Personalized MCI electrical stimulation intervention method based on dynamic brain network
By employing a personalized MCI electrical stimulation intervention method based on dynamic brain networks, EEG datasets were constructed for normal and cognitive impairment groups. Specific frequency bands were selected and dynamic weights were optimized to identify abnormal frequency bands and core nodes. This approach addresses the issue of efficacy differences in existing technologies and achieves more efficient electrical stimulation therapy.
Patent Information
- Application Number
- CN202511191115.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-25
- Publication Date
- 2025-11-14
AI Technical Summary
Current transcranial electrical stimulation protocols neglect the individual heterogeneity of patients' brain functional networks, leading to significant differences in efficacy.
The personalized MCI electrical stimulation intervention method based on dynamic brain networks constructs EEG datasets for normal and cognitive impairment groups, selects effective data in specific frequency bands, and uses a dynamic time-varying weight optimization adaptive method based on reaction time to construct average dynamic brain network connectivity, identifying the main abnormal frequency bands and core nodes as targets.
It improved the reliability of target selection, enhanced the personalized effect of electrical stimulation, and improved the therapeutic effect on cognitive function.
Smart Images

Figure CN120939450A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of electrical stimulation target determination technology, specifically involving a personalized MCI electrical stimulation intervention method based on dynamic brain networks. Background Technology
[0002] With the increasing aging of the population, early intervention for Alzheimer's Disease (AD) has become a research hotspot. Mild Cognitive Impairment (MCI), as a precursor stage of AD, is crucial for slowing disease progression through effective intervention. Transcranial electrical stimulation (TCS), a non-invasive neuromodulation technique, improves cognitive function by modulating cortical activity and has shown potential in MCI intervention in recent years. However, current TCS protocols often use fixed targets and uniform stimulation parameters, neglecting the individual heterogeneity of patients' brain functional networks, leading to significant differences in efficacy. Summary of the Invention
[0003] To address the problems mentioned in the background section, this invention provides a personalized MCI electrical stimulation intervention method based on dynamic brain networks. This addresses the issue that existing transcranial electrical stimulation protocols often use fixed targets and uniform stimulation parameters, neglecting the individual heterogeneity of patients' brain functional networks, leading to significant differences in therapeutic effects.
[0004] To achieve the above objectives, the present invention provides the following technical solution:
[0005] A personalized MCI electrical stimulation intervention method based on dynamic brain networks includes the following steps:
[0006] S1: Construct EEG datasets for normal and cognitively impaired groups when performing multiple trials of working memory tasks, and record the reaction time for each trial of the working memory task;
[0007] S2: Preprocess the constructed EEG dataset and select EEG data in specific frequency bands from the EEG dataset based on prior knowledge to form an effective dataset;
[0008] S3: Based on the effective dataset, an adaptive method based on dynamic time-varying weight optimization of reaction time is used to construct the average dynamic brain network connectivity of the normal group and the cognitive impairment group for the total number of trials.
[0009] S4: Based on the differences in average dynamic brain network connectivity between the normal group and the cognitive impairment group, determine the main abnormal frequency bands in the normal group and the cognitive impairment group;
[0010] S5: Identify non-invasive stimuli suitable for mild cognitive impairment by targeting the core nodes of the average dynamic brain network connectivity in the main abnormal frequency bands.
[0011] Compared with the prior art, the beneficial effects of the present invention are:
[0012] This application considers the individual heterogeneity of patients' brain functional networks. By selecting specific frequency bands from EEG datasets of normal and cognitively impaired groups performing multi-trial working memory tasks, an effective dataset is constructed. Based on the effective dataset, an adaptive method based on dynamic time-varying weights optimized by reaction time is used to construct the average dynamic brain network connectivity of the normal and cognitively impaired groups for the total number of trials. Then, the main abnormal frequency bands are determined by the difference in the average dynamic brain network connectivity between the two groups. Target points are then determined based on the core nodes of the average dynamic brain network connectivity of the main abnormal frequency bands. Compared with traditional single-frequency band analysis, this application proposes to construct an effective dataset with specific frequency bands to determine the main abnormal frequency bands. At the same time, this application also locates stimulation targets based on the core nodes of the average dynamic brain network connectivity of the abnormal frequency bands, improving the reliability of target selection. Attached Figure Description
[0013] Figure 1 This is a flowchart illustrating the process of this application. Detailed Implementation
[0014] To facilitate understanding of the technical content of this invention by those skilled in the art, the invention will be further described in detail below with reference to the accompanying drawings and specific examples. It should be understood that the specific examples described herein are merely illustrative and not intended to limit the scope of the invention.
