Decoding method and device for minimally invasive intracranial brain electrical signals and treatment equipment

Abstract

The embodiment of the application provides a kind of minimally invasive intracranial brain electrical signal decoding method, device and treatment equipment, method includes: using the space-time characteristics of minimally invasive intracranial brain electrical signal to filter out multiple effective bands;Utilize multiple effective bands to jointly construct covariance feature;The covariant feature matrix of the covariant feature matrix is embedded into the tangent plane space of the geometric center point of manifold with AIRM Riemann distance measurement, and uses Euclidean distance measurement in the tangent plane space;The covariant feature matrix is projected to tangent plane to obtain the eigenvector of the covariant, and the eigenvector is processed based on machine learning classification to realize the decoding classification of interface data;The embodiment of the application combines affine invariance Riemann geometry measurement, solves the problem of few minimally invasive brain electrode signal channels and limited signal bandwidth, realizes long-term stable and accurate decoding of minimally invasive brain-computer interface.

Description

Technical Field

[0001] The present application relates to the field of brain-computer interface decoding, and specifically to a method, device and treatment equipment for decoding minimally invasive intracranial electroencephalogram (EEG) signals. Background Art

[0002] Compared to commonly used scalp EEG, intracranial EEG has a higher signal-to-noise ratio and a wider effective frequency band, enabling the realization of high-accuracy and low-latency brain-computer interface control paradigms. However, it is also limited by its high invasiveness. Minimally invasive intracranial EEG is a new EEG recording method between intracranial and scalp EEG. It uses fewer electrodes to record signals from outside the dura mater of the brain, maintaining the wide bandwidth and high signal-to-noise ratio characteristics of intracranial EEG without disrupting the brain's internal environment.

[0003] Generally speaking, the spatial resolution, amplitude, and frequency characteristics of minimally invasive epidural EEG are intermediate between those of scalp EEG and intracranial EEG. Existing decoding methods for intracranial or scalp EEG cannot meet the needs of minimally invasive intracranial EEG. Taking the invention patent application number CN202210046576.6 as an example, existing intracranial brain-computer interface decoding methods primarily focus on exploiting EEG features in the time and frequency dimensions, underutilizing signal spatial information and poorly resolving noise signals with spatial patterns that differ from the target signal. Furthermore, traditional machine learning methods often employ traditional linear classification methods, which lack robustness to changes in signal scale. Deep learning methods, on the other hand, have poor interpretability, limited clinical application, and low generalization capabilities. They require a large amount of high-quality annotated clinical data, which is difficult to obtain in clinical settings. Existing scalp EEG decoding methods primarily focus on spatial patterns in EEG. Due to the narrow frequency band and limited frequency domain information of scalp EEG, directly applying scalp EEG decoding methods to minimally invasive EEG decoding results in a significant loss of valuable frequency domain information, significantly reducing decoding accuracy. Summary of the Invention

[0004] In response to the problems in the existing technology, the present application provides a decoding method, device and treatment equipment for minimally invasive intracranial EEG signals, which can construct multi-frequency integrated covariance space-time frequency characteristics based on the space-time frequency characteristics of minimally invasive intracranial electrode signals, and combine them with affine invariant Riemannian geometry metrics to solve the problems of small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and realize long-term stable and accurate decoding of minimally invasive brain-computer interfaces.

[0005] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0006] According to a first aspect of the embodiments of the present application, the present application provides a minimally invasive intracranial EEG signal decoding method, comprising:

[0007] Utilize the spatiotemporal characteristics of minimally invasive intracranial EEG signals to screen out multiple effective frequency bands;

[0008] Using multiple effective frequency bands to jointly construct a covariance feature, the covariance feature is used to characterize the spatial pattern of minimally invasive electroencephalogram signals;

[0009] Embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and use the Euclidean distance metric in the tangent plane space;

[0010] The characteristic matrix of the covariance is projected onto a tangent plane to obtain a characteristic vector of the covariance, and the characteristic vector is processed based on machine learning classification to achieve decoding classification of the interface data.

[0011] According to any embodiment of the present application, the method of screening out multiple effective frequency bands using the spatiotemporal characteristics of minimally invasive intracranial EEG signals includes:

[0012] The ratio of the inter-class distance to the intra-class distance of the covariance matrix of each frequency band in motion and static state is calculated based on the affine invariant Riemannian distance metric to determine the separability of each frequency band;

[0013] The Rayleigh quotient metric of linear discriminant analysis is extended to the Riemannian geometry distance metric, and a category separability metric index in the sense of Riemannian metric is obtained to screen the effective frequency bands that meet the classification contribution criteria.

[0014] According to any embodiment of the present application, the step of jointly constructing a covariance feature using a plurality of the effective frequency bands includes:

[0015] According to the channel signal obtained after each effective frequency band is band-pass filtered, a covariance feature is constructed by jointly calculating the covariance of multiple frequency bands.

[0016] According to any embodiment of the present application, after constructing the covariance feature by jointly calculating the covariance of multiple frequency bands, the method further includes:

[0017] Regularization processing is performed on the covariance feature.

[0018] According to any embodiment of the present application, the differentiable manifold formed by the characteristic matrix of the covariance is embedded into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and the Euclidean distance metric is used in the tangent plane space, including:

[0019] Calculate the geometric center point of the differentiable manifold formed by the characteristic matrix of the covariance using AIRM;

[0020] Embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent space corresponding to the geometric center point;

[0021] A Euclidean distance metric is used in the tangent space defined at the geometric center point of the differentiable manifold, wherein the Euclidean distance metric is isomorphic to AIRM.

