A method and system for selecting parameters of the magnetoencephalogram (MEG) subspace projection algorithm based on data.

Abstract

This invention discloses a method and system for selecting parameters for a magnetic resonance imaging (MRI) subspace projection algorithm based on data. The method includes: S1. Measuring the original multi-channel MRI signal; S2. Setting a range of parameters for subspace projection, sequentially substituting the parameters within the range into the subspace projection to denoise the multi-channel MRI signal, obtaining a denoised multi-channel MRI signal corresponding to each parameter; S3. Constructing a first sub-objective function describing the degree of distortion of the neural signal based on the neural signal of interest, and constructing a second sub-objective function describing the degree of noise suppression based on the signal power of the noise; S4. Constructing a total objective function based on the first and second sub-objective functions, inputting the denoised multi-channel MRI signal into the total objective function, and selecting the optimal denoising parameter based on the function output value. This invention, after setting the parameter range, can automatically find the optimal parameters within the range based on the characteristics of the MRI signal.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of magnetoencephalogram signal denoising processing, and particularly relates to a method and system for selecting parameters of a magnetoencephalogram subspace projection algorithm based on data. BACKGROUND

[0002] Magnetoencephalography is a brain neural imaging technology with high time and spatial resolution. With the development of optically pumped magnetometers in recent years, the wearable optically pumped magnetometer-magnetoencephalography system has better signal sensitivity level while making up for the shortcomings of traditional superconducting quantum interference device-based magnetoencephalography equipment, such as the need for low-temperature operation, fixed probe, and relatively far distance from the human body. Therefore, the wearable optically pumped magnetometer-magnetoencephalography system has broad prospects in brain neuroscience and clinical applications. However, the magnetoencephalography signal is very weak, and even in a room with magnetic shielding, there are inevitably significant environmental noise disturbances stronger than the neural signal in the signal. Therefore, noise suppression algorithms are very important in magnetoencephalography signal processing.

[0003] The subspace projection denoising algorithm based on the subspace projection principle is a very important algorithm in the current magnetoencephalography denoising algorithm. The subspace projection algorithm distinguishes between internal neural signals and external interference signals by projecting the magnetoencephalography signal into the divided internal and external subspace. The external interference subspace is usually defined by singular value decomposition of noise data and selecting the eigenvectors corresponding to the larger singular values. Therefore, the number of singular values, i.e., the noise dimension of the external subspace, is an important parameter of the subspace projection algorithm. If the parameter is too small, the interference subspace cannot contain enough noise, and the noise suppression effect is poor. If the parameter is too large, some internal neural signals will be incorrectly projected into the interference subspace, resulting in signal distortion.

[0004] Currently, for the selection of parameters, most studies observe the changes in the waveform and power spectral density after denoising using different parameters, and rely on experience to select approximately appropriate parameters, which is not objective and time-consuming. Some studies optimize a specific subspace projection algorithm, but these methods are usually complex in principle and require simulation and iterative calculation, and are not suitable for all subspace projection algorithms. Therefore, it is necessary to design an optimal parameter selection method suitable for all subspace projection algorithms and easy to implement. SUMMARY

[0005] The present application aims to solve the problems of the prior art, and provides a method and system for selecting parameters of a brain magnetic subspace projection algorithm based on data, which set signals of specific frequency bands and time periods as neural signals of interest and noise in multi-channel brain magnetic signals, construct a sub-objective function for describing the distortion degree of neural signals and a sub-objective function for describing the noise suppression degree by using signal powers of the neural signals of interest and the noise respectively, and construct a total objective function by using a weighted average and a hyperbolic tangent function, so that parameter values corresponding to a set of signal powers maximizing the objective function are optimal subspace projection algorithm parameters.

[0006] To achieve the above object, the present application provides the following scheme:

[0007] A method for selecting parameters of a brain magnetic subspace projection algorithm based on data, comprising the following steps:

[0008] S1. measuring an original multi-channel brain magnetic signal;

[0009] S2. setting a range of parameters of subspace projection, substituting parameters in the range into the subspace projection in sequence, and denoising the multi-channel brain magnetic signal to obtain a denoised multi-channel brain magnetic signal corresponding to each parameter;

[0010] S3. constructing a first sub-objective function for describing the distortion degree of neural signals based on neural signals of interest, and constructing a second sub-objective function for describing the noise suppression degree based on signal powers of noise;

[0011] S4. constructing a total objective function based on the first sub-objective function and the second sub-objective function, inputting the denoised multi-channel brain magnetic signal into the total objective function, and selecting an optimal denoising parameter based on function output values.