[0015] Example 1
[0016] like Figure 1 As shown, the personalized MCI electrical stimulation intervention method based on dynamic brain networks includes the following steps:
[0017] S1: Construct EEG datasets for the normal group and the cognitive impairment group during the performance of a multi-trial working memory task, and record the reaction time for each trial of the working memory task, specifically:
[0018] Employs a 32-channel electrode system (compatible with international 10-20 / 10-10 standards); amplifier type: 24-bit ADC; input noise <0.4μVrms (0.1-100Hz); sampling rate: 250Hz; common-mode rejection ratio: ≥110dB; electrode type: Ag / AgCl active electrode (with integrated preamplifier); bandpass filter: 0.01~1000Hz; reference electrode: earlobe electrode; ground electrode: FPz; real-time impedance monitoring: impedance <5kΩ.
[0019] Working memory EEG data were collected from 100 individuals with mild cognitive impairment and 103 individuals in a normal cognitive control group. The working memory paradigm employed a delayed-matching sample paradigm, which simulates real-world working memory applications by presenting specific stimuli and response requirements. This paradigm is one of the main paradigms in working memory research. The specific process of the working memory paradigm includes a memory phase, a memory delay phase, a recall phase, and a rest phase. In the memory phase, four black-and-white geometric images are displayed consecutively on the computer screen, each image presented for 1 second, for a total of 4 seconds. In the memory delay phase, a "+" symbol appears on the screen for 3 seconds. In the recall phase, a black-and-white geometric image is presented on the screen, with a 50% probability that it is an image from the previous memory phase, presented for 2.5 seconds. Participants are required to judge whether the current image is from the memory phase by pressing the "space" key if yes and the "ENTER" key if no. The participants' key press response times are recorded. Finally, the word "rest" appears on the screen for 4 seconds. This is one trial of the complete working memory task; a total of 20 trials were conducted. The participants' reaction times were recorded after the entire working memory task was completed.
[0020] S2: Preprocess the constructed EEG dataset and select EEG data in specific frequency bands from the dataset based on prior knowledge to form a valid dataset. Preprocessing includes:
[0021] Mean reference values; 1-30Hz bandpass filtering; Fast Independent Component Analysis (ICA) performed using built-in functions in EEGLAB v2022.0 and MATLAB R2018b (MathWorks Inc., USA). As a data decomposition technique, ICA decomposes the EEG signal into multiple independent components, each representing a unique brain source or artifact source (such as eye movement, muscle activity, etc.). ICA is used to remove artifacts with a recognition artifact component greater than 80%, including blinking (>80%), vertical / horizontal eye movements (>80%), muscles (>80%), and electrocardiogram (>80%); baseline correction; data segmentation for each memory stage (memory stage, memory delay stage, memory retrieval stage, and rest stage).
[0022] EEG data in specific frequency bands were selected from the theta band (4-7 Hz) and the alpha band (8-13 Hz). Existing research shows that the frequency bands mainly involved during working memory tasks are the theta band (4-7 Hz) and the alpha band (8-13 Hz). In addition, considering that the subjects do not need to press buttons during the memory phase of the working memory test, and the subjects need to memorize pictures, their attention is relatively focused. Therefore, the data quality in this phase is relatively good and can more effectively reflect the subjects' cognitive status and obtain a stable brain network map.
[0023] S3: Based on the effective dataset, an adaptive method using dynamic time-varying weights optimized according to reaction time is used to construct the average dynamic brain network connectivity for the normal group and the cognitive impairment group across all trials. Specifically:
[0024] S3.1: For each trial, the 4-second EEG data is segmented using a strictly non-overlapping 1-second sliding window (t1: 0-1s, t2: 1-2s, t3: 2-3s, t4: 3-4s);
[0025] S3.2: Calculate the directional transfer function (DTF) matrices for the θ band and α band for each time window. The DTF reflects the directed causal connections between nodes.
[0026] S3.3: Time-varying weight optimization stage: Designing an exponentially decaying weight function: Where t∈{1,2,3,4} corresponds to 4 time windows, and λ is the decay coefficient. The weights are normalized so that their sum is 1. Based on the characteristics of the memory encoding stage, in trials with fast responses (high-efficiency encoding), the early window is more important and requires a smaller λ, while in trials with slow responses (deep processing), the late window is more important and requires a larger λ. The optimization of λ based on the subject's behavioral data and reaction time is as follows:
[0027] S3.31: Calculate the mean reaction time RT_mean and standard deviation RT_std for all trials of the subject;
[0028] S3.32: Standardized reaction time; z-score standardization is performed on the RT of each trial to obtain zRT.
[0029] ;
[0030] zRT<0 indicates that the reaction rate of this trial is higher than the average level, and zRT>0 indicates that the reaction rate of this trial is slower than the average level.
