Time-frequency multi-sparse prior champagne method

Abstract

The invention discloses a time-frequency multi-sparse prior champagne method, and belongs to the technical field of neural oscillation activity reconstruction. The method comprises the steps that a sensor is used for collecting and preprocessing original data of cranial nerve activity; dividing the cerebral cortex source space into a plurality of parcel regions; processing the data of the plurality of parcel regions by using a Bayesian sparse prior algorithm, and estimating an optimal source covariance matrix and an optimal noise covariance matrix through a maximum likelihood principle; and based on space prior, introducing a result after Bayesian sparse prior processing under a TFC framework, respectively obtaining an optimal sensor covariance matrix and an optimal source variance matrix of a control section, a task section and a comprehensive section, calculating and comparing changes of source power, and realizing estimation of oscillation activity. The method has the performance superior to that of other reference algorithms, and particularly shows the potential and practicability in neural oscillation activity reconstruction under the conditions of low signal-to-noise ratio and multiple sources.

Description

Technical Field

[0001] The present invention belongs to the technical field of neural oscillation activity reconstruction, and particularly relates to a time-frequency multi-sparse prior champagne method. Background Art

[0002] With the development of neuroscience and brain imaging technologies, non-invasive brain functional imaging technologies such as magnetoencephalography (MEG), optically pumped magnetoencephalography (OPM-MEG), and electroencephalogram (EEG) are being increasingly widely used in fields such as cognitive neuroscience and neurological disease diagnosis. These technologies measure the electromagnetic activities in the brain to help researchers and clinicians analyze and reconstruct the neural oscillation activities of the brain. However, in practical applications, existing brain source reconstruction algorithms face various challenges, especially in the case of multiple sources, strongly correlated sources, and low signal-to-noise ratio conditions, where it is very difficult to accurately reconstruct brain activities.

[0003] In clinical applications, such as neurological diseases like epilepsy, Parkinson's disease, and Alzheimer's disease, abnormal changes in neural oscillation activities can serve as biomarkers for disease diagnosis and monitoring. Therefore, accurately reconstructing and analyzing neural oscillation activities in the time-frequency domain has become crucial. However, due to the presence of background brain activities, neural oscillation signals are usually very weak with an extremely low signal-to-noise ratio. At the same time, the simultaneous activation of multiple strongly correlated sources can cause problems such as energy leakage, false sources, or localization errors in traditional reconstruction algorithms, thus affecting the accuracy of diagnosis.

[0004] Existing brain source reconstruction algorithms, such as beamformers and sparse Bayesian learning (SBL) algorithms, have limitations in dealing with these problems. Traditional beamforming algorithms rely on a preset forward model and are sensitive to noise. Especially when faced with multiple strongly correlated sources, the reconstructed energy is prone to diffusion, resulting in low reconstruction accuracy. Although sparse Bayesian learning can solve the sparsity problem to a certain extent, it still faces the risks of overfitting and false sources in low signal-to-noise ratio or multi-source cases. Therefore, how to achieve high-precision neural oscillation activity reconstruction in low signal-to-noise ratio and complex source environments has become a key technical problem. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a time-frequency multi-sparse prior champagne method, which mainly includes pre-segmenting the cerebral cortex source space into multiple parcel regions; introducing a spatial prior to improve the accuracy of source reconstruction; introducing a Bayesian sparse prior hypothesis, and using a hierarchical Bayesian model to estimate the source covariance matrix and the noise covariance matrix through the maximum likelihood principle; using the TFC framework to achieve the estimation of oscillation activities; using the Bayesian sparse prior algorithm to obtain the sensor coherence matrices and source variances of the control segment, task segment, and comprehensive segment respectively; and superimposing the coherence matrices of different trials through Bayesian network learning to obtain the reconstruction results of oscillation activities at different times and frequencies.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A time-frequency multi-sparse prior champagne method, the method comprising:

[0008] Step S1: Using a sensor to collect raw data of cranial nerve activities and perform preprocessing;

[0009] Step S2: Dividing the cerebral cortex source space into multiple parcel regions;

[0010] Step S3: Using the Bayesian sparse prior algorithm to process the data of the multiple parcel regions, and estimating the optimal source covariance matrix and the optimal noise covariance matrix through the principle of maximum likelihood;

[0011] Step S4: Based on the spatial prior, introducing the result of the Bayesian sparse prior processing under the TFC framework, respectively obtaining the optimal sensor covariance matrix and the optimal source covariance matrix of the control segment, the task segment and the comprehensive segment, calculating and comparing the changes in source power, and realizing the estimation of oscillatory activities.

