CN113425312A

CN113425312A - Electroencephalogram data processing method and device

Info

Publication number: CN113425312A
Application number: CN202110873934.6A
Authority: CN
Inventors: 覃小雅; 袁媛; 王志燕; 胡迎炳; 郝红伟; 李路明
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2021-07-30
Filing date: 2021-07-30
Publication date: 2021-09-24
Anticipated expiration: 2041-07-30
Also published as: CN113425312B

Abstract

The present disclosure relates to an electroencephalogram data processing method and apparatus, the method comprising: acquiring electroencephalogram data of a target object; determining a first feature vector from the brain electrical data of the target object, the first feature vector comprising a plurality of brain network attributes determined from the brain electrical data of the target object; and inputting the first feature vector into a trained prediction model to obtain a prediction result, wherein the prediction result represents whether the target object is suitable for receiving specific medical treatment. According to the electroencephalogram data processing method disclosed by the embodiment of the disclosure, the prediction accuracy of the adaptability of the target object to specific medical treatment can be improved.

Description

Electroencephalogram data processing method and device

Technical Field

The present disclosure relates to the field of data analysis, and in particular, to a method and an apparatus for processing electroencephalogram data.

Background

Epilepsy as a disease affects the life of patients, most of whom can be controlled for seizure by combination therapy with one or more drugs, but some patients are not sensitive to drug therapy and are called drug refractory epilepsy patients. Vagal nerve stimulation surgery (VNS) has fewer side effects on cognition, nerves, and the whole body, and was approved in 1997 for the treatment of drug refractory epilepsy. There is great uncertainty about the post-operative efficacy of the patient's VNS. According to the existing research data, by treating drug-refractory epilepsy through VNS, about 60% of patients reach a clinically effective level (the seizure control rate is more than or equal to 50%) 2 years after operation, partial patients still cannot effectively control the seizures, and the proportion of patients with complete epilepsy without seizures is only about 10%. Thus, treatment of drug refractory epilepsy with VNS poses a significant challenge-ensuring that patients receiving VNS treatment receive a high rate of seizure control after treatment, rather than no effect at all or a low rate of seizure control.

Therefore, when predicting whether a patient is suitable for receiving a specific medical treatment, for example, a drug-refractory epilepsy patient is used as a target object, it is a research hotspot in the related field to predict whether the target object is suitable for receiving a vagus nerve stimulation operation by a preoperative assessment means in view of the problems of uncertainty and large individual difference in the therapeutic effect of treating the target object by a therapeutic means using vagus nerve stimulation.

Disclosure of Invention

In view of this, the present disclosure provides an electroencephalogram data processing method and an electroencephalogram data processing device, and according to the electroencephalogram data processing method of the embodiments of the present application, the prediction accuracy of the adaptability of a target object to a specific medical treatment can be improved.

According to an aspect of the present disclosure, there is provided a brain electrical data processing method, the method including: acquiring electroencephalogram data of a target object; determining a first feature vector from the brain electrical data of the target object, the first feature vector comprising a plurality of brain network attributes determined from the brain electrical data of the target object; and inputting the first feature vector into a trained prediction model to obtain a prediction result, wherein the prediction result represents whether the target object is suitable for receiving specific medical treatment.

In one possible implementation, the determining the first feature vector includes determining the first feature vector based on the target object's electroencephalogram data, including: determining a plurality of attribute types as inputs to the trained predictive model; determining the plurality of brain network attributes corresponding to the plurality of attribute types from artifact-free inter-episode brain electrical data of the target subject; and obtaining the first feature vector according to the plurality of brain network attributes.

In one possible implementation, the prediction model is trained from electroencephalogram data of a sample object, and the method further includes: determining a plurality of brain network attributes of the sample object according to the electroencephalogram data of the sample object; according to the effect of the sample object after the specific medical treatment, determining a plurality of brain network attributes of the sample object of which the effect meets the preset condition as positive samples, and determining a plurality of brain network attributes of the sample object of which the effect does not meet the preset condition as negative samples; performing difference significance analysis on the positive sample and the negative sample to obtain brain network attributes with significance difference, wherein the difference between the brain network attributes with significance difference of the positive sample and the negative sample is larger than a threshold value; sorting the importance of the brain network attributes with the significant difference to obtain a second feature vector; and training to obtain the prediction model according to the brain network attributes with the same attribute type as the attributes in the second feature vector in the plurality of brain network attributes of the sample object.

In one possible implementation, the importance ranking of the brain network attributes with significant differences to obtain a second feature vector includes: obtaining a feature subset according to all brain network attributes with significant differences, and determining the prediction accuracy of a prediction model to the feature subset; judging whether the current feature subset is an empty set; and when the current feature subset is an empty set, according to the importance scores of the brain network attributes in the feature subset with the highest prediction accuracy, performing importance ranking on the brain network attributes in the feature subset with the highest prediction accuracy to obtain a second feature vector.

In a possible implementation manner, the importance ranking is performed on the brain network attributes with significant differences to obtain a second feature vector, and the method further includes: when the current feature subset is not an empty set, repeating the following operations: inputting the current feature subset into a preset random forest model to obtain the importance score of each brain network attribute in the current feature subset; eliminating at least one brain network attribute with the lowest importance score in the current feature subset, taking the collection of the brain network attributes left in the current feature subset as a new current feature subset, and determining the prediction accuracy of the prediction model on the current feature subset; and judging whether the current feature subset is an empty set again.

In one possible implementation manner, the training to obtain the prediction model according to the brain network attributes of the plurality of brain network attributes of the sample object, which are of the same type as the attributes in the second feature vector, includes: obtaining a plurality of third feature vectors by permutation and combination according to different brain network attributes in the second feature vectors, wherein the combinations of the brain network attributes in the different third feature vectors are different, or the combinations of the brain network attributes in the different third feature vectors and the ranks of the brain network attributes in the third feature vectors are different; calculating the prediction accuracy of the prediction model when the brain network attributes with the same attribute types as those in the third feature vectors in the plurality of brain network attributes of the sample object are taken as the input of the prediction model; and determining the input attribute type of the prediction model according to the third feature vector with the highest prediction accuracy to obtain the trained prediction model.

In one possible implementation, the prediction model is trained from electroencephalogram data of a sample object, and the method further includes: determining a plurality of brain network attributes of the sample object according to the electroencephalogram data of the sample object; taking the effect of the sample object after the specific medical treatment as a mark, taking a plurality of brain network attributes of the sample object as influence factors, inputting the influence factors into a preset linear regression model, and screening and outputting at least one brain network attribute from the input brain network attributes by the linear regression model; and training to obtain the prediction model according to the at least one brain network attribute.

In one possible implementation, the sample object's brain electrical data includes artifact-free inter-episode brain electrical data of the sample object, and determining a plurality of brain network attributes of the sample object from the sample object's brain electrical data includes: obtaining preprocessed electroencephalogram data of the sample object according to the electroencephalogram data of the sample object, wherein the preprocessing comprises at least one of filtering, denoising, re-referencing and segmenting; determining indexes of the sample object according to the preprocessed electroencephalogram data of the sample object, wherein the indexes of the sample object indicate the communication relation among all channels of electroencephalogram signals of the sample object, and the indexes of the sample object comprise at least one of coherence, a phase locking value, a phase delay index, a weighted phase delay index, a Glan's cause and effect and synchronous likelihood; obtaining at least one synchronicity matrix of the sample object according to the indexes of the sample object, wherein each synchronicity matrix corresponds to one index; obtaining a threshold correlation matrix of the sample object according to the synchronicity matrix and a preset threshold condition; establishing a brain network of the sample object according to the threshold incidence matrix; determining a plurality of brain network attributes of the sample object from the brain network of the sample object, the plurality of brain network attributes of the sample object including one or more of average synchronicity, clustering coefficient, characteristic path length, betweenness centrality, global efficiency, and local efficiency.

According to another aspect of the present disclosure, there is provided a brain electrical data processing apparatus, the apparatus including: the acquisition module is used for acquiring electroencephalogram data of a target object; a determining module for determining a first feature vector from the brain electrical data of the target object, the first feature vector comprising a plurality of brain network attributes determined from the brain electrical data of the target object; and the prediction module is used for inputting the first feature vector into a trained prediction model to obtain a prediction result, wherein the prediction result represents whether the target object is suitable for receiving specific medical treatment.

According to another aspect of the present disclosure, there is provided an electroencephalogram data processing apparatus, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the above method.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the above-described method.

According to the electroencephalogram data processing method, the electroencephalogram data of the target object are collected and processed to obtain the first feature vector, when the first feature vector is input into the prediction model, the prediction model can give a prediction result indicating whether the target object is suitable for receiving specific medical treatment, and therefore the adaptability of the target object to the specific medical treatment can be predicted in advance before the specific medical treatment is received. And the first characteristic vector comprises a plurality of brain network attributes, so that the prediction result is associated with the brain network attributes, the prediction results of different target objects can reflect the individual difference of susceptibility, and the accuracy of the prediction result is ensured.

Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

Fig. 1 shows an exemplary application scenario of the electroencephalogram data processing method according to an embodiment of the present application.

Fig. 2 shows an exemplary schematic diagram of an exemplary acquisition method of a predictive model according to an embodiment of the application.

FIG. 3 shows an example of a pre-processing manner of an electroencephalogram signal according to an embodiment of the present application.

Fig. 4 is a diagram illustrating an exemplary method for obtaining a phase-locked value according to an embodiment of the present application.

Fig. 5 illustrates an example of a synchronization matrix according to an embodiment of the present application.

Fig. 6 shows a schematic diagram of an exemplary method for obtaining a weighted phase delay index according to an embodiment of the present application.

Fig. 7 illustrates an example of a synchronization matrix according to an embodiment of the present application.

FIG. 8 illustrates an exemplary method for obtaining coherency according to an embodiment of the present application.

Fig. 9 shows a schematic diagram of an exemplary method for obtaining synchronization likelihood according to an embodiment of the present application.

FIG. 10 illustrates an exemplary method schematic for achieving a Greenger cause and effect in accordance with an embodiment of the present application.

Fig. 11 illustrates one example of an implementation of screening and ranking brain network attributes according to an embodiment of the present application.

Fig. 12 shows an example of an implementation manner of ranking brain network attributes according to the scoring result from high to low in the embodiment of the present application.

Fig. 13 shows an example of obtaining a trained predictive model according to an embodiment of the present application.

Fig. 14 shows an exemplary schematic diagram of a brain electrical data processing method according to an embodiment of the present application.

FIG. 15 illustrates an exemplary diagram of a trained predictive model according to an embodiment of the present application.

Fig. 16 shows an exemplary schematic diagram of an electrocardiographic data processing device according to an embodiment of the present application.

Fig. 17 illustrates an exemplary block diagram of an apparatus 800 in accordance with an embodiment of the present application.

Fig. 18 illustrates an exemplary block diagram of an apparatus 1900 according to an embodiment of the application.