[0031] Set the minimum value of λmin = 0.1 and the maximum value of λmax = 0.9;
[0032] Design the mapping relationship between λ and zRT: Use the Sigmoid function to map zRT to the λ interval:
[0033] ;
[0034] k controls the slope of the Sigmoid (k=0.5);
[0035] S3.4: Calculate the weighted DTF network independently for each frequency band: ;
[0036] S3.5: Network Normalization Processing: Perform Fisher-Z Transform on the weighted DTF network for each time window to improve its distribution characteristics.
[0037] S3.6: Final As network parameters for each subject in each trial at each time window, these indicators describe the causal connectivity strength of the network. They are used for each frequency band and corresponding time window in both the normal group and the cognitively impaired group. By performing superposition and averaging, n frequencies of the α band in the normal group are finally obtained. and n frequency bands of the θ band and n alpha bands in the cognitive impairment group and n frequency bands of the θ band .
[0038] S4: Statistically analyze the brain network connectivity corresponding to the time windows of the two frequency bands respectively. This yields multiple differential network maps for two groups across two frequency bands. At the group level, based on the differential network maps of directed weighted brain networks spanning two frequency bands and four time windows, the most anomalous frequency bands are identified. The specific steps are as follows:
[0039] S4.1: Statistics on differential network density:
[0040] For each frequency band f, calculate the percentage of significant edge differences (p<0.05) in all time window difference networks:
[0041] ;
[0042] in, This represents the difference p-value of inter-group network connectivity for edge i→j in frequency band f and time window t, where N represents the number of nodes in the network, N(N−1) represents the total number of edges in the directed network, and D f It is the global difference density of frequency band f. The larger the value, the more significant the difference between groups in that frequency band.
[0043] S4.2: Identify abnormal frequency bands: Select D f Larger frequency bands (e.g., 50% of the θ band has a significant edge, compared to 20% of the α band);
[0044] S5: The differential network connections of the above main abnormal frequency bands are sorted for the core nodes of the differential network for each time window. First, the weights of the network nodes in the differential network graph are sorted based on the PageRank algorithm, and the weights are normalized to [0,1] by Min-Max normalization to obtain the weight of each network node.
[0045] Secondly, calculate the out-degree centrality of each node and normalize the weights using Min-Max to [0,1] to obtain the weights of each network node;
[0046] Then, the normalized PageRank and out-degree centrality weights are combined with 50% weights to generate a comprehensive score and sort it.
[0047] Based on the node rankings obtained for each time window, we further analyze the stability of each node across multiple time windows and calculate the average ranking of each node. For example, for a node k, we take the average ranking from t1 to t4 (e.g., rank 1 gets 4 points, rank 2 gets 3 points, and so on) to calculate the average score of node k and obtain the average ranking of each node across all time windows.
[0048] The nodes ranked in the top 10% on average were selected as stable nodes across time windows and then used as core nodes. Finally, based on prior knowledge, at least two core nodes were selected as targets for non-invasive stimulation.
[0049] In this embodiment, the present application considers the individual heterogeneity of the patient's brain functional network. By selecting specific frequency bands from the EEG datasets of the normal group and the cognitive impairment group when performing multi-trial working memory tasks, an effective dataset is constructed. Based on the effective dataset, an adaptive method based on dynamic time-varying weight optimization of reaction time is used to construct the average dynamic brain network connectivity of the total trials for the normal group and the cognitive impairment group, respectively. Then, the main abnormal frequency bands are determined by the difference in the average dynamic brain network connectivity between the two groups. Target points are then determined for the core nodes of the average dynamic brain network connectivity of the main abnormal frequency bands. Compared with the traditional single-frequency band analysis, the present application proposes to construct an effective dataset with specific frequency bands to determine the main abnormal frequency bands. At the same time, the present application also locates the stimulation target points based on the core nodes of the average dynamic brain network connectivity of the abnormal frequency bands, which improves the reliability of target point selection.
Claims
1. A personalized MCI electrical stimulation intervention method based on dynamic brain networks, characterized in that, Includes the following steps: S1: Construct EEG datasets for normal and cognitively impaired groups when performing multiple trials of working memory tasks, and record the reaction time for each trial of the working memory task; S2: Preprocess the constructed EEG dataset and select EEG data in specific frequency bands from the EEG dataset based on prior knowledge to form an effective dataset; S3: Based on the effective dataset, an adaptive method based on dynamic time-varying weight optimization of reaction time is used to construct the average dynamic brain network connectivity of the normal group and the cognitive impairment group for the total number of trials. S4: Based on the differences in average dynamic brain network connectivity between the normal group and the cognitive impairment group, determine the main abnormal frequency bands in the normal group and the cognitive impairment group; S5: Identify non-invasive stimuli suitable for mild cognitive impairment by targeting the core nodes of the average dynamic brain network connectivity in the main abnormal frequency bands.
2. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 1, characterized in that, A single trial of the working memory task consists of a memory phase, a memory delay phase, a recall phase, and a rest phase. Specifically, in the memory phase, participants are presented with multiple geometric images in sequence, each image for a preset duration. In the memory delay phase, participants are presented with content unrelated to the geometric images for a preset duration. In the recall phase, participants are presented with a geometric image, which has a certain probability of being the same image presented in the memory phase, for a preset duration. Participants are then asked to determine whether the image is the same one presented in the memory phase, and their reaction time is recorded. Finally, a rest phase of a preset duration is initiated.
3. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 2, characterized in that, The preprocessing steps are as follows: average reference is applied to the EEG signals in the EEG dataset; 1-30Hz bandpass filtering is performed; the EEG signals are decomposed into multiple independent components using the data analysis tool ICA, with each component representing a unique brain source or artifact source; artifacts with a component content greater than 80% are removed using the data analysis tool ICA; baseline correction is performed; and finally, the EEG signals are segmented based on the various stages of the working memory task.
4. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 3, characterized in that, In S2, the effective dataset based on prior knowledge includes EEG signals in the θ band of 4-7 Hz and EEG signals in the α band of 8-13 Hz.
5. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 4, characterized in that, S3 includes the following steps: S3.1: For each EEG signal in the valid dataset, a non-overlapping sliding window is used to divide each memory stage into n segments, each segment being a time window, where n equals the number of geometric images in the memory stage; S3.2: Calculate the Directional Transfer Function (DTF) matrix for the θ band and α band for each time window; S3.3: Set the exponential decay weighting function: Where t∈{1,…,n} corresponds to n time windows, e is the natural constant, λ is the decay coefficient, and the weights are normalized so that their sum is 1; S3.4: Calculate the weighted DTF network for each time window of each frequency band: ; S3.5: Perform Fisher-Z transform on the weighted DTF network for each time window to obtain... ; S3.6: For each frequency band and corresponding time window, respectively, the normal group and the cognitive impairment group... By performing superposition and averaging, n frequencies of the α band in the normal group are finally obtained. and n frequency bands of the θ band and n alpha bands in the cognitive impairment group and n frequency bands of the θ band .
6. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 5, characterized in that, The λ is optimized based on reaction time, specifically by calculating the mean reaction time RT_mean and standard deviation RT_std for all trials, and then standardizing the reaction time RT for each trial using z-score to obtain zRT. ; When zRT<0, it means that the reaction speed of this trial is higher than the average level; when zRT>0, it means that the reaction speed of this trial is slower than the average level. Set the minimum value λmin and the maximum value λmax of λ. Mapping zRT to the λ interval using the Sigmoid function is expressed as: ; Where k is the slope of the Sigmoid function.
7. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 6, characterized in that, S4 specifically includes the following steps: S4.1: Statistically analyze the time windows corresponding to the two frequency bands respectively. Multiple differential network diagrams were obtained; S4.2: Let frequency band f represent either the α-band or the θ-band. Calculate the percentage of significant edges in the difference network graph for all time windows: ; in, This represents the difference p-value of inter-group network connectivity for edge i→j in frequency band f and time window t, where N represents the number of nodes in the network, N(N−1) represents the total number of edges in the directed network, and D f It is the global difference density of frequency band f. The larger the value, the more significant the difference between groups in that frequency band. S4.3: Select D f The largest frequency band, f, is considered the main anomalous frequency band.
8. The personalized MCI electrical stimulation intervention method based on dynamic brain networks according to claim 7, characterized in that, S5 specifically involves: based on the differential network graph of the main abnormal frequency bands, sorting all network nodes of the differential network graph for each time window; firstly, sorting the network node weights of the differential network graph based on the PageRank algorithm, and normalizing the weights to [0,1] using Min-Max, to obtain the weight of each network node; Secondly, calculate the out-degree centrality of each node and normalize the weights using Min-Max to [0,1] to obtain the weights of each network node; Then, the normalized PageRank and out-degree centrality weights are combined with 50% weights to generate a comprehensive score and sort it. Based on the node ranking of each time window obtained above, the stability of each node in multiple time windows is further analyzed, the average ranking of each node is calculated, and the nodes with the top 10% average ranking are selected as stable nodes across time windows as core nodes. Finally, based on prior knowledge, two or more core nodes are obtained as targets for non-invasive stimulation.
Citation Information
Cited By
Head electromagnetic field coherent phase stimulation method and system
CN122183001A