[0022] According to any embodiment of the present application, before projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, the method further includes:

[0023] A tangent plane is constructed at the Frechet mean point of the differentiable manifold.

[0024] According to a second aspect of the embodiments of the present application, the present application provides a minimally invasive intracranial EEG signal decoding device, comprising:

[0025] The frequency band screening module is used to: screen out multiple effective frequency bands by utilizing the spatiotemporal characteristics of minimally invasive intracranial EEG signals;

[0026] A covariance feature construction module is used to: jointly construct a covariance feature using a plurality of the effective frequency bands, wherein the covariance feature is used to characterize the spatial pattern of the minimally invasive EEG signal;

[0027] A spatial metric module is used to: embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the manifold geometric center point using the AIRM Riemann distance metric, and use the Euclidean distance metric in the tangent plane space;

[0028] The projection transformation module is used to: project the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, and process the characteristic vector based on machine learning classification to realize decoding classification of interface data.

[0029] According to any embodiment of the present application, the frequency band screening module is specifically used to:

[0030] The ratio of the inter-class distance to the intra-class distance of the covariance matrix of each frequency band in motion and static state is calculated based on the affine invariant Riemannian distance metric to determine the separability of each frequency band;

[0031] The Rayleigh quotient metric of linear discriminant analysis is extended to the Riemannian geometry distance metric, and a category separability metric index in the sense of Riemannian metric is obtained to screen the effective frequency bands that meet the classification contribution criteria.

[0032] According to any embodiment of the present application, the covariance feature construction module is specifically used to:

[0033] According to the channel signal obtained after each effective frequency band is band-pass filtered, a covariance feature is constructed by jointly calculating the covariance of multiple frequency bands.

[0034] According to any embodiment of the present application, after the covariance feature construction module constructs the covariance feature by jointly calculating the covariance of multiple frequency bands, it further includes a regularization processing unit for:

[0035] Regularization processing is performed on the covariance feature.

[0036] According to any embodiment of the present application, the spatial measurement module is specifically configured to:

[0037] Calculate the geometric center point of the differentiable manifold formed by the characteristic matrix of the covariance using AIRM;

[0038] Embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent space corresponding to the geometric center point;

[0039] A Euclidean distance metric is used in the tangent space defined at the geometric center point of the differentiable manifold, wherein the Euclidean distance metric is isomorphic to AIRM.

[0040] According to any embodiment of the present application, before the projection transformation module projects the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, a tangent plane construction module is further included, which is used to:

[0041] A tangent plane is constructed at the Frechet mean point of the differentiable manifold.

[0042] According to a third aspect of the embodiments of the present application, the present application provides a therapeutic device, comprising:

[0043] A signal receiving module is used to receive intracranial EEG signals from a patient with spinal cord injury and decode the patient's intracranial EEG signals based on the minimally invasive intracranial EEG signal decoding method;

[0044] The peripheral control module is used to control the external device according to the obtained decoding result, wherein the external device includes at least one of a mechanical device and a computer device.

[0045] According to any embodiment of the present application, the peripheral control module mechanical equipment control unit is used to:

[0046] The motion state of the mechanical device is controlled according to the multi-dimensional control instructions in the decoding result.

[0047] According to any embodiment of the present application, the peripheral control module includes a computer control unit configured to:

[0048] The eight-directional motion state and click state of the cursor of the computer device are controlled according to the multi-dimensional control instructions in the decoding result.

[0049] According to a fourth aspect of the embodiments of the present application, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the decoding method of the minimally invasive intracranial brain electrical signal when executing the program.

[0050] According to a fifth aspect of the embodiments of the present application, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executable on a processor to implement the steps of the decoding method of the minimally invasive intracranial brain electrical signal.

[0051] According to a sixth aspect of the embodiments of the present application, the present application provides a computer program product comprising computer programs / instructions, wherein the computer programs / instructions are executable on a processor to implement the steps of the decoding method of the minimally invasive intracranial brain electrical signal.

[0052] According to the above technical solutions, the present application provides a decoding method, device and treatment equipment of minimally invasive intracranial brain electrical signal, which screens a plurality of effective frequency bands by using the space-time characteristics of the minimally invasive intracranial brain electrical signal; constructs a covariance feature by using the plurality of effective frequency bands in combination, wherein the covariance feature is used to represent the spatial mode of the minimally invasive brain electrical signal; embeds the differentiable manifold formed by the feature matrix of the covariance into the tangent plane space of the manifold geometric center point by using the AIRM Riemann distance measurement, and uses the Euclidean distance measurement in the tangent plane space; projects the feature matrix of the covariance to the tangent plane to obtain the feature vector of the covariance, and processes the feature vector based on machine learning classification to realize the decoding classification of the interface data. The space-time frequency characteristics of the minimally invasive intracranial electrode signal can be used to construct a multi-frequency integrated covariance space-time frequency feature, and the affine invariance Riemann geometric measurement is combined to solve the problems of the small number of minimally invasive brain electrode signal channels and the limited signal bandwidth, so as to realize the long-term stable and accurate decoding of the minimally invasive brain-computer interface. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0054] Figure 1 The minimally invasive brain electrical time-frequency feature large average (N=996) schematic diagram provided by the embodiments of the present application.

[0055] Figure 2 One of the flowcharts of the decoding method of the minimally invasive intracranial brain electrical signal in the embodiments of the present application;

[0056] Figure 3 This is a second flow chart of the minimally invasive intracranial EEG signal decoding method in an embodiment of the present application;

[0057] Figure 4 This is a third flow chart of the minimally invasive intracranial EEG signal decoding method in an embodiment of the present application;

[0058] Figure 5 This is a fourth flow chart of the minimally invasive intracranial EEG signal decoding method in an embodiment of the present application;

[0059] Figure 6 This is a structural diagram of a minimally invasive intracranial EEG signal decoding device in an embodiment of the present application;

[0060] Figure 7 Schematic diagram of the structure of the electronic device in the embodiment of the present application. DETAILED DESCRIPTION

[0061] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0062] The acquisition, storage, use, and processing of data in this application's technical solution comply with relevant national laws and regulations.