[0012] Preferably, the original multi-channel brain magnetic signal is a multi-channel brain magnetic measurement signal containing brain magnetic evoked neural signals obtained by stimulating a subject using an experimental paradigm, which has not been denoised.

[0013] Preferably, the S2 comprises:

[0014] The range of a parameter x for defining the dimension of an external subspace in the subspace projection is set as [1, n];

[0015] The original multi-channel brain magnetic signal is denoised by using the subspace projection algorithm to obtain n denoised multi-channel brain magnetic signals;

[0016] Wherein, a brain magnetic signal is represented as the sum of an internal neural signal of the brain and external interference noise:

[0017] Y=B+N

[0018]

[0019] wherein Y is the measured total magnetoencephalic signal, B is the brain neural signal, N is the external interference signal, is the denoised total magnetoencephalic signal, is the denoised brain neural signal, is the denoised external interference signal.

[0020] Preferably, the first sub-objective function is the ratio of the neural signal before and after denoising, and the construction method of the first sub-objective function comprises:

[0021] calculating the total power value of the magnetoencephalic signal:

[0022]

[0023] wherein f represents the set frequency range of interest, t represents the set time range of interest, p is the total number of data segments, k represents the kth data segment, y k represents the signal amplitude in the kth data segment;

[0024] calculating the power average of the time segment before the stimulation time point which does not contain obvious neural signal corresponding to the frequency:

[0025]

[0026] wherein m represents the total number of time sampling points of the noise baseline segment, t' represents the t'th sampling point;

[0027] calculating the neural signal in the multi-channel data based on the total power value and the power average:

[0028]

[0029] wherein l represents the number of channels;

[0030] calculating the first sub-objective function based on the neural signal:

[0031]

[0032] wherein, represents the denoised neural signal.

[0033] Preferably, the second sub-objective function is represented as the ratio of the noise before and after denoising, and the construction method of the second sub-objective function comprises:

[0034] calculating the total signal power ratio:

[0035]

[0036] Wherein, T represents total sampling points of brain magnetic signal, y represents amplitude of brain magnetic signal before denoising, y represents amplitude of brain magnetic signal after denoising.

[0037] The second sub-target function is calculated based on the total signal power ratio:

[0038]

[0039]

[0040] Wherein, w i represents channel weight, σ i is standard deviation of each channel signal.

[0041] Preferably, the construction method of the total target function comprises:

[0042] The result F B is normalized to [0, 1];

[0043] The result F N is normalized to [0, 2], and then substituted into the hyperbolic tangent function tanh, so that the result range is reduced to [0, 1];

[0044] The total target function is obtained by weighted summation:

[0045] F all (x)=αF B (x)+(1-α)tanh(F N (x))

[0046] Wherein, x represents parameter of subspace projection algorithm, and alpha represents parameter for balancing two target functions.

[0047] Preferably, the selection method of the optimal denoising parameter comprises: inputting the denoised multi-channel brain magnetic signal into the total target function, so that the parameter making the total target function reach maximum value is the optimal denoising parameter.

[0048] The application also provides a system for selecting parameters of brain magnetic subspace projection algorithm based on data, wherein the parameter selection system applies the parameter selection method in any one of the above aspects, and comprises: a signal collection module, a denoising module, a sub-target function construction module and a parameter selection module.

[0049] The signal collection module is used for measuring original multi-channel brain magnetic signal.

[0050] The denoising module is configured to set a parameter range of the subspace projection, sequentially substitute parameters in the range into the subspace projection, and denoise the multi-channel magnetoencephalogram to obtain a denoised multi-channel magnetoencephalogram corresponding to each parameter.

[0051] The sub-target function construction module is configured to construct a first sub-target function describing a distortion degree of the neural signal based on the neural signal of interest, and construct a second sub-target function describing a noise suppression degree based on a signal power of the noise.

[0052] The parameter selection module is configured to construct a total target function based on the first sub-target function and the second sub-target function, input the denoised multi-channel magnetoencephalogram into the total target function, and select an optimal denoising parameter based on a function output value.

