Biomarker-driven closed-loop brain-controlled rTMS stimulation system

Abstract

The invention provides a closed-loop brain-controlled rTMS stimulation system driven by a biomarker. Relates to the technical field of closed-loop rTMS diagnosis and treatment. According to the invention, a single and double pulse TMS technology is used, interaction mechanisms in hemispheres and between hemispheres are considered, and beta rhythm spontaneous oscillation with biomarker attributes in index finger isotonic expansion movement of a stroke patient is individually detected. Beta rhythm spontaneous oscillation with biomarker attributes of different patients is precisely traced to a specific brain region, the brain region is a target spot of exercise rehabilitation rTMS intervention, and a foundation is laid for personalized and precise intervention of follow-up stroke upper limb exercise rehabilitation.

Description

Technical Field

[0001] The present invention relates to the technical field of closed-loop rTMS diagnosis and treatment, and in particular to a biomarker-driven closed-loop brain-controlled rTMS stimulation system. Background Art

[0002] The closed-loop rTMS system relies on in vitro electrophysiological measurements to characterize disease states. In the past, the physiological indicators monitored included action potentials, local field potentials, and other evoked potentials that depend on external stimuli (Price et al., 2020). However, subsequent studies have shown that diseases such as schizophrenia and stroke are associated with abnormalities in the spontaneous oscillation of the EEG beta rhythm (Briley et al., 2020; Espenhahn et al., 2020). Unlike evoked potentials, spontaneous rhythmic oscillations refer to stable and continuous outputs that can be produced by the subject according to voluntary intentions or movements without external stimulation, and are expected to become the core parameters of the new generation of closed-loop brain control technology (Espenhahn et al., 2020). Movement-Related Beta Desynchronization (MRBD) and Post-Movement Beta Synchronization (PMBS) are two beta rhythm spontaneous oscillations that appear successively during movement. MRBD is related to the excitability of the primary motor cortex (M1), while PMBS is related to the inhibition of M1. Both are modulated by the neural pathway mediated by gamma-aminobutyric acid (GABA) neurons (Zhang et al., 2024; Rossiter et al., 2014). MRBD and PMBS in the M1 region alternately release excitatory and inhibitory signals, jointly modulating the excitability and inhibition of the spinal nerve fiber bundles, thereby achieving precise control of upper limb movements. Previous studies have shown that MRBD and PMBS can be stably induced by index finger extension movements (Pfurtscheller et al., 1996), and abnormal MRBD and PMBS in stroke patients have a crucial impact on upper limb motor control impairment (Rossiter et al., 2014). Therefore, it is highly feasible to use beta rhythm spontaneous oscillations as a biomarker to characterize upper limb motor control impairment in stroke patients.

[0003] However, the use of beta rhythm spontaneous oscillations in closed-loop brain control still faces huge challenges. There are two main reasons. First, the relationship between the brain and motor tasks is complex and diverse, and the modulation relationship between beta rhythm spontaneous oscillations in the sensorimotor cortex and motor behavior parameters is unclear; second, the pathological mechanism of the nervous system is complex, and the regulatory mechanism between sensorimotor cortical lesions and motor control is not clear, which makes the already unclear relationship between beta rhythm spontaneous oscillations and motor control even more vague, and ultimately leads to the application of these electrophysiological parameters in closed-loop brain stimulation. Be limited (Bouthour et al., 2019; Price et al., 2020). If it is impossible to accurately identify the physiological indicators with biomarker properties of each stroke patient, targeted regulation cannot be achieved, and therefore the accuracy of closed-loop rTMS brain control cannot be achieved, which ultimately leads to poor homogeneity of patients' prognostic responses and difficulty in achieving universalization (Bouthour et al., 2019; Price et al., 2020).

[0004] Based on this, the present invention aims to construct a TMS system, combining the single-pulse and double-pulse TMS paradigm to personalize the intra-hemispheric and inter-hemispheric interaction mechanisms during the index finger extension isotonic exercise, accurately identify and trace the beta rhythm spontaneous oscillation with biomarker properties, and lay the foundation for building a closed-loop brain-controlled rTMS stimulation system driven by biomarkers and formulating a rehabilitation program to enhance upper limb motor function after stroke. Summary of the invention

[0005] The purpose of the present invention is to provide a biomarker-driven closed-loop brain-controlled rTMS stimulation system in order to solve the above-mentioned problems.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] According to one aspect of the present invention, a biomarker-driven closed-loop brain-controlled rTMS stimulation system is provided, comprising: a data acquisition module, which collects eligible stroke patients according to inclusion criteria and exclusion annotations, collects functional magnetic resonance imaging (fMRI) data of these patients, and collects EEG and electromyography (EMG) data during the execution of motor tasks;

[0008] The online denoising module performs online denoising on the collected EEG data in conjunction with the online EEG analysis algorithm during the TMS examination, including time domain filtering, online PCA, and online ICA and CSP filtering;

[0009] The real-time update module dynamically monitors biomarkers and movement speed during TMS examinations, and updates the source localization of active brain areas in real time.

