阅读说明：本技术 一种基于非侵入式脑电信号的抑郁症生物标记物辨识方法 (Depression biomarker identification method based on non-invasive electroencephalogram signals ) 是由 陈盛博 于 2019-09-12 设计创作，主要内容包括：本发明提供一种基于非侵入式脑电信号的抑郁症生物标记物辨识方法。该方法包括：按照预设数据采集规则获取被试的抑郁症标记物辨识用数据,包括HDRS-17分数和/或IMS分数,以及与HDRS-17分数和/或IMS分数对应的EEG信号；设计k轮交叉检验,在每轮交叉检验中,将抑郁症标记物辨识用数据分为训练集和测试集；根据训练集,建立低维EEG-HDRS-17模型和/或低维EEG-IMS模型,以及高维EEG-低维EEG模型；利用高维EEG-低维EEG模型将测试集的高维EEG特征向量转化为低维EEG特征向量,利用低维EEG-HDRS-17模型和/或低维EEG-IMS模型预测测试集的HDRS-17分数和/或IMS分数；比较预测得到的HDRS-17分数与真实测得的HDRS-17分数,和/或预测得到的IMS分数与真实测得的IMS分数,根据比较结果判断是否辨识出被试的抑郁症生物标记物。(The invention provides a depression biomarker identification method based on non-invasive electroencephalogram signals. The method comprises the following steps: acquiring tested depression marker identification data according to a preset data acquisition rule, wherein the data comprises an HDRS-17 score and/or an IMS score and an EEG signal corresponding to the HDRS-17 score and/or the IMS score; designing k rounds of cross examination, and dividing data for identifying the depression markers into a training set and a testing set in each round of cross examination; establishing a low-dimensional EEG-HDRS-17 model and/or a low-dimensional EEG-IMS model and a high-dimensional EEG-low-dimensional EEG model according to the training set; converting the high-dimensional EEG feature vector of the test set into a low-dimensional EEG feature vector by using a high-dimensional EEG-low-dimensional EEG model, and predicting the HDRS-17 score and/or the IMS score of the test set by using a low-dimensional EEG-HDRS-17 model and/or a low-dimensional EEG-IMS model; and (3) comparing the predicted HDRS-17 score with the truly measured HDRS-17 score and/or the predicted IMS score with the truly measured IMS score, and judging whether the tested depression biomarker is identified according to the comparison result.)

1. A depression biomarker identification method based on non-invasive electroencephalogram signals is characterized by comprising the following steps:

step 1: acquiring tested depression marker identification data according to a preset data acquisition rule, wherein the depression marker identification data comprises an HDRS-17 score and/or an IMS score and an EEG signal corresponding to the HDRS-17 score and/or the IMS score;

step 2: designing k rounds of cross-over tests, and dividing the data for identifying the depression markers into a training set and a testing set in each round of cross-over tests;

and step 3: establishing a low-dimensional EEG-HDRS-17 model between the low-dimensional EEG feature vector and the HDRS-17 score and/or a low-dimensional EEG-IMS model between the low-dimensional EEG feature vector and the IMS score, and a high-dimensional EEG-low-dimensional EEG model between the high-dimensional EEG feature vector and the low-dimensional EEG feature vector according to the training set;

and 4, step 4: transforming the high-dimensional EEG feature vectors of the test set into low-dimensional EEG feature vectors using the high-dimensional EEG-low-dimensional EEG model, and predicting the HDRS-17 scores and/or IMS scores of the test set using the low-dimensional EEG-HDRS-17 model and/or the low-dimensional EEG-IMS model from the low-dimensional EEG feature vectors of the test set;

and 5: and (3) comparing the predicted HDRS-17 score of the test set with the actually-measured HDRS-17 score of the test set, and/or the predicted IMS score of the test set with the actually-measured IMS score of the test set, and judging whether the tested depression biomarker is identified or not according to the comparison result.

2. The method according to claim 1, wherein the preset data collection rule in step 1 is specifically:

performing m groups of EEG records on a tested object in each preset time period, wherein each group of EEG records is divided into h subgroups, and every two adjacent groups of EEG records are separated by d days;

prior to each group of EEG recordings, the subjects were tested to fill out a cognitive behavioral assessment quantification chart comprising the hamilton depression scale according to which the HDRS-17 scores of the subjects were recorded;

wherein, the data acquisition process of each subgroup is as follows:

filling an instant emotion quantization table in a tested object, and recording the IMS score of the tested object according to the instant emotion quantization table;

after being tested to fill out the instant emotion quantization table, 3 minutes of the tested open-eye resting EEG signal and 3 minutes of the tested closed-eye resting EEG signal were collected and then entered the next subgroup.

