Methods, devices, equipment, and storage media for classifying users of motor imagery brain-computer interfaces.

Abstract

This application discloses a method, apparatus, device, and storage medium for classifying users of a motor imagery brain-computer interface. The method includes using independent component analysis (ICA) to decompose the user's target EEG signal, obtaining multiple initial independent components. Feature data is extracted from each initial independent component. Based on the feature data, a first target independent component corresponding to contralateral event-related desynchronization and a second target independent component corresponding to ipsilateral event-related synchronization are determined from the initial independent components. Finally, the classification result of the motor imagery brain-computer interface user is obtained. This scheme improves the accuracy and reliability of classification by using ICA-related neurodynamic modeling and classifying users based on multi-dimensional neural indicators obtained from feature data. Furthermore, by combining multiple feature data for automated selection of independent components, the subjectivity issues caused by manual intervention are avoided, improving the applicability and scalability of the classification method.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for classifying users of a motor imagination brain-computer interface. Background Technology

[0002] Brain-Computer Interface (BCI) is an innovative control technology that enables human-computer interaction without the need for peripheral nerve and muscle activity. Motor Imagery (MI)-BCI is an important branch, allowing users to interact with the outside world simply by imagining limb movements. However, there are significant individual differences in the motor imagery abilities of BCI users. Some users are unable to effectively generate distinguishable brainwave patterns, resulting in poor BCI performance; this is referred to as "BCI illiteracy."

[0003] Currently, existing BCI illiteracy discrimination methods typically categorize users as "excellent users" or "BCI illiterate" based on classification accuracy. However, numerous studies have shown that approximately 70% of users fall into the "intermediate user" range, with classification accuracy between 60% and 80%. While their performance exceeds random levels, it still falls short of reliable BCI control. Furthermore, existing research has found that these users exhibit an advantage in unilateral hand motor imagery while performing poorly on the other side. This unilateral advantage limits the training and classification effectiveness of existing BCI models.

[0004] Therefore, existing methods for classifying users of motor imagery brain-computer interfaces suffer from low classification accuracy. Summary of the Invention

[0005] This application aims to at least solve the technical problems existing in the prior art. To this end, the first aspect of this application proposes a method for classifying users of a motor imagery brain-computer interface, the method comprising:

[0006] For the user's left-hand motor imagery task and right-hand motor imagery task, the independent component analysis algorithm was used to decompose the user's target EEG signal to obtain multiple initial independent components.

[0007] For each initial independent component, feature extraction is performed to obtain feature data; the feature data includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent component.

[0008] Based on feature data, the first target independent component corresponding to the desynchronization of contralateral event correlation is determined from the initial independent components, and the second target independent component corresponding to the synchronization of same-side event correlation is determined.

[0009] Based on the independent components of the first objective, the independent components of the second objective, feature data, and preset performance indicators of the motor imagery brain-computer interface, the classification results of motor imagery brain-computer interface users are determined; among them, the classification results of motor imagery brain-computer interface users include the first classification result for the left-hand motor imagery task and the second classification result for the right-hand motor imagery task.

[0010] In one possible implementation, the process of generating the dipole position includes:

[0011] The initial independent components are fitted with equivalent current dipoles to generate the dipole positions corresponding to the initial independent components.

[0012] In one possible implementation, the process of generating relative mu / beta power includes:

[0013] Obtain the neural power spectral density corresponding to the initial independent components;

[0014] The neural power spectral density is parameterized to obtain at least one periodic component and one non-periodic component.

[0015] After removing interference from non-periodic components, the component with the largest amplitude among the periodic components is extracted as the relative mu / beta power.

[0016] In one possible implementation, the process of generating dynamic power changes includes:

[0017] The initial independent components are decomposed into time-frequency components using a pre-defined wavelet transform algorithm to obtain the decomposition results;

[0018] Within a preset time window, dynamic power changes are calculated based on the power value at a preset time and the average power value of a preset reference interval.

[0019] In one possible implementation, the process of generating salient event-related desynchronized event graphs and salient event-related synchronization graphs includes:

[0020] Random sampling is performed within the preset reference interval and the preset task interval to calculate the target statistic and target distribution.

[0021] Based on the target statistic and target distribution, the significance level corresponding to each voxel unit in the brain is estimated;

[0022] Based on the saliency level, the connected regions corresponding to adjacent salient voxel units are determined;

[0023] Based on connected regions, determine the desynchronized event graph and the synchronization graph related to significant events.

[0024] In one possible implementation, the method further includes:

[0025] Acquire raw brainwave signals; among which, raw brainwave signals are signals related to motor imagery;

[0026] Bandpass filtering was performed on the raw EEG signal to extract the oscillation signals in the mu and beta bands;

[0027] Extract time-domain segments within a preset time period from the oscillation signal; wherein the preset time period includes a preset reference interval and a preset task interval;

[0028] Artifact removal is performed on the time-domain segment to obtain the target EEG signal.

[0029] In one possible implementation, the method further includes:

[0030] Based on the first classification result and the second classification result, the corresponding target training strategy is determined;

[0031] The pre-defined brain-computer interface model is trained based on the target training strategy to obtain the training results.

[0032] A second aspect of this application discloses a user classification device for a motor imagery brain-computer interface, the device comprising:

[0033] The decomposition module is used to decompose the user's target EEG signal using independent component analysis algorithms for the user's left-hand motor imagery task and right-hand motor imagery task, respectively, to obtain multiple initial independent components.

[0034] The extraction module is used to perform feature extraction processing on each initial independent component to obtain feature data. The feature data includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent component.