Detailed Description

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.

Existing studies for adaptive prediction for specific medical treatments are mostly based on clinical data of the target subject, including etiology, epileptic type or epileptic focus location for adaptive (especially complete epileptic control) analysis. However, such analysis methods have low accuracy and cannot be used for clinical popularization. Moreover, such analysis is predictive analysis for a target object population with certain characteristics, and adaptive prediction of specific medical treatment is still uncertain for target object individuals.

Some prior art proposes VNS efficacy prediction by means of electroencephalography (EEG). Electroencephalograms are gold standards for monitoring and judging the physical state of a target subject, and are one of the necessary examinations for clinical state monitoring of the target subject. Finding the adaptability of a target object to a specific medical treatment based on electroencephalography may employ a method of visual analysis, which generates an electroencephalography through collected electroencephalogram data, and acquires visual information by an electroencephalograph technician or researcher directly observing the electroencephalography through the naked eye. However, the accuracy of the visual inspection is limited by the ability and experience of the operator (electroencephalograph or researcher), subjectivity is high, and the visual analysis can analyze only images that can be seen with the naked eye, neglecting electroencephalogram features that cannot be seen with the naked eye, and sensitivity is poor.

In order to solve the above problems in visual analysis, a quantitative electroencephalogram analysis method is proposed, electroencephalogram characteristics invisible to naked eyes can be monitored by processing acquired electroencephalogram data by a computer through a quantitative method (such as time domain and frequency domain calculation), such as synchronism (the consistency degree of the appearance or disappearance of electroencephalogram activity in corresponding areas of two hemispheres of a brain or each area of the same hemisphere on time), a power spectrum (a graph or a histogram with the abscissa as frequency and the ordinate as amplitude and capable of representing electroencephalogram frequency change caused by electroencephalogram activity) and symmetry (the similarity degree of certain electroencephalogram activity with synchronism appearing in corresponding areas of two hemispheres on the aspects of waveform, frequency and amplitude), and the quantitative electroencephalogram analysis is more objective and sensitive relative to the visual analysis. Various electroencephalographic features obtained using quantitative electroencephalographic analysis can be used to construct brain networks.

Brain networks of target objects for a particular medical treatment (e.g., VNS) are more prone to more irregularities or randomness than the structure of healthy human brain networks. VNS can facilitate brain network reorganization of a target subject, and reorganization proceeds in a direction toward a normal brain network state. Therefore, it is proposed that a sample object (for example, an epilepsy control rate is greater than or equal to a preset threshold value) in which epilepsy is controlled after a specific medical treatment (for example, VNS) is called a valid object, a sample object (for example, an epilepsy control rate is less than a preset threshold value) in which epilepsy is not controlled after the specific medical treatment (for example, VNS) is called an invalid object, pre-operation brain networks and post-operation brain networks of the valid object and the invalid object are respectively constructed by electroencephalogram data acquired before and after the valid object and the invalid object are subjected to the specific medical treatment (for example, VNS), and based on this, a brain network difference change between the valid object and the invalid object caused by the specific medical treatment is determined. This differential change may reflect the inter-individual differences in the (non-specific) susceptibility to changes in the brain network caused by external stimuli. This susceptibility may be based on the target subject's suitability for a particular medical treatment, and may be manifested to some extent by the brain network state of the target subject itself.

Based on the conclusion, the electroencephalogram data processing method and the electroencephalogram data processing device can utilize the sample object to receive the electroencephalogram data before specific medical treatment, obtain the brain network and the brain network characteristics of the sample object through quantitative analysis, and obtain a prediction model by combining a machine learning method; the method comprises the steps of receiving electroencephalogram data before specific medical treatment by using a target object, obtaining a brain network and brain network characteristics of the target object through quantitative analysis, and accurately and efficiently predicting the adaptability of the target object to the specific medical treatment by combining a prediction model.

Fig. 1 shows an exemplary application scenario of the electroencephalogram data processing method according to an embodiment of the present application. As shown in fig. 1, the electroencephalogram data processing method according to the embodiment of the present application may be executed by a server or a terminal device, and the server or the terminal device may include, but is not limited to, a smartphone, a personal computer, a tablet computer, and other various electronic devices. The server or the terminal device can be deployed with a trained prediction model. The target object can be worn or is worn with data acquisition equipment capable of acquiring electroencephalogram data, the server or the terminal equipment can receive the electroencephalogram data from the data acquisition equipment in a wired or wireless communication mode, and the terminal equipment can also be the data acquisition equipment. Before the target object receives specific medical treatment, the data acquisition equipment continuously acquires the electroencephalogram data of the target object and sends the electroencephalogram data to the server or the terminal equipment within a certain time period.

According to the electroencephalogram data processing method, on the server or the terminal device, electroencephalogram data of the target object are subjected to a series of analysis processing, a feature vector used for being input into the prediction model can be obtained, the feature vector is input into the prediction model, and a prediction result of the target object can be obtained. From the prediction result, the suitability of the target subject for a specific medical treatment such as a vagus nerve stimulation operation can be determined.

An exemplary manner of obtaining a predictive model according to an embodiment of the present application is described below in conjunction with fig. 2-13. Fig. 2 shows an exemplary schematic diagram of an exemplary acquisition method of a predictive model according to an embodiment of the application. As shown in fig. 2, in one possible implementation, the method includes:

s11, acquiring artifact-free inter-episode electroencephalogram data of the sample object.

For example, the sample object may include the valid object and the invalid object described above, and the artifact-free inter-episode brain electrical data of the sample object may be extracted from the pre-operative brain electrical data of the sample object. The preoperative electroencephalogram data of the sample object can be electroencephalogram data acquired by adopting clinical electroencephalogram equipment before the sample object is subjected to specific medical treatment. For example, a plurality of sample subjects are subjected to conventional long-range electroencephalogram data acquisition by using a data acquisition device before being subjected to a specific medical treatment, the data acquisition device can be a Japanese photoelectric brand electroencephalograph, electrodes of the electroencephalograph can be pure silver/silver chloride plated disc-shaped electrodes, the electrode positions can be arranged according to international standard lead 10-20 system standard, a reference electrode can be set as a CZ electrode placed at the top of the head, the real-time sampling rate of the electroencephalograph is 250Hz for example, band-pass filtering parameters are 0.5Hz to 70Hz for example, and all electrode impedances are lower than 10K omega for example. Each electrode can be used as a collection channel of electroencephalogram signals, electroencephalogram data can comprise multi-channel electroencephalogram signals collected by a plurality of electrodes, the long range can be continuous for several hours, in order to improve accuracy of electroencephalogram data, each sample object can be monitored for electroencephalogram data for a period of time (for example, several days) before receiving specific medical treatment, and the electroencephalogram data before operation of a plurality of long ranges can be collected to ensure that the electroencephalogram data before operation meeting requirements are obtained.

In one possible implementation, pre-operative electroencephalographic data of a sample subject may be manually screened by visual analysis (visual inspection). For example, when pre-operation electroencephalogram data is acquired, a sample object may be in an epileptic seizure stage, in this case, the physiological state of the sample object is unstable, and the electroencephalogram activity may not be regular, so that the sample object cannot be used as a data basis for processing to obtain a prediction model. In this case, an electroencephalogram generated according to the acquired preoperative electroencephalogram data can be observed, the characteristics shown by the electroencephalogram can be analyzed, and the inter-seizure segment corresponding to the electroencephalogram data acquired between two epileptic seizures can be determined. Furthermore, the head movements, the electromyographic signals, and the electro-ocular signals of the sample objects all cause certain errors in the signal amplitude, frequency, phase, and the like in the electroencephalogram, and appear as artifacts in the electroencephalogram. The part without obvious artifacts in the inter-episode segment can be extracted as the non-artifact inter-episode electroencephalogram data of the sample object. The duration of the artifact-free inter-episode electroencephalogram data can be preset, for example, 10 min. Visual analysis is respectively carried out on the preoperative electroencephalogram data of each sample object, and artifact-free inter-attack electroencephalogram data of a plurality of sample objects can be obtained.

And S12, preprocessing the artifact-free episode-interval electroencephalogram data to obtain processed electroencephalogram data.

For example, the artifact-free inter-episode electroencephalogram data screened for visual analysis may still have errors invisible to the naked eye, and to obtain a clean, smooth signal, one or more pre-processing steps may be applied to the artifact-free inter-episode electroencephalogram data prior to quantitative electroencephalogram analysis. Wherein the one or more pre-processing steps may include one or more of prior art filtering processing, de-noising processing, re-referencing processing, and segmentation processing.

FIG. 3 shows an example of a pre-processing manner of an electroencephalogram signal according to an embodiment of the present application. As shown in fig. 3, in a possible implementation manner, an execution order of the steps for preprocessing the artifact-free inter-episode-electroencephalogram data may be preset, for example, when the preprocessing steps include filtering, denoising, re-referencing, and segmentation, the steps may be performed sequentially according to the order of filtering, denoising, re-referencing, and segmentation.

For example, if the data acquisition device has N electrodes (N >1 and is an integer), the artifact-free inter-episode brain data processed brain data may include N channels of artifact-free inter-episode brain signals P1-PN.

First, a filtering process is performed on the artifact-free inter-episode brain electrical signals P1-PN for the N channels. The filtering process may be, for example, band-pass filtering, and the filtering parameters may be preset, for example, set to 2Hz to 30Hz, that is, after band-pass filtering, data with a frequency less than 2Hz and a frequency greater than 30Hz are removed, and the frequency of the remaining data is between 2Hz and 30 Hz. By performing band-pass filtering on the electroencephalogram data, baseline drift, power frequency interference and high-frequency noise in the electroencephalogram data can be removed. After filtering the artifact-free attack interval electroencephalogram signals P1-PN of the N channels, filtered electroencephalogram signals P11-P1N of the N channels can be obtained.

After the filtering process, denoising processing can be performed on the filtered electroencephalogram signals P11-P1N of the N channels. The denoising process can be performed by an Independent Component Analysis (ICA) method in the prior art, which separates independent signal components from each other according to a statistically independent principle for a plurality of signal components from mutually independent signal sources mixed in the same signal. The electroencephalogram signals of each channel can also comprise signal components generated by the electro-ocular activity, the head movement, the electromyographic activity and the like besides the signal components generated by the electroencephalographic activity, and because the signal sources of the signal components are different, the signal components meet the independent condition of the signal source required by the independent component analysis method, the independent component analysis method is executed, the signal components generated by the electro-ocular activity, the head movement and the electromyographic activity can be automatically identified and can be removed from the electroencephalographic signals of each channel, so that the proportion occupied by the signal components generated by the electroencephalographic activity in the electroencephalographic data of each channel after denoising processing is larger, and further denoising processing is completed. After the N channels of filtered electroencephalogram signals P11-P1N are subjected to denoising processing, N channels of denoised electroencephalogram signals P21-P2N can be obtained.