3. The method according to claim 1 wherein the building of the high-dimensional EEG-low-dimensional EEG model in step 3 comprises:

removing non-neural signals in the EEG signal using frequency domain filtering and an independent component analysis algorithm;

calculating a high dimensional EEG feature vector of the EEG signal after removal of the non-neural signals;

and performing space-time dimensionality reduction on the high-dimensional EEG feature vector, and extracting to obtain a low-dimensional EEG feature vector.

4. The method according to claim 3, wherein said calculating a high dimensional EEG feature vector of said EEG signal after removal of said non-neural signal is in particular:

calculating the power of each EEG channel in each relevant frequency band;

calculating the coherence of each two EEG channels in the relevant frequency band;

calculating the phase amplitude coupling in the relevant frequency band for each two EEG channels;

combining the power, the coherence, and the phase amplitude coupling as the high-dimensional EEG feature vector.

5. The method according to claim 3, wherein the space-time dimensionality reduction of the high-dimensional EEG feature vector to obtain a low-dimensional EEG feature vector is specifically:

obtaining a low-dimensional vector y after linear conversion processing is carried out on the high-dimensional feature vector by PCA_t：

y_t＝Pz_t (1)

Wherein the content of the first and second substances,z_ta high-dimensional feature vector representing the time t,p is represented by the covariance matrix of the high-dimensional eigenvector corresponding to the top n_yThe feature vector of each feature value;

establishing a low-dimensional state space model:

wherein e is_tIs noise; a, K, C are model parameters; x is the number of_tRepresenting the extracted low-dimensional EEG feature vectors,n_x<n_y＜＜n_z。

6. the method according to claim 1, wherein the low dimensional EEG-HDRS-17 model or the low dimensional EEG-IMS model in step 3 is:

s_t＝f(x_t)+ε_t (3)

wherein s is_tHDRS-17 or IMS score, x, at time t_tLow dimensional EEG feature vector, ε, representing time t_tRepresenting the noise, and f represents the fitting function obtained using a machine learning algorithm.

7. The method of claim 1, wherein the step of determining whether the tested depression biomarker is identified in step 5 according to the comparison result is specifically:

calculating a root mean square prediction error for each test set, the root mean square prediction error comprising an HDRS-17 fractional root mean square prediction error and/or an IMS fractional root mean square prediction error;

performing a statistical test to determine whether the tested depression biomarker is identified based on the root mean square prediction errors for the k test sets.

Technical Field

The invention relates to the technical field of electroencephalogram signal analysis, in particular to a depression biomarker identification method based on non-invasive electroencephalogram signals.

Background

Depression (MDD) is one of the most serious mental diseases in the world, the incidence rate of depression in China is as high as 8-9%, and the number of depression increases year by year, and about 30% of depression is refractory depression (TRD). Standard drug therapy is rarely effective in patients with refractory depression. Therefore, current leading-edge treatment studies for refractory depression focus on developing new treatment modalities, such as neurofeedback therapy, to replace or assist standard drug therapy.

One key issue in developing new forms of treatment for depression is the search for and identification of effective biomarkers for depression. Research shows that nervous activities of depression patients are different from healthy people, so the electroencephalogram signal-based depression biomarker identification is becoming the leading research direction in the field of neuroscience. An effective neural marker should be able to: (1) distinguishing depression patients from healthy people; (2) differentiating whether the depression patient is effectively responding to standard drug therapy; (3) predicting a change in the degree of depression over time in a patient with depression; (4) predicting mood swings of a depressed patient over time; (5) can effectively reflect the treatment effect of a novel treatment mode.