[0035] The first determining module is used to determine, based on feature data, the first target independent component corresponding to the desynchronization of the opposite event and the second target independent component corresponding to the synchronization of the same event.

[0036] The second determining module is used to determine the classification results of motor imagery brain-computer interface users based on the first target independent component, the second target independent component, feature data, and preset motor imagery brain-computer interface performance indicators; wherein, the motor imagery brain-computer interface user classification results include a first classification result for the left-hand motor imagery task and a second classification result for the right-hand motor imagery task.

[0037] In one possible implementation, the above-described motor imagery brain-computer interface user classification device is further used for:

[0038] The process of generating dipole positions includes:

[0039] The initial independent components are fitted with equivalent current dipoles to generate the dipole positions corresponding to the initial independent components.

[0040] In one possible implementation, the above-described motor imagery brain-computer interface user classification device is further used for:

[0041] Obtain the neural power spectral density corresponding to the initial independent components;

[0042] The neural power spectral density is parameterized to obtain at least one periodic component and one non-periodic component.

[0043] After removing interference from non-periodic components, the component with the largest amplitude among the periodic components is extracted as the relative mu / beta power.

[0044] In one possible implementation, the above-described motor imagery brain-computer interface user classification device is further used for:

[0045] The initial independent components are decomposed into time-frequency components using a pre-defined wavelet transform algorithm to obtain the decomposition results;

[0046] Within a preset time window, dynamic power changes are calculated based on the power value at a preset time and the average power value of a preset reference interval.

[0047] In one possible implementation, the above-described motor imagery brain-computer interface user classification device is further used for:

[0048] Random sampling is performed within the preset reference interval and the preset task interval to calculate the target statistic and target distribution.

[0049] Based on the target statistic and target distribution, the significance level corresponding to each voxel unit in the brain is estimated;

[0050] Based on the saliency level, the connected regions corresponding to adjacent salient voxel units are determined;

[0051] Based on connected regions, determine the desynchronized event graph and the synchronization graph related to significant events.

[0052] In one possible implementation, the above-described motor imagery brain-computer interface user classification device is further used for:

[0053] Acquire raw brainwave signals; among which, raw brainwave signals are signals related to motor imagery;

[0054] Bandpass filtering was performed on the raw EEG signal to extract the oscillation signals in the mu and beta bands;

[0055] Extract time-domain segments within a preset time period from the oscillation signal; wherein the preset time period includes a preset reference interval and a preset task interval;

[0056] Artifact removal is performed on the time-domain segment to obtain the target EEG signal.

[0057] In one possible implementation, the above-described motor imagery brain-computer interface user classification device is further used for:

[0058] Based on the first classification result and the second classification result, the corresponding target training strategy is determined;

[0059] The pre-defined brain-computer interface model is trained based on the target training strategy to obtain the training results.

[0060] A third aspect of this application provides an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the motor imagery brain-computer interface user classification method as described in the first aspect.

[0061] The fourth aspect of this application proposes a computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the motor imagery brain-computer interface user classification method as described in the first aspect.

[0062] The embodiments of this application have the following beneficial effects:

[0063] The method for classifying users of a motor imagery brain-computer interface provided in this application includes: for a user's left-hand motor imagery task and right-hand motor imagery task, using an independent component analysis algorithm to decompose the user's target EEG signal to obtain multiple initial independent components; for each initial independent component, performing feature extraction processing to obtain feature data, wherein the feature data includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent components; based on the feature data, determining the first target independent component corresponding to contralateral event-related desynchronization from the initial independent components, and determining the second target independent component corresponding to ipsilateral event-related synchronization; and determining the motor imagery brain-computer interface user classification result based on the first target independent component, the second target independent component, the feature data, and preset motor imagery brain-computer interface performance indicators. This solution improves the accuracy and reliability of user classification by modeling independent component-related brain dynamics and classifying users based on multi-dimensional neural indicators obtained from feature data. In addition, by combining multiple feature data for automated screening of independent components, it avoids the subjectivity problems caused by human intervention and improves the applicability and scalability of the classification method. Attached Figure Description

[0064] Figure 1 A block diagram of a computer device provided in an embodiment of this application;

[0065] Figure 2 A flowchart illustrating the steps of a brain-computer interface user classification method for motor imagery provided in this application embodiment;

[0066] Figure 3 A flowchart illustrating the steps for generating a target electroencephalogram (EEG) signal, provided in an embodiment of this application;

[0067] Figure 4 A flowchart illustrating the steps for extracting relative mu / beta power is provided in this application embodiment;

[0068] Figure 5 A flowchart illustrating the steps for extracting dynamic power changes is provided in this embodiment of the application.

[0069] Figure 6 A flowchart illustrating the steps for extracting a desynchronized event graph and a synchronization graph related to salient events, as provided in this application embodiment;

[0070] Figure 7 This is a structural block diagram of a motor imagery brain-computer interface user classification device provided in an embodiment of this application. Detailed Implementation

[0071] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0072] Brain-Computer Interface (BCI) is an innovative control technology that enables human-computer interaction without the need for peripheral nerve and muscle activity. Motor Imagery (MI)-BCI is an important branch, allowing users to interact with the outside world simply by imagining limb movements. However, there are significant individual differences in the motor imagery abilities of BCI users. Some users cannot effectively generate distinguishable EEG patterns, resulting in poor BCI performance; this is referred to as "BCI illiteracy." Currently, existing BCI illiteracy identification methods typically categorize users into "excellent users" and "BCI illiterate" based on classification accuracy. However, numerous studies have shown that approximately 70% of users fall into the "intermediate user" range with 60%-80% classification accuracy. While their performance exceeds random levels, they still cannot achieve reliable BCI control. Furthermore, existing research has found that these users exhibit an advantage in unilateral hand motor imagery while performing poorly on the other side. This unilateral advantage limits the training and classification effectiveness of existing BCI models.

[0073] Furthermore, Event-Related Desynchronization (ERD) and Event-Related Synchronization (ERS) reflect task-related decreases and increases in EEG power, respectively, and are important markers of cortical activation and deactivation. In actual and imagined movement, ERD typically appears in the mu (8-13Hz) and beta (13-30Hz) bands of the contralateral motor region, while ERS may appear in the ipsilateral region associated with movement cessation or attention shift and is widely considered a reliable indicator of MI involvement. However, most existing studies rely on channel-level analysis. Some studies have introduced an entropy method based on vector quantization patterns to estimate ERD / ERS. In addition, researchers have attempted to combine sensorimotor rhythm power with phase synchronicity in the mu and beta bands to improve assessment accuracy, further demonstrating that directional connectivity patterns in the mu and beta bands are closely related to individual MI capacity.

[0074] In recent years, some studies have attempted to use Independent Component Analysis (ICA) to separate the sources of EEG signals and extract independent components (ICs) related to motor imagery by combining time-frequency features and power spectrum analysis. These studies extracted the topological map, time-frequency features, and dipole location of each IC, and used clustering methods to divide them into multiple groups, selecting ERD-related groups through visual testing. Other studies have ranked ICs according to their discrimination ability in left-hand motor imagery, right-hand motor imagery, and resting state, selecting the ICs with the best classification performance. Still other studies have used stimulus-response-based trial coherence to classify ICs into sensorimotor, sensory, motor, or unspecified groups, proposing several criteria for selecting sensorimotor ICs: central scalp distribution, dipole localization in the precentral and postcentral gyri, and mu band inhibition.

[0075] However, existing BCI assessments of motor imagery mostly use overall classification accuracy as the standard, failing to fully consider the neural characteristics of users in the intermediate performance range (60%-80% accuracy), thus hindering the targeted improvement of training strategies. Traditional ERD / ERS assessments largely rely on the EEG channel level, which is susceptible to interference from volume conduction and other factors, making it difficult to accurately reflect the actual neural activity in the cortex and affecting assessment accuracy. Most existing technologies fail to effectively distinguish between periodic neural oscillations and non-periodic background signals, leading to interference with ERD / ERS analysis results and affecting the extraction and classification of user features. Existing ICA methods mostly rely on manual experience or simple statistical indicators for IC selection, lacking an automated analysis process that combines spatial localization, ERD / ERS dynamic characteristics, and power spectrum estimation, making it difficult to stably extract independent components directly related to motor imagery, thus limiting the universality and interpretability of the methods.

[0076] Based on this, this application proposes a method for classifying users of a motor imagery brain-computer interface. The method includes: for a user's left-hand and right-hand motor imagery tasks, using independent component analysis (ICA) to decompose the user's target EEG signals to obtain multiple initial independent components; for each initial independent component, performing feature extraction to obtain feature data, which includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent components; based on the feature data, determining the first target independent component corresponding to contralateral event-related desynchronization and the second target independent component corresponding to ipsilateral event-related synchronization from the initial independent components; and determining the user classification result of the motor imagery brain-computer interface based on the first target independent component, the second target independent component, the feature data, and preset motor imagery brain-computer interface performance indicators. This solution improves the accuracy and reliability of user classification by modeling independent component-related brain dynamics and classifying users based on multi-dimensional neural indicators obtained from feature data. In addition, by combining multiple feature data for automated screening of independent components, it avoids the subjectivity problems caused by human intervention and improves the applicability and scalability of the classification method.

[0077] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more. Furthermore, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0078] The motor imagery brain-computer interface user classification method provided in this application can be applied to computer devices (electronic devices). The computer device can be a server or a terminal. The server can be a single server or a server cluster composed of multiple servers. This application does not specifically limit this. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices.

[0079] Taking a computer device as an example, Figure 1 A block diagram of a server is shown, such as Figure 1As shown, the server may include a processor and memory connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. When the computer program is executed by the processor, it implements a motor imagery brain-computer interface user classification method.

[0080] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the server to which the present application is applied. Optionally, the server may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0081] It should be noted that the execution subject of the embodiments of this application can be a computer device or a motor imagery brain-computer interface user classification device. The following method embodiments will be described with a computer device as the execution subject.

[0082] Figure 2 This is a flowchart illustrating the steps of a user classification method for a motor imagery brain-computer interface provided in an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0083] Step 202: For the user's left-hand motor imagery task and right-hand motor imagery task, the independent component analysis algorithm is used to decompose the user's target EEG signal to obtain multiple initial independent components.

[0084] The user's target EEG signal is obtained by preprocessing the raw EEG signal. In some optional embodiments, such as... Figure 3 As shown, Figure 3 A flowchart illustrating the steps for generating a target electroencephalogram (EEG) signal, as provided in this application embodiment, includes:

[0085] Step 302: Acquire the raw EEG signals.

[0086] Step 304: Perform bandpass filtering on the original EEG signal to extract the oscillation signals in the mu and beta bands.

[0087] Step 306: Extract time-domain segments within a preset time period from the oscillation signal.

[0088] Step 308: Perform artifact removal processing on the time domain segment to obtain the target EEG signal.

[0089] Among them, the raw EEG signals are signals related to motor imagery, and corresponding signals can be collected from channels related to motor imagery, including the frontal lobe, central parietal lobe, and posterior parietal lobe.