After the denoising processing, re-reference processing can be performed on the denoised electroencephalogram signals P21-P2N of the N channels. The re-reference process means resetting the reference electrode, and REST technology in the prior art can be adopted. The reference electrode may be generally configured as a nose tip electrode, a vertex electrode, or the like. The brain electrical signal for each channel may be the potential difference between the active electrode and the reference electrode (e.g., the parietal electrode) corresponding to that channel. In this case, for the two active electrodes closer to and farther from the reference electrode, when the potential changes at the two active electrodes caused by the electroencephalogram activity are the same, the potential difference between the active electrode farther from the reference electrode and the reference electrode is naturally greater than the potential difference between the active electrode closer to the reference electrode and the reference electrode, and the difference is generated by the selection of the reference electrode and is not related to the electroencephalogram activity, so that certain errors appear in the electroencephalogram data. The reference electrode can be reselected through re-reference processing, and errors of electroencephalogram data caused by selection of the reference electrode are reduced. After the de-noised electroencephalogram signals P21-P2N of the N channels are subjected to re-reference processing, re-referenced electroencephalogram signals P31-P3N of the N channels can be obtained.

After the re-reference processing, segmentation processing can be performed on the re-referenced electroencephalogram signals P31-P3N of the N channels. The segmentation processing may be to mark a segment of electroencephalogram data according to a preset duration, so that the segment of electroencephalogram data can be marked as a plurality of segments of data, for example, when the total duration of the electroencephalogram data is 10min and the preset duration is 2s, the segment of electroencephalogram data with the total duration of 10min is marked according to 2s, and 300 segments of electroencephalogram data can be obtained. After the N channels of the re-referenced electroencephalogram signals P31-P3N are subjected to segmentation processing, according to the marked segment number M, N × M segmented electroencephalogram signals P311-P31M, P321-P32M, … … and P3N1-P3NM of the N channels can be obtained. In this case, the N × M segmented electroencephalogram signals can be used as processed electroencephalogram data. The segmentation processing enables partial operations in the process of obtaining the synchronization matrix according to the processed electroencephalogram data to be completed in parallel, the following description of step S13 is referred to as an exemplary implementation manner of obtaining the synchronization matrix, and the segmentation processing enables the accuracy of data processing to be improved.

It should be understood by those skilled in the art that the filtering process, the de-noising process, the re-referencing process, and the segmenting process may be performed in other orders, and the order of performing the steps of the pre-processing stage is not limited in the present application. In addition, the preprocessing stage may further include other possible steps, such as windowing, changing a sampling rate, etc. in the prior art, and a specific selection manner of the preprocessing step may be determined according to the artifact-free inter-attack electroencephalogram data quality after the visual analysis processing, which is not limited in this application.

And S13, obtaining a synchronization matrix according to the processed electroencephalogram data.

Taking the application scenario of fig. 3 as an example, the processed electroencephalogram data may include N × M segmented electroencephalogram signals of N channels, according to the prior art, electroencephalogram signals of each channel corresponding to a specific frequency band of each segment may be respectively extracted from the electroencephalogram signals of each channel corresponding to each segment, an index indicating a synchronization relationship of a combination of every two channels in the multiple channels under the specific frequency band is calculated, and a synchronization matrix under the specific frequency band is obtained according to all combinations of the two channels and the corresponding indexes.

The specific frequency band may be one or more preset frequency bands. For example, the common frequency bands are the delta frequency band (1-4Hz), the theta frequency band (4-8Hz), the alpha frequency band (8-14Hz), the beta1 frequency band (14-20Hz) and the beta2 frequency band (20-30Hz), and the specific frequency range of each frequency band can be adjusted adaptively according to the age of the patient. The particular frequency bands may include one or more of a delta frequency band, a theta frequency band, an alpha frequency band, a beta1 frequency band, and a beta2 frequency band.

The index obtained from the electroencephalogram signal includes, but is not limited to, one or more of coherence, Phase Locking Value (PLV), phase delay index (PLI), weighted phase delay index (wPLI), Granger Causality (GC), and synchronization likelihood (SI).

Fig. 4 is a diagram illustrating an exemplary method for obtaining a phase-locked value according to an embodiment of the present application. An exemplary method of calculating the phase-lock value is described below in conjunction with fig. 3 and 4.

The basic theoretical assumption of phase lock values is: the phase locking value can reflect the interaction between neurons, under the condition of avoiding using signal amplitude, the synchronism of two channels under specific frequency is predicted only by using phase information, and the channel is directly used as a brain network node, so that the phase locking value can directly reflect the variability of phase difference between two nodes in the brain network node. For example, if the phase locking value is 1, it can indicate that the directions of the phase vectors of the signals of the two channels are consistent, and the phase difference is constant, i.e. the two channels are completely synchronous; if the phase locking value is 0, it can be shown that the phases of the signals of the two channels are randomly distributed on the unit circle, and the two electrode channels cannot be completely synchronized.

As shown in fig. 4, band-pass filtering may be performed on the electroencephalogram signals of the N channels corresponding to each segment, respectively, to obtain electroencephalogram signals of the N channels at a specific frequency band of each segment. For the electroencephalogram signals of the N channels under the specific frequency bands of the segments, Hilbert transform can be carried out, and analytic signals of the N channels under the specific frequency bands of the segments are obtained respectively. According to the analytic signals under the specific frequency band of each segment of the N channels, the phase of the analytic signals under the specific frequency band of each segment of the N channels can be determined. In this case, the phase difference between the analytic signals of any two channels (e.g., channel x and channel y, 1 ≦ x ≦ N, 1 ≦ y ≦ N) in the specific frequency band (e.g., theta frequency band) of the same segment (e.g., segment M) is substituted into equation (1), and the phase locking value in the specific frequency band of one segment of the two channels can be calculated.

In the formula (1), x and y respectively represent two channels,

respectively representing the phases of the channels x and y at the t-th time point of the same section under a specific frequency band, n representing the number of time points, PLV_xyIndicating the phase lock values of the combination of channels x, v at a particular frequency band of the same segment.

The phase difference between the analytic signals of the channel x and the channel y in the specific frequency band (e.g., theta frequency band) of each segment (e.g., segment 1-M) is sequentially substituted into the formula (1), so as to obtain the phase locking value of the combination of the channel x and the channel y in the specific frequency band of each segment. And averaging the sum of the phase locking values of the combination of the channels x and y under the specific frequency band of each segment according to the number (M) of the segments to obtain the phase locking value of the combination of the channels x and y under the specific frequency band. The phase locking value of the combination of the channels x and y in a specific frequency band can be used as one element in the synchronization matrix. And calculating to obtain phase locking values of all channel combinations under the specific frequency band by adopting the same method, thereby obtaining all elements in the matrix.

Under the application scene that the segmentation processing is not carried out in the preprocessing stage, the processed electroencephalogram data can comprise processed electroencephalogram signals of N channels, and the processed electroencephalogram signals of the N channels can be subjected to band-pass filtering respectively to obtain electroencephalogram signals of the N channels under the specific frequency bands. For the electroencephalogram signals of the N channels under the specific frequency band, Hilbert transformation can be carried out, and analytic signals of the N channels under the specific frequency band are obtained respectively. According to the analytic signals under the specific frequency band of the N channels, the phase of the analytic signals under the specific frequency band of the N channels can be determined. In this case, the phase difference of the analytic signals in the specific frequency band (e.g., theta frequency band) of any two channels (e.g., channel x and channel y, 1 ≦ x ≦ N, 1 ≦ y ≦ N) is substituted into equation (1), and the phase locking value in the specific frequency band of the two channels can be calculated. The phase locking value of the combination of the channels x and y in a specific frequency band can be used as one element in the synchronization matrix. And calculating to obtain phase locking values of all channel combinations under the specific frequency band by adopting the same method, thereby obtaining all elements in the matrix.

Fig. 5 illustrates an example of a synchronization matrix according to an embodiment of the present application. As shown in FIG. 5, the elements PLV11, PLV of the synchronization matrix₁₂……PLV_1N，PLV₂₁、PLV₂₂……PLV_2N，……，PLV_N1、PLV_N2……PLV_NNMay be a phase locked value, and the elements on the main diagonal of the synchronization matrix may represent the synchronicity of each channel in combination with itself at a particular frequency band, and thus the value of the element on the main diagonal (PLV)₁₁，PLV₂₂，……，PLV_NN) May be 1; besides the main diagonal line, each of the other elements may represent the synchronicity of the combination of the channel corresponding to the row in which the element is located and the channel corresponding to the column in which the element is located in the specific frequency band, which can be calculated by the above formula (1).

Fig. 6 shows a schematic diagram of an exemplary method for obtaining a weighted phase delay index according to an embodiment of the present application. An exemplary method of calculating the weighted phase delay index is described below in conjunction with fig. 3 and 6.

The weighted phase delay index can be used for measuring the asymmetry of the distribution of the phase difference sequence with 0 as the center, the synchronism of two channels under specific frequency can be predicted through the asymmetry of the phase difference distribution, and the weighted phase delay index can directly reflect the asymmetry of the phase difference distribution between every two nodes in the brain network nodes by directly using the channels as the brain network nodes. For example, if the value of the weighted phase delay index is between 0 and 1, and if the value of the weighted phase delay index is 0, it may indicate that the phase difference distribution of the signals of the two channels is completely symmetric, and the two channels may be considered to be completely asynchronous; if the weighted phase delay index has a value of 1, it can indicate that the phase of the signal of one channel always lags (or leads) the other channel, and the two channels can be considered to be completely synchronous.

As shown in fig. 6, band-pass filtering may be performed on the electroencephalogram signals of the N channels corresponding to each segment, respectively, to obtain electroencephalogram signals of the N channels at a specific frequency band of each segment. For the electroencephalogram signals of the N channels under the specific frequency bands of the segments, Hilbert transform can be carried out, and analytic signals of the N channels under the specific frequency bands of the segments are obtained respectively. According to the analytic signals of the N channels under the specific frequency bands of the segments, cross spectrums of the two channels under the specific frequency bands of the segments can be obtained respectively. According to the cross spectrum, the amplitude of the analytic signal under the specific frequency band of each segment of every two channels and the phase difference of the analytic signal under the specific frequency band corresponding to the same segment of every two channels at different time can be determined. And substituting the phase difference and the amplitude into a formula (2), so that the weighted phase delay index of one segment of the two channels under the specific frequency band can be calculated.

In the formula (2), x and y respectively represent two channels, L₁And L₂Respectively representing the amplitude of the analytic signals of the channels x and y under the specific frequency band of the same segment, N representing the number of sample points, wherein theta_kAnd the phase difference of the analytic signals of the channels x and y in the specific frequency band of the same segment at the corresponding time point of the kth sample point is represented. wPLI_xyIndicating the weighted phase delay index of the combination of channels x, y in a particular frequency band of the same segment.