[0010] Furthermore, the data acquisition module includes acquiring fMRI data and obtaining voxel points of each patient for tracing analysis of EEG data.

[0011] Furthermore, the data acquisition module includes collecting EMG data of the first dorsal interosseous muscle FDI and the extensor digitorum communis muscle EDC during TMS examination.

[0012] Furthermore, the data acquisition module includes the following steps when performing TMS inspection:

[0013] Step 1: First determine the resting motor threshold (RMT). When at least 5 out of 10 consecutive stimulations induce MEP values ​​not less than 50µV, the minimum TMS panel stimulation intensity that meets this condition is the RMT.

[0014] Step 2: Perform Task 1: When the white circle appears, the patient performs two index finger extension movements at different speeds, namely fast movement and slow movement, and performs an index finger tapping movement during the induced PMBS burst and after the baseline is restored; the index finger tapping movement is only performed when the green circle appears, and the index finger remains still when the red circle appears. During the execution of Task 1, SICI and LICI are measured and calculated as follows:

[0015] (1)

[0016] (2)

[0017] Step 3: The patient performs Task 2: A white circle appears, and the patient performs two index finger extension movements at different speeds, namely fast movement and slow movement. Double pulses are triggered during the PMBS burst and after the baseline is restored. The suprathreshold stimulus CS is delivered to the ipsilateral primary motor cortex IM1, which is the best skull surface location in the M1 area of ​​the right hemisphere; 10ms later, the suprathreshold stimulus TS is triggered to the contralateral primary motor cortex CM1, which is the best skull surface location in the M1 area of ​​the left hemisphere. IHI is measured during Task 2, and the formula is as follows:

[0018] (3)

[0019] Furthermore, the data acquisition module includes synchronously collecting EEG data of 64 electrode leads during TMS examination, the electrode type is wet electrode, the sampling rate is 1KHz, the electrode impedance during the experiment is reduced to below 10KΩ, the reference electrode is bilateral mastoid M1&M2, and the notch filter is set to 50Hz.

[0020] Furthermore, the online denoising module comprises the following steps:

[0021] Step 1: Perform time domain filtering on the synchronously collected EEG data. To exclude artifacts such as TMS discharge artifacts and spike pulses, remove the signal within 80ms after TMS stimulation, use a notch filter to remove power frequency noise, and downsample the EEG data to 500Hz;

[0022] Step 2: Perform online PCA analysis on the synchronously collected EEG data, as shown below:

[0023] (4)

[0024] in, is the data matrix after dimensionality reduction, To represent the EEG raw data into a The matrix of is the projection matrix;

[0025] Step 3: Perform online ICA analysis on the synchronously collected EEG data, as shown below:

[0026] (5)

[0027] in, is the observation signal matrix, is the unknown mixing matrix, is the source signal matrix. The purpose of ICA is to find the source signal matrix ;

[0028] Step 4: Perform online CSP filtering on the synchronously collected EEG data to generate a set of CSP filters, each of which corresponds to a different brain activity pattern, some of which are used for enhancement and some for inhibition;

[0029] Step 5: Use a spatial domain filter to remove eye artifacts and use the pure EEG signal for online analysis.

[0030] The real-time update module is used to dynamically monitor biomarkers and movement speed during rTMS intervention and to update the source localization of active brain areas in real time.

[0031] Furthermore, in the real-time update module, the parsing process of EEG data monitoring biomarkers includes the following steps:

[0032] Step 1: The spectrum estimation of the EEG time series signal S(n) is calculated as follows:

[0033] (6)

[0034] in, is the number of Slepian data windows (also called order), For the Slepian data window;

[0035] Step 2: Spectral energy is concentrated on The time-bandwidth product is:

[0036] (7)

[0037] reflects the degrees of freedom of the variance of the coordinated spectrum estimate, the order of the orthogonal window and degrees of freedom Provides a trade-off between frequency domain resolution and spectral estimation variance;

[0038] Step 3: Usually only A Slepian data window can reduce spectrum leakage, so it is often used orthogonal windows. If the kth characteristic spectrum corresponds to The eigenvalue is, then the minimum sidelobe leakage spectrum estimation result is:

[0039] (8)

[0040] Step 4: The event-related spectrum perturbation is calculated as:

[0041] (9)

[0042] Furthermore, in the real-time update module, the movement speed is obtained according to the EMG data, and the average result of the integration of the horizontal axial movement acceleration output by the MPU-6050 accelerometer sensor is recorded as the movement speed.