[0012] On the other hand, the present invention provides a time-frequency multi-sparse prior champagne device, comprising:

[0013] A preprocessing module, configured to use a sensor to collect raw data of cranial nerve activities and perform preprocessing;

[0014] A segmentation module, configured to divide the cerebral cortex source space into multiple parcel regions;

[0015] An estimation module, configured to use the Bayesian sparse prior algorithm to process the data of the multiple parcel regions, and estimate the optimal source covariance matrix and the optimal noise covariance matrix through the principle of maximum likelihood;

[0016] A calculation module, configured to, based on the spatial prior, introduce the result of the Bayesian sparse prior processing under the TFC framework, respectively obtain the optimal sensor covariance matrix and the optimal source covariance matrix of the control segment, the task segment and the comprehensive segment, calculate and compare the changes in source power, and realize the estimation of oscillatory activities.

[0017] In a third aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the foregoing time-frequency multi-sparse prior champagne method.

[0018] Fourthly, the present invention provides a computer-readable storage medium, on which executable instructions are stored. When the instructions are executed by a processor, the processor can implement the foregoing time-frequency multi-sparse prior champagne method.

[0019] The beneficial effects of the present invention are as follows:

[0020] The time-frequency multi-sparse prior champagne algorithm of the present invention shows significant advantages in the reconstruction of neural oscillation activities, especially in the multi-source, strongly correlated source, and low signal-to-noise ratio environments; by introducing source space pre-segmentation (parcel segmentation) and sparse prior assumptions, combined with the adaptive learning of the source and noise covariance matrices in the Bayesian framework, it can effectively reduce the mutual interference between multiple strongly correlated sources and significantly improve the accuracy of source reconstruction. This method can still ensure the concentration of the reconstructed energy in the multi-source case and avoid the problem of energy leakage. Through diagonal heteroscedastic noise learning and multi-modal prior information fusion (such as the joint use of MEG and fMRI data), it can still ensure high-quality source reconstruction under extremely low signal-to-noise ratio conditions, and has strong robustness and anti-noise ability. Under the TFC framework, combined with the time-frequency beamformer (TFBF) for source activity reconstruction in the time-frequency domain, it can accurately capture the oscillation activities of different frequency bands of the brain, especially the estimation of instantaneous power changes is more accurate, avoiding the centering error brought by traditional methods. Description of the Drawings

[0021] Figure 1 It is a flowchart of a time-frequency multi-sparse prior champagne method of the present invention;

[0022] Figure 2 It is a flowchart of the DDP algorithm for oscillation sources;

[0023] Figure 3 It is a performance comparison diagram of the time-frequency multi-sparse prior champagne method and the benchmark algorithm under different signal-to-noise ratios and source numbers. Detailed Embodiments

[0024] The present invention will be further described below with reference to the drawings and embodiments.

[0025] In the following description, the terms "first" and "second" are only for the purpose of description and cannot be construed as indicating or implying relative importance. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0026] To address the problem of achieving high-precision reconstruction of neural oscillation activities in low signal-to-noise ratio and complex source environments, some researchers have proposed the coherent source dynamic imaging (DICS) technique. Some researchers have also reduced the possibility of generating false sources by calculating the average covariance matrices of the task time period and the control time period and introducing the pseudo-statistic F-ratio to compare the rhythm variation differences. Some researchers have proposed the TFC framework by combining the SBL-BF method and the TFBF method, which learns the coherence matrix through a multi-layer Bayesian structure and improves the robustness of the algorithm. However, the above methods will have energy leakage and false sources in low signal-to-noise ratio or strong correlated sources, and the imaging effect is not good under low signal-to-noise ratio and multiple correlated source activities. Combining the work done by previous researchers, the present invention discloses a time-frequency multi-sparse prior champagne method, which can efficiently and accurately reconstruct the neural oscillation activities of the brain in the time-frequency domain, and is particularly applicable to complex environments with multiple sources, strong correlated sources and low signal-to-noise ratio. The process is as Figure 1 shown. By fusing multi-modal prior information, such as MEG, OPM-MEG and fMRI data, the algorithm significantly improves the spatial resolution and temporal accuracy of neural oscillation imaging, and can provide a more reliable tool for neuroscience research and clinical applications. Specifically, it includes the following steps:

[0027] Step S1: Use sensors to collect brain nerve activity data, and preprocess the collected raw data, such as filtering, artifact removal and time locking, etc. to reduce noise and enhance signal quality.