The amplitude and phase difference of the analytic signals of the channel x and the channel y in the specific frequency band (for example, theta frequency band) of each segment (for example, segments 1 to M) are sequentially substituted into the formula (2), and the weighted phase delay index of the combination of the channel x and the channel y in the specific frequency band of each segment can be obtained. And averaging the sum of the weighted phase delay indexes of the combination of the channels x and y under the specific frequency band of each segment according to the number (M) of the segments to obtain the weighted phase delay index of the combination of the channels x and y under the specific frequency band. The weighted phase delay index of the combination of the channels x and y in a specific frequency band can be used as one element in the synchronization matrix. And calculating to obtain the weighted phase delay indexes of all the channel combinations under the specific frequency band by adopting the same method, thereby obtaining all the elements in the matrix.

Under the application scene that the segmentation processing is not carried out in the preprocessing stage, the processed electroencephalogram data can comprise processed electroencephalogram signals of N channels, and the processed electroencephalogram signals of the N channels can be subjected to band-pass filtering respectively to obtain electroencephalogram signals of the N channels under the specific frequency bands. For the electroencephalogram signals of the N channels under the specific frequency band, Hilbert transformation can be carried out, and analytic signals of the N channels under the specific frequency band are obtained respectively. According to the analytic signals of the N channels under the specific frequency band, cross spectrums of every two channels under the specific frequency band can be obtained respectively. According to the cross spectrum, the amplitude of the analytic signal under the specific frequency band of each two channels and the phase difference of the analytic signal under the specific frequency band of each two channels at different moments can be determined, and the weighted phase delay index under the specific frequency band of the two channels can be calculated by substituting the amplitude and the phase difference into the formula (2). The weighted phase delay index of the combination of the channels x and y in a specific frequency band can be used as one element in the synchronization matrix. And calculating to obtain the weighted phase delay indexes of all the channel combinations under the specific frequency band by adopting the same method, thereby obtaining all the elements in the matrix.

Fig. 7 illustrates an example of a synchronization matrix according to an embodiment of the present application. As shown in fig. 7Shown, element wPLI of the synchronization matrix₁₁、wPLI₁₂……wPLI_1N，wPLI₂₁、wPLI₂₂……wPLI_2N，……，wPLI_N1、wPLI_N2……wPLI_NNMay be a weighted phase delay index, the element on the main diagonal of the synchronization matrix (wPLI)₁₁、wPLI₂₂、……、wPLI_NN) In the calculation of (2), the channels x, y are the same channel, e.g. wPLI₁₁The corresponding channels x, y are both channel 1, so θ_kIs 0, so that the denominator on the right side of equation (2) is 0, so that wPLI₁₁、wPLI₂₂、……、wPLI_NNCannot be calculated. The element on the main diagonal line can be set to 0 in advance; each of the other elements, except the main diagonal line, may represent the synchronicity of the combination of the channel corresponding to the row in which the element is located and the channel corresponding to the column in which the element is located in the specific frequency band, and may be calculated by the above formula (2).

In a possible implementation manner, the manner of obtaining the synchronization matrix according to the indexes of coherence, phase delay index, glange cause and synchronization likelihood may be implemented based on the prior art, wherein an exemplary manner of obtaining the phase delay index may refer to the related description of fig. 6, and equation (2) may be replaced by an existing equation for calculating the phase delay index. Exemplary methods of obtaining coherency, synchronization likelihood, granger cause and effect are briefly described below in connection with fig. 8, 9, and 10.

As shown in fig. 8, band-pass filtering may be performed on the electroencephalogram signals of the N channels corresponding to each segment, respectively, to obtain electroencephalogram signals of the N channels at a specific frequency band of each segment. For the electroencephalogram signals of the N channels under the specific frequency bands of the segments, fast Fourier transform or continuous wavelet transform can be carried out, and frequency domain information of the N channels under the specific frequency bands of the segments is obtained respectively. According to the frequency domain information of the segments of the N channels under the specific frequency band, the power spectral density of each frequency under the specific frequency band of each segment of the N channels and the cross power spectral density under the specific frequency band of each segment of every two channels can be obtained through power spectral density calculation. From the power spectral densities calculated above, the coherence of each frequency at a particular frequency band for each segment of each two channels can be calculated. The coherence of every two channels under the specific frequency band of each segment can be obtained by averaging the sum of the coherence of every two channels under the specific frequency band of each segment according to the number of the segments (M). The coherence of each two channels at a specific frequency band can be used as an element in the synchronization matrix, and the position of the coherence of each two channels in the synchronization matrix can refer to the examples of fig. 5 and fig. 7.

Fig. 9 shows a schematic diagram of an exemplary method for obtaining synchronization likelihood according to an embodiment of the present application. Here, the synchronization likelihood of the time point a of the channel 1 is calculated as an example.

As shown in fig. 9, the electroencephalogram signal of the channel 1 may include a plurality of time points, for example, if synchronization likelihood of a certain time point of the channel 1 is desired, for example, the time point a, a time point c which is earlier than the time point a and whose starting point is the time point c and whose ending point is the time point a may be determined according to the electroencephalogram signal of the channel 1 (which may be after segmentation before segmentation) and the preset time delay parameter, so as to obtain a time window whose starting point is the time point a. And segmenting the time window according to a preset embedding dimension parameter to enable the number of segments to be equal to the numerical value of the embedding dimension parameter, wherein the amplitude numerical value of the electroencephalogram signal corresponding to the time point at the dividing position of each two segments is used as one element of an embedding vector, the amplitude value of the electroencephalogram signal corresponding to the time point c is used as the first element of the embedding vector, the amplitude value of the electroencephalogram signal corresponding to the time point at the dividing position of the first segment and the second segment is used as the second element of the embedding vector, and so on, and the amplitude value of the electroencephalogram signal corresponding to the time point a is used as the last element of the embedding vector, so that the embedding vector of the time point a is obtained. Optionally, a time point of the channel 1 other than the time point a, for example, the time point b, may be obtained by repeating the above operations on the time point b according to the electroencephalogram signal (which may be segmented before and after the segmentation) of the channel 1, a preset time delay parameter, and a preset embedding dimension parameter. According to the embedding vector of the time point a, the embedding vector of the time point b and the preset evaluation probability, the critical distance between the embedding vector of the time point a and the embedding vector of the time point b can be calculated, and according to the electroencephalogram signals of other channels (channels 2-N), the preset time delay parameter and the preset embedding dimension parameter, the distance between the embedding vector of the time point a and the embedding vector of the time point b of the other channels can be calculated respectively. After the calculation is completed, the number of channels whose distance between the embedding vector of the time point a and the embedding vector of the time point b of other channels satisfies the critical distance less than or equal to the embedding vector of the time point a and the embedding vector of the time point b is determined, the determined number of channels is substituted into the existing synchronization likelihood formula, the synchronization likelihood of the time point a and the time point b of the channel 1 can be obtained, the synchronization likelihood of the time point a of the channel 1 and all the time points of the channel 1 can be calculated by referring to the above method, the obtained synchronization likelihood sum of the time point a of the channel 1 and all the time points of the channel 1 is averaged, and the synchronization likelihood of the time point a of the channel 1 can be obtained.

Further, the synchronization likelihood of each time point of the channel 1, the synchronization likelihood of each time point of the channel 2, … …, and the synchronization likelihood of each time point of the channel N may be calculated, and then the synchronization matrix may be obtained according to the synchronization likelihood of each time point of each channel.

As shown in fig. 10, the order W of the autoregressive model to be obtained subsequently may be determined according to the electroencephalogram signal (which may be obtained before segmentation) of the channel 1 and the akabane information criterion or the bayesian information criterion, and a time point of the channel 1, for example, a time point a, is selected, and a time sequence between the time point a of the channel 1 and W time points before the time point is obtained according to the time point a and W time points before the time point in the electroencephalogram signal of the channel 1. By analogy, the time sequence between the time point a of the N channels and the W previous time points can be obtained. And performing autoregressive fitting according to the time sequence between the time point a of every two channels and the previous W time points to obtain two W-order autoregressive models of the combined time sequence of the two channels. Taking a time sequence between the time point a of the channel 1 and W previous time points as an independent variable, and taking a time sequence between the time point a of the channel 2 and W previous time points as a dependent variable, so as to obtain a W-order autoregressive model. And taking the time sequence between the time point a of the channel 2 and the previous W time points as an independent variable, and taking the time sequence between the time point a of the channel 1 and the previous W time points as a dependent variable to obtain a W-order autoregressive model. And (3) according to a W-order autoregressive model obtained by autoregressive fitting of the time series between the time point a of all the channel combinations and the previous W time points, processing to obtain the W-order autoregressive model of the time series of the whole brain network. The method comprises the steps that each time of autoregressive fitting, a W-order autoregressive model of a time sequence of the whole brain network comprises a 0-order residual error item, and the variance of the residual error items can be calculated according to all the residual error items obtained from the beginning of autoregressive fitting to the end of autoregressive fitting, wherein the variance is the time domain granger cause and effect. If the frequency domain Glan cause and effect needs to be calculated, Fourier transform can be carried out on a coefficient matrix of the autoregressive model obtained through fitting, and the obtained numerical value is the frequency domain Glan cause and effect. The synchronicity matrix may be derived from either a time domain glancing cause or a frequency domain glancing cause.

In a possible implementation manner, there may be one or more synchronization matrices for each two channel combinations, for example, when there are multiple calculated indexes indicating synchronization relationships of each two channel combinations in the multiple channels in a specific frequency band, each index may correspond to obtain one synchronization matrix. For another example, when there are multiple specific frequency bands, a synchronization matrix may be obtained for each frequency band. After steps S14 and S15 are executed, each synchronization matrix may obtain a plurality of brain network attributes, and when there are a plurality of synchronization matrices combined for each two channels, each brain network attribute may also correspond to a plurality of values, in which case, the plurality of values may be averaged. In order to further improve the data processing efficiency, a smaller number of indexes and/or a smaller number of frequency bands may be selected, and in order to further improve the accuracy, a larger number of indexes and/or a larger number of frequency bands may be selected, that is, the specific acquisition mode of the synchronization matrix may be flexibly set according to the requirements of the application scenario, which is not limited in this application.

And S14, obtaining a threshold correlation matrix according to the synchronicity matrix and the threshold.

The threshold selection mode is flexible and can be determined according to elements in the synchronization matrix, and the threshold can be selected according to the constraint conditions of removing weak connection edges (noise edges), ensuring smooth network (no isolated brain region), network density and the like in the prior art. Taking the determination of the threshold value according to the removal of the weak connection edge as an example, if there is a connection edge between two nodes in the network, it indicates that the two nodes are not completely out of synchronization. If the elements in the synchronization matrix are phase-locked values, all the phase-locked values in the matrix can be arranged from large to small, the phase-locked values with the first 80% of the values can be retained, i.e. the element values of the positions in the synchronization matrix corresponding to the first 80% of the phase-locked values are set to 1, and the phase-locked values with the last 20% of the values can be removed, i.e. the element values of the positions in the synchronization matrix corresponding to the last 20% of the phase-locked values are set to 0, in which case all the elements in the matrix are 0 or 1, thereby constructing a binary matrix, i.e. a threshold correlation matrix. If the layout of the elements in the synchronization matrix is as in the example of fig. 5, the elements on the main diagonal of the threshold correlation matrix obtained from the synchronization matrix may be set to 0, because there is no connecting edge between the same node in the network and itself. The threshold determined in this way makes the selection of the threshold more suitable for a specific application scenario, and can improve the closeness of the brain network constructed according to the threshold incidence matrix and the real brain network of the sample object.