[0043] Furthermore, in the real-time update module, the real-time source location analysis includes the following steps:

[0044] Step 1: Perform preprocessing analysis such as slicing, resampling, and correction on fMRI data;

[0045] Step 2: Use the following formula to calculate the volume conduction model to characterize how current propagates through brain tissue;

[0046] (10)

[0047] in, yes The electric potential at for The conductivity at is the Green’s function;

[0048] Step 3: Establish a source model to estimate the source of neural activity from the observed EEG signal using the following formula:

[0049] (11)

[0050] in, For the measured EEG data, is the leading field matrix, is the source distribution, is the regularization parameter;

[0051] Step 4: Use a registration algorithm to align the fMRI-derived model with the EEG electrode locations. The formula is as follows:

[0052] (12)

[0053] Among them is EEG electrode placement, is the corresponding position in the fMRI data, is the transformation matrix;

[0054] Step 5: Calculate the source information through the forward matrix using the weighted minimum norm estimation algorithm (wMNE). There is a linear correlation between the EEG signal and the cortical source signal:

[0055] (13)

[0056] in, is the EEG signal matrix, is the source signal, is the gain matrix (forward solution), Then represents the noise matrix of the measured data;

[0057] Step 6: The source signal can be estimated by the wMNE algorithm: .in,

[0058] (14)

[0059] in, is the covariance matrix of the source signal, is the covariance matrix of the noise signal, is the regularization parameter;

[0060] Through this process, the source analysis of the EEG signal after online artifact removal is achieved, and the EEG information after TMS induction is obtained, so as to find the traced brain area as the target brain area for subsequent intervention.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention uses single and double pulse TMS technology, taking into account the interaction mechanism within and between hemispheres, to detect the beta rhythm spontaneous oscillation with biomarker properties in the index finger isotonic extension exercise of stroke patients in a personalized way. The beta rhythm spontaneous oscillation with biomarker properties of different patients is accurately traced to a specific brain area, which is the target of rTMS intervention for motor rehabilitation, laying the foundation for personalized and precise intervention of upper limb motor rehabilitation after stroke. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flow chart of the biomarker-driven closed-loop brain-controlled rTMS stimulation system provided by the present invention;

[0064] Figure 2 A flowchart of task one provided by the present invention;

[0065] Figure 3 The flowchart of Task 2 provided by the present invention;

[0066] Figure 4 The present invention provides a diagram of 64 sampling electrodes arranged according to the 10-20 international standard lead system.

[0067] Figure 5 This is a flow chart of the TMS synchronized EEG online tracing algorithm based on the weighted minimum norm estimation model provided by the present invention. DETAILED DESCRIPTION

[0068] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0069] The present invention provides a biomarker-driven closed-loop brain-controlled rTMS stimulation system, comprising a data acquisition module, an online denoising module, and a real-time update module.

[0070] The data acquisition module collects eligible stroke patients, collects their functional magnetic resonance imaging (fMRI) data, and collects EEG and electromyogram (EMG) data during motor tasks.

[0071] The inclusion criteria for stroke patients included: ① patients with first stroke on computerized tomography (CT) or magnetic resonance imaging (MRI); ② age range between 18-65 years old; ③ first onset time more than 6 months; ④ mild to moderate functional deficits in hand motor function graded on the Activity Index (AI) scale; ⑤ motor evoked potentials (MEP) can be induced in the first dorsal interosseous muscle (FDI) of the affected hand. ⑥ All recruited subjects had no previous experience in brain-computer interaction or brain-computer interface training. ⑦ Able to understand and follow simple verbal instructions; ⑧ Able to participate in an examination lasting at least 30 minutes; ⑨ Normal vision or corrected normal vision. ⑩ The lateral index finger movement distance independent of the trunk is at least 8 cm.

[0072] Exclusion criteria for stroke patients include: ① hand spasm, hand mirror overflow movement, history of epileptic seizures; ② metal implants, pacemakers and other devices in the brain or body; ③ taking drugs that can change brain excitability; ④ postural hypotension; ⑤ concurrent neurological diseases such as aphasia and dysphagia caused by stroke; ⑥ severe visual impairment; ⑦ severe mental disorders; ⑧ recurrent stroke; ⑨ skull defects that affect the wearing of EEG caps and the placement of TMS handle coils. Those who meet any of the above criteria will be excluded.