[0028] Step S2: Divide the cerebral cortex source space into multiple parcel regions. Based on prior models or data-driven methods, use the point cloud growth method and the adjacency relationship of cortical vertices to generate parcel regions. Through parcel segmentation, the model parameters to be estimated are reduced, and spatial priors are introduced to improve the accuracy of source reconstruction.

[0029] In particular, use the pre-segmentation method for non-phase-locked oscillation data - the oscillatory source DDP algorithm to divide the cerebral cortex source space into multiple parcel regions, as Figure 2 shown. Assume that multiple non-phase-locked oscillation trials are collected, and multiple band-pass filters are designed according to the frequency bands of interest, and each trial is filtered. Considering that the design of the band-pass filter may cause phase distortion, a 200th-order FIR filter is used for filtering, and the edge effect and the time delay caused by the FIR filter are alleviated by zero-padding extension and bidirectional filtering. Use the MSP method to calculate the activation probability after filtering, and take the average score of each dipole, and then use the point cloud growth method and the adjacency relationship of cortical vertices to generate multiple parcel regions.

[0030] For the multiple parcel regions, at different times Under the condition that the prior distribution of the source is defined as an independent Gaussian distribution, represents the total number of time instants, where the source prior covariance , represents the number of parcel regions, represents the th parcel sub-covariance matrix, represents the th weight corresponding to the th parcel region; where

[0031] Step S3. Use the Bayesian sparse prior algorithm to describe the optimization problem as:

[0032] ,

[0033] where, is the sensor covariance, is the noise covariance matrix, represents the conditional probability distribution of and under the condition of ;

[0034] The objective function is expressed as:

[0035] ,

[0036] In the formula, the sensor covariance estimate , represents the matrix trace operation, represents the lead field matrix, and the sensor prior covariance with the superscript T representing the transpose of the matrix, ensures the sparsity of the model structure, reflects the similarity measure between the sensor covariance estimate value and the model learning value.

[0037] For ease of calculation, define the auxiliary matrix , represents the matrix cholesky decomposition, represents the matrix rearrangement encoding, represents the non-singular covariance sub-matrix generated by the index corresponding to the th parcel sub-covariance matrix , represents the th dimension of the parcel sub-covariance matrix;

[0038] The parcel regions obtained by different modalities or data-driven methods are denoted by symbols as , denotes the set of dipole indices constituted by the th parcel region;

[0039] Extract the submatrix corresponding to the set of dipole indices from , where . Vectorize the elements of the submatrix column-wise into a vector:

[0040] ,

[0041] The superscript T represents the transpose of the matrix;

[0042] Generate the row indices and column indices required to fill the matrix according to the indices of the set of dipole indices :

[0043] ,

[0044] where denotes the all-ones vector. Fill each element of into the matrix . , and obtain the prior integration matrix by combining the matrices generated by different parcels;

[0045] Alternately update the source prior covariance and the noise covariance matrix for parameter optimization, including: First, update the source prior covariance , fixing :

[0046] ,

[0047] ,

[0048] In the formula, denotes the estimated value of the source at time. The matrix denotes , is the lead field matrix. The superscript of the variable denotes the th updated value. The matrix denotes the set of dipole indices All column elements of the corresponding lead field matrix ;

[0049] Then update and fix :

[0050] ,

[0051] wherein represents the diagonal element of matrix ; represents the elements of the -th row of the lead field matrix ; represents the -th diagonal element;

[0052] Estimate the brain activity and noise time series by iterating and two equations until the objective loss function converges, and then output the optimal source prior covariance matrix and the optimal noise covariance matrix;

[0053] Step S4: Use spatial priors to improve the accuracy of source reconstruction, and at the same time introduce the results processed by Bayesian sparse priors in the TFC framework to improve the robustness of source reconstruction, and obtain the sensor coherence matrix and source variance of the control segment, task segment and comprehensive segment respectively, and finally realize the estimation of oscillatory activity, calculate and compare the changes in source power.