At present, the study of biomarkers of electroencephalogram signals for depression is mainly divided into invasive electroencephalogram signals and non-invasive electroencephalogram signals. Non-invasive electroencephalography-based studies have focused on the use of Extracranial Electroencephalograms (EEG). Due to the low signal-to-noise ratio of the extracranial electroencephalogram, the existing standard modeling technology can only identify the biomarker which can distinguish depression patients from healthy people, and the biomarker has limited effectiveness and cannot be directly and effectively applied to treatment of depression. Invasive electroencephalographic signal based studies have emerged in the last two years. These studies employ a better signal quality intracranial electroencephalogram (ECoG) that can effectively predict mood swings over time in depressed patients. Studies have shown that these higher quality biomarkers can even effectively reflect the efficacy of electrical stimulation therapy for depression. However, the operation technology of intracranial electroencephalogram is complex, and the intracranial electroencephalogram is mainly used for monitoring epilepsy in the existing clinical standard and is not widely applied to detection of depression. Also, intracranial electroencephalograms are more costly than extracranial electroencephalograms, and are therefore difficult to spread to a wide range of patients with depression.

Collecting EEG and identifying depression biomarkers can provide a more economical means of depression monitoring and treatment. However, current technologies (e.g., Palmiero. M. and Picchardi. L. front EEG Asymmetry of Mood: A Mini-Review [ J ]. Frontiers in Beiviroral neuroscience.2017.11:224) have not been able to identify effective non-invasive brain electrical signal-based depression biomarkers for the following reasons: (1) the prior art is not effective in removing noise from EEG and extracting the sub-signals that are truly related to the extent of depression; (2) the prior art fails to identify its unique depression biomarker for each patient; (3) the prior art does not acquire changes in the degree of depression of a patient over time while acquiring EEG; (4) the prior art does not collect real-time mood swings of a patient over time while collecting EEG, which are closely related to but different from the degree of depression, because the degree of depression measures the mental state of the patient over a long period of time rather than the real-time mood swings; (5) the prior art cannot predict the change of the depression degree of the depression patient along with time by utilizing EEG, and cannot predict the mood fluctuation of the depression patient along with time; (6) the prior art does not perform a strict cross-check.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a depression biomarker identification method based on non-invasive electroencephalogram signals.

The invention provides a depression biomarker identification method based on non-invasive electroencephalogram signals, which comprises the following steps:

step 1: acquiring tested depression marker identification data according to a preset data acquisition rule, wherein the depression marker identification data comprises an HDRS-17 score and/or an IMS score and an EEG signal corresponding to the HDRS-17 score and/or the IMS score;

step 2: designing k rounds of cross-over tests, and dividing the data for identifying the depression markers into a training set and a testing set in each round of cross-over tests;

and step 3: establishing a low-dimensional EEG-HDRS-17 model between the low-dimensional EEG feature vector and the HDRS-17 score and/or a low-dimensional EEG-IMS model between the low-dimensional EEG feature vector and the IMS score, and a high-dimensional EEG-low-dimensional EEG model between the high-dimensional EEG feature vector and the low-dimensional EEG feature vector according to the training set;

and 4, step 4: transforming the high-dimensional EEG feature vectors of the test set into low-dimensional EEG feature vectors using the high-dimensional EEG-low-dimensional EEG model, and predicting the HDRS-17 scores and/or IMS scores of the test set using the low-dimensional EEG-HDRS-17 model and/or the low-dimensional EEG-IMS model from the low-dimensional EEG feature vectors of the test set;

and 5: and (3) comparing the predicted HDRS-17 score of the test set with the actually-measured HDRS-17 score of the test set, and/or the predicted IMS score of the test set with the actually-measured IMS score of the test set, and judging whether the tested depression biomarker is identified or not according to the comparison result.

Further, the preset data acquisition rule in step 1 specifically includes:

performing m groups of EEG records on a tested object in each preset time period, wherein each group of EEG records is divided into h subgroups, and every two adjacent groups of EEG records are separated by d days;

prior to each group of EEG recordings, the subjects were tested to fill out a cognitive behavioral assessment quantification chart comprising the hamilton depression scale according to which the HDRS-17 scores of the subjects were recorded;

wherein, the data acquisition process of each subgroup is as follows:

filling an instant emotion quantization table in a tested object, and recording the IMS score of the tested object according to the instant emotion quantization table;

after being tested to fill out the instant emotion quantization table, 3 minutes of the tested open-eye resting EEG signal and 3 minutes of the tested closed-eye resting EEG signal were collected and then entered the next subgroup.

Further, the process of establishing the high-dimensional EEG-low-dimensional EEG model in step 3 comprises:

removing non-neural signals in the EEG signal using frequency domain filtering and an independent component analysis algorithm;

calculating a high dimensional EEG feature vector of the EEG signal after removal of the non-neural signals;

and performing space-time dimensionality reduction on the high-dimensional EEG feature vector, and extracting to obtain a low-dimensional EEG feature vector.