[0073] fMRI data were collected to obtain voxel points of each patient for source analysis of EEG data. The fMRI equipment was a Siemens 3T Allegra magnetic resonance scanner. The parameter settings included repetition time (TR) = 2s, echo time (TE) = 15ms, flip angle (FA) = 90 。 , slice thickness 4mm, field of view (FOA) = 240mm × 240mm, number of scanning slices is 33. Head translation exceeds 2mm, rotation exceeds 2 。 The data of the patients will be removed. The fMRI images will be registered to the Montreal MNE standard space template and resampled to 3mm×3mm×3mm.

[0074] The present invention uses two Magstim Rapid2 magnetic stimulators (Magstim, Whitland, Carmarthenshire, UK) produced by British company Magstim, to form single pulse, double pulse, repetitive transcranial magnetic stimulation mode through BiStim module, and can act on corresponding brain area through one or two figure eight coils (coil outer diameter is 90mm). When performing TMS inspection, the EMG data of the first dorsal interosseous muscle (FDI) and the extensor digitorum communis (EDC) are collected. The electromyographic activity of FDI and EDC muscles is recorded using Ag-AgC (diameter 10mm) surface self-adhesive disc electrodes. Electromyographic signal is amplified and filtered (5-2KHz) with a bioamplifier (Neurolog System, Digitimer, UK), and then digitally processed with an analog-to-digital (A / D) converter (CED Micro 1401, Cambridge Electronic Design, UK).

[0075] Determine the resting motor threshold (RMT). The patient's muscles are in a relaxed state, and the TMS panel intensity is adjusted. When at least 5 of the 10 consecutive stimulations induce MEP values ​​of no less than 50µV, the minimum TMS panel stimulation intensity that meets this condition is the resting motor threshold. In each trial experiment when the muscles are in a resting state, the trials contaminated by electromyographic activity are eliminated.

[0076] like Figure 2 As shown, the patient performed task 1. When the white circle appeared, the patient performed two index finger extension movements at different speeds (fast movement and slow movement), and performed an index finger tapping movement during the induced PMBS burst and after the baseline was restored. The index finger tapping movement was only performed when the green circle appeared, and the index finger remained still when the red circle appeared. Combined with paired-pulse transcranial magnetic stimulation (pTMS) technology, double pulses were triggered during the PMBS burst and after the baseline was restored. To evaluate the influence of PMBS on GABA A Short-interval intracortical inhibition (SICI) or GABA BWhat neural pathways regulate long-interval intracortical inhibition (LICI) mediated by leukemia? To prevent muscle fatigue, each condition task was separated by 15 minutes. Each task contained four groups, each containing 21 white circles (fast and slow index finger extension movements), 14 green circles and 7 red circles (green and red appeared randomly, and the index finger tapping task was performed or not). The above four experimental conditions were carried out in sequence without any particular order.

[0077] SICI is measured during the execution of Task 1. In the present invention, the intensity of the conditioned stimulation (CS) is set to 70% of the RMT, and the intensity of the test stimulation (TS) is set to 120% of the RMT. The interval between CS and TS is 2ms. In addition, a single-pulse TMS is used to check whether the TS in the paired pulse stimulation needs to be corrected. Therefore, the protocol of the single-pulse TMS trigger mode is: trigger TS alone in the M1 area of ​​the left hemisphere. In the process of evaluating SICI, it is necessary to ensure that there is no statistical difference in the MEP induced by a single pulse. If there is a significant difference in the single-pulse MEP under different experimental conditions, it is necessary to adjust the intensity of the TS pulse to ensure that the induced MEP amplitude is between 1-2 mV. The calculation method of SICI is as follows:

[0078] (1)

[0079] LICI was measured during the execution of Task 1. In the present invention, the stimulation intensity of CS and TS of LICI was 120%RMT, and the interval time was 100ms. Similarly, it is necessary to use a single pulse triggering protocol to check whether the TS of pTMS needs to be corrected. It must be ensured that the difference in MEP induced by single pulse TMS under different experimental conditions is not statistically different, otherwise the TS intensity needs to be adjusted so that the MEP intensity induced by a single pulse is within 1-2mV. The calculation method of LICI is as follows:

[0080] (2)