[0054] Specifically, use the TFC framework to output the original data through a band-pass filter of the target frequency band and segment the period of interest to obtain the control segment data b c (t) and the task segment data b t (t), and construct the comprehensive data b o (t)=[b c (t),b t (t)] to enhance the robustness of the algorithm and solve the zero-division error of the traditional time-frequency beamformer.

[0055] Calculate the output power of the control segment and the task segment, and calculate the pseudo-statistics to mitigate the influence caused by the beamformer centering, and compare the rhythm changes in different time periods. The results are as Figure 3 shown:

[0056] ,

[0057] wherein represents the estimated rhythm power change of the n-th source, , , Describe the expected sensor covariance for multiple trials respectively, where the subscript o corresponds to the comprehensive segment, the subscript t corresponds to the task segment, the subscript c corresponds to the control segment, f refers to a specific frequency, and the optimal source prior covariance matrix and the optimal noise covariance matrix are iterated until the objective function converges to obtain the sensor covariance matrix. Represents the expected variance of multiple trials corresponding to the source. By learning through a Bayesian network, the coherence matrices of different trials are superimposed to improve the accuracy of estimation.

[0058] On the other hand, the present invention provides a time-frequency multi-sparse prior champagne device, and each module included therein can execute the steps of the foregoing method, including:

[0059] A preprocessing module for collecting raw data of cranial nerve activities using sensors and performing preprocessing;

[0060] A segmentation module for dividing the cerebral cortex source space into multiple parcel regions;

[0061] An estimation module for processing the data of the multiple parcel regions using the Bayesian sparse prior algorithm, and estimating the optimal source covariance matrix and the optimal noise covariance matrix through the principle of maximum likelihood;

[0062] A calculation module for introducing the results processed by the Bayesian sparse prior under the TFC framework based on the spatial prior, respectively obtaining the optimal sensor covariance matrix and the optimal source covariance matrix of the control segment, the task segment, and the comprehensive segment, calculating and comparing the changes in source power, and realizing the estimation of oscillatory activities.

[0063] In a third aspect, the present invention provides an electronic device, including: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the foregoing time-frequency multi-sparse prior champagne method.

[0064] In a fourth aspect, the present invention provides a computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by a processor, the processor can implement the foregoing time-frequency multi-sparse prior champagne method.

[0065] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and does not limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A time-frequency multi-sparse prior champagne method, characterized in that: The method comprises: Step S1: using sensors to collect raw data of brain nerve activity and preprocessing; Step S2: dividing the cerebral cortex source space into multiple parcel areas; Step S3: using the Bayesian sparse prior algorithm to process the plurality of parcel area data, and estimating the optimal source covariance matrix and the optimal noise covariance matrix by the maximum likelihood principle; Step S4: Based on the spatial prior, the results of Bayesian sparse prior processing are introduced under the TFC framework to obtain the optimal sensor covariance matrix and the optimal source covariance matrix of the control segment, task segment and comprehensive segment respectively, calculate and compare the changes in source power, and realize the estimation of oscillation activity.

2. The time-frequency multi-sparse prior champagne method according to claim 1, characterized in that: The preprocessing in step S1 includes filtering, artifact removal and time locking.

3. The time-frequency multi-sparse prior champagne method according to claim 1, characterized in that: In the step S2, the cerebral cortex source space is divided into a plurality of parcel areas based on the oscillation source DDP algorithm.

4. The time-frequency multi-sparse prior champagne method according to claim 3, characterized in that: The method for dividing the cerebral cortical source space into multiple parcel areas based on the oscillation source DDP algorithm includes: collecting multiple non-phase-locked oscillation trials, designing multiple bandpass filters according to the frequency bands of interest, filtering each trial, calculating the activation probability after filtering using the MSP method, calculating the average score of each dipole, and generating multiple parcel areas using the point cloud growth method and the adjacency relationship of cortical vertices.

5. The time-frequency multi-sparse prior champagne method according to claim 4, characterized in that: The bandpass filter uses a 200-order FIR filter for filtering, and alleviates edge effects and time delay caused by the FIR filter through zero-filling extension and bidirectional filtering.