In a possible implementation manner, the threshold may also be a preset numerical value, the magnitude relationship between the numerical value of each element and the threshold is determined, and the element whose magnitude relationship between the numerical value and the threshold satisfies the condition is set to 1, and the element which does not satisfy the condition is set to 0. Therefore, the steps of arranging the elements in the matrix from small to large and determining the threshold can be omitted, and the data processing efficiency can be improved.

And S15, obtaining at least one brain network attribute according to the threshold correlation matrix.

The brain network can be established according to the threshold incidence matrix, nodes of the brain network can correspond to the N channels respectively, the connection relation among the nodes can be determined through the threshold incidence matrix, an element in the threshold incidence matrix is 1, which indicates that a connection edge exists between the node pairs corresponding to the element, and an element in the threshold incidence matrix is 0, which indicates that a connection edge does not exist between the node pairs corresponding to the element. After the brain network is established, brain network attributes including but not limited to one or more of average synchronicity, characteristic path length, clustering coefficient, betweenness centrality, global efficiency and local efficiency can be extracted based on the graph theory of the prior art. The brain network attributes are described separately below.

The brain network attribute comprises average synchronicity, wherein the average synchronicity refers to the average value of other elements except diagonal elements on a synchronicity matrix, each element of the synchronicity matrix represents synchronicity of each node pair of the network, and the average synchronicity represents synchronicity of the whole network.

The brain network attribute includes a feature path length, which is a global feature of the network and refers to an average value of all shortest path lengths between all node pairs of the brain network, where the shortest path length represents the number of upper edges of the shortest paths between the node pairs, for example, if there are two paths from node a to node B, one includes 3 edges, and one includes 4 edges, the shortest path length from node a to node B may be 3. A smaller characteristic path length indicates a faster information transmission speed of the network. One exemplary way to calculate the characteristic path length is as follows:

firstly, the shortest path lengths among all nodes in the network are summed to obtain the shortest path length sum C of the network_T. Sum of shortest path lengths C_TAn example of the calculation method of (c) is shown in formula (3).

C_T＝∑_i,jD(i,j) (3)

In formula (3), D (i, j) represents the shortest path length of the node pair from node i to node j, i is greater than or equal to 1 and less than or equal to N, j is greater than or equal to 1 and less than or equal to N, and N represents the total number of nodes. A node and all other nodes except itself can form a node pair, so that each node can correspond to N-1 node pairs at most, and N nodes correspond to N (N-1) node pairs in total (node i to node j and node j to node i are regarded as two node pairs), then the characteristic path length C can be calculated by the following exemplary manner of formula (4):

the brain network attributes include clustering coefficients. Clustering coefficient is a measure of the clustering degree of the brain network, which is associated with the clustering coefficient. According to the threshold incidence matrix, determining that connecting edges are arranged between any three nodes in the brain network, and when the three nodes and the connecting edges can form a triangle, the three nodes in the brain network are considered to be clustered, and the clustering coefficient of the brain network represents the clustered triangle number of any three nodes in the brain network as vertexes. The larger the clustering coefficient of the brain network is, the larger the number of triangles in the network is, the larger the clustering degree of the brain network is. Meanwhile, information transmission can be independently completed between every two three nodes forming the triangle, and the larger the clustering coefficient of the brain network is, the stronger the information transmission capability of the network is. An example of the calculation formula of the clustering coefficient is shown in formula (5):

in the formula (5), C_iRepresenting the clustering coefficient of a single node i, N representing the total number of all nodes in the network, N representing the set of all nodes in the network, t_iNumber of triangles, k, representing clustering with node i as vertex_iIndicating the degree of the node i (the number of nodes having connecting edges with the node i). Clusthe ending Coefficient represents the clustering Coefficient of the brain network, and is equal to the mean value of the clustering coefficients of all nodes in the network.

Brain network attributes include betweenness centrality. Betweenness centrality is a metric that measures the connectivity between different nodes connected to a node. For the shortest paths between all the node pairs in the network, the betweenness centrality of the given node can be obtained by calculating the ratio of the number of the shortest paths passing through the given node to the total number of the shortest paths between all the node pairs. An example of a specific calculation method of the betweenness centrality is shown in formula (6):

wherein i, j, k belongs to N, i ≠ j ≠ k, and N denotes a set including all nodes in the network. Determining the shortest path and the number D (j, k) of the shortest paths between the node j and the node k according to all the paths between the node j and the node k by taking the node j as a path starting point (or an end point) and taking the node k as a path end point (or a starting point); determining the number of paths D through node i in the shortest path between node j and node k_i(j，k)。

The brain network attributes include global efficiency. The global efficiency is a metric index for measuring information flow, is a scalar quantity, and can be defined as the inverse of all shortest path lengths in the network. An example of a calculation formula of the global efficiency is shown in formula (7):

in formula (7), N represents a set including all nodes in the network, N represents the total number of all nodes in the network, and d_ijRepresenting the shortest path length between node i and node j. As can be seen from equation (7), the global efficiency E_gloIs an average of the inverse shortest path lengths, the improvement of the global efficiency is mainly affected by the small shortest path length.

Brain network attributes include local efficiency. The local efficiency indicates the efficiency of information transfer on a sub-network, where n indicates the total number of all nodes in the network, and a sub-network may be a network composed of some of the n nodes. An example of a calculation formula for the local efficiency of the network is shown in formula (8):

in the above formula, H is the set of all nodes of the brain network, d_jhIndicating the shortest path length between node j and node h. k is a radical of_iIndicating the degree of the node i (the number of nodes having connecting edges with the node i). w is a_iiRepresenting the weight of the connection between node i and node j in the network, w_ihRepresents the weight of the connection between node i and node h in the network, w_ijAnd w_ihIs in the range of 0 to 1, w_ij(or w)_ih) Closer to 1, the stronger the connection between node i and node j (or node i and node h), w_ij(or w)_ih) The closer to 0, the weaker the connection between the node i and the node j (or the node i and the node h).

It will be appreciated by those skilled in the art that the brain network attributes that can be determined from the threshold correlation matrix should be more than the attributes described above, and may also include brain network attributes that can be derived by those skilled in the art from the prior art. The specific selection of brain network attributes is not limited by the present application.

And S16, obtaining a prediction model according to the at least one brain network attribute.

For example, according to at least one brain network attribute, brain network attributes related to adaptability of a specific medical treatment can be screened out through significance difference analysis, and importance degrees of adaptability are ranked for the screened brain network attributes through a machine learning method. The prediction model may be derived from the ranking results. Fig. 11 illustrates one example of an implementation of screening and ranking brain network attributes according to an embodiment of the present application. Referring to FIG. 11, step S16 may, for example, include the following steps S121-S123:

and S121, classifying the brain network attributes of the sample object into a positive sample and a negative sample according to the effect of the sample object after receiving the specific medical treatment.

As can be seen from the above description, the sample object has already undergone a specific medical treatment, and the post-treatment effect has a certain correlation with the brain network properties of the sample object. The effect of a sample subject after a specific medical treatment may be predetermined as a valid subject with a reduced post-operative seizure frequency compared to the pre-operative frequency. The effect of a sample subject after a specific medical treatment can be predetermined as a non-effective subject after surgery where the frequency of seizures does not significantly change before surgery. In this case, a positive sample may include brain network attributes corresponding to valid subjects, and a negative sample may include brain network attributes corresponding to invalid subjects.

And S122, performing inter-group difference significance analysis on the brain network attributes of the positive and negative samples, and determining the brain network attributes with statistically significant differences.

For example, significance difference (significance difference) is a statistical term that can be used to evaluate the difference in data. If there is a significant difference between certain types of data, e.g., the difference between the data involved in the alignment is greater than or equal to a threshold, the data involved in the alignment may be considered not to be from the same Population (Population), but from two different populations that differ in one or more respects. If there is no significant difference between certain types of data, e.g., the difference between the data involved in the alignment is less than a threshold, the data involved in the alignment can be considered to be from the same population. In embodiments of the present application, the data involved in the alignment has been determined to be from two different populations with differences, in which case if there is a significant difference between a type of data, the type of data is associated with a difference that may be associated with both populations, and if there is no significant difference between a type of data, the type of data is not associated with a difference that may be associated with both populations.

For example, the two populations may be, for example, positive and negative samples respectively, and when the brain network attribute is average synchronicity, the difference of the populations may be, for example, the difference of the effect of the sample object after receiving a specific medical treatment, and if the analysis determines that the average synchronicity of the positive and negative samples is greatly different, for example, greater than a preset threshold, the average synchronicity for the analysis may be considered to have a significant difference therebetween, and the average synchronicity is associated with the effect of the sample object after receiving the specific medical treatment. When the brain network attribute is the characteristic path length, if the analysis determines that the characteristic path lengths of the positive sample and the negative sample are slightly different, for example, less than another preset threshold, it can be considered that there is no significant difference between the characteristic path lengths used for the analysis, and the characteristic path length is irrelevant to the effect of the sample object after being subjected to a specific medical treatment.

Step S122 can be implemented based on the mann-whitney U test method in the prior art. It should be understood by those skilled in the art that step S122 may be implemented by other methods as long as the brain network attributes with statistically significant difference can be analyzed and determined by a statistical method from the brain network attributes of the positive and negative samples, and the specific determination manner of the brain network attributes with statistically significant difference is not limited in the present application.

And S123, calculating importance scores of the brain network attributes with statistically significant differences, and sequencing the brain network attributes from high to low according to the scoring results.

Fig. 12 shows an example of an implementation manner of ranking brain network attributes according to the scoring result from high to low in the embodiment of the present application. Referring to FIG. 12, step S123 may include, for example, the following steps S21-S25:

s21, inputting the initial feature subset into a random forest machine learning model to obtain the importance score of each brain network attribute; and determining the prediction accuracy corresponding to the initial feature subset by a cross validation method.

Wherein the initial feature subset may include all brain network attributes with statistically significant differences, and T represents the number of brain network attributes in the set. For example, assuming that the brain network attributes with statistically significant differences determined in step S122 are a clustering coefficient, an betweenness centrality, a global efficiency, and a local efficiency, respectively, in this case, the initial feature subset may be S0 ═ clustering coefficient, betweenness centrality, global efficiency, and local efficiency, and T ═ 4.