[0081] like Figure 3As shown, the patient performed task 2. The white circle appeared, and the patient performed two index finger extension movements at different speeds (fast movement and slow movement). Combined with pTMS technology, two magnetic stimulators were connected to two figure-of-eight coils via the BiStim module. Double pulses were triggered during the PMBS burst and after the baseline was restored. The suprathreshold stimulation CS was delivered to the ipsilateral primary motor cortex (Ipsilateral Motor Cortex, IM1), which is the best cranial surface location of the M1 area in the right hemisphere; 10ms later, the suprathreshold stimulation TS was triggered to the contralateral primary motor cortex (Contralateral Motor Cortex, CM1), which is the best cranial surface location of the M1 area in the left hemisphere. Each condition had four groups, each including 20 trials. Among them, the probability of single and double pulses appearing was 1:1, and the two appeared randomly. In order to prevent muscle fatigue during the experiment, the interval between each condition was 15 minutes.

[0082] The interhemispheric inhibition (IHI) was measured during Task 2. The single pulse trigger mode was to deliver only the suprathreshold stimulus TS to the optimal cranial position of CM1. The stimulation intensity of CS and TS was kept constant, and the TS adjustment range was 110%RMT-120%RMT. It was necessary to ensure that the TS intensity induced MEP values ​​between 1.5-2mV in at least 5 out of 10 consecutive pulses in the right hand FDI muscle. The interhemispheric inhibition IHI of IM1 to CM1 was calculated as follows:

[0083] (3)

[0084] When the stroke patients performed Task 1 and Task 2, EEG data from 64 electrode leads were collected synchronously. Ag / AgCl electrode caps (Quickcap64) produced by Neuroscan were used. The electrode type was wet electrode and the placement was as follows: Figure 4 Neuroscan Synamps2 amplifier was used to amplify the EEG signal, with a sampling rate of 1KHz. During the experiment, the electrode impedance was reduced to below 10KΩ, the reference electrodes were bilateral mastoids (M1 & M2), and the notch filter was set to 50Hz.

[0085] like Figure 5 As shown in FIG, the online denoising module performs online denoising on the collected EEG data in combination with the online EEG analysis algorithm during the rTMS intervention process, including time domain filtering, online PCA, and online ICA and CSP filtering.

[0086] For time domain filtering, in order to exclude artifacts such as TMS discharge artifacts and spike pulses, the signal within 80 ms after TMS stimulation was removed, and a notch filter was used to remove power frequency noise, and the EEG data was downsampled to 500 Hz.

[0087] Online PCA analysis of synchronously collected EEG data includes the following steps:

[0088] Step 1: When collecting EEG signals, there are n samplings, each sampling has m electrodes (in the present invention, the number of electrodes is 64), and the EEG data is represented as a The matrix , the covariance matrix The calculation formula is as follows:

[0089] (4)

[0090] in, is the mean vector of the data.

[0091] Step 2: Use the following formula to perform eigenvalue decomposition on the covariance matrix:

[0092] (5)

[0093] in, is the eigenvector matrix, is a diagonal matrix, and the elements on the diagonal are the eigenvalues.

[0094] Step 3: Select the eigenvectors corresponding to the first k eigenvalues ​​to form the projection matrix , project the data into the principal component space:

[0095] (6)

[0096] in, is the data matrix after dimensionality reduction.

[0097] The basic steps of online ICA are as follows: First, construct an EEG data matrix containing m electrodes , the size is ,in is the number of sampling points. ICA is expressed mathematically as follows:

[0098] (7)

[0099] in, is the observation signal matrix, is the unknown mixing matrix, is the source signal matrix. The purpose of ICA is to find the source signal matrix , specifically including the following steps:

[0100] Step 1: For the data matrix , first use the following formula to decentralize:

[0101] (8)

[0102] in, is the original data matrix The mean of each feature column.

[0103] Step 2: Whiten the data using the following formula:

[0104] (9)

[0105] The covariance matrix is ​​calculated using the following formula Perform eigendecomposition:

[0106] (10)

[0107] in, is the eigenvector matrix, is a diagonal eigenvalue matrix. The whitening transformation can be expressed as follows:

[0108] (11)

[0109] Step 3: Based on maximizing non-Gaussianity, negative entropy is selected as the measure of non-Gaussianity, using a nonlinear function To approximate the maximization of negative entropy and find the projection weight that maximizes negative entropy, its iterative update rule can be expressed by the following formula:

[0110] (12)

[0111] Afterwards, the new weight vector is calculated using the following formula: Normalize:

[0112] (13)

[0113] Step 4: Orthogonalization and convergence, using the Gram-Schmidt process. components, the update rule will take into account the previous Quantity:

[0114] (14)

[0115] right After normalization, check for convergence.

