ) frequency band, record the number of clustering clusters within the second frequency band range and obtain the average frequency of the clustering clusters, and obtain the coupling degree between the average frequency of the clustering clusters in the first frequency band range and the average frequency of the clustering clusters in the second frequency band range; Preferably, the training of the first cognitive decline risk assessment model according to the spectral characteristics of multiple conventional frequency bands and the spectral characteristics of the individualized alpha frequency band for each individual in the target population includes: Train the first cognitive decline risk assessment model according to the relative energy of multiple conventional frequency bands, the power spectral density ratio, the number of clustering clusters in the first frequency band range, the average frequency of the clustering clusters, and the coupling degree between the average frequency of the clustering clusters in the first frequency band range and the average frequency of the clustering clusters in the second frequency band range.

4. The method for constructing a cognitive decline risk assessment model according to claim 3, wherein The training of the second cognitive decline risk assessment model by using PLV to construct a brain network based on the EEG data of multiple conventional frequency bands after tracing and performing a correlation analysis between the constructed brain network and the behavioral data of the target population includes: Evaluate the phase difference time series distribution of the EEG signals between brain regions through the PLV expression to construct a brain network; Perform a correlation analysis between the constructed brain network and the behavioral data of the target population to obtain network loops with statistical differences, divide the network loops with statistical differences into positive correlation loops and negative correlation loops according to the frequency band, and normalize each network loop and perform a regression analysis on the behavioral data to train the second cognitive decline risk assessment model; Preferably, the PLV expression is: ； In the PLV expression, Indicates the time series of phase differences of EEG signals between brain regions distribution, Indicate and phase difference And The electroencephalogram signals representing different brain regions, N represents The length of.

5. The method for constructing a cognitive decline risk assessment model according to claim 4, characterized in that The behavioral data of the target population is the MoCa scale characteristics of the target population. The performing a correlation analysis between the constructed brain network and the behavioral data of the target population to obtain network loops with statistical differences and dividing the network loops with statistical differences into positive correlation loops and negative correlation loops according to the frequency band includes: Perform a correlation analysis between the constructed brain network and the behavioral data of the target population, and select the network loops with statistical parameters less than the preset threshold as the network loops with statistical differences; According to the frequency band, the network loops with a Pearson correlation greater than 0 with the behavioral data are regarded as positive correlation loops, and the network loops with a Pearson correlation less than 0 with the behavioral data are regarded as negative correlation loops.

6. The method for constructing a cognitive decline risk assessment model according to claim 5, wherein The brain functional network includes any one of the following or any combination thereof: limbic network, visual network, somatosensory network, fronto-parietal network, dorsal attention network, ventral attention network, default network. The electroencephalogram data of multiple conventional frequency bands after source tracing are subjected to network segmentation to obtain multiple brain functional networks, and the self-time correlation window, permutation entropy, and neuron variability ratio of each brain functional network in the multiple brain functional networks are obtained. The third cognitive decline risk assessment model is trained based on the self-time correlation window, permutation entropy, and neuron variability ratio, including: The self-time correlation window of each brain functional network is obtained through the calculation formula of the self-time correlation window; The permutation entropy of each brain functional network is obtained through the calculation formula of the permutation entropy; The neuron variability ratio of each brain functional network is obtained by calculating the standard deviation of the electroencephalogram signal of each brain functional network.

7. The method for constructing a cognitive decline risk assessment model according to claim 6, wherein The calculation formula of the self-time correlation window is: ； In the calculation formula of the time-related window, ACW represents the time-related window, Denotes the result of the continuous Fourier transform, i.e., , where \(t\) represents time and \(x\) represents the electroencephalogram signals of different brain functional networks. represents the frequency; Preferably, the calculation formula of the permutation entropy is: ； In the calculation formula of permutation entropy, represents permutation entropy, and x represents the EEG signals of each brain functional network. represents the probability of the x-th sample point appearing in the electroencephalogram signal.

8. A cognitive decline risk assessment system, characterized in that, includes: A data receiving module, configured to: receive the electroencephalogram data of the subject, where the subject is a healthy person, a patient suspected of having a cognitive decline risk, or a patient with a cognitive decline risk; An evaluation module, configured to: according to the electroencephalogram data of the subject, predict the first cognitive decline risk value of the subject through the first cognitive decline risk assessment model. When the first cognitive decline risk value of the subject is higher than or equal to the preset high-risk threshold, output a prediction result that the first cognitive decline risk of the subject is high risk. When the first cognitive decline risk value of the subject is less than the preset high-risk threshold and equal to or greater than the preset medium-risk threshold, output a prediction result that the first cognitive decline risk of the subject is medium risk. When the first cognitive decline risk value of the subject is less than the preset medium-risk threshold, predict the second cognitive decline risk value of the subject through the second cognitive decline risk assessment model. When the second cognitive decline risk value of the subject is greater than or equal to the preset low-risk threshold, output a prediction result that the second cognitive decline risk of the subject is low risk. When the second cognitive decline risk value of the subject is less than the preset low-risk threshold, cluster the electroencephalogram data of the subject through the third cognitive decline risk assessment model. When the distance between the electroencephalogram data of the subject and the cluster of patients with a cognitive decline risk is less than or equal to the preset distance threshold, output a prediction result that the third cognitive decline risk of the subject is low risk. When the distance between the electroencephalogram data of the subject and the cluster of patients with a cognitive decline risk is greater than the preset distance threshold, output a prediction result that the subject has no cognitive decline risk for the time being.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for constructing the cognitive decline risk assessment model according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for constructing the cognitive decline risk assessment model according to any one of claims 1 to 7

Images