[0090] Next, the raw EEG signal can be bandpass filtered to extract the oscillation signals in the mu and beta bands. Then, time-domain segments within a preset time period corresponding to each trial are extracted. The preset time period includes a preset reference interval and a preset task interval. For example, the preset time period can be -1 second to +4 seconds, the preset reference interval can be -1 second to 0 seconds, and the preset task interval can be 0 seconds to +4 seconds.

[0091] Finally, artifact removal processing can be performed on the time-domain segment to obtain the target EEG signal. Optionally, the artifact removal process can include, but is not limited to, artifact removal, trend correction, and outlier detection. In this way, the signal quality of the obtained target EEG signal can be guaranteed.

[0092] Therefore, for left-hand and right-hand motor imagery tasks, independent component analysis (ICA) algorithms can be used to decompose the user's target EEG signals to obtain multiple initial independent components. Specifically, ICA algorithms can separate independent neural activities from highly overlapping scalp signals using statistical methods. Optionally, an algorithm based on maximizing information transmission can be used to achieve signal separation by maximizing signal entropy. This algorithm calculates the unmixing matrix W based on the principle of information maximization to recover the potential source signal from the recorded target EEG signals.

[0093] Then, for each user's two different tasks—a left-hand motor imagery task and a right-hand motor imagery task—the target EEG signals from the selected preset number of channels are decomposed and processed to obtain the corresponding preset number of independent components. Optionally, the preset number is determined by the number of channels in the EEG acquisition device; for example, the preset number can be 30.

[0094] Step 204: For each initial independent component, perform feature extraction processing to obtain feature data.

[0095] The feature data may include the dipole position corresponding to the initial independent component, the relative mu / beta power, the dynamic power change, the significant event-related desynchronization event map, and the significant event-related synchronization map.

[0096] In some optional embodiments, when extracting dipole positions, the dipole positions corresponding to the initial independent components can be generated by performing equivalent current dipole fitting on the initial independent components. Optionally, the DIPFIT2 plugin can be used to perform equivalent current dipole fitting on the initial independent components. The specific implementation process of this plugin can use existing technology, which will not be elaborated here, thereby generating the dipole positions corresponding to the initial independent components. By combining dipole localization technology, the automated screening and localization of spatially independent motion-related neural sources are achieved, reducing the interference of volume conduction at the source and significantly improving the spatial resolution and interpretability of the signal.

[0097] Additionally, initial independent components with residual variance greater than 15% can be removed, as these are unlikely to originate from a single neural source. Furthermore, initial independent components with localized activation limited to a single electrode (such as electromyographic artifacts, which are high-frequency electrical activities exhibiting spike-like peaks) can also be removed.

[0098] In some alternative embodiments, when extracting the relative mu / beta power, such as Figure 4 As shown, Figure 4 A flowchart illustrating the steps for extracting relative mu / beta power provided in this application embodiment includes:

[0099] Step 402: Obtain the neural power spectral density corresponding to the initial independent components.

[0100] Step 404: Parameterize the neural power spectral density to obtain at least one periodic component and one non-periodic component.

[0101] Step 406: After removing the interference of non-periodic components, extract the component with the largest amplitude among the periodic components as the relative mu / beta power.

[0102] The neural power spectral density can be parameterized using the Fitting Oscillations & One-Over-F (FOOOF) algorithm to obtain at least one periodic component and an aperiodic component. The oscillation peak is extracted by identifying the spectral peaks that exceed the modeled aperiodic background, thereby obtaining a relative narrowband power estimate with aperiodic interference removed, thus eliminating the interference of the aperiodic component.

[0103] Therefore, after removing interference from aperiodic components, the component with the largest amplitude among the extracted periodic components can be used as the relative mu / beta power. By effectively separating the periodic neural oscillation component from the aperiodic background signal, the extracted feature data is more realistic and stable, solving the problem of traditional narrowband analysis being easily interfered with by background signals.

[0104] In some alternative embodiments, when extracting dynamic power changes, such as Figure 5 As shown, Figure 5 A flowchart illustrating the steps for extracting dynamic power changes provided in this application embodiment includes:

[0105] Step 502: Use a preset wavelet transform algorithm to perform time-frequency decomposition on the initial independent components to obtain the decomposition results.

[0106] Step 504: Within a preset time window, calculate the dynamic power change based on the power value at a preset time and the average power value of a preset reference interval.

[0107] Among them, dynamic power change can be denoted as ERD%, ERS%, and relative ERD power can be defined as the percentage decrease in power relative to the preset reference range, and relative ERS power can be defined as the percentage increase in power relative to the preset reference range.

[0108] Optionally, the preset wavelet transform algorithm can be the Morlet wavelet transform algorithm, which can be set to 3 periods and use the Hanning window to perform time-frequency decomposition on the initial independent component time series and obtain the decomposition result. The specific time-frequency decomposition process can refer to the existing technology, and will not be elaborated here.

[0109] Therefore, dynamic power changes can be calculated within a preset time window based on the power value at a preset time and the average power value of a preset reference interval. Optionally, ERD% and ERS% are equal to the power value at the preset time t minus the average power value of the preset reference interval, then divided by the average power value of the preset reference interval, and then multiplied by 100%.

[0110] In some optional embodiments, when extracting salient event-related desynchronization event graphs and salient event-related synchronization graphs, such as Figure 6 As shown, Figure 6 A flowchart illustrating the steps for extracting salient event-related desynchronized event graphs and salient event-related synchronization graphs, as provided in this application embodiment, includes:

[0111] Step 602: Random sampling is performed within the preset reference interval and the preset task interval to calculate the target statistic and target distribution.

[0112] Step 604: Based on the target statistic and target distribution, estimate the significance level of each voxel unit in the brain.

[0113] Step 606: Based on the saliency level, determine the connected regions corresponding to adjacent salient voxel units.

[0114] Step 608: Based on the connected regions, determine the desynchronized event graph and the synchronization graph related to significant events.

[0115] By conducting extensive random sampling within a preset reference interval and a preset task interval, the target statistic and target distribution can be calculated. The target statistic is a pseudo-t-statistic, and the target distribution is the distribution of pseudo-t-values.

[0116] Next, based on the target statistics and target distribution, the significance level of each voxel unit in the brain can be estimated. For each significant voxel unit, the connectivity of its adjacent voxels in the horizontal direction (representing time) and vertical direction (representing frequency) is further analyzed, and adjacent significant voxel units are classified as connected regions.

[0117] Then, based on the connected regions, the average power change value of each 4-connected region can be calculated. For regions with an average value less than zero, they are defined as significant event-related desynchronization event regions, i.e., significant ERD regions; for regions with an average value greater than zero, they are defined as significant event-related synchronization regions, i.e., significant ERS regions.

[0118] The area of ​​each significant ERD region or significant ERS region is represented by the number of significant voxel units it contains. For each initial independent component, the connected region with the largest area can be selected as the representative significant ERD region or significant ERS region, thus determining the final significant event-related desynchronization event graph and significant event-related synchronization graph.

[0119] Step 206: Based on the feature data, determine the first target independent component corresponding to the desynchronization of the opposite side event from the initial independent components, and determine the second target independent component corresponding to the synchronization of the same side event.

[0120] In order to identify independent components related to hand motor imagery, that is, independent components that can present contralateral ERD or ipsilateral ERS, the first target independent component corresponding to contralateral event-related desynchronization can be determined from the initial independent components based on feature data, and the second target independent component corresponding to ipsilateral event-related synchronization can be determined, thereby achieving automated screening of the initial independent components.

[0121] Optionally, the selection rules for the first objective independent component corresponding to contra-event-related desynchronization include:

[0122] 1. The dipole location corresponding to the initial independent component must be located in the brain regions related to contralateral motor imagery, including the contralateral somatosensory cortex (BA1-3), primary motor cortex (BA4), superior parietal lobule (BA5, BA7), premotor and supplementary motor areas (BA6), anterior cingulate cortex (BA24, BA32), fusiform gyrus somatic selection area (BA37), inferior parietal lobule (BA40), insular operculum (BA44, BA45), and middle frontal gyrus (BA9, BA46). The dipole location is usually represented by three-dimensional coordinates, and different brain regions also have corresponding three-dimensional coordinate ranges, thereby determining the first independent component candidate set located in the brain regions related to contralateral motor imagery.

[0123] 2. In the first independent component candidate set mentioned above, select the top three independent components with relatively high mu / beta power to form the second independent component candidate set.

[0124] 3. In the aforementioned second independent component candidate set, the independent component with the largest 4-connected region area in the significant event-related desynchronization event graph is selected as the first target independent component. If there is no corresponding significant event-related desynchronization event graph in the aforementioned second independent component candidate set, then the independent component with the highest relative mu / beta power is selected as the first target independent component.

[0125] Optionally, the selection rules for the second objective independent component corresponding to the synchronization of events on the same side include:

[0126] 1. The dipole positions corresponding to the initial independent components must be located in the brain regions related to ipsilateral motor imagery, thus obtaining a third set of independent component candidates located in the brain regions related to ipsilateral motor imagery.

[0127] 2. From the aforementioned third independent component candidate set, select the independent components that exhibit significant oscillation peaks within the mu / beta frequency range relative to the mu / beta power as the fourth independent component candidate set. The component exhibiting significant oscillation peaks is defined as having at least one periodic component.

[0128] 3. In the aforementioned fourth independent component candidate set, the independent component with the largest 4-connected region area in the significant event-related synchronization graph is selected as the second target independent component. If there is no corresponding significant event-related synchronization graph in the aforementioned fourth independent component candidate set, then the independent component with the highest relative mu / beta power is selected as the second target independent component.

[0129] Step 208: Based on the first target independent component, the second target independent component, feature data, and preset motor imagery brain-computer interface performance indicators, determine the motor imagery brain-computer interface user classification results.

[0130] The classification results of the motor imagery brain-computer interface users can include a first classification result for the left-hand motor imagery task and a second classification result for the right-hand motor imagery task.

[0131] For the left-hand motor imagery task, the ERD% of the first target independent component corresponding to the contralateral event-related desynchronization during the left-hand motor imagery can be selected as the input features, along with the relative ERS power obtained from the extracted feature data and the preset motor imagery brain-computer interface performance index ClassDis. Then, the k-means clustering algorithm (k=2) is used for clustering.

[0132] To quantify the performance of motor imagery brain-computer interface, a preset performance index, ClassDis, also known as the class separability index, can be used. The class separability index between two EEG patterns is measured by Fisher's discriminant criterion applicable to the covariance matrix Riemannian geometric space. A higher ClassDis value indicates a greater degree of separation between categories and less variability within categories, thus reflecting better MI-BCI performance.

[0133] Therefore, users with higher ClassDis values ​​are classified as high performers in left-hand motor imagery, while users with lower ClassDis values ​​are classified as low performers in left-hand motor imagery.

[0134] The same method was also used in the right-hand motor imagery task. Based on the clustering results, users who performed well in both left-hand and right-hand motor imagery were marked as "excellent performers"; those who performed well in the left hand but poorly in the right hand were marked as "left-excellent and right-poor users"; those who performed well in the right hand but poorly in the left hand were marked as "right-excellent and left-poor users"; and those who performed poorly in both hands were marked as "poor performers".

[0135] In some optional embodiments, personalized training schemes can be formulated for different groups of users. That is, based on the first classification result and the second classification result, the corresponding target training strategy can be determined, and then the preset brain-computer interface model can be trained based on the target training strategy to obtain the training result.

[0136] The target training strategies include: 1. For users with equal ambidextrous strength, no additional training is required to effectively control the MI-BCI system; 2. For users with unilateral dominance, training is conducted on the weaker hand, using the ipsilateral relative mu power of the weaker hand as the real-time feedback object; 3. For users with equal ambidextrous weakness, methods such as emotional state BCI or visual evoked potential paradigms are recommended. Finally, after determining the target training strategies, the preset brain-computer interface model can be trained based on these strategies to obtain training results. By developing personalized training programs for different user groups, the overall applicability and universality of the BCI system are improved.

[0137] In addition, this application conducted specific experiments. The results showed that 27 users (24.8% of the total) were classified as high performers. The "left-handed high-right-handed poor" group consisted of 30 users (27.5%), the "right-handed high-left-handed poor" group consisted of 13 users (11.9%), and the remaining 39 users (35.8%) were classified as low performers. High performers showed higher values ​​across all four dimensions, demonstrating stronger ERD% and relative ERS power responses. In contrast, low performers had the lowest values ​​across all dimensions. The "left-handed high-right-handed poor" group showed higher relative ERS power during the left-handed motor imagery task and lower values ​​during the right-handed motor imagery task; while the "right-handed high-left-handed poor" group showed the opposite pattern, with higher relative ERS power during the right-handed task and lower relative ERS power during the left-handed task.

[0138] This application provides a method for classifying users of a motor imagery brain-computer interface (BCI). The method includes: for left-hand and right-hand motor imagery tasks, using independent component analysis (ICA) to decompose the user's target EEG signals to obtain multiple initial independent components; for each initial independent component, performing feature extraction to obtain feature data, which includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent components; based on the feature data, determining the first target independent component corresponding to contralateral event-related desynchronization and the second target independent component corresponding to ipsilateral event-related synchronization from the initial independent components; and determining the classification result of the motor imagery BCI user based on the first target independent component, the second target independent component, the feature data, and preset motor imagery BCI performance indicators. This scheme improves the accuracy and reliability of classification by using independent component-related neurodynamic modeling and classifying users based on multi-dimensional neural indicators obtained from feature data. Furthermore, by combining multiple feature data for automated selection of independent components, it avoids the subjectivity problems caused by manual intervention, improving the applicability and scalability of the classification method.

[0139] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0140] Figure 7 This is a structural block diagram of a motor imagery brain-computer interface user classification device provided in an embodiment of this application.

[0141] like Figure 7 As shown, the motor imagery brain-computer interface user classification device 700 includes:

[0142] The decomposition module 702 is used to decompose the user's target EEG signal using an independent component analysis algorithm for the user's left-hand motor imagery task and right-hand motor imagery task, respectively, to obtain multiple initial independent components.

[0143] The extraction module 704 is used to perform feature extraction processing on each initial independent component to obtain feature data; wherein, the feature data includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent component.

[0144] The first determining module 706 is used to determine, based on feature data, a first target independent component corresponding to the desynchronization of the opposite event from the initial independent components, and a second target independent component corresponding to the synchronization of the same event.

[0145] The second determining module 708 is used to determine the classification results of the motor imagery brain-computer interface user based on the first target independent component, the second target independent component, feature data and preset motor imagery brain-computer interface performance indicators; wherein, the motor imagery brain-computer interface user classification results include a first classification result for the left-hand motor imagery task and a second classification result for the right-hand motor imagery task.

[0146] Regarding the apparatus in the above embodiments, the specific methods by which each module performs its operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here. Each module in the above-described motor imagery brain-computer interface user classification device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations of each module.

[0147] In one embodiment of this application, a computer device is provided, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0148] For the user's left-hand motor imagery task and right-hand motor imagery task, the independent component analysis algorithm was used to decompose the user's target EEG signal to obtain multiple initial independent components.

[0149] For each initial independent component, feature extraction is performed to obtain feature data; the feature data includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent component.

[0150] Based on feature data, the first target independent component corresponding to the desynchronization of contralateral event correlation is determined from the initial independent components, and the second target independent component corresponding to the synchronization of same-side event correlation is determined.

[0151] Based on the independent components of the first objective, the independent components of the second objective, feature data, and preset performance indicators of the motor imagery brain-computer interface, the classification results of motor imagery brain-computer interface users are determined; among them, the classification results of motor imagery brain-computer interface users include the first classification result for the left-hand motor imagery task and the second classification result for the right-hand motor imagery task.

[0152] In one embodiment of this application, the processor further performs the following steps when executing the computer program:

[0153] The initial independent components are fitted with equivalent current dipoles to generate the dipole positions corresponding to the initial independent components.

[0154] In one embodiment of this application, the processor further performs the following steps when executing the computer program:

[0155] Obtain the neural power spectral density corresponding to the initial independent components;

[0156] The neural power spectral density is parameterized to obtain at least one periodic component and one non-periodic component.

[0157] After removing interference from non-periodic components, the component with the largest amplitude among the periodic components is extracted as the relative mu / beta power.

[0158] In one embodiment of this application, the processor further performs the following steps when executing the computer program:

[0159] The initial independent components are decomposed into time-frequency components using a pre-defined wavelet transform algorithm to obtain the decomposition results;

[0160] Within a preset time window, dynamic power changes are calculated based on the power value at a preset time and the average power value of a preset reference interval.

[0161] In one embodiment of this application, the processor further performs the following steps when executing the computer program:

[0162] Random sampling is performed within the preset reference interval and the preset task interval to calculate the target statistic and target distribution.

[0163] Based on the target statistic and target distribution, the significance level corresponding to each voxel unit in the brain is estimated;

[0164] Based on the saliency level, the connected regions corresponding to adjacent salient voxel units are determined;

[0165] Based on connected regions, determine the desynchronized event graph and the synchronization graph related to significant events.

[0166] In one embodiment of this application, the processor further performs the following steps when executing the computer program:

[0167] Acquire raw brainwave signals; among which, raw brainwave signals are signals related to motor imagery;

[0168] Bandpass filtering was performed on the raw EEG signal to extract the oscillation signals in the mu and beta bands;

[0169] Extract time-domain segments within a preset time period from the oscillation signal; wherein the preset time period includes a preset reference interval and a preset task interval;

[0170] Artifact removal is performed on the time-domain segment to obtain the target EEG signal.

[0171] In one embodiment of this application, the processor further performs the following steps when executing the computer program:

[0172] Based on the first classification result and the second classification result, the corresponding target training strategy is determined;

[0173] The pre-defined brain-computer interface model is trained based on the target training strategy to obtain the training results.

[0174] The computer device provided in this application embodiment has a similar implementation principle and technical effect to the above method embodiment, and will not be described again here.

[0175] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, it performs the following steps:

[0176] For the user's left-hand motor imagery task and right-hand motor imagery task, the independent component analysis algorithm was used to decompose the user's target EEG signal to obtain multiple initial independent components.

[0177] For each initial independent component, feature extraction is performed to obtain feature data; the feature data includes the dipole position, relative mu / beta power, dynamic power change, significant event-related desynchronization event map, and significant event-related synchronization map corresponding to the initial independent component.

[0178] Based on feature data, the first target independent component corresponding to the desynchronization of contralateral event correlation is determined from the initial independent components, and the second target independent component corresponding to the synchronization of same-side event correlation is determined.

[0179] Based on the independent components of the first objective, the independent components of the second objective, feature data, and preset performance indicators of the motor imagery brain-computer interface, the classification results of motor imagery brain-computer interface users are determined; among them, the classification results of motor imagery brain-computer interface users include the first classification result for the left-hand motor imagery task and the second classification result for the right-hand motor imagery task.

[0180] In one embodiment of this application, the computer program, when executed by a processor, further performs the following steps:

[0181] The initial independent components are fitted with equivalent current dipoles to generate the dipole positions corresponding to the initial independent components.

[0182] In one embodiment of this application, the computer program, when executed by a processor, further performs the following steps:

[0183] Obtain the neural power spectral density corresponding to the initial independent components;

[0184] The neural power spectral density is parameterized to obtain at least one periodic component and one non-periodic component.

[0185] After removing interference from non-periodic components, the component with the largest amplitude among the periodic components is extracted as the relative mu / beta power.

[0186] In one embodiment of this application, the computer program, when executed by a processor, further performs the following steps:

[0187] The initial independent components are decomposed into time-frequency components using a pre-defined wavelet transform algorithm to obtain the decomposition results;

[0188] Within a preset time window, dynamic power changes are calculated based on the power value at a preset time and the average power value of a preset reference interval.

[0189] In one embodiment of this application, the computer program, when executed by a processor, further performs the following steps:

[0190] Random sampling is performed within the preset reference interval and the preset task interval to calculate the target statistic and target distribution.

[0191] Based on the target statistic and target distribution, the significance level corresponding to each voxel unit in the brain is estimated;

[0192] Based on the saliency level, the connected regions corresponding to adjacent salient voxel units are determined;

[0193] Based on connected regions, determine the desynchronized event graph and the synchronization graph related to significant events.

[0194] In one embodiment of this application, the computer program, when executed by a processor, further performs the following steps:

[0195] Acquire raw brainwave signals; among which, raw brainwave signals are signals related to motor imagery;

[0196] Bandpass filtering was performed on the raw EEG signal to extract the oscillation signals in the mu and beta bands;

[0197] Extract time-domain segments within a preset time period from the oscillation signal; wherein the preset time period includes a preset reference interval and a preset task interval;

[0198] Artifact removal is performed on the time-domain segment to obtain the target EEG signal.

[0199] In one embodiment of this application, the computer program, when executed by a processor, further performs the following steps:

[0200] Based on the first classification result and the second classification result, the corresponding target training strategy is determined;

[0201] The pre-defined brain-computer interface model is trained based on the target training strategy to obtain the training results.

[0202] The computer-readable storage medium provided in this embodiment is similar in principle and technical effect to the method embodiment described above, and will not be repeated here.

[0203] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0204] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0205] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A motor imagery brain-computer interface user classification method, characterized in that, The method comprises: For the left-hand motor imagery task and the right-hand motor imagery task of a user, an independent component analysis algorithm is used to decompose and process target brain wave signals of the user, to obtain a plurality of initial independent components; For each of the initial independent components, feature extraction processing is performed on the initial independent components to obtain feature data; wherein the feature data includes a dipole position corresponding to the initial independent component, a relative mu / beta power, a dynamic power change, a significant event-related desynchronization event map, and a significant event-related synchronization map; Based on the feature data, a first target independent component corresponding to contralateral event-related desynchronization is determined from the initial independent components, and a second target independent component corresponding to ipsilateral event-related synchronization is determined; wherein the selection rule of the first target independent component includes: the dipole position corresponding to the initial independent component is located in the brain region related to contralateral motor imagery, to determine a first independent component candidate set located in the brain region related to contralateral motor imagery; in the first independent component candidate set, the first three independent components with high relative mu / beta power are selected to form a second independent component candidate set; in the second independent component candidate set, the independent component with the largest 4-connected region area in the significant event-related desynchronization event map is selected as the first target independent component, and if there is no corresponding significant event-related desynchronization event map in the second independent component candidate set, the independent component with the highest relative mu / beta power is selected as the first target independent component; the selection rule of the second target independent component includes: the dipole position corresponding to the initial independent component is located in the brain region related to ipsilateral motor imagery, to determine a third independent component candidate set located in the brain region related to ipsilateral motor imagery; in the third independent component candidate set, the independent component with a significant oscillation peak in the mu / beta frequency range in the relative mu / beta power is selected as a fourth independent component candidate set; in the fourth independent component candidate set, the independent component with the largest 4-connected region area in the significant event-related synchronization map is selected as the second target independent component, and if there is no corresponding significant event-related synchronization map in the fourth independent component candidate set, the independent component with the highest relative mu / beta power is selected as the second target independent component; Based on the first target independent component, the second target independent component, the feature data, and a preset motor imagery brain-computer interface performance indicator, a motor imagery brain-computer interface user classification result is determined; wherein the motor imagery brain-computer interface user classification result includes a first classification result for the left-hand motor imagery task and a second classification result for the right-hand motor imagery task.

2. The method of claim 1, wherein, The generation process of the dipole position comprises: Equivalent current dipole fitting processing is performed on the initial independent component to generate a dipole position corresponding to the initial independent component.

3. The method of claim 1, wherein, The generation process of the relative mu / beta power comprises: The neural power spectral density corresponding to the initial independent component is obtained; Parameterize the neural power spectral density to obtain at least one periodic component and a non-periodic component; After removing the interference of the non-periodic component, extract the component with the largest amplitude in the periodic component as the relative mu / beta power.

4. The method of claim 1, wherein, The generation process of the dynamic power change includes: Perform time-frequency decomposition on the initial independent components using a preset wavelet transform algorithm to obtain a decomposition result; In a preset time window, the dynamic power change is calculated based on the power value at a preset time and the power mean value of a preset reference interval.

5. The method of claim 1, wherein, The generation process of the significant event-related desynchronization event map and the significant event-related synchronization map includes: Random sampling is performed in a preset reference interval and a preset task interval to calculate a target statistic and a target distribution; Based on the target statistic and the target distribution, the significance level corresponding to each voxel unit in the brain is estimated; Based on the significance level, the connected region corresponding to adjacent significant voxel units is determined; Based on the connected region, the significant event-related desynchronization event map and the significant event-related synchronization map are determined.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Obtain the collected original brain wave signal; wherein the original brain wave signal is a signal related to motor imagery; Perform band-pass filtering on the original brain wave signal to extract the oscillation signal in the mu and beta frequency bands; Segment the time domain segment in the preset time period from the oscillation signal; wherein the preset time period includes a preset reference interval and a preset task interval; Perform artifact removal on the time domain segment to obtain the target brain wave signal.

7. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Based on the first classification result and the second classification result, determine the corresponding target training strategy; Based on the target training strategy, train the preset brain-computer interface model to obtain a training result.

8. A motor imagery brain-computer interface user classification apparatus characterized by, The device includes: A decomposition module configured to, for left-hand motor imagery tasks and right-hand motor imagery tasks of a user, respectively perform decomposition processing on the target brain wave signal of the user using an independent component analysis algorithm to obtain a plurality of initial independent components; An extraction module configured to, for each of the initial independent components, perform feature extraction processing on the initial independent component to obtain feature data; wherein the feature data includes the dipole position, the relative mu / beta power, the dynamic power change, the significant event-related desynchronization event map, and the significant event-related synchronization map corresponding to the initial independent component. The first determining module is configured to determine a first target independent component corresponding to contralateral event-related desynchronization and a second target independent component corresponding to ipsilateral event-related synchronization from the initial independent components based on the feature data; wherein the selection rule of the first target independent component comprises that the dipole position corresponding to the initial independent component is located in a brain region related to contralateral motor imagery, a first independent component candidate set located in the brain region related to contralateral motor imagery is determined, the first independent component candidate set is selected to form a second independent component candidate set, and the second independent component candidate set is selected to determine the first target independent component with the largest 4-connected region area in a significant event-related desynchronization event graph, if none of the second independent component candidate set has a corresponding significant event-related desynchronization event graph, the independent component with the highest mu / beta power is selected as the first target independent component; the selection rule of the second target independent component comprises that the dipole position corresponding to the initial independent component is located in a brain region related to ipsilateral motor imagery, a third independent component candidate set located in the brain region related to ipsilateral motor imagery is obtained, the independent component with a significant oscillation peak in the mu / beta frequency range is selected from the third independent component candidate set to form a fourth independent component candidate set, and the fourth independent component candidate set is selected to determine the second target independent component with the largest 4-connected region area in a significant event-related synchronization graph, if none of the fourth independent component candidate set has a corresponding significant event-related synchronization graph, the independent component with the highest mu / beta power is selected as the second target independent component. The second determining module is configured to determine a motor imagery brain-computer interface user classification result based on the first target independent component, the second target independent component, the feature data, and a preset motor imagery brain-computer interface performance index; wherein the motor imagery brain-computer interface user classification result comprises a first classification result for the left-hand motor imagery task and a second classification result for the right-hand motor imagery task.

9. An electronic device, comprising: The electronic device comprises a processor and a memory, and the memory stores at least one instruction, at least one program, a code set, or an instruction set, which are loaded and executed by the processor to implement the motor imagery brain-computer interface user classification method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, which are loaded and executed by the processor to implement the motor imagery brain-computer interface user classification method according to any one of claims 1-7.

Images