[0116] Step 5: Independent component analysis, find all weight vectors:

[0117] (15)

[0118] The independent components are estimated as follows:

[0119] (16)

[0120] According to the PCA and ICA results, the EMG and other noises are removed.

[0121] CSP filtering includes the following steps:

[0122] Step 1: Create different covariance matrices for different categories of EEG signals. The matrices are as follows:

[0123]

[0124] in, Represents the variance of the factor and is calculated using the following formula:

[0125] (17)

[0126] in, For different types of EEG signals, is the mean vector of the data.

[0127] Represents the covariance of two factors and is calculated using the following formula:

[0128] (18)

[0129] in, For different types of EEG signals, is the mean vector of the data.

[0130] Step 2: Generalized eigenvalue decomposition is used on these covariance matrices. The basic idea is as follows: For a set of matrices , ... and ,in, yes dimensional matrix, yes -dimensional positive definite matrix. The generalized eigenvalue problem is of the form:

[0131] (19)

[0132] in is a non-zero eigenvector, is the generalized eigenvalue. The generalized eigenvalue decomposition in this general form can be expressed as follows:

[0133] (20)

[0134] in, yes dimensional matrix, yes -dimensional matrix whose columns are eigenvectors combination. is a diagonal matrix whose diagonal elements are generalized eigenvalues . Finally, the generalized eigenvalue decomposition will generate a set of eigenvalues ​​and corresponding eigenvectors.

[0135] Step 3: Select the eigenvectors with the largest or smallest eigenvalues ​​to construct filters. Select the eigenvector with the largest eigenvalue to enhance the brain activity of interest, and select the eigenvector with the smallest eigenvalue to suppress other brain activities. Finally, a set of CSP filters is generated, each filter corresponds to a different brain activity pattern, some of which are used for enhancement and some for suppression.

[0136] Spatial domain filters were used to remove eye artifacts and the pure EEG signals were used for online analysis.

[0137] like Figure 5 As shown, the real-time update module is used to dynamically monitor biomarkers and movement speed during rTMS intervention and update the source localization of active brain areas in real time.

[0138] The multi-window spectrum estimation of the Slepian data window is used to perform online analysis on the EEG signal after denoising and artifact removal, and the spectrum estimation of the time series signal S(n) is calculated as follows:

[0139] (twenty one)

[0140] in, is the number of Slepian data windows (also called order), is the kth Slepian data window. The spectrum energy is concentrated between 2W, and the time-bandwidth product is:

[0141] (twenty two)

[0142] reflects the degrees of freedom of the variance of the coordinated spectrum estimate, the order of the orthogonal window and degrees of freedom It provides a trade-off between frequency domain resolution and spectral estimation variance. A Slepian data window can reduce spectrum leakage, so it is often used orthogonal windows. If the kth characteristic spectrum corresponds to The eigenvalue is, then the minimum sidelobe leakage spectrum estimation result is:

[0143] (twenty three)

[0144] The bandpass filter range used in the present invention is 8-35Hz, and the slice data length is 0.25s before the task command signal to 2.5s after the task signal, a total of 2.75s. The data preprocessing process of denoising and removing artifacts is the same as the above method. A third-order Slepian orthogonal window is used to control spectrum leakage, and the frequency smoothing window is , and use the time window to advance in steps of 25ms. The time-frequency representation (TFR) of the power can be obtained by averaging the absolute value of the square of the convolution of all trials in each channel task. For the calculation of TFR, we choose the 250ms before the instruction task as the baseline for correction, and the event-related spectrum perturbation is calculated as:

[0145] (twenty four)

[0146] Beta band event-related power changes are usually intuitively displayed by reducing the dimension of the beta band (14-30Hz) data in TFR to the time domain, including two processes: desynchronization and synchronization. The present invention focuses on the changes near the peaks of MRBD and PMBS for the target brain area. MRBD is defined as the average value within a 200ms window centered on the peak of the event-related power change trough after the start of movement (negative value, showing desynchronization phenomenon), and PMBS is defined as the average value within a 200ms window centered on the peak of the event-related power change after the end of movement (positive value, showing synchronization phenomenon).

[0147] In the process of obtaining the movement speed based on the EMG data, the average result of the integration of the horizontal axial movement acceleration output by the MPU-6050 accelerometer sensor is recorded as the movement speed.

[0148] Real-time source location analysis includes the following steps:

[0149] Step 1: Preprocess and analyze the fMRI data by slicing, resampling, and correction.