Further, the calculating the high-dimensional EEG feature vector of the EEG signal after removing the non-neural signal is specifically:

calculating the power of each EEG channel in each relevant frequency band;

calculating the coherence of each two EEG channels in the relevant frequency band;

calculating the phase amplitude coupling in the relevant frequency band for each two EEG channels;

combining the power, the coherence, and the phase amplitude coupling as the high-dimensional EEG feature vector.

Further, the performing space-time dimension reduction on the high-dimensional EEG feature vector to obtain a low-dimensional EEG feature vector by extraction specifically includes:

obtaining a low-dimensional vector y after linear conversion processing is carried out on the high-dimensional feature vector by PCA_t：

y_t＝Pz_t (1)

Wherein the content of the first and second substances,z_ta high-dimensional feature vector representing the time t,p is represented by the covariance matrix of the high-dimensional eigenvector corresponding to the top n_yThe feature vector of each feature value;

establishing a low-dimensional state space model:

wherein e is_tIs noise; a, K, C are model parameters; x is the number of_tRepresenting the extracted low-dimensional EEG feature vectors,n_x<n_y＜＜n_z。

further, the low dimensional EEG-HDRS-17 model or the low dimensional EEG-IMS model in step 3 is:

s_t＝f(x_t)+ε_t (3)

wherein s is_tHDRS-17 or IMS score, x, at time t_tLow dimensional EEG feature vector, ε, representing time t_tRepresenting the noise, and f represents the fitting function obtained using a machine learning algorithm.

Further, the step 5 of determining whether the tested depression biomarker is identified according to the comparison result is specifically as follows:

calculating a root mean square prediction error for each test set, the root mean square prediction error comprising an HDRS-17 fractional root mean square prediction error and/or an IMS fractional root mean square prediction error;

performing a statistical test to determine whether the tested depression biomarker is identified based on the root mean square prediction errors for the k test sets.

The invention has the beneficial effects that:

(1) by removing the non-nerve signals in the EEG signals by using frequency domain filtering and an independent component analysis algorithm, the noise of the EEG can be obviously removed, and the sub-signals really related to the depression degree can be extracted, thereby being beneficial to the depression neuromechanism research and the marker identification research of other mental diseases;

(2) by calculating a high-dimensional EEG feature vector of an EEG signal and establishing a high-dimensional EEG-low-dimensional EEG model between the high-dimensional EEG feature vector and a low-dimensional EEG feature vector, a depression marker specific to each test can be identified by the EEG signal and the effectiveness of the depression marker can be strictly tested by cross check, so that the depression marker for precise medical treatment for clinical application is provided;

(3) by designing a data acquisition rule and adopting a brand-new clinical test design, the multichannel extracranial electroencephalogram signals of the patient, the depression state of the patient and real-time mood fluctuation can be acquired simultaneously;

(4) because the HDRS-17 score is used to reflect the degree of depression, the HDRS-17 score of a subject is predicted through the established low-dimensional EEG-HDRS-17 model, so that depression markers which can predict the degree of depression with time can be identified according to the HDRS-17 score, thereby assisting in guiding clinical treatment of depression;

(5) because the IMS score is used for reflecting the mood fluctuation condition, the IMS score of the tested person is predicted through the established low-dimensional EEG-IMS model, so that depression markers which can predict real-time mood changes along with the change of time can be identified according to the IMS score, real-time mood measurement and control are provided for the tested person, and the quality of life of the tested person is improved.

Drawings

Fig. 1 is a schematic flow chart of a method for identifying a biomarker for depression based on a non-invasive electroencephalogram signal according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly described below with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, an embodiment of the present invention provides a method for identifying biomarkers for depression based on non-invasive electroencephalogram signals, including the following steps:

s101: acquiring tested depression marker identification data according to a preset data acquisition rule, wherein the depression marker identification data comprises an HDRS-17 score and/or an IMS score and an EEG signal corresponding to the HDRS-17 score and/or the IMS score;

s102: designing k rounds of cross-over tests, and dividing the data for identifying the depression markers into a training set and a testing set in each round of cross-over tests;

s103: establishing a low-dimensional EEG-HDRS-17 model between the low-dimensional EEG feature vector and the HDRS-17 score and/or a low-dimensional EEG-IMS model between the low-dimensional EEG feature vector and the IMS score, and a high-dimensional EEG-low-dimensional EEG model between the high-dimensional EEG feature vector and the low-dimensional EEG feature vector according to the training set;