The random forest machine learning model and the cross validation method can be realized based on the prior art. Step S21 may obtain, for example, a cluster coefficient importance score of 0.5, an betweenness centrality importance score of 0.2, a global efficiency importance score of 0.2, and a local efficiency importance score of 0.1 through a prior art random forest learning model. And the prediction accuracy corresponding to the initial feature subset S0 may be determined to be 60%, for example, by a prior art cross-validation method.

And S22, removing a plurality of brain network attributes with the lowest importance scores from the current feature subset, and obtaining a new feature subset according to the rest brain network attributes.

Wherein if there are a plurality of brain network attributes with the lowest importance scores, at least one of the plurality of brain network attributes may be removed. In this case, step S22 may remove the local efficiency (0.1) with the lowest importance score in the initial feature subset S0, and obtain a new feature subset S1 as { clustering coefficient, betweenness centrality, global efficiency }, where T is 3, according to the remaining clustering coefficient, betweenness centrality, and global efficiency.

S23, inputting the new feature subset into a random forest machine learning model to obtain the importance score of each brain network attribute; and determining the prediction accuracy corresponding to the new feature subset by a cross validation method.

The random forest machine learning model and the cross validation method used in step S23 may be the same as those used in step S21. Step S23 may obtain, for example, an importance score of 0.6 for the clustering coefficient, an importance score of 0.2 for the betweenness centrality, and an importance score of 0.2 for the global efficiency through a prior art random forest learning model. And the prediction accuracy corresponding to the feature subset S1 may be determined to be 70%, for example, by a prior art cross-validation method.

And S24, judging whether the current feature subset is an empty set, if not, repeating the steps S22-S24, and if so, executing the step S25.

For example, the current feature subset determined in step S24 is actually obtained by performing step S22, and as can be seen from the above description, the current feature subset S1 is { clustering coefficient, betweenness centrality, global efficiency }, and T is 3, so that it can be determined that the current feature subset is not an empty set, and the process proceeds to the first repeated execution stage of step S22-step S24. At this stage, step S22 is executed to remove the betweenness centrality (score 0.2) and the global efficiency (score 0.2) with the lowest importance score in the feature subset S1, and obtain a new feature subset S2 ═ clustering coefficient, where T is 1, according to the remaining clustering coefficients. Executing step S23, for example, the importance score of the clustering coefficient of the feature subset S2 may be obtained as 1.0, and the prediction accuracy corresponding to the feature subset S2 may be obtained as 65%. Step S24 is executed to determine that the current feature subset S2 ═ { clustering coefficient } is not an empty set, and the process proceeds to the second repeated execution stage of steps S22 to S24. In this stage, step S22 is executed to remove the clustering coefficient (score 1.0) with the lowest importance score in the feature subset S2, the obtained feature subset S3 is an empty set, step S23 is executed to fail to calculate the importance score, step S24 is executed to determine that the current feature subset S3 is an empty set, and in this case, step S25 below is executed.

For another example, after the step S24 determines that the current feature subset S1 is not an empty set { clustering coefficient, betweenness centrality, global efficiency }, the process proceeds to the first iterative execution stage of steps S22 to S24. At this stage, step S22 is executed to remove the betweenness centrality (score 0.2) in the feature subset S1 with the lowest importance score and the betweenness centrality in the global efficiency (score 0.2), retain the betweenness centrality, and obtain a new feature subset S4 ═ { clustering coefficient, betweenness centrality }, T ═ 2 according to the remaining clustering coefficients and the betweenness centrality. Executing step S23, for example, an importance score of 0.6 for the clustering coefficient of the feature subset S4, an importance score of 0.4 for the betweenness center may be obtained, and a prediction accuracy corresponding to the feature subset S4 may be obtained as 75%. Step S24 is executed to determine that the current feature subset S4 is not an empty set { clustering coefficient, betweenness centrality }, and the process proceeds to the second repeated execution stage of steps S22 to S24. At this stage, step S22 is executed to remove the centrality of the betweenness with the lowest importance score (score 0.4) in the feature subset S4, and obtain a new feature subset S5 ═ clustering coefficient, and T ═ 1 according to the remaining clustering coefficients. Executing step S23, for example, the importance score of the clustering coefficient of the feature subset S5 may be obtained as 1.0, and the prediction accuracy corresponding to the feature subset S5 may be obtained as 65%. Step S24 is executed to determine that the current feature subset S5 ═ { clustering coefficient } is not an empty set, and the process proceeds to the third repeated execution stage of steps S22 to S24. In this stage, step S22 is executed to remove the clustering coefficient (score 1.0) with the lowest importance score in the feature subset S5, the obtained feature subset S6 is an empty set, step S23 is executed to fail to calculate the importance score, step S24 is executed to determine that the current feature subset S6 is an empty set, and in this case, step S25 below is executed.

Those skilled in the art should understand that the present application is not limited to the specific removing manner of the brain network attribute with the lowest importance score as long as it can satisfy that at least one brain network attribute with the lowest importance score is removed from the current feature subset each time S22 is executed.

S25, selecting one feature subset with the highest prediction accuracy in all the feature subsets determined in the steps S22-S24, and sorting the brain network attributes in the feature subsets according to the sequence of the importance scores of the brain network attributes in the feature subsets from high to low to obtain a sorting result.

For example, of the total feature subsets S1, S2, S3 determined in steps S22-S24, the feature subset with the highest prediction accuracy may be, for example, the feature subset S1, and the corresponding prediction accuracy is 70%. The feature subset S1 includes a clustering coefficient, an betweenness centrality, and a global efficiency, where the importance score of the clustering coefficient is 0.6, the importance score of the betweenness centrality is 0.2, and the importance score of the global efficiency is 0.2. In this case, the indexes in the feature subset S1 are sorted, and a result of sorting the clustering coefficient, the betweenness centrality, the global efficiency, or the clustering coefficient, the global efficiency, and the betweenness centrality can be obtained. The embodiment of the present application does not limit the ordering order of the brain network attributes with the same importance scores.

For another example, of the total feature subsets S1, S4, S5 and S6 determined in steps S22-S24, the feature subset with the highest prediction accuracy may be, for example, the feature subset S4, and the corresponding prediction accuracy is 75%. The feature subset S4 includes an index clustering coefficient, an betweenness centrality, an importance score of the clustering coefficient of 0.6, and an importance score of the betweenness centrality of 0.4. In this case, the indexes in the feature subset S4 are sorted, and a sorting result of the clustering coefficient and the betweenness centrality can be obtained.

Step S123 may be implemented based on the recursive feature elimination feature selection method of the prior art. It should be understood by those skilled in the art that step S123 may also be implemented by other methods as long as the ranking of the brain network attributes with statistically significant difference can be completed, and the importance of any brain network attribute in the ranking result is not higher than the previous brain network attribute of the brain network attribute, and the present application does not limit the specific ranking manner of the brain network attributes with statistically significant difference.

In a possible implementation manner, after determining the importance ranking of the brain network attributes, a prediction model may be trained based on the ranking result and the brain network attributes of the sample object, and the model may be used to predict the target object to determine a prediction result of the target object.

For example, according to the ranking result of the brain network attributes, adding the brain network attributes ranked later one by one from the first brain network attribute ranked to obtain a vector, which can be used as an input feature vector of the prediction model to be trained. Training of the predictive model may be accomplished based on the input feature vectors. Fig. 13 shows an example of obtaining a trained predictive model according to an embodiment of the present application.

For example, as shown in fig. 13, after receiving an input feature vector, the prediction model to be trained performs permutation and combination according to a plurality of features in the feature vector, so as to obtain a plurality of processed feature vectors including combinations of different features. For example, the input feature vector may be { clustering coefficient, betweenness centrality, global efficiency }, and if the order of the elements in the processed feature vector is not limited, the processed feature vector may have

The number of the clusters can be { clustering coefficient }, { betweenness centrality }, { global efficiency }, { clustering coefficient, betweenness centrality }, { clustering coefficient, global efficiency }, { betweenness centrality, global efficiency }, { clustering coefficient, betweenness centrality, and global efficiency }; if the elements in the feature vector after the limiting process are arranged in sequence, the feature vector after the limiting process can have

The number of the cells can be respectively { clustering coefficient }, { betweenness centrality }, { global efficiency }, { clustering coefficient, betweenness centrality }, { betweenness centrality, global efficiency }, { clustering coefficient, global efficiency }, { betweenness centrality, clustering coefficient }, { global efficiency, betweenness centrality }, { clustering coefficient, betweenness centrality, global efficiency }, { clustering coefficient, global efficiency, betweenness centrality }, { betweenness centrality, global efficiency, clustering coefficient }, { global efficiency, clustering coefficient, betweenness centrality }, { betweenness centrality, clustering coefficient, global efficiency }, { global efficiency, betweenness centrality, and clustering coefficient }. And taking the brain network attribute with the same type as the brain network attribute of each processed feature vector in the plurality of brain network attributes of the sample object as the input of the prediction model to be trained, wherein the prediction model to be trained can obtain a prediction accuracy, the optimal feature vector can be obtained according to the processed feature vector corresponding to the input of the prediction model when the prediction accuracy is highest, and the prediction model determined according to the optimal feature vector is taken as the trained prediction model.

The prediction accuracy of the prediction model can be determined by comparing the prediction result obtained by inputting the brain network attribute of the sample object into the prediction model as a vector with the effect of the sample object after receiving a specific medical treatment (for example, vagus nerve stimulation surgery). For example, the prediction model may obtain a sample with a predicted result of responders and a sample with a predicted result of non-responders, wherein the predicted result is that the number of epileptic seizures of responders after vagus nerve stimulation surgery is reduced by more than or equal to 50%, and the predicted result is that the number of epileptic seizures of non-responders after vagus nerve stimulation surgery is reduced by less than 50%. The number of samples satisfying that the prediction result is a responder in the positive samples and the number of samples satisfying that the prediction result is a non-responder in the negative samples can be respectively counted, and the number of samples having an accurate prediction result can be obtained according to the sum of the number of samples satisfying that the prediction result is a non-responder.

The above-described manner of determining the prediction accuracy of the prediction model is merely an example. The embodiment of the present application does not limit the determination method of the prediction accuracy of the prediction model.

In one possible implementation, a linear regression model may also be established by a linear regression method, and the screening of brain network attributes related to the suitability of a particular medical treatment is accomplished according to the linear regression model. For example, the linear regression model may extract the brain network attributes related to the adaptability of the specific medical treatment, with the adaptability of the valid object and the invalid object to the specific medical treatment as the label, and the at least one brain network attribute obtained in step S16 as the influence factor of the adaptability. According to the specific numerical values of the extracted brain network attributes of the effective object and the ineffective object, the threshold corresponding to the extracted brain network attributes can be obtained, so that the effective object and the ineffective object can be distinguished according to the relation between the numerical values of the extracted brain network attributes and the threshold. In this case, the prediction model may be determined according to a threshold value.