[0150] Slicing fMRI data involves dividing the 3D fMRI data into 2D slices, as follows:

[0151] (25)

[0152] in, Represents a 3D coordinate point The activation strength of the voxel at .

[0153] Resampling involves changing the size or number of voxels using interpolation to couple with the EEG data, as follows: (26)

[0154] in, Represents the coordinate voxel activation intensity generated after resampling, Represents the original coordinate voxel activation intensity. is a function that uses a Gaussian kernel to smoothly transform for The contribution is weighted.

[0155] Correction includes motion correction, temporal filtering and spatial normalization.

[0156] When performing motion correction, the subject's movement during scanning is corrected by aligning all images to a reference image using the following formula:

[0157] (27)

[0158] in, is the transformation matrix, which transforms the original volume Pair and reference volume Alignment.

[0159] Temporal filtering involves removing noise and artifacts from a time series using the following formula:

[0160] (28)

[0161] in, represent The voxel activation intensity after filtering at each moment, represent The original voxel activation intensity at time, Representative from Time has come The average voxel activation at time instant.

[0162] The spatial normalization process aligns the brain image to a standard template, using the MNE template, with the following formula:

[0163] (29)

[0164] in, is the normalized voxel activation intensity after alignment with the MNE template, Describing operators to find transformations , so that the difference between the original brain image and the template is minimized, represents the voxel activation intensity in the original brain image, Represents the MNE standard brain template.

[0165] Step 2: Use the following formula to calculate the volume conduction model to characterize how current propagates through brain tissue.

[0166] (30)

[0167] in, yes The electric potential at for The conductivity at is the Green's function.

[0168] Step 3: Establish a source model to estimate the source of neural activity from the observed EEG signal using the following formula:

[0169] (31)

[0170] in, For the measured EEG data, is the leading field matrix, is the source distribution, is the regularization parameter.

[0171] Step 4: Use a registration algorithm to align the fMRI-derived model with the EEG electrode locations. The formula is as follows:

[0172] (32)

[0173] in is the EEG electrode location, is the corresponding position in the fMRI data, is the transformation matrix.

[0174] Step 5: Calculate the source information through the forward matrix using the weighted minimum norm estimation algorithm (wMNE). There is a linear correlation between the EEG signal and the cortical source signal:

[0175] (33)

[0176] in, is the EEG signal matrix, is the source signal, is the gain matrix (forward solution), Then represents the noise matrix of the measured data.

[0177] Step 6: The source signal can be estimated by the wMNE algorithm: .in,

[0178] (34)

[0179] in, is the covariance matrix of the source signal, is the covariance matrix of the noise signal, is the regularization parameter.

[0180] Through this process, the source analysis of the EEG signal after online artifact removal is achieved, and the EEG information after TMS induction is obtained, so as to find the traced brain area as the target brain area for subsequent intervention.

[0181] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0182] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A biomarker-driven closed-loop brain-controlled rTMS stimulation system, characterized by: include: The data acquisition module collects eligible stroke patients according to the inclusion criteria and exclusion annotations, collects functional magnetic resonance imaging (fMRI) data of these patients, and collects EEG and electromyography (EMG) data during the execution of motor tasks; The online denoising module performs online denoising on the collected EEG data in conjunction with the online EEG analysis algorithm during the TMS examination, including time domain filtering, online PCA, and online ICA and CSP filtering; The real-time update module dynamically monitors biomarkers and movement speed during TMS examinations, and updates the source localization of active brain areas in real time.

2. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: The data acquisition module includes acquiring fMRI data and obtaining voxel points of each patient for tracing analysis of EEG data.

3. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: The data acquisition module includes collecting EMG data of the first dorsal interosseous muscle FDI and the extensor digitorum communis muscle EDC during TMS examination.

4. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: The data acquisition module includes the following steps when performing TMS inspection: Step 1: First determine the resting motor threshold (RMT). When at least 5 out of 10 consecutive stimulations induce MEP values ​​not less than 50µV, the minimum TMS panel stimulation intensity that meets this condition is the RMT. Step 2: Perform Task 1: When the white circle appears, the patient performs two index finger extension movements at different speeds, namely fast movement and slow movement, and performs an index finger tapping movement during the induced PMBS burst and after the baseline is restored; the index finger tapping movement is only performed when the green circle appears, and the index finger remains still when the red circle appears. During the execution of Task 1, SICI and LICI are measured and calculated as follows: (1) (2) Step 3: The patient performs Task 2: A white circle appears, and the patient performs two index finger extension movements at different speeds, namely fast movement and slow movement. Double pulses are triggered during the PMBS burst and after the baseline is restored; the suprathreshold stimulus CS is delivered to the ipsilateral primary motor cortex IM1, which is the best skull surface position in the M1 area of ​​the right hemisphere; 10ms later, the suprathreshold stimulus TS is triggered to the contralateral primary motor cortex CM1, which is the best skull surface position in the M1 area of ​​the left hemisphere; IHI is measured during Task 2, and the formula is as follows: (3)。 5. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: The data acquisition module includes synchronously collecting EEG data of 64 electrode leads during TMS examination, the electrode type is wet electrode, the sampling rate is 1KHz, the electrode impedance during the experiment is reduced to below 10KΩ, the reference electrode is bilateral mastoid M1&M2, and the notch filter is set to 50Hz.

6. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: The online denoising module comprises the following steps: Step 1: Perform time domain filtering on the synchronously collected EEG data. To exclude artifacts such as TMS discharge artifacts and spike pulses, remove the signal within 80ms after TMS stimulation, use a notch filter to remove power frequency noise, and downsample the EEG data to 500Hz; Step 2: Perform online PCA analysis on the synchronously collected EEG data, as shown below: (4) in, is the data matrix after dimensionality reduction, To represent the EEG raw data into a The matrix of is the projection matrix; Step 3: Perform online ICA analysis on the synchronously collected EEG data, as shown below: (5) in, is the observation signal matrix, is the unknown mixing matrix, is the source signal matrix. The purpose of ICA is to find the source signal matrix ; Step 4: Perform online CSP filtering on the synchronously collected EEG data to generate a set of CSP filters, each of which corresponds to a different brain activity pattern, some of which are used for enhancement and some for inhibition; Step 5: Use a spatial domain filter to remove eye artifacts and use the pure EEG signal for online analysis; The real-time update module is used to dynamically monitor biomarkers and movement speed during rTMS intervention and to update the source localization of active brain areas in real time.

7. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: In the real-time update module, the analysis process of EEG data monitoring biomarkers includes the following steps: Step 1: The spectrum estimation of the EEG time series signal S(n) is calculated as follows: (6) in, is the number of Slepian data windows (also called order), For the Slepian data window; Step 2: Spectral energy is concentrated on The time-bandwidth product is: (7) reflects the degrees of freedom of the variance of the coordinated spectrum estimate, the order of the orthogonal window and degrees of freedom Provides a trade-off between frequency domain resolution and spectral estimation variance; Step 3: Usually only A Slepian data window can reduce spectrum leakage, so it is often used orthogonal windows, if the kth feature spectrum corresponds to The eigenvalue is, then the minimum sidelobe leakage spectrum estimation result is: (8) Step 4: The event-related spectrum perturbation is calculated as: (9)。 8. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: In the real-time update module, the movement speed is obtained according to the EMG data, and the average result of the integration of the horizontal axial movement acceleration output by the MPU-6050 accelerometer sensor is recorded as the movement speed.

9. The biomarker-driven closed-loop brain-controlled rTMS stimulation system according to claim 1, characterized in that: In the real-time update module, the real-time source location analysis includes the following steps: Step 1: Perform preprocessing analysis such as slicing, resampling, and correction on fMRI data; Step 2: Use the following formula to calculate the volume conduction model to characterize how current propagates through brain tissue; (10) in, yes The electric potential at for The conductivity at is the Green’s function; Step 3: Establish a source model to estimate the source of neural activity from the observed EEG signal using the following formula: (11) in, For the measured EEG data, is the leading field matrix, is the source distribution, is the regularization parameter; Step 4: Use a registration algorithm to align the fMRI-derived model with the EEG electrode locations. The formula is as follows: (12) Among them is EEG electrode placement, is the corresponding position in the fMRI data, is the transformation matrix; Step 5: Calculate the source information through the forward matrix using the weighted minimum norm estimation algorithm (wMNE). There is a linear correlation between the EEG signal and the cortical source signal: (13) in, is the EEG signal matrix, is the source signal, is the gain matrix (forward solution), Then represents the noise matrix of the measured data; Step 6: The source signal can be estimated by the wMNE algorithm: ;in, (14) in, is the covariance matrix of the source signal, is the covariance matrix of the noise signal, is the regularization parameter; Through this process, the source analysis of the EEG signal after online artifact removal is achieved, and the EEG information after TMS induction is obtained, so as to find the traced brain area as the target brain area for subsequent intervention.

Images