6. The time-frequency multi-sparse prior champagne method according to claim 4, characterized in that: For the multiple parcel areas, different times In this case, the prior distribution of the sources is defined as independent Gaussian distributions, , represents the total number of moments, where the source prior covariance , Indicates the number of parcel areas. Indicates parcel sub-covariance matrix, Indicates The weight corresponding to each parcel area.

7. The time-frequency multi-sparse prior champagne method according to claim 6, characterized in that: In step S3, the optimization problem is described as follows using the Bayesian sparse prior algorithm: ， in, is the sensor covariance, is the noise covariance matrix, Indicates that in a given and Under the conditions, The conditional probability distribution of ; Objective Function It is expressed as: ， In the formula, the sensor covariance estimation , represents the matrix trace operation, represents the lead field matrix, sensor prior covariance , the superscript T indicates the transpose of the matrix, This ensures the sparsity of the model structure. Reflects the similarity measure between the sensor covariance estimate and the model learned value; Define the auxiliary matrix , represents the matrix Cholesky decomposition, represents the matrix rearrangement encoding, Indicated by Parcel sub-covariance matrix The non-singular covariance submatrix generated by the corresponding index, Indicates The dimensions of the parcel sub-covariance matrix; The parcel regions obtained by different modal or data-driven methods are represented by symbols: , Indicates A set of dipole indices consisting of parcel regions; from Extract the dipole index from the collection The corresponding sub-matrix ,in , the sub-matrix The elements of are vectorized in column-major fashion into a A vector of: ， Index collection by dipole The indices of generate the row indices needed to fill the matrix and column index : ； in, represents a vector of all 1s, Fill each element of the matrix middle, , by generating matrices with different parcels Merge to get the prior integration matrix ; Alternating update source prior covariance , the noise covariance matrix Perform parameter optimization, including: first update the source prior covariance ,fixed : ， ， In the formula, express The estimated value of the moment source, matrix express , is the lead field matrix, the variable Superscript indicates the The updated value of the matrix Represents a collection of dipole indices Corresponding lead field matrix All column elements of ; Then update ,fixed : ， In the formula, Representation Matrix The diagonal elements of Represents the lead field matrix No. row elements, Indicates diagonal elements; Through iteration and The two equations estimate brain activity and noise time series until the objective loss function converges, and then output the optimal source prior covariance matrix and the optimal noise covariance matrix.

8. A time-frequency multi-sparse prior champagne method according to claim 1, characterized in that: In step S4, the original data is output through the bandpass filter of the target frequency band using the TFC framework and the time period of interest is divided into control segment data of the same length. and task segment data , construct comprehensive data ; Calculate the output power of the control segment, the task segment, and pseudo-statistics, and compare the rhythm changes in different time periods: ， In the formula, represents the rhythmic power variation estimate of the nth source, , , Describe the expected sensor covariance of multiple trials respectively, where subscript o corresponds to the comprehensive segment, subscript t corresponds to the task segment, subscript c corresponds to the control segment, and f refers to a specific frequency. The optimal source prior covariance matrix and the optimal noise covariance matrix are iterated until the objective function converges to obtain the sensor covariance matrix. It represents the expected variance of multiple trials of the corresponding source, and superimposes the coherence matrices of different trials through Bayesian network learning to improve the accuracy of estimation.

9. A time-frequency multi-sparse prior champagne device, characterized in that: include: A preprocessing module, used to collect raw data of brain nerve activity using sensors and perform preprocessing; The segmentation module is used to segment the cerebral cortex source space into multiple parcel areas; An estimation module, used to process the plurality of parcel area data using a Bayesian sparse prior algorithm, and estimate an optimal source covariance matrix and an optimal noise covariance matrix by using a maximum likelihood principle; The calculation module is used to introduce the results of Bayesian sparse prior processing under the TFC framework based on spatial prior, obtain the optimal sensor covariance matrix and the optimal source covariance matrix of the control segment, task segment and comprehensive segment respectively, calculate and compare the changes in source power, and realize the estimation of oscillation activity.

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; Wherein, when one or more programs are executed by the one or more processors, the one or more processors implement the time-frequency multi-sparse prior champagne method described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, which, when executed by a processor, enable the processor to implement a time-frequency multi-sparse prior champagne method as described in any one of claims 1-8.

Images