In one possible implementation, after obtaining the prediction model, the prediction model may be used to accomplish prediction of the suitability of the target subject for a particular medical treatment.

For example, the target object may be an object that has not received a particular medical treatment. The method of step S11-15 above may be performed by first obtaining artifact-free inter-episode-electroencephalogram data of the target object, preprocessing the artifact-free inter-episode-electroencephalogram data to obtain processed electroencephalogram data, obtaining a synchronicity matrix according to the processed electroencephalogram data, obtaining a threshold correlation matrix according to the synchronicity matrix and a threshold, and obtaining at least one brain network attribute according to the threshold correlation matrix. In this way, brain network attributes of the target object may be derived. The at least one brain network attribute obtained according to the threshold incidence matrix at least comprises a brain network attribute indicated by the optimal feature vector input by the prediction model, and after the brain network attribute is obtained, the plurality of brain network attributes can be freely combined to obtain the feature vector. In this case, the feature vector is used as input data of the prediction model, and the prediction model can output a prediction result of the target object. The prediction result may be used to determine the suitability of the target subject for a particular medical treatment, such as a vagal nerve stimulation procedure.

The prediction model described above is a two-class prediction model capable of obtaining two kinds of prediction results of a responder or a non-responder by classification, and it will be understood by those skilled in the art that instead of the two-class prediction model, a multi-class prediction model may be obtained by training, and the multi-class prediction model is used to process the feature vector and output the prediction result of the target object. The present application does not limit the specific form of the prediction model used to obtain the prediction result of the target object.

Fig. 14 shows an exemplary schematic diagram of a brain electrical data processing method according to an embodiment of the present application. As shown in fig. 14, an embodiment of the present application provides an electroencephalogram data processing method, where the method includes:

and S31, acquiring electroencephalogram data of the target object. The target object may be an adaptive prediction object of a specific medical treatment such as a vagus nerve stimulation operation, and an exemplary manner of acquiring electroencephalogram data of the target object may be as described above with reference to fig. 1.

S32, determining a first feature vector according to the electroencephalogram data of the target object, wherein the first feature vector comprises a plurality of brain network attributes determined by the electroencephalogram data of the target object. Exemplary implementations of step S32 may refer to the above description of when the prediction model is used to accomplish prediction of the suitability of the target object for a particular medical treatment after the prediction model is derived. The brain network attribute may be an index that is determined according to the brain network of the target object and can represent the brain network state of the target object, for example, the brain network attributes mentioned in step S15, such as average synchronicity, characteristic path length, clustering coefficient, betweenness centrality, global efficiency, and local efficiency, may also be other indexes that indicate the brain network state and can be obtained by processing electroencephalogram data according to the prior art. The first feature vector may be a feature vector obtained by freely combining a plurality of brain network attributes in the description of the corresponding part above.

S33, inputting the first feature vector into a trained prediction model to obtain a prediction result, wherein the prediction result represents whether the target object is suitable for receiving a specific medical treatment. The trained prediction model may be the prediction model obtained in step S16 above. The specific medical treatment may be a medical treatment means such as the above vagus nerve stimulation surgery. Exemplary implementations of step S33 may refer to the above description of when the prediction model is used to accomplish prediction of the suitability of the target object for a particular medical treatment after the prediction model is derived.

In one possible implementation, the target subject's brain electrical data includes artifact-free inter-episode brain electrical data of the target subject,

determining a first feature vector according to the electroencephalogram data of the target object, wherein the determining comprises:

determining a plurality of attribute types as inputs to the trained predictive model; determining the plurality of brain network attributes corresponding to the plurality of attribute types from artifact-free inter-episode brain electrical data of the target subject; and obtaining the first feature vector according to the plurality of brain network attributes.

In the above description, the exemplary determination manner of the plurality of attribute types input by the trained prediction model and the exemplary determination manner of the first feature vector may refer to the relevant description when the prediction model is used to complete the prediction of the adaptability of the target object to the specific medical treatment after the prediction model is obtained.

By the method, the attribute type in the first feature vector is consistent with the input attribute types corresponding to the trained prediction model, so that the accuracy of the prediction result can be ensured when the first feature vector is input into the prediction model for prediction.

In a possible implementation manner, the prediction model is obtained by training electroencephalogram data of a sample object, and fig. 15 shows an exemplary schematic diagram of the prediction model obtained by training according to the embodiment of the present application.

As shown in fig. 15, the method further includes:

s41, determining a plurality of brain network attributes of the sample object according to the electroencephalogram data of the sample object. An exemplary implementation thereof may refer to the description above regarding steps S11-S15.

And S42, according to the effect of the sample object after the specific medical treatment, determining the plurality of brain network attributes of the sample object with the effect meeting the preset condition as positive samples, and determining the plurality of brain network attributes of the sample object with the effect not meeting the preset condition as negative samples. An example of a preset condition may be that the postoperative seizure frequency described above is reduced compared to that before the operation. An exemplary implementation of step S42 may be referred to above in connection with step S121.

And S43, performing difference significance analysis on the positive sample and the negative sample to obtain brain network attributes with significant differences, wherein the difference between the brain network attributes with significant differences of the positive sample and the negative sample is larger than a threshold value. An exemplary implementation thereof may refer to the related description of step S122 above.

And S44, performing importance ranking on the brain network attributes with the significant differences to obtain a second feature vector. The second feature vector may be, for example, a result of sorting the brain network attributes in the above related description of step S123. An exemplary implementation of step S44 may be found in relation to step S123 above.

And S45, training to obtain the prediction model according to the brain network attributes with the same attribute type as the attributes in the second feature vector in the plurality of brain network attributes of the sample object. An exemplary implementation thereof may refer to the related description of step S124 above.

In this way, a well-trained predictive model can be obtained. The prediction model is obtained by training according to a plurality of brain network attributes, the brain network attributes can represent the brain network state of the target object, and the brain network state can reflect the individual difference of susceptibility to a certain extent, so that the prediction result obtained by the prediction model is the prediction result of the susceptibility of different target objects to external stimuli, and the accuracy of the prediction result obtained by the prediction model is higher.

In one possible implementation, the importance ranking of the brain network attributes with significant differences to obtain a second feature vector includes:

obtaining a feature subset according to all brain network attributes with significant differences, and determining the prediction accuracy of a prediction model to the feature subset; judging whether the current feature subset is an empty set; and when the current feature subset is an empty set, according to the importance scores of the brain network attributes in the feature subset with the highest prediction accuracy, performing importance ranking on the brain network attributes in the feature subset with the highest prediction accuracy to obtain a second feature vector.

The feature subset obtained from all brain network attributes with significant differences may be, for example, the initial feature subset in step S21 above. An exemplary manner of obtaining the second feature vector may refer to the above description of steps S21, S25.

In this way, the attribute type of the second feature vector can be made to correspond to the attribute type in the feature subset with the highest accuracy, so that the prediction accuracy of the prediction model is higher.

In a possible implementation manner, the importance ranking is performed on the brain network attributes with significant differences to obtain a second feature vector, and the method further includes:

when the current feature subset is not an empty set, repeating the following operations:

inputting the current feature subset into a preset random forest model to obtain the importance score of each brain network attribute in the current feature subset; eliminating at least one brain network attribute with the lowest importance score in the current feature subset, taking the collection of the brain network attributes left in the current feature subset as a new current feature subset, and determining the prediction accuracy of the prediction model on the current feature subset; and judging whether the current feature subset is an empty set again.

Wherein the current feature subset may be, for example, the initial feature subset or the new feature subset described above. When the current feature subset is not an empty set, an exemplary implementation of the electroencephalogram data processing method may refer to the above description relating to steps S22-S24.

By eliminating the attribute with the lowest importance score every time, the attribute with higher importance score can be reserved in the feature subset after elimination processing, and the prediction accuracy of the feature subset can be ensured on the basis of reducing the dimensionality of the feature subset.

In one possible implementation manner, the training to obtain the prediction model according to the brain network attributes of the plurality of brain network attributes of the sample object, which are of the same type as the attributes in the second feature vector, includes:

obtaining a plurality of third feature vectors by permutation and combination according to different brain network attributes in the second feature vectors, wherein the combinations of the brain network attributes in the different third feature vectors are different, or the combinations of the brain network attributes in the different third feature vectors and the ranks of the brain network attributes in the third feature vectors are different; calculating the prediction accuracy of the prediction model when the brain network attributes with the same attribute types as those in the third feature vectors in the plurality of brain network attributes of the sample object are taken as the input of the prediction model; and determining the input attribute type of the prediction model according to the third feature vector with the highest prediction accuracy to obtain the trained prediction model.

Wherein the second feature vector may be the input feature vector in the correlation description of step S16 above, and the third feature vector may be the processed feature vector in the correlation description of step S16 above. An exemplary implementation of training the predictive model may be as described above in connection with step S16.

And further screening attributes included by the second feature vectors after the importance sorting, and arranging and combining the screening results to obtain a third feature vector with higher accuracy, so that the prediction accuracy of the prediction model is further improved when the prediction model is obtained according to the third feature vector. And the permutation and combination has a plurality of selection modes, thereby improving the flexibility of obtaining the prediction model.

In one possible implementation, the prediction model is trained from electroencephalogram data of a sample object, and the method further includes:

determining a plurality of brain network attributes of the sample object according to the electroencephalogram data of the sample object; taking the effect of the sample object after the specific medical treatment as a mark, taking a plurality of brain network attributes of the sample object as influence factors, inputting the influence factors into a preset linear regression model, and screening and outputting at least one brain network attribute from the input brain network attributes by the linear regression model; and training to obtain the prediction model according to the at least one brain network attribute.

In this way, the flexibility of the training mode of the prediction model can be improved.

In one possible implementation, the sample object's brain electrical data includes artifact-free inter-episode brain electrical data of the sample object,

determining a plurality of brain network attributes of the sample object from the electroencephalogram data of the sample object, including:

obtaining preprocessed electroencephalogram data of the sample object according to the electroencephalogram data of the sample object, wherein the preprocessing comprises at least one of filtering, denoising, re-referencing and segmenting; determining indexes of the sample object according to the preprocessed electroencephalogram data of the sample object, wherein the indexes of the sample object indicate the communication relation among all channels of electroencephalogram signals of the sample object, and the indexes of the sample object comprise at least one of coherence, a phase locking value, a phase delay index, a weighted phase delay index, a Glan's cause and effect and synchronous likelihood; obtaining at least one synchronicity matrix of the sample object according to the indexes of the sample object, wherein each synchronicity matrix corresponds to one index; obtaining a threshold correlation matrix of the sample object according to the synchronicity matrix and a preset threshold condition; establishing a brain network of the sample object according to the threshold incidence matrix; determining a plurality of brain network attributes of the sample object from the brain network of the sample object, the plurality of brain network attributes of the sample object including one or more of average synchronicity, clustering coefficient, characteristic path length, betweenness centrality, global efficiency, and local efficiency.