s104: transforming the high-dimensional EEG feature vectors of the test set into low-dimensional EEG feature vectors using the high-dimensional EEG-low-dimensional EEG model, and predicting the HDRS-17 scores and/or IMS scores of the test set using the low-dimensional EEG-HDRS-17 model and/or the low-dimensional EEG-IMS model from the low-dimensional EEG feature vectors of the test set;

s105: and (3) comparing the predicted HDRS-17 score of the test set with the actually-measured HDRS-17 score of the test set, and/or the predicted IMS score of the test set with the actually-measured IMS score of the test set, and judging whether the tested depression biomarker is identified or not according to the comparison result.

According to the depression biomarker identification method based on the non-invasive electroencephalogram signals, firstly, the EEG signals can be used for identifying the depression marker specific to each tested object, and the effectiveness of the depression marker can be strictly tested by using cross check, so that the depression marker for accurate medical treatment and provided for clinical application is provided; secondly, because the HDRS-17 score is used for reflecting the depression degree, the HDRS-17 score of a tested person is predicted through the established low-dimensional EEG-HDRS-17 model, so that depression markers which can predict the depression degree along with the change of time can be identified according to the HDRS-17 score, and the clinical treatment of depression is guided in an auxiliary way; thirdly, because the IMS score is used for reflecting the mood fluctuation condition, the IMS score of the tested person is predicted through the established low-dimensional EEG-IMS model, so that depression markers which can predict real-time mood changes along with the change of time can be identified according to the IMS score, real-time mood measurement and control are provided for the tested person, and the quality of life of the tested person is improved.

On the basis of the foregoing embodiment, the difference between the embodiment of the present invention and the foregoing embodiment is that, in the embodiment of the present invention, the preset data acquisition rule in step S101 specifically includes:

performing m groups of EEG records on a tested object in each preset time period, wherein each group of EEG records is divided into h subgroups, and every two adjacent groups of EEG records are separated by d days; prior to each group of EEG recordings, the subjects were tested to fill out a cognitive behavioral assessment quantification chart comprising the hamilton depression scale according to which the HDRS-17 scores of the subjects were recorded; wherein, the data acquisition process of each subgroup is as follows: filling an instant emotion quantization table in a tested object, and recording the IMS score of the tested object according to the instant emotion quantization table; after being tested to fill out the instant emotion quantization table, 3 minutes of the tested open-eye resting EEG signal and 3 minutes of the tested closed-eye resting EEG signal were collected and then entered the next subgroup.

For example, to ensure that proper and sufficient data is acquired, the acquisition of data may be performed as follows:

firstly, a suitable test is determined, and the selection criteria of the test can be:

(1) the subject age is between 18-50 years;

(2) patients with depression;

the tissue is then tested for the following preparatory tasks:

(1) obtaining written informed consent;

(2) recording the associated concomitant medication;

(3) recording age, sex, drinking history, smoking history and nationality;

(4) recording the history of the previous diseases;

(5) recording the time and course of depression;

(6) performing a Hamilton Depression Scale (HDRS-17) score to detect the extent of Depression;

(7) performing an Immediate Mood Scale (IMS) score to measure real-time mood swings;

(8) recording vital signs, heart rate and blood pressure (blood pressure: sitting blood pressure after resting for 5 minutes);

(9) plasma samples were taken on an empty stomach for laboratory examinations including blood routine and biochemical examinations (liver and kidney function, blood lipids, blood glucose).

And finally, acquiring or recording data: the subjects were subjected to 10-week EEG recordings by the following specific procedure:

(1) weekly performing 2-3 EEG recordings, one hour per experiment, and 2-3 days apart;

(2) HDRS-17 and other related cognitive behavioral assessments were filled in before each group of EEG recordings;

(2) the patient wears an EEG cap and prepares to record multi-channel whole brain EEG signals;

(4) each group of EEG recordings was divided into 6 subgroups, each subgroup lasting 10 minutes. IMS scores were filled out and recorded before each subgroup started (approximately 4 minutes), after which 3 minutes of open-eye resting EEG, 3 minutes of closed-eye resting EEG were recorded, and then the next subgroup was entered.