[0053] Compared with the prior art, the present application has the following advantages:

[0054] After setting the parameter range, the present application can automatically find the optimal parameter in the range according to the characteristics of the magnetoencephalogram, and through construction and solution of the target function, the algorithm parameter most meeting the requirements in terms of denoising and neural signal preservation can be accurately found. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed to be used in the embodiments, and obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0056] Figure 1 The method flowchart of the embodiment of the present application is shown in the figure.

[0057] Figure 2 The method flowchart of the embodiment of the present application is shown in the figure.

[0058] Figure 3 The system structure diagram of the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application, and obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0060] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0061] Embodiment one

[0062] In this embodiment, as shown in Figure 1 , Figure 2 A method for selecting parameters of a brain magnetic subspace projection algorithm based on data, comprising the following steps:

[0063] S1. Measure the original multi-channel brain magnetic signal.

[0064] The original multi-channel brain magnetic signal is a multi-channel brain magnetic measurement signal containing brain magnetic evoked neural signals without de-noising processing obtained by stimulating the subject using an experimental paradigm. In this embodiment, the original multi-channel brain magnetic signal is measured using a sensor, which is an optically pumped magnetometer. The optically pumped magnetometer is a very weak magnetic measurement sensor based on the spin exchange relaxation effect.

[0065] S2. Set the range of the parameters of the subspace projection, and substitute the parameters in the range into the subspace projection in turn to de-noise the multi-channel brain magnetic signal, and obtain the de-noised multi-channel brain magnetic signal corresponding to each parameter.

[0066] S2 includes:

[0067] The range of the parameter x used to define the dimension of the external subspace in the subspace projection is set to [1, n];

[0068] The original multi-channel brain magnetic signal is de-noised using the subspace projection algorithm to obtain n de-noised multi-channel brain magnetic signals; wherein the brain magnetic signal is represented as the sum of the internal neural signal of the brain and the external interference noise:

[0069] Y=B+N

[0070]

[0071] Where Y is the measured brain magnetic signal, B is the brain neural signal, N is the external interference signal, is the de-noised measured brain magnetic signal, is the de-noised brain neural signal, is the de-noised external interference signal.

[0072] S3. Construct a first sub-objective function describing the distortion degree of the neural signal based on the neural signal of interest, and construct a second sub-objective function describing the noise suppression degree based on the signal power of the noise.

[0073] The construction method of the first sub-objective function comprises the following steps: in the embodiment, a sub-objective function for describing the distortion degree of the neural signal is first constructed, and the distortion degree of the neural signal can be defined as the ratio of the neural signal before and after denoising:

[0074]

[0075] After the task-state magnetoencephalogram data signal is segmented according to the stimulation cycle, the signal in a specific time period and frequency period in the data segment usually needs to be studied, and the signal is defined as the time period and frequency period signal of interest. Therefore, Y is defined as the total power value of the signal. N is defined as the power average value of the time period (usually referred to as a time baseline) before the stimulation moment of the corresponding frequency, which does not contain obvious neural signals.

[0076] The total power value of the magnetoencephalogram signal is calculated as follows:

[0077]

[0078] Where f represents the set frequency period of interest, t represents the set time period of interest, p is the total number of data segments, k represents the kth data segment, y k represents the signal amplitude in the kth data segment;

[0079] The power average value of the time period before the stimulation moment of the corresponding frequency, which does not contain obvious neural signals, is calculated as follows:

[0080]

[0081] Where m represents the number of time sampling points of the noise baseline segment, and t' represents the t'th sampling point.

[0082] The neural signal in the multi-channel data is calculated based on the total power value and the power average value, and the neural signal in the magnetoencephalogram multi-channel data is defined as the absolute value of the difference between the two:

[0083]

[0084] Where l represents the number of channels.

[0085] The first sub-objective function is calculated based on the neural signal:

[0086]

[0087] Where, represents the neural signal after denoising. The F B is saved as [F B (1), F B (2), …, F B (n)].