By preprocessing the electroencephalogram data, the accuracy of the electroencephalogram data can be improved, the accuracy of the index of the sample object obtained according to the preprocessed electroencephalogram data is higher, and the accuracy of establishing a brain network according to the index of the sample object and obtaining the brain network attribute of the sample object according to the established brain network is further higher. The preprocessing mode, the index type of the sample object and the brain network attributes are selected in various ways, so that the flexibility of the mode of determining the brain network attributes of the sample object according to the electroencephalogram data of the sample object can be improved.

As shown in fig. 16, the present application provides an electroencephalogram data processing apparatus including:

the acquisition module 100 is used for acquiring electroencephalogram data of a target object;

a determining module 200, configured to determine a first feature vector according to the electroencephalogram data of the target object, where the first feature vector includes a plurality of brain network attributes determined by the electroencephalogram data of the target object;

a prediction module 300, configured to input the first feature vector into a trained prediction model to obtain a prediction result, where the prediction result indicates whether the target object is suitable for receiving a specific medical treatment.

determining a plurality of brain network attributes of the sample object according to the electroencephalogram data of the sample object; according to the effect of the sample object after the specific medical treatment, determining a plurality of brain network attributes of the sample object of which the effect meets the preset condition as positive samples, and determining a plurality of brain network attributes of the sample object of which the effect does not meet the preset condition as negative samples; performing difference significance analysis on the positive sample and the negative sample to obtain brain network attributes with significance difference, wherein the difference between the brain network attributes with significance difference of the positive sample and the negative sample is larger than a threshold value; sorting the importance of the brain network attributes with the significant difference to obtain a second feature vector; and training to obtain the prediction model according to the brain network attributes with the same attribute type as the attributes in the second feature vector in the plurality of brain network attributes of the sample object.

In one possible implementation, the sample object's brain electrical data includes artifact-free inter-episode brain electrical data of the sample object, and determining a plurality of brain network attributes of the sample object from the sample object's brain electrical data includes:

In one possible implementation, the present application provides an electroencephalogram data processing apparatus, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: and calling the instruction stored in the memory to execute the electroencephalogram data processing method.

In one possible implementation, the present application provides a non-transitory computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the electroencephalogram data processing method described above.

Fig. 17 illustrates an exemplary block diagram of an apparatus 800 in accordance with an embodiment of the present application. The apparatus 800 may be an electrocardiographic data processing apparatus as described above, for example, the apparatus 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and the like.

Referring to fig. 17, the apparatus 800 may include one or more of the following components: processing component 802, memory 804, power component 806, multimedia component 808, audio component 810, input/output (I/O) interface 812, sensor component 814, and communication component 816.

The processing component 802 generally controls overall operation of the device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing components 802 may include one or more processors 820 to execute instructions to perform all or a portion of the steps of the methods described above. Further, the processing component 802 can include one or more modules that facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate interaction between the multimedia component 808 and the processing component 802.

The memory 804 is configured to store various types of data to support operations at the apparatus 800. Examples of such data include instructions for any application or method operating on device 800, contact data, phonebook data, messages, pictures, videos, and so forth. The memory 804 may be implemented by any type or combination of volatile or non-volatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

Power components 806 provide power to the various components of device 800. The power components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the apparatus 800.

The multimedia component 808 includes a screen that provides an output interface between the device 800 and a user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the device 800 is in an operating mode, such as a shooting mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component 810 is configured to output and/or input audio signals. For example, the audio component 810 includes a Microphone (MIC) configured to receive external audio signals when the apparatus 800 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may further be stored in the memory 804 or transmitted via the communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

The I/O interface 812 provides an interface between the processing component 802 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor assembly 814 includes one or more sensors for providing various aspects of state assessment for the device 800. For example, the sensor assembly 814 may detect the open/closed status of the device 800, the relative positioning of components, such as a display and keypad of the device 800, the sensor assembly 814 may also detect a change in the position of the device 800 or a component of the device 800, the presence or absence of user contact with the device 800, the orientation or acceleration/deceleration of the device 800, and a change in the temperature of the device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 816 is configured to facilitate communications between the apparatus 800 and other devices in a wired or wireless manner. The device 800 may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the apparatus 800 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 804, is also provided that includes computer program instructions executable by the processor 820 of the device 800 to perform the above-described methods.

Fig. 18 illustrates an exemplary block diagram of an apparatus 1900 according to an embodiment of the application. Where the device 1900 may be a brain electrical data processing device as described above, for example, the device 1900 may be provided as a server. Referring to FIG. 18, the device 1900 includes a processing component 1922 further including one or more processors and memory resources, represented by memory 1932, for storing instructions, e.g., applications, executable by the processing component 1922. The application programs stored in memory 1932 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1922 is configured to execute instructions to perform the above-described method.

The device 1900 may also include a power component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input/output (I/O) interface 1958. The device 1900 may operate based on an operating system stored in memory 1932, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.

In an exemplary embodiment, a non-transitory computer readable storage medium, such as the memory 1932, is also provided that includes computer program instructions executable by the processing component 1922 of the apparatus 1900 to perform the above-described methods.

The present disclosure may be systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present disclosure.

The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terms used herein were chosen in order to best explain the principles of the embodiments, the practical application, or technical improvements to the techniques in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for processing electroencephalogram data, the method comprising:

acquiring electroencephalogram data of a target object;

determining a first feature vector from the brain electrical data of the target object, the first feature vector comprising a plurality of brain network attributes determined from the brain electrical data of the target object;

and inputting the first feature vector into a trained prediction model to obtain a prediction result, wherein the prediction result represents whether the target object is suitable for receiving specific medical treatment.

2. The electroencephalographic data processing method of claim 1, wherein the electroencephalographic data of the target object includes artifact-free inter-episode electroencephalographic data of the target object,

determining a plurality of attribute types as inputs to the trained predictive model;

determining the plurality of brain network attributes corresponding to the plurality of attribute types from artifact-free inter-episode brain electrical data of the target subject;

and obtaining the first feature vector according to the plurality of brain network attributes.

3. The electroencephalographic data processing method of claim 2, wherein the predictive model is trained from electroencephalographic data of a sample object, the method further comprising:

determining a plurality of brain network attributes of the sample object according to the electroencephalogram data of the sample object;

according to the effect of the sample object after the specific medical treatment, determining a plurality of brain network attributes of the sample object of which the effect meets the preset condition as positive samples, and determining a plurality of brain network attributes of the sample object of which the effect does not meet the preset condition as negative samples;

performing difference significance analysis on the positive sample and the negative sample to obtain brain network attributes with significance difference, wherein the difference between the brain network attributes with significance difference of the positive sample and the negative sample is larger than a threshold value;

sorting the importance of the brain network attributes with the significant difference to obtain a second feature vector;

and training to obtain the prediction model according to the brain network attributes with the same attribute type as the attributes in the second feature vector in the plurality of brain network attributes of the sample object.

4. The method of claim 3, wherein the ranking of importance of the brain network attributes with significant differences, resulting in a second feature vector, comprises:

obtaining a feature subset according to all brain network attributes with significant differences, and determining the prediction accuracy of a prediction model to the feature subset;

judging whether the current feature subset is an empty set;

and when the current feature subset is an empty set, according to the importance scores of the brain network attributes in the feature subset with the highest prediction accuracy, performing importance ranking on the brain network attributes in the feature subset with the highest prediction accuracy to obtain a second feature vector.

5. The method of claim 4, wherein ranking the brain network attributes with significant differences for importance, resulting in a second feature vector, further comprises:

inputting the current feature subset into a preset random forest model to obtain the importance score of each brain network attribute in the current feature subset;

eliminating at least one brain network attribute with the lowest importance score in the current feature subset, taking the collection of the brain network attributes left in the current feature subset as a new current feature subset, and determining the prediction accuracy of the prediction model on the current feature subset;

and judging whether the current feature subset is an empty set again.

6. The method of claim 3, wherein training the predictive model based on the brain network attributes of the plurality of brain network attributes of the sample object that are of the same type as the attributes in the second feature vector comprises:

obtaining a plurality of third feature vectors by permutation and combination according to different brain network attributes in the second feature vectors, wherein the combinations of the brain network attributes in the different third feature vectors are different, or the combinations of the brain network attributes in the different third feature vectors and the ranks of the brain network attributes in the third feature vectors are different;

calculating the prediction accuracy of the prediction model when the brain network attributes with the same attribute types as those in the third feature vectors in the plurality of brain network attributes of the sample object are taken as the input of the prediction model;

and determining the input attribute type of the prediction model according to the third feature vector with the highest prediction accuracy to obtain the trained prediction model.

7. The electroencephalographic data processing method of claim 2, wherein the predictive model is trained from electroencephalographic data of a sample object, the method further comprising:

taking the effect of the sample object after the specific medical treatment as a mark, taking a plurality of brain network attributes of the sample object as influence factors, inputting the influence factors into a preset linear regression model, and screening and outputting at least one brain network attribute from the input brain network attributes by the linear regression model;

and training to obtain the prediction model according to the at least one brain network attribute.

8. The method of any one of claims 3-7, wherein the sample subject's electroencephalographic data includes artifact-free inter-episode electroencephalographic data of the sample subject,

obtaining preprocessed electroencephalogram data of the sample object according to the electroencephalogram data of the sample object, wherein the preprocessing comprises at least one of filtering, denoising, re-referencing and segmenting;

determining indexes of the sample object according to the preprocessed electroencephalogram data of the sample object, wherein the indexes of the sample object indicate the communication relation among all channels of electroencephalogram signals of the sample object, and the indexes of the sample object comprise at least one of coherence, a phase locking value, a phase delay index, a weighted phase delay index, a Glan's cause and effect and synchronous likelihood;

obtaining at least one synchronicity matrix of the sample object according to the indexes of the sample object, wherein each synchronicity matrix corresponds to one index;

obtaining a threshold correlation matrix of the sample object according to the synchronicity matrix and a preset threshold condition;

establishing a brain network of the sample object according to the threshold incidence matrix;

determining a plurality of brain network attributes of the sample object from the brain network of the sample object, the plurality of brain network attributes of the sample object including one or more of average synchronicity, clustering coefficient, characteristic path length, betweenness centrality, global efficiency, and local efficiency.

9. An electroencephalogram data processing apparatus, characterized in that the apparatus comprises:

the acquisition module is used for acquiring electroencephalogram data of a target object;

a determining module for determining a first feature vector from the brain electrical data of the target object, the first feature vector comprising a plurality of brain network attributes determined from the brain electrical data of the target object;

and the prediction module is used for inputting the first feature vector into a trained prediction model to obtain a prediction result, wherein the prediction result represents whether the target object is suitable for receiving specific medical treatment.

10. An electroencephalogram data processing apparatus, characterized by comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

invoking the memory-stored instructions to perform the method of any of claims 1-8.

11. A non-transitory computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the method of any of claims 1 to 8.