According to the method for identifying the depression biomarker based on the non-invasive electroencephalogram signals, the data are acquired by designing a brand-new data acquisition rule, the multichannel extracranial electroencephalogram signals of the patient and the depression state and real-time mood fluctuation of the patient can be acquired simultaneously, the sample size and applicability of the data are ensured, and therefore a proper data basis is provided for subsequent data processing.

On the basis of the above embodiments, the present invention provides another method for identifying a depression biomarker based on a non-invasive electroencephalogram signal, where the depression biomarker includes two markers, namely a depression marker capable of predicting a degree of depression varying with time and a depression marker capable of predicting real-time mood variation varying with time, and the identification processes of the two markers are similar, and the method includes the following steps:

s201: acquiring data for identifying the tested depression marker according to a preset data acquisition rule;

for example, following the data acquisition process mentioned above, after the entire acquisition process is complete, there will be approximately 30 HDRS-17 scores and/or 180 IMS scores per subject and corresponding EEG signals.

S202: designing k rounds of cross-over tests, and dividing the data for identifying the depression markers into a training set and a testing set in each round of cross-over tests;

for example, 10 rounds of cross-testing were performed on each patient's IMS and EEG data, each round using 162 IMS and corresponding EEG signals (3 minutes open-eye resting EEG data and 3 minutes closed-eye resting EEG data) as a training set, and the remaining 18 IMS and EEG signals as a test set. The k-th round of examination uses the 18 IMS and EEG data 18k-17 to 18k as the test set, and the remaining data as the training set.

S203: establishing a low-dimensional EEG-IMS model between a low-dimensional EEG feature vector and an IMS score and a high-dimensional EEG-low-dimensional EEG model between a high-dimensional EEG feature vector and a low-dimensional EEG feature vector according to the training set;

specifically, the process of establishing the high-dimensional EEG-low-dimensional EEG model comprises the following steps:

(1): EEG data preprocessing: removing non-neural signals in the EEG signal using frequency domain filtering and an independent component analysis algorithm;

specifically, the non-neural signals include: 50Hz noise, blink noise, muscle movement noise, and the like. The EEG noise can be obviously removed and the true sub-signals related to the depression degree can be extracted by utilizing frequency domain filtering and Independent Component Analysis (ICA) analysis technology, thereby being beneficial to the depression neuromechanism research and the marker identification research of other mental diseases.

(2): calculating a high dimensional EEG feature vector of the EEG signal after removal of the non-neural signals;

specifically, the steps are as follows: calculating the power of each EEG channel in each relevant frequency band using spectral analysis; the relevant frequency bands include: 1-3Hz (delta),3-7Hz (theta),8-12Hz (alpha),12-30Hz (beta),30-50Hz (low gamma) and 50-100Hz (high gamma); calculating the coherence of each two EEG channels in the relevant frequency band; calculating the phase amplitude coupling in the relevant frequency band for each two EEG channels; combining the power, the coherence, and the phase amplitude coupling as the high-dimensional EEG feature vector.

(3): and performing space-time dimensionality reduction on the high-dimensional EEG feature vector, and extracting to obtain a low-dimensional EEG feature vector.

Specifically, a high-dimensional EEG-low-dimensional EEG model is established by extracting low-dimensional EEG feature vectors using Principal Component Analysis (PCA) and low-dimensional state-space modeling (state-space modeling). The method specifically comprises the following steps: firstly, carrying out PCA linear transformation on high-dimensional feature vectors to obtain low-dimensional vectors y_t：

y_t＝Pz_t (1)

Wherein the content of the first and second substances,z_ta high-dimensional feature vector representing the time t,p is represented by the covariance matrix of the high-dimensional eigenvector corresponding to the top n_yThe feature vector of each feature value;

then, establishing a low-dimensional state space model:

wherein e is_tIs noise; a, K, C are model parameters; x is the number of_tRepresenting the extracted low-dimensional EEG feature vectors,n_x<n_y＜＜n_z。

taking the establishment of the low-dimensional EEG-IMS model as an example, the low-dimensional EEG-IMS model can be established by utilizing machine learning algorithms such as regularized linear regression, convolutional neural network and the like. Namely: the low dimensional EEG-IMS model is:

s_t＝f(x_t)+ε_t (3)

wherein s is_tHDRS-17 or IMS score, x, at time t_tLow dimensional EEG feature vector, ε, representing time t_tRepresenting the noise, and f represents the fitting function obtained using a machine learning algorithm (regularized linear regression, convolutional neural networks, etc.).