[0088] The construction method of the second sub-objective function includes: the second sub-objective function is expressed as the proportion of the noise before and after denoising, and the order of magnitude of the total signal is about several ten thousand fT, and is as high as several thousand fT after band-pass filtering, while the induced brain magnetic signal is generally only about 500 fT, and the noise is much larger than the brain magnetic signal, so the proportion of the noise is directly simplified as the ratio of the total signal power, and the total signal power ratio is calculated as follows:

[0089]

[0090] Wherein, T represents the total sampling point number of the brain magnetic signal, y represents the amplitude of the brain magnetic signal before denoising, y represents the amplitude of the brain magnetic signal after denoising.

[0091] The second sub-objective function is calculated based on the total signal power ratio: in actual data, the noise level of each channel is different; a small amount of random noise can usually be eliminated through subsequent superposition average processing, and therefore, in order to pay attention to the channel with relatively more noise, the channel weight w N is added to the objective function F i :

[0092]

[0093]

[0094] Wherein, w i represents the channel weight, and σ i is the standard deviation of each channel signal. The F N corresponding to the n parameters in the parameter range is calculated. N The results are saved as [F N (1), F N (2), …, F B (n)].

[0095] S4. Based on the first sub-objective function and the second sub-objective function, a total objective function is constructed, the multi-channel brain magnetic signal after denoising is input into the total objective function, and the optimal denoising parameter is selected based on the function output value.

[0096] The construction method of the total objective function includes:

[0097] The results [F B (1), F B (2), …, F N (n)] obtained by the first sub-objective function are normalized to [0, 1];

[0098] The results [F N (1), F N(n)] is normalized to [0, 2], and then substituted into the hyperbolic tangent function tanh, so that the result is reduced to [0, 1], in order to reduce the influence of excessive signal suppression on Fall;

[0099] The total objective function is obtained by weighted summation:

[0100] F all (x) = aF B (x) + (1-a) tanh(F N (x))

[0101] Where x represents the parameter of the subspace projection algorithm, and a represents the parameter for balancing the two objective functions.

[0102] The method for selecting the optimal denoising parameter includes: inputting the denoised multi-channel magnetoencephalogram signal into the total objective function, so that the parameter that makes the total objective function maximum is the optimal denoising parameter.

[0103] Embodiment Two

[0104] In this embodiment, as shown in Figure 3 A system for selecting parameters of a magnetoencephalogram subspace projection algorithm based on data includes a signal collection module, a denoising module, a sub-objective function construction module, and a parameter selection module.

[0105] The signal collection module is used to measure the original multi-channel magnetoencephalogram signal.

[0106] The original multi-channel magnetoencephalogram signal is an un-denoised multi-channel magnetoencephalogram measurement signal containing magnetoencephalogram evoked neural signals obtained by stimulating the subject using an experimental paradigm. In this embodiment, the original multi-channel magnetoencephalogram signal is measured using a sensor, which is an optically pumped magnetometer. The optically pumped magnetometer is a very weak magnetic measurement sensor based on the spin exchange relaxation effect.

[0107] The denoising module is used to set the range of the parameters of the subspace projection, and substitute the parameters in the range into the subspace projection in turn to denoise the multi-channel magnetoencephalogram signal, thereby obtaining the denoised multi-channel magnetoencephalogram signal corresponding to each parameter.

[0108] In this embodiment, the workflow of the denoising module includes: setting the range of the parameter x used to define the dimension of the external subspace in the subspace projection as [1, n]; denoising the original multi-channel magnetoencephalogram signal using the subspace projection algorithm to obtain n denoised multi-channel magnetoencephalogram signals; wherein the magnetoencephalogram signal is represented as the sum of the internal neural signal of the brain and the external interference noise:

[0109] Y = B + N

[0110]

[0111] Wherein, Y is the measured total magnetoencephalogram signal, B is the brain neural signal, N is the external interference signal, is the denoised total magnetoencephalogram signal, is the denoised brain neural signal, is the denoised external interference signal.

[0112] The sub-objective function construction module is configured to construct a first sub-objective function describing the distortion degree of the neural signal based on the neural signal of interest, and construct a second sub-objective function describing the noise suppression degree based on the signal power of the noise. In the embodiment, the sub-objective function construction module includes a first construction unit and a second construction unit.

[0113] The working process of the first construction unit includes: in the embodiment, a sub-objective function for describing the distortion degree of the neural signal is first constructed, and the distortion degree of the neural signal can be defined as the ratio of the neural signal before and after denoising:

[0114]

[0115] After the task-state magnetoencephalogram data signal is segmented according to the stimulation cycle, it is usually necessary to study the signal in a specific time period and frequency period in the data segment, which is defined as the signal of interest in the time period and frequency period. Therefore, Y is defined as the total power value of this signal. N is defined as the power average value of the time period before the stimulation moment of the corresponding frequency without containing obvious neural signal (usually referred to as time baseline).

[0116] The total power value of the magnetoencephalogram signal is calculated as:

[0117]

[0118] Wherein, f represents the set frequency period of interest, t represents the set time period of interest, p is the total number of data segments, k represents the kth data segment, y k represents the signal amplitude in the kth data segment;

[0119] The power average value of the time period before the stimulation moment of the corresponding frequency without containing obvious neural signal is calculated as:

[0120]

[0121] Wherein, m represents the number of time sampling points of the noise baseline segment, t' represents the t'th sampling point;

[0122] The neural signal in the multi-channel data is calculated based on the total power value and the power average value, and the neural signal in the magnetoencephalogram multi-channel data is defined as the absolute value of the difference between the two:

[0123]

[0124] wherein, I represents the number of channels;

[0125] The first sub-objective function is calculated based on the neural signal:

[0126]

[0127] wherein, represents the denoised neural signal. The F B results are saved as [F B (1), F B (2), …, F B (n)].

[0128] The workflow of the second construction unit includes that the second sub-objective function is represented as the proportion of the noise before and after denoising. Generally, the order of magnitude of the total signal is about several ten thousand fT, and after band-pass filtering, it is also as high as several thousand fT, while the induced brain magnetic signal is generally only about 500 fT, and the noise is much larger than the brain magnetic signal. Therefore, the proportion of the noise is directly simplified as the ratio of the total signal power, and the total signal power ratio is calculated:

[0129]

[0130] wherein, T represents the total sampling point number of the brain magnetic signal, y represents the amplitude of the brain magnetic signal before denoising, represents the amplitude of the brain magnetic signal after denoising;

[0131] The second sub-objective function is calculated based on the total signal power ratio: in actual data, the noise level of each channel is not the same; a small amount of random noise can usually be eliminated through subsequent superposition average processing, and therefore, in order to focus on the channels with relatively more noise, the channel weight w N is added to the noise suppression degree objective function F i :

[0132]

[0133]

[0134] wherein, w i represents the channel weight, and σ i is the standard deviation of each channel signal. The F N results corresponding to the n parameters in the parameter range are saved as [F N (1), F N (2), …, F N (n)].

[0135] The parameter selection module is used to construct a total objective function based on the first and second sub-objective functions, input the denoised multi-channel magnetoencephalography (MEG) signals into the total objective function, and select the optimal denoising parameters based on the function output value. In this embodiment, the parameter selection module includes: a total objective function construction unit and a selection unit;

[0136] The workflow of the overall objective function construction unit includes: taking the result obtained from the first sub-objective function [F] B (1), F B (2), ..., F B Normalize (n)] to [0,1]; normalize the result of the second sub-objective function [F N (1), F N (2), ..., F N Normalize (n) to [0,2], then substitute it into the hyperbolic tangent function tanh to reduce the result range to [0,1], thus reducing the impact of excessive signal suppression on Fall; obtain the total objective function through weighted summation:

[0137] F all (x)=αF B (x)+(1-α)tanh(F N (x))

[0138] Where x represents the parameters of the subspace projection algorithm, and α represents the parameters for balancing the two objective functions.

[0139] The process of selecting a unit includes: inputting the denoised multi-channel magnetoencephalogram (MEG) signal into the overall objective function, and the parameter that maximizes the overall objective function is the optimal denoising parameter.

[0140] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for selecting parameters of a brain magnetic subspace projection algorithm based on data, characterized by, The method comprises the following steps: S1. measuring original multi-channel magnetoencephalogram signals; S2. setting a range of parameters of subspace projection, substituting parameters in the range into the subspace projection in sequence, and denoising the multi-channel magnetoencephalogram signals to obtain denoised multi-channel magnetoencephalogram signals corresponding to each parameter; S3. constructing a first sub-target function describing the distortion degree of neural signals based on neural signals of interest, and constructing a second sub-target function describing the noise suppression degree based on the signal power of noise; S4. constructing a total target function based on the first sub-target function and the second sub-target function, inputting the denoised multi-channel magnetoencephalogram signals into the total target function, and selecting an optimal denoising parameter based on the function output value.

2. The method of claim 1, wherein the method is based on data selection. The original multi-channel magnetoencephalogram signals are multi-channel magnetoencephalogram measurement signals containing magnetically induced neural signals without denoising processing obtained by stimulating subjects using experimental paradigms.

3. The method of claim 1, wherein the method is based on data-driven selection of parameters for the MEG subspace projection algorithm. The S2 comprises: Setting parameters in the subspace projection for defining the dimension of the outer subspace x ranging from ; The original multi-channel brain magnetic signal is denoised by using the subspace projection algorithm to obtain a denoised multi-channel brain magnetic signal. n The original multi-channel brain magnetic signal is denoised by using the subspace projection algorithm to obtain a denoised multi-channel brain magnetic signal. Wherein, the magnetoencephalogram signal is represented as the sum of internal neural signals and external interference noise: wherein, is the measured total brain magnetic signal, is the brain neural signal, is the external interference signal, is the denoised measured total brain magnetic signal, is the denoised brain neural signal, is the denoised external interference signal.

4. The method of claim 3, wherein the method is based on data selection. The first sub-target function is the ratio of neural signals before and after denoising, and the construction method of the first sub-target function comprises: Calculating the total power value of the magnetoencephalogram signal: wherein, f represents a set frequency band of interest, t represents a set time band of interest, p is the total number of data bands, k represents the kth data band, represents the signal amplitude within the kth data band; Calculating the power average value of a time period before the stimulation time point of the corresponding frequency without containing obvious neural signals: wherein, m denotes the total number of time sample points of the noise baseline segment, t’ denotes the i-th sample point of the noise baseline segment, t’ denotes the i-th sample point of the noise baseline segment, Calculating the neural signals in the multi-channel data based on the total power value and the power average value: wherein l denotes the number of channels; Calculating the first sub-target function based on the neural signals: wherein, denotes the denoised neural signal.

5. The method of claim 4, wherein the method is based on data selection. The second sub-target function is represented as the ratio of noise before and after denoising, and the construction method of the second sub-target function comprises: Calculating the total signal power ratio: wherein, T denotes the total number of sampling points of the magnetoencephalography signal, denotes the amplitude of the magnetoencephalography signal before denoising, denotes the amplitude of the magnetoencephalography signal after denoising; Calculating the second sub-target function based on the total signal power ratio: wherein, represents a channel weight, is the standard deviation of each channel signal.

6. The method of claim 5, wherein the method is based on data selection. The construction method of the total target function comprises: the result obtained from the first sub-objective function normalized to [0, 1]; The result obtained by the second sub-objective function Normalized to [0, 2], and then substituted into the hyperbolic tangent function tanh, so that the result is reduced to the range [0, 1]; Obtaining the total target function by weighted summation: wherein, x denotes a parameter of the subspace projection algorithm, denotes a parameter balancing the two objective functions.

7. The method of claim 6, wherein the method is based on data selection. The selection method of the optimal denoising parameter comprises: inputting the denoised multi-channel magnetoencephalogram signals into the total target function, so that the parameter making the total target function reach the maximum value is the optimal denoising parameter.

8. A system for selecting parameters of a brain magnetic subspace projection algorithm based on data, the system applying the method of any one of claims 1 to 7, characterized in that, It comprises a signal collection module, a denoising module, a sub-target function construction module, and a parameter selection module. The signal collection module is used to measure original multi-channel magnetoencephalogram signals; The denoising module is used to set a range of parameters of subspace projection, substitute parameters in the range into the subspace projection in sequence, and denoise the multi-channel magnetoencephalogram signals to obtain denoised multi-channel magnetoencephalogram signals corresponding to each parameter; The sub-target function construction module is used to construct a first sub-target function describing the distortion degree of neural signals based on neural signals of interest, and construct a second sub-target function describing the noise suppression degree based on the signal power of noise; The parameter selection module is used to construct a total target function based on the first sub-target function and the second sub-target function, input the denoised multi-channel magnetoencephalogram signals into the total target function, and select an optimal denoising parameter based on the function output value.

Images