[0063] Taking into account the inherent problems of the existing main methods for decoding intracranial brain-computer interfaces and the problems of small number of minimally invasive brain electrode signal channels and limited signal bandwidth, the present application provides a minimally invasive intracranial EEG signal decoding method, device and treatment equipment to solve the problems of small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and to achieve long-term, stable and accurate decoding of minimally invasive brain-computer interfaces.

[0064] Minimally invasive intracranial EEG has a wider frequency band than scalp EEG. Figure 1 As shown in the data, during motor imagery, the main manifestations are event-related desynchronization (ERD) in the beta (15-35 Hz) and low-gamma (35-50 Hz) frequency bands and energy enhancement in the high-gamma (High-gamma) frequency band above 50 Hz.

[0065] like Figure 2As shown, the complete technical solution of this application fully utilizes the spatiotemporal and frequency characteristics of minimally invasive EEG, optimizes the effective frequency band through machine learning methods, and expands the single-channel information bandwidth of minimally invasive EEG; fully utilizes the spatial information of minimally invasive EEG by constructing covariance features; and solves the long-term stability of minimally invasive EEG by introducing affine invariant Riemannian geometry metrics. Combining these technologies, a set of long-term, efficient and stable intracranial minimally invasive brain-computer interface decoding methods has been constructed.

[0066] In order to construct multi-frequency integrated covariance spatiotemporal frequency features based on the spatiotemporal frequency characteristics of minimally invasive intracranial electrode signals, and combine them with affine invariant Riemannian geometry metrics to solve the problems of a small number of channels and limited signal bandwidth of minimally invasive brain electrode signals, and to achieve long-term stable and accurate decoding of minimally invasive brain-computer interfaces, this application provides an embodiment of a decoding method for minimally invasive intracranial EEG signals, see Figure 3 The minimally invasive intracranial EEG signal decoding method specifically includes the following contents:

[0067] Step S101: Filter out multiple effective frequency bands using the spatiotemporal characteristics of minimally invasive intracranial EEG signals.

[0068] The ability to distinguish EEG signals across different frequency bands is assessed using the affine invariant Riemann distance metric (AIRM). AIRM measures the differences between covariance matrices, thereby evaluating the spatial distribution characteristics of the signal across different frequency bands. By calculating the covariance matrix for each frequency band and evaluating its separability using AIRM, frequency bands that contribute significantly to classification can be identified. These effective frequency bands have stronger discriminatory power, helping to improve decoding accuracy.

[0069] Step S102: using a plurality of the effective frequency bands to jointly construct a covariance feature, wherein the covariance feature is used to characterize the spatial pattern of the minimally invasive EEG signal.

[0070] Based on the selected effective frequency bands, covariance features are constructed by combining multiple frequency bands. The channel signals obtained after bandpass filtering in each frequency band are used to calculate the covariance matrix, which reflects the spatial distribution pattern of the EEG signal. Combining information from multiple frequency bands can comprehensively capture the characteristics of the EEG signal across different frequency ranges, thereby more accurately characterizing the spatial pattern of the EEG signal and improving the accuracy and stability of decoding.

[0071] Step S103: embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and use the Euclidean distance metric in the tangent plane space.

[0072] The differentiable manifold formed by the covariance feature matrix is ​​embedded in a high-dimensional Euclidean space. This manifold is then embedded in the tangent plane space, centered at the manifold's geometric center (the Fréchet mean point), and the Euclidean distance metric is used within the tangent space. This Euclidean distance is isomorphic to the AIRM metric in local regions; that is, within a small range, the two metrics produce the same results. This approach simplifies computational complexity and allows the advantages of Riemannian geometry to be leveraged within traditional machine learning frameworks, ensuring computational accuracy and stability.

[0073] Step S104: Projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, and processing the characteristic vector based on machine learning classification to achieve decoding classification of the interface data.

[0074] When building a classifier, a tangent plane is constructed through a reference point, and all data points are projected onto this tangent plane. The projected feature matrix is ​​converted into eigenvectors, simplifying the data structure. Finally, these eigenvectors can be classified and processed using classic machine learning classification algorithms to decode the brain-computer interface data. This process not only retains the advantages of Riemannian geometry but also leverages the efficiency and accuracy of traditional machine learning.

[0075] From the above description, it can be seen that the decoding method of minimally invasive intracranial EEG signals provided in the embodiment of the present application can construct multi-frequency integrated covariance space-time-frequency features based on the space-time-frequency characteristics of minimally invasive intracranial electrode signals, and combine them with affine invariant Riemannian geometry metrics to solve the problems of small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and realize long-term stable and accurate decoding of minimally invasive brain-computer interfaces.

[0076] In one embodiment of the minimally invasive intracranial EEG signal decoding method of the present application, see Figure 4 , the spatiotemporal characteristics of minimally invasive intracranial EEG signals are used to screen out multiple effective frequency bands, including:

[0077] Step S101A: Calculate the ratio of the inter-class distance to the intra-class distance of the covariance matrix of each frequency band in motion and static states based on the affine invariance Riemannian distance metric to determine the separability of each frequency band.