S204: transforming the high-dimensional EEG feature vectors of the test set into low-dimensional EEG feature vectors using the high-dimensional EEG-low-dimensional EEG model, and predicting the IMS score of the test set using the low-dimensional EEG-IMS model;

specifically, first, EEG preprocessing and high-dimensional EEG feature vector calculation are the same as the training set; then, converting the high-dimensional EEG feature vector into a low-dimensional EEG feature vector by using a high-dimensional EEG-low-dimensional EEG model established on the training set; finally, a predicted IMS score is calculated from the low-dimensional EEG feature vectors using the low-dimensional EEG-IMS model that has been built on the training set.

S205: and comparing the predicted IMS score of the test set with the actually measured IMS score of the test set, and judging whether the tested depression biomarker is identified or not according to the comparison result.

Specifically, first, the IMS score root mean square prediction error of each test set is calculated; then, a statistical test is performed to determine whether the subject depression biomarker is identified based on the root mean square prediction errors for the k test sets.

For example, the predicted IMS scores are compared to the actual measured IMS, the root mean square prediction errors for 18 IMS are calculated, the root mean square prediction errors for all ten test sets are collected, and statistical tests are performed to determine whether a depression marker that can predict real-time mood swings over time is identified.

Similarly, in identifying a depression marker that can predict the degree of depression over time, reference may be made to the above process, which differs from the above process in that: the IMS score needs to be converted to an HDRS-17 score and the corresponding EEG data is all EEG data recorded for each set of EEG (3 minutes/sub-set 6-18 minutes open-eye resting EEG data and 18 minutes closed-eye resting EEG data).

In addition, the embodiment of the invention also provides a method for establishing a causal relationship between the depression marker and depression, and the method can promote the development of novel depression therapy. The method comprises the following specific steps:

10 patients of 100 patients were randomly selected for neurofeedback treatment, and the causal relationship between the depression markers and depression was preliminarily studied. For each patient, neurofeedback treatment was performed using the found depression markers as feedback signals. The electroencephalogram neurofeedback treatment builds on the plasticity of the central nervous system, which usually allows the patient to look at his own depression markers, self-adjust (self-adjustment) in real time, and remodel his own brain wave pattern.

10 patients continued for 4 weeks of neurofeedback treatment. The treatment is performed 3 times per week, with intervals of 2-3 days, for 4 weeks (12 times). Before each treatment, the patient's IMS and HDRS-17 were recorded, with the eye open resting electroencephalogram for 3 minutes and the eye closed resting electroencephalogram for 3 minutes. After each treatment, the IMS and the 3 minute open-eye resting brain electricity and the 3 minute closed-eye resting brain electricity were recorded again.

The neurofeedback treatment was investigated for its ability to slow the extent of depression using standard linear, non-linear correlation analysis and to alter the specific depression markers for each patient accordingly. If neurofeedback treatment is a statistically significant corresponding change in depression markers specific to each patient, there is a causal relationship between the depression markers and depression.

From the above, the present invention can achieve the following objects:

(1) the invention can obviously remove EEG noise and extract the sub-signals really related to depression degree, thereby being beneficial to the depression neuromechanism research and the marker identification research of other mental diseases;

(2) the EEG signal can be used to identify the depression marker specific to each subject and the effectiveness of the depression marker is rigorously tested using a cross-test, thereby providing a depression marker for precision medicine for clinical use;

(3) by designing a data acquisition rule and adopting a brand-new clinical test design, the invention can simultaneously acquire multichannel extracranial electroencephalogram signals of a patient, the depression state of the patient and real-time mood fluctuation;

(4) because the HDRS-17 score is used to reflect the degree of depression, the HDRS-17 score of a subject is predicted through the established low-dimensional EEG-HDRS-17 model, so that depression markers which can predict the degree of depression with time can be identified according to the HDRS-17 score, thereby assisting in guiding clinical treatment of depression;

(5) because the IMS score is used for reflecting the mood fluctuation condition, the IMS score of the tested person is predicted through the established low-dimensional EEG-IMS model, so that depression markers which can predict real-time mood changes along with the change of time can be identified according to the IMS score, real-time mood measurement and control are provided for the tested person, and the quality of life of the tested person is improved.

(6) The causal relationship between depression markers and depression can be examined, thus facilitating the development of new depression therapies.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.