Micro-state analysis-oriented electroencephalogram few-channel discovery processing method, system and device, processor and readable storage medium thereof

Abstract

The invention relates to a method for realizing electroencephalogram few-channel discovery processing for micro-state analysis. The method comprises the following steps: preprocessing collected resting-state electroencephalogram data; a micro-state monitoring module is constructed, training is carried out according to sequence samples, a channel importance generation sorting module is constructed, and few channels which are non-uniformly distributed are found in combination with an attribution analysis technology based on input gradients; and the retest reliability of the few-channel micro-state is verified. By adopting the method, the system, the device, the processor and the computer readable storage medium for realizing electroencephalogram few-channel discovery processing for micro-state analysis, the electroencephalogram few-channel discovery method for micro-state analysis provided by the invention has the advantages that an electroencephalogram micro-state monitoring task is modeled from the perspective of interpretable deep learning for the first time; compared with all the channels and the empirically selected symmetric few channels, the discovered few-channel electroencephalogram micro-state retest reliability can reach a relatively high level and is superior to the empirically selected symmetric few channels.

Description

Technical Field

[0001] The present invention relates to the field of medical data analysis technology, and in particular to the field of electroencephalogram (EEG) microstates, and specifically refers to a method, system, device, processor, and computer-readable storage medium thereof for realizing EEG few-channel discovery processing for microstate analysis. Background Art

[0002] In recent years, resting EEG microstates have been widely analyzed as a potential biomarker for disease diagnosis. EEG microstates are short-lived and stable scalp voltage distribution patterns during electrical activity in the brain, which can reflect the instantaneous functional state of the brain's neural network and usually last for tens to more than a hundred milliseconds. These dynamically changing microstates are closely related to the brain's information processing process and can reflect cognition, perception, and disease states. In the process of EEG microstate analysis, it is usually necessary to first use a clustering algorithm to identify different microstate topological templates, and then assign the EEG data at each moment to the most matching microstate template to obtain the final microstate sequence, and calculate indicators such as the duration, coverage, frequency, and transition probability of each microstate.

[0003] Microstate analysis relies heavily on densely and evenly distributed EEG channel settings, generally no less than 10-20 EEG systems with 19 electrode channels, and has certain requirements for EEG equipment. In order to reduce equipment costs and simplify the experimental process, some researchers are currently trying to use fewer channels to analyze microstates. The results show that the retest reliability of microstates with fewer channels is relatively high, proving the prospects of microstate analysis based on a few channels. However, these methods usually use a spatially symmetrical distribution of few channels, and the number of subjects analyzed is small. On the one hand, the practice of selecting symmetrical channels is usually based on experience rather than systematic optimization, which may result in the selected electrode channel layout not being able to well represent the global characteristics of EEG signals, and may not be able to adapt to the brain structure and functional characteristics of all subjects, reducing the universality of the results. On the other hand, the number of subjects used for the analysis of few-channel microstates is insufficient, which is easily affected by individual specificity, resulting in overfitting or deviation from the true distribution of the results, making it difficult to generalize the conclusions to a wider population.

[0004] In summary, the few-channel distribution schemes that may be applicable in microstate analysis have not been fully explored, and there is still a lack of a highly universal, non-empirical EEG few-channel discovery method. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and to provide a method, system, device, processor and computer-readable storage medium thereof for realizing EEG few-channel discovery processing by microstate analysis, which is highly universal, non-empirical dependent and has a wide range of applications.

[0006] In order to achieve the above objectives, the method, system, device, processor and computer-readable storage medium thereof for realizing EEG few-channel discovery processing oriented to microstate analysis of the present invention are as follows:

[0007] The method for realizing EEG few-channel discovery and processing oriented to microstate analysis is mainly characterized in that the method comprises the following steps:

[0008] (1) Preprocessing the collected resting-state EEG data;

[0009] (2) Using the preprocessed EEG data, a microstate monitoring module is constructed, which is trained according to the sequence samples, and a channel importance generation and ranking module is constructed. Combined with the attribution analysis technology based on the input gradient, a non-uniformly distributed minority channel is discovered;

[0010] (3) Use the few-channel discovery results to verify the retest reliability of the few-channel microstates.