[0078] First, the covariance matrix corresponding to each frequency band is calculated. The affine-invariant Riemannian distance metric is then used to measure the ratio of the inter-class to intra-class distances for each frequency band's covariance matrix in motion and at rest. The affine-invariant Riemannian distance metric is a method that can measure the similarity between covariance matrices and maintains invariance to linear transformations. Specifically, the calculation formula for the Riemannian distance metric includes the eigenvalues ​​of the matrices, reflecting the differences in the spatial distribution of signals in different frequency bands. The inter-class distance represents the average distance between the covariance matrices in motion and at rest. A larger inter-class distance indicates more distinct signal differences between the different states. The intra-class distance represents the average distance between the covariance matrices of different samples in the same state. A smaller intra-class distance indicates more consistent signals in the same state. The separability of each frequency band can be determined by calculating the ratio of the inter-class to intra-class distances. A high inter-class to intra-class distance ratio indicates that the frequency band is more effective in distinguishing between motion and rest. Based on this ratio, the effective frequency band that contributes most to the classification task can be screened.

[0079] Step S101B: Extending the Rayleigh quotient metric of the linear decision analysis to the Riemannian geometry distance metric to obtain a category separability metric index in the sense of the Riemannian metric to screen effective frequency bands that meet the classification contribution standard.

[0080] After evaluating the separability of frequency bands, the team extended the Rayleigh quotient metric from linear discriminant analysis (LDA) to Riemannian geometry to propose a new class separability metric. The Rayleigh quotient metric is a commonly used method for evaluating classifier performance. By introducing it into Riemannian geometry, it is possible to perform frequency band separability analysis in a more complex space.

[0081] Exemplarily, in the current step, a machine learning method is used to screen effective frequency band components that contribute more to classification, so as to effectively improve the single-channel information bandwidth of minimally invasive EEG.

[0082] This application evaluates the separability of each frequency band based on the Riemannian metric and the category separability evaluation method.

[0083] Based on the affine invariance Riemann distance metric, see formula 1:

[0084]

[0085] in‖·‖ F represents the Frobenius matrix norm, λ i is a matrix The eigenvalue of .

[0086] Based on this distance metric, the Rayleigh quotient used in linear discriminant analysis (LDA) is extended to Riemannian geometry, and a category separability metric in the sense of the Riemannian metric can be proposed, see Formula 2:

[0087]

[0088] in, represents the mean covariance under the Riemannian metric, σ C represents the covariance standard deviation under the Riemannian metric, Superscript C (A / B) Indicates the corresponding category.

[0089] In one embodiment of the minimally invasive intracranial EEG signal decoding method of the present application, the method of jointly constructing a covariance feature using a plurality of effective frequency bands includes:

[0090] According to the channel signal obtained after each effective frequency band is band-pass filtered, a covariance feature is constructed by jointly calculating the covariance of multiple frequency bands.

[0091] After constructing the covariance feature by jointly calculating the covariance of multiple frequency bands, it also includes:

[0092] Regularization processing is performed on the covariance feature.

[0093] In this application, covariance features are used to characterize the spatial patterns of EEG signals, which can make up for the shortcomings of traditional machine learning methods in utilizing the spatial patterns of minimally invasive EEG signals. The typical EEG response pattern of the motor imagery paradigm has multi-band information, which is usually manifested as the energy variance of multiple bands. Therefore, this application constructs covariance features by jointly calculating covariance across multiple bands to more accurately capture the spatial characteristics of EEG signals.

[0094] Specifically, we first select multiple valid frequency bands, identified in the previous step. Bandpass filtering is then performed on each frequency band to extract the channel signals within that frequency band. The bandpass-filtered channel signals contain information about EEG activity within each frequency band. These channel signals are then combined and their covariance matrix is ​​calculated.

[0095] Multi-band covariance calculation combines the channel signals of each frequency band into a large covariance matrix, which can simultaneously capture EEG signal characteristics in multiple frequency bands. The covariance features constructed in this way can comprehensively reflect the spatial patterns of EEG signals in multiple frequency bands, improving the accuracy and effectiveness of signal processing.

[0096] For example, when designing the covariance feature, the present application adopts a method of jointly calculating the covariance of multiple frequency bands, see Formula 3:

[0097]

[0098] in, is the channel signal after bandpass filtering in different frequency bands, and F is the number of frequency bands selected.

[0099] Preferably, after constructing the covariance feature, in order to further improve its robustness and accuracy, the present application uses the Ledoit-Wolf algorithm to regularize the covariance matrix. The algorithm reduces the noise effect in the estimation process and enhances the robustness of the covariance feature by smoothing the covariance matrix.

[0100] In one embodiment of the minimally invasive intracranial EEG signal decoding method of the present application, see Figure 5 , embedding the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the manifold geometric center point using the AIRM Riemann distance metric, and using the Euclidean distance metric in the tangent plane space, including:

[0101] Step S103A: Calculate the geometric center point of the differentiable manifold formed by the characteristic matrix of the covariance using AIRM;

[0102] Step S103B: embedding the differentiable manifold formed by the characteristic matrix of the covariance into the tangent space corresponding to the geometric center point;

[0103] Step S103C: Using the Euclidean distance metric in the tangent space defined at the geometric center point of the differentiable manifold, wherein the Euclidean distance metric is isomorphic to the AIRM.

[0104] Because covariance matrices are symmetric positive definite matrices (SPDs), the distance between these matrices can be measured using the affine invariant Riemannian metric (AIRM). Because AIRM is invariant to linear transformations, it maintains a consistent distance metric across linear transformations. This property is particularly important for EEG signal processing, ensuring the stability and consistency of the covariance matrix under different sensor configurations and source space conditions. Using AIRM for measurement can significantly improve the stability and accuracy of the decoder.

[0105] Specifically, because the AIRM metric has the following properties: the covariance matrix S formed by any two samplings in the source space is A and S B , which is mapped by a certain conduction field matrix L (L is required to be reversible), and the covariance matrix C of the sensor space is obtained A =LS A L T ,C B =LSB L T , which is easy to prove, as shown in Formula 4:

[0106] δ r (C A ,C B )=δ r (S A ,S B ) (4)

[0107] Under the AIRM metric, the distance metric between the covariance matrices of the source and sensor spaces is identical. Regardless of how the signal is spatially transformed, the separability of the source and sensor spaces remains consistent, which means that AIRM guarantees the stability of its metric.

[0108] Similarly, with prolonged use, physiological changes in the brain-computer interface can cause some changes in the conduction field matrix, L′. However, the AIRM distance metric ensures that separability remains unchanged before and after these changes. However, brain-computer interface decoding algorithms that use spatial filters will fail due to changes in the conduction field matrix. Therefore, brain-computer interface decoding algorithms using the AIRM metric possess sensor-spatial robustness and long-term stability that traditional methods lack.

[0109] In this application, in order to adapt the Riemannian geometry metric method to the traditional machine learning classification model, the Riemannian geometry metric method is adapted to the traditional machine learning classification model.

[0110] The differentiable manifold formed by the n×n dimensional SPD matrix is ​​defined as It can be embedded into the n(n+1) / 2 dimensional Euclidean space. The tangent space of C passing through any point on this manifold is defined as In this tangent space, the distance metric is Euclidean distance, and the distance in the neighborhood of X is the same as that in the manifold Riemann distance is used to measure homomorphic distance. i , the corresponding point T on the tangent plane can be found through the following mapping relationship i , as shown in Formula 5:

[0111]

[0112] Through this mapping, the Euclidean distance can be calculated on the tangent plane without the need to calculate the Riemann distance on the original complex manifold.

[0113] In an optional embodiment, before projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, the method further includes:

[0114] A tangent plane is constructed at the Frechet mean point of the differentiable manifold.

[0115] Among them, when building a classifier, it is necessary to find a reference point C ref , construct a tangent plane through this point and project all data points onto this tangent plane. This application takes the tangent plane formed by the Frechet mean points on the manifold, as shown in Formula 6:

[0116]

[0117] The Fréchet mean point is the average position of all points on the manifold, determined by minimizing the sum of the squared distances between the points and the mean point. The tangent plane constructed at this reference point optimizes the distance measurement of other points on the manifold projected onto this plane.

[0118] By this method, all data points can be projected onto the tangent plane to obtain n(n+1) / 2-dimensional feature vectors, where n is the dimension of the constructed covariance feature.

[0119] The dimension of this feature vector depends on the dimension of the constructed covariance features. By converting the high-dimensional covariance matrix features into low-dimensional feature vectors, classic machine learning classification methods can be used for classification. This method not only retains the advantages of Riemannian geometry metrics but is also compatible with traditional machine learning models, simplifying the computational process and improving classification efficiency and accuracy.

[0120] The spatiotemporal characteristics of minimally invasive EEG signals are between those of conventional scalp EEG and fully invasive intracranial EEG, with a higher signal-to-noise ratio and a wider effective frequency band, which can capture more subtle brain activities. However, compared with fully invasive intracranial EEG, minimally invasive EEG signals have a slightly lower spatial resolution, but are also less invasive and will not damage the internal environment of the brain. Conventional scalp EEG signals have a narrow frequency band, limited frequency domain information, and a weak ability to capture spatial patterns. The solution of this application constructs multi-band joint covariance features, utilizes affine invariant Riemannian metrics for projection, and uses Euclidean distance metrics in tangent space. This special method not only maintains the advantages of high signal-to-noise ratio and wide frequency band, but also simplifies the computational complexity, ensuring the stability and accuracy of signal processing.

[0121] Furthermore, this application selects and combines multiple valid frequency bands to construct covariance features, fully utilizing the frequency domain information of each band to compensate for the lack of information in a single frequency band. AIRM is used to evaluate and process the covariance matrix between different frequency bands, maintaining the separability of the signal in different spaces and improving decoding accuracy. Furthermore, by projecting the covariance feature matrix onto the tangent plane and using the Euclidean distance metric to simplify the calculation process, not only is the computational complexity reduced, but efficient feature extraction and classification performance are also maintained.

[0122] In summary, by converting the Riemannian manifold distance calculation into the Euclidean distance calculation and using the projection method on the tangent space, this application can effectively combine the advantages of Riemannian geometry and traditional machine learning models to achieve efficient classification and decoding of EEG signals, providing a new, powerful and stable decoding solution for brain-computer interface technology.

[0123] According to another aspect of the embodiments of the present application, the present application provides a therapeutic device, comprising:

[0124] A signal receiving module is used to receive intracranial EEG signals from a patient with spinal cord injury and decode the patient's intracranial EEG signals based on the minimally invasive intracranial EEG signal decoding method;

[0125] The peripheral control module is used to control the external device according to the obtained decoding result, wherein the external device includes at least one of a mechanical device and a computer device.

[0126] According to any embodiment of the present application, the peripheral control module mechanical equipment control unit is used to:

[0127] The motion state of the mechanical device is controlled according to the multi-dimensional control instructions in the decoding result.

[0128] According to any embodiment of the present application, the peripheral control module includes a computer control unit configured to:

[0129] The eight-directional motion state and click state of the cursor of the computer device are controlled according to the multi-dimensional control instructions in the decoding result.

[0130] Exemplarily, the control of the external device according to the obtained decoding result includes controlling the motion state of the mechanical device according to the multi-dimensional control instructions in the decoding result. Its application scenarios are as follows:

[0131] For patients with spinal cord injuries, motor imagery generates EEG signals, which are then decoded and analyzed by a decoding device to generate control instructions for controlling the gripping motion of the robotic arm. Through the principles of neuroplasticity, ascending sensory signals and descending control signals converge in the central nervous system, enabling precise control of the robotic arm.

[0132] In addition, other peripheral control is also an application area of ​​this application. Users can control external devices through paradigms such as motor imagery, and the decoding device can classify and output multi-dimensional control instructions to achieve control. Specific operations include:

[0133] Using the above-mentioned EEG signal decoding technology of this application, the user's motor imagery signals are analyzed and corresponding multi-dimensional control instructions are generated. These control instructions can be used to control various external devices, such as wheelchairs, robotic arms, etc., enabling users to achieve precise control of the equipment through EEG signals.

[0134] In addition, controlling the external device according to the obtained decoding result also includes controlling the eight-directional motion state and click state of the cursor of the computer device according to the multi-dimensional control instructions in the decoding result.

[0135] The specific applications are as follows:

[0136] Users can generate EEG signals through multi-part movement imagination, which are then analyzed by a decoding device to generate multi-dimensional control instructions to control the eight-directional movement and clicking operations of the computer cursor, allowing users to operate the computer through EEG signals.

[0137] In other words, the device control method described in this application enables precise control of mechanical and computer equipment. Whether controlling a manipulator for rehabilitation treatment of spinal cord injury patients, controlling peripherals in daily life, or operating a computer, this method can efficiently analyze EEG signals and generate multi-dimensional control commands, thereby achieving comprehensive control of external devices and improving the user's quality of life and autonomy.

[0138] In order to construct multi-frequency integrated covariance space-time frequency features based on the space-time frequency characteristics of minimally invasive intracranial electrode signals, and combine them with affine invariant Riemannian geometry metrics to solve the problems of a small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and to achieve long-term stable and accurate decoding of minimally invasive brain-computer interfaces, the present application provides an embodiment of a minimally invasive intracranial EEG signal decoding device for implementing all or part of the content of the minimally invasive intracranial EEG signal decoding method, see Figure 6 The minimally invasive intracranial EEG signal decoding device specifically includes the following contents:

[0139] The frequency band screening module 1101 is configured to screen a plurality of effective frequency bands by using the spatial and temporal characteristics of the minimally invasive intracranial electroencephalogram signal.

[0140] The covariance feature construction module 1102 is configured to jointly construct a covariance feature by using the plurality of effective frequency bands, and the covariance feature is used to represent the spatial mode of the minimally invasive electroencephalogram signal.

[0141] The spatial metric module 1103 is configured to embed a differentiable manifold formed by the feature matrix of the covariance into the tangent plane space of the geometric center point of the manifold by using AIRM Riemann distance measurement, and use Euclidean distance measurement in the tangent plane space.

[0142] The projection transformation module 1104 is configured to project the feature matrix of the covariance into the tangent plane to obtain a feature vector of the covariance, and process the feature vector based on machine learning classification to realize decoding and classification of the interface data.

[0143] According to any one of the embodiments of the present application, the frequency band screening module is specifically configured to:

[0144] The class distance and the class distance ratio of the moving and static states of each frequency band covariance matrix are calculated based on the affine invariance Riemann distance measurement, so as to determine the separability of each frequency band.

[0145] The Rayleigh quotient measurement of linear decision analysis is extended to Riemann geometric distance measurement, so as to obtain a class separability measurement index in the sense of Riemann measurement, so as to screen the effective frequency band meeting the classification contribution standard.

[0146] According to any one of the embodiments of the present application, the covariance feature construction module is specifically configured to:

[0147] The covariance feature is constructed by using the multi-band joint covariance calculation mode according to the channel signal obtained by band-pass filtering of each effective frequency band.

[0148] According to any one of the embodiments of the present application, after the covariance feature construction module constructs the covariance feature by using the multi-band joint covariance calculation mode, the method further comprises a regularization processing unit, configured to:

[0149] The covariance feature is regularized.

[0150] According to any one of the embodiments of the present application, the spatial metric module is specifically configured to:

[0151] The geometric center point of the differentiable manifold formed by the feature matrix of the covariance is calculated by using AIRM.

[0152] The differentiable manifold formed by the feature matrix of the covariance is embedded into the tangent space corresponding to the geometric center point.

[0153] A Euclidean distance metric is used in the tangent space defined at the geometric center point of the differentiable manifold, wherein the Euclidean distance metric is isomorphic to AIRM.

[0154] According to any embodiment of the present application, before the projection transformation module projects the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, a tangent plane construction module is further included, which is used to:

[0155] A tangent plane is constructed at the Frechet mean point of the differentiable manifold.

[0156] From the above description, it can be seen that the decoding device for minimally invasive intracranial EEG signals provided in the embodiment of the present application can construct multi-frequency integrated covariance space-time-frequency features based on the space-time-frequency characteristics of minimally invasive intracranial electrode signals, and combine them with affine invariant Riemannian geometry metrics to solve the problems of small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and realize long-term stable and accurate decoding of minimally invasive brain-computer interfaces.

[0157] From a hardware perspective, in order to construct multi-frequency integrated covariance space-time frequency features based on the space-time frequency characteristics of minimally invasive intracranial electrode signals, combined with affine invariant Riemannian geometry metrics, to solve the problems of a small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and to achieve long-term stable and accurate decoding of minimally invasive brain-computer interfaces, the present application provides an embodiment of an electronic device for implementing all or part of the content of the minimally invasive intracranial EEG signal decoding method, and the electronic device specifically includes the following content:

[0158] Processor (processor), memory (memory), communication interface (Communications Interface) and bus; wherein, the processor, memory, and communication interface complete mutual communication through the bus; the communication interface is used to realize information transmission between the decoding device of minimally invasive intracranial EEG signals and related equipment such as core business systems, user terminals and related databases; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, etc., and this embodiment is not limited to this. In this embodiment, the logic controller can be implemented with reference to the embodiment of the decoding method of minimally invasive intracranial EEG signals in the embodiment, and the embodiment of the decoding device of minimally invasive intracranial EEG signals, the contents of which are incorporated herein, and the repeated parts are not repeated.

[0159] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0160] In practical applications, part of the minimally invasive intracranial EEG signal decoding method can be performed on the electronic device side as described above, or all operations can be completed in the client device. The specific selection can be based on the processing capabilities of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0161] The client device may include a communication module (i.e., a communication unit) that can establish a communication connection with a remote server to implement data transmission with the server. The server may include a server on the task scheduling center side, and in other implementation scenarios, may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0162] Figure 7 Schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Figure 7 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that the Figure 7 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0163] In one embodiment, the minimally invasive intracranial EEG signal decoding method function may be integrated into the central processing unit 9100. The central processing unit 9100 may be configured to perform the following control:

[0164] Step S101: Filtering out multiple effective frequency bands using the spatiotemporal characteristics of minimally invasive intracranial EEG signals;

[0165] Step S102: using a plurality of the effective frequency bands to jointly construct a covariance feature, wherein the covariance feature is used to characterize the spatial pattern of the minimally invasive EEG signal;

[0166] Step S103: embedding the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and using the Euclidean distance metric in the tangent plane space;

[0167] Step S104: Projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, and processing the characteristic vector based on machine learning classification to achieve decoding classification of the interface data.

[0168] From the above description, it can be seen that the electronic device provided in the embodiment of the present application constructs multi-frequency integrated covariance space-time-frequency characteristics based on the space-time-frequency characteristics of minimally invasive intracranial electrode signals, and combines them with affine invariant Riemannian geometry metrics to solve the problems of small number of minimally invasive brain electrode signal channels and limited signal bandwidth, and realize long-term stable and accurate decoding of minimally invasive brain-computer interfaces.

[0169] In another embodiment, the decoding device for minimally invasive intracranial EEG signals can be configured separately from the central processing unit 9100. For example, the decoding device for minimally invasive intracranial EEG signals can be configured as a chip connected to the central processing unit 9100, and the function of the decoding method for minimally invasive intracranial EEG signals can be realized through the control of the central processing unit.

[0170] like Figure 7 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Figure 7 In addition, the electronic device 9600 may also include all components shown in Figure 7 For components not shown, reference may be made to the prior art.

[0171] like Figure 7 As shown, the central processing unit 9100 is sometimes also referred to as a controller or operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0172] Memory 9140 can be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It can store the aforementioned failure-related information and also store programs that execute the relevant information. The CPU 9100 can execute the programs stored in memory 9140 to implement information storage or processing.

[0173] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 may be, for example, a keypad or touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display objects such as images and text. The display may be, for example, an LCD display, but is not limited thereto.

[0174] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), or a SIM card. Alternatively, it may be a memory that retains information even when power is off, can be selectively erased, and is provided with more data. Examples of such memory are sometimes referred to as EPROMs. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 by the central processing unit 9100.

[0175] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various driver programs for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0176] The communication module 9110 is a transmitter / receiver 9110 that transmits and receives signals via an antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as in a conventional mobile communication terminal.

[0177] Based on different communication technologies, multiple communication modules 9110 can be provided in the same electronic device, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby implementing common telecommunication functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to the central processing unit 9100, enabling local recording via the microphone 9132 and playback of stored audio via the speaker 9131.

[0178] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all steps of the minimally invasive intracranial EEG signal decoding method in the above-mentioned embodiment, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the computer program implements all steps of the minimally invasive intracranial EEG signal decoding method in the above-mentioned embodiment, where the execution subject is a server or a client. For example, when the processor executes the computer program, the following steps are implemented:

[0179] Step S101: Filtering out multiple effective frequency bands using the spatiotemporal characteristics of minimally invasive intracranial EEG signals;

[0180] Step S102: using a plurality of the effective frequency bands to jointly construct a covariance feature, wherein the covariance feature is used to characterize the spatial pattern of the minimally invasive EEG signal;

[0181] Step S103: embedding the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and using the Euclidean distance metric in the tangent plane space;

[0182] Step S104: Projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, and processing the characteristic vector based on machine learning classification to achieve decoding classification of the interface data.

[0183] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application constructs multi-frequency integrated covariance space-time-frequency features based on the space-time-frequency characteristics of minimally invasive intracranial electrode signals, and combines them with affine invariant Riemannian geometry metrics to solve the problems of the small number of channels and limited signal bandwidth of minimally invasive brain electrode signals, and realizes long-term stable and accurate decoding of minimally invasive brain-computer interfaces.

[0184] The embodiments of the present application also provide a computer program product capable of implementing all steps of the minimally invasive intracranial EEG signal decoding method in the above-mentioned embodiment, where the execution subject is a server or a client. When the computer program / instructions are executed by a processor, the steps of the minimally invasive intracranial EEG signal decoding method are implemented. For example, the computer program / instructions implement the following steps:

[0185] Step S101: Filtering out multiple effective frequency bands using the spatiotemporal characteristics of minimally invasive intracranial EEG signals;

[0186] Step S102: using a plurality of the effective frequency bands to jointly construct a covariance feature, wherein the covariance feature is used to characterize the spatial pattern of the minimally invasive EEG signal;

[0187] Step S103: embedding the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and using the Euclidean distance metric in the tangent plane space;

[0188] Step S104: Projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, and processing the characteristic vector based on machine learning classification to achieve decoding classification of the interface data.

[0189] From the above description, it can be seen that the computer program product provided in the embodiment of the present application constructs multi-frequency integrated covariance space-time-frequency features based on the space-time-frequency characteristics of minimally invasive intracranial electrode signals, and combines them with affine invariant Riemannian geometry metrics to solve the problems of the small number of channels and limited signal bandwidth of minimally invasive brain electrode signals, and realizes long-term stable and accurate decoding of minimally invasive brain-computer interfaces.

[0190] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0191] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0192] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0193] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0194] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A minimally invasive intracranial electroencephalogram (EEG) signal decoding method, characterized in that: The method comprises: Utilize the spatiotemporal characteristics of minimally invasive intracranial EEG signals to screen out multiple effective frequency bands; Using multiple effective frequency bands to jointly construct a covariance feature, the covariance feature is used to characterize the spatial pattern of minimally invasive electroencephalogram signals; Embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and use the Euclidean distance metric in the tangent plane space; Projecting the characteristic matrix of the covariance onto a tangent plane to obtain a characteristic vector of the covariance, and processing the characteristic vector based on machine learning classification to achieve decoding classification of the interface data; The method of screening out multiple effective frequency bands by utilizing the spatiotemporal characteristics of minimally invasive intracranial EEG signals includes: The ratio of the inter-class distance to the intra-class distance of the covariance matrix of each frequency band in motion and static state is calculated based on the affine invariant Riemannian distance metric to determine the separability of each frequency band; The Rayleigh quotient metric of linear discriminant analysis is extended to the Riemannian geometry distance metric, and a category separability metric index in the sense of Riemannian metric is obtained to screen the effective frequency bands that meet the classification contribution criteria.

2. The minimally invasive intracranial EEG signal decoding method according to claim 1, characterized in that: The method of jointly constructing a covariance feature using a plurality of effective frequency bands includes: According to the channel signal obtained after each effective frequency band is band-pass filtered, a covariance feature is constructed by jointly calculating the covariance of multiple frequency bands.

3. The minimally invasive intracranial EEG signal decoding method according to claim 2, characterized in that: After constructing the covariance feature by jointly calculating the covariance of multiple frequency bands, it also includes: Regularization processing is performed on the covariance feature.

4. The minimally invasive intracranial EEG signal decoding method according to claim 1, characterized in that: The method of embedding the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the geometric center point of the manifold using the AIRM Riemann distance metric, and using the Euclidean distance metric in the tangent plane space, includes: Calculate the geometric center point of the differentiable manifold formed by the characteristic matrix of the covariance using AIRM; Embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent space corresponding to the geometric center point; A Euclidean distance metric is used in the tangent space defined at the geometric center point of the differentiable manifold, wherein the Euclidean distance metric is isomorphic to AIRM.

5. The minimally invasive intracranial EEG signal decoding method according to claim 4, characterized in that: Before projecting the characteristic matrix of the covariance onto the tangent plane to obtain the characteristic vector of the covariance, the method further includes: A tangent plane is constructed at the Frechet mean point of the differentiable manifold.

6. A minimally invasive intracranial EEG signal decoding device, characterized in that: The device comprises: The frequency band screening module is used to: screen out multiple effective frequency bands by utilizing the spatiotemporal characteristics of minimally invasive intracranial EEG signals; A covariance feature construction module is used to: jointly construct a covariance feature using a plurality of the effective frequency bands, wherein the covariance feature is used to characterize the spatial pattern of the minimally invasive EEG signal; A spatial metric module is used to: embed the differentiable manifold formed by the characteristic matrix of the covariance into the tangent plane space of the manifold geometric center point using the AIRM Riemann distance metric, and use the Euclidean distance metric in the tangent plane space; A projection transformation module, configured to project the characteristic matrix of the covariance onto a tangent plane to obtain a characteristic vector of the covariance, and process the characteristic vector based on machine learning classification to implement decoding classification of the interface data; The frequency band screening module is specifically used for: The ratio of the inter-class distance to the intra-class distance of the covariance matrix of each frequency band in motion and static state is calculated based on the affine invariant Riemannian distance metric to determine the separability of each frequency band; The Rayleigh quotient metric of linear discriminant analysis is extended to the Riemannian geometry distance metric, and a category separability metric index in the sense of Riemannian metric is obtained to screen the effective frequency bands that meet the classification contribution criteria.

7. A therapeutic device, characterized in that include: A signal receiving module, configured to: receive intracranial EEG signals from a patient with spinal cord injury, and decode the patient's intracranial EEG signals based on the decoding method according to any one of claims 1 to 5; The peripheral control module is used to control the external device according to the obtained decoding result, wherein the external device includes at least one of a mechanical device and a computer device.

8. The therapeutic device according to claim 7, characterized in that The mechanical equipment control unit of the peripheral control module is used to: The motion state of the mechanical device is controlled according to the multi-dimensional control instructions in the decoding result.

9. The therapeutic device according to claim 7, characterized in that The peripheral control module includes a computer control unit for: The eight-directional motion state and click state of the cursor of the computer device are controlled according to the multi-dimensional control instructions in the decoding result.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the minimally invasive intracranial electroencephalogram signal decoding method according to any one of claims 1 to 5 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the minimally invasive intracranial electroencephalogram signal decoding method according to any one of claims 1 to 5 are implemented.

12. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the minimally invasive intracranial electroencephalogram signal decoding method according to any one of claims 1 to 5 are implemented.

Images