[0011] Preferably, the step (2) specifically comprises the following steps:

[0012] (2.1) constructing an expert annotation module to construct EEG time series data based on the preprocessed EEG data;

[0013] (2.2) randomly sampling at least one time series from the training data set consisting of EEG time series data;

[0014] (2.3) constructing a learnable micro-state monitoring module and performing training according to the time series;

[0015] (2.4) Using the loss function to update the parameters of the micro-state monitoring module until the micro-state monitoring module converges, the training of the micro-state monitoring module is completed. In the training process, Adam is used as the optimizer, and cosine annealing is used as the learning rate scheduling strategy. The early stopping strategy is used to prevent overfitting.

[0016] (2.5) Construct a channel importance generation and ranking module, use the trained microstate monitoring module, and combine it with the input gradient-based attribution analysis technology to generate the corresponding channel importance ranking for each microstate;

[0017] (2.6) Based on the channel importance ranking, sparse channels are selected for each microstate, resulting in non-uniformly distributed few-channel discovery.

[0018] Preferably, the step (2.1) specifically includes the following steps:

[0019] (2.1.1) The preprocessed resting-state EEG data were used as input, and the robust K-means MOD clustering algorithm was used to identify four basic topological patterns and obtain four microstate templates;

[0020] (2.1.2) Based on the microstate template, the EEG data is annotated with microstates, and the sliding window size is specified to segment the EEG data to construct EEG time series data.

[0021] Preferably, the microstate monitoring module of step (2.3) includes a projection layer, a Transformer encoder layer and a classifier in sequence, and the projection layer, Transformer encoder layer and classifier are connected in sequence, the projection layer is a layer of linear transformation, which is used to project the input time series into the latent variable space; the Transformer encoder layer includes a multi-layer encoder, each layer includes a multi-head self-attention mechanism and a feedforward neural network, which is used to capture long-distance dependencies in the sequence data; the classifier is used to map the output of the Transformer to the final microstate category space.

[0022] Preferably, in the step (2.4), the loss function is used to update the microstate monitoring module parameters, wherein the expression of the loss function is specifically:

[0023] The loss function L of the entire time series is calculated according to the following formula total :

[0024]

[0025] Among them, L total Represents the loss function of the entire time series, representing the average cross entropy loss of all time steps T. is the cross entropy loss for each time step t;

[0026] The cross entropy loss is calculated at each time step t according to the following formula

[0027]

[0028] Among them, {1,2,…,C} represents the microstate category, represents the true category at the t-th time step, Indicates indicator function, indicates condition If true, it takes 1, otherwise it takes 0, y (t,c) represents the output of the classifier for category c at the tth time step, y (t,c′) Represents the output of the classifier for category c′ at the tth time step.

[0029] Preferably, the step (2.5) specifically comprises the following steps:

[0030] (2.5.1) Using the trained microstate monitoring module, input all EEG time series data and infer the monitoring results;

[0031] (2.5.2) Based on the attribution analysis technology of input gradient, the gradient map of each microstate monitoring result to the input feature map in each EEG time series data is calculated respectively;

[0032] (2.5.3) The gradient map of the input feature map is averaged in the EEG channel dimension to obtain the gradient value of each EEG channel. Under each microstate monitoring result, the sum of the EEG channel gradient values ​​of all EEG time series data is counted to obtain the channel importance corresponding to each microstate and sort them.

[0033] Preferably, the step (3) specifically comprises the following steps:

[0034] (3.1) Construct a microstate test-retest reliability verification module;

[0035] (3.2) The EEG data of all channels, the empirically selected symmetric few-channel EEG data, and the few-channel EEG data found in (2) are input into the microstate retest reliability verification module, and the intraclass correlation coefficient ICC is output to evaluate the microstate retest reliability.

[0036] Preferably, the microstate retest reliability verification module includes a microstate indicator calculation unit and an indicator consistency evaluation unit; the microstate indicator calculation unit is used to calculate microstate indicators such as duration, frequency, proportion, and transition probability; the indicator consistency evaluation unit is used to use the microstate indicators to calculate the intra-group correlation coefficient ICC to evaluate the consistency of the microstate indicators.

[0037] The system for realizing EEG few-channel discovery and processing for microstate analysis has the main characteristics that the system includes a preprocessing module, an expert annotation module, a microstate monitoring module, a channel importance generation and ranking module, and a microstate retest reliability verification module, which are used to realize the above-mentioned EEG few-channel discovery method for microstate analysis.

[0038] The main feature of the device for realizing microstate analysis-oriented EEG few-channel discovery processing is that the device comprises:

[0039] a processor configured to execute computer executable instructions;

[0040] The memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the method for realizing EEG few-channel discovery processing oriented to microstate analysis are implemented.

[0041] The processor for realizing microstate analysis-oriented EEG few-channel discovery processing has the main feature that the processor is configured to execute computer-executable instructions, and when the computer-executable instructions are executed by the processor, the various steps of the above-mentioned method for realizing microstate analysis-oriented EEG few-channel discovery processing are realized.

[0042] The main feature of the computer-readable storage medium is that a computer program is stored thereon, and the computer program can be executed by a processor to implement the various steps of the above-mentioned method for realizing EEG few-channel discovery processing oriented to microstate analysis.

[0043] The method, system, device, processor and computer-readable storage medium thereof are used to implement EEG few-channel discovery processing for microstate analysis. The provided EEG few-channel discovery method for microstate analysis models the EEG microstate monitoring task from the perspective of explainable deep learning for the first time, extracts the classification representation of the EEG spatial topological structure, and uses the attribution analysis technology based on input gradient to generate a corresponding channel importance ranking for each microstate, thereby obtaining a non-uniformly distributed few-channel discovery. Compared with all channels and empirically selected symmetric few channels, the discovered few-channel EEG microstate retest reliability can reach a high level and is better than the empirically selected symmetric few channels. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The present invention is a flowchart of a method for realizing EEG few-channel discovery processing for microstate analysis.

[0045] Figure 2 This is a schematic diagram of the overall structure of the system for realizing EEG few-channel discovery processing for microstate analysis of the present invention. DETAILED DESCRIPTION

[0046] In order to more clearly describe the technical content of the present invention, further description is given below in conjunction with specific embodiments.

[0047] The method for realizing EEG few-channel discovery processing oriented to microstate analysis of the present invention comprises the following steps:

[0048] (1) Preprocessing the collected resting-state EEG data;

[0049] (2) Using the preprocessed EEG data, a microstate monitoring module is constructed, which is trained according to the sequence samples, and a channel importance generation and ranking module is constructed. Combined with the attribution analysis technology based on the input gradient, a non-uniformly distributed minority channel is discovered;

[0050] (3) Use the few-channel discovery results to verify the retest reliability of the few-channel microstates.

[0051] As a preferred embodiment of the present invention, the step (2) specifically comprises the following steps:

[0052] (2.1) constructing an expert annotation module to construct EEG time series data based on the preprocessed EEG data;

[0053] (2.2) randomly sampling at least one time series from the training data set consisting of EEG time series data;

[0054] (2.3) constructing a learnable micro-state monitoring module and performing training according to the time series;

[0055] (2.4) Using the loss function to update the parameters of the micro-state monitoring module until the micro-state monitoring module converges, the training of the micro-state monitoring module is completed. In the training process, Adam is used as the optimizer, and cosine annealing is used as the learning rate scheduling strategy. The early stopping strategy is used to prevent overfitting.

[0056] (2.5) Construct a channel importance generation and ranking module, use the trained microstate monitoring module, and combine it with the input gradient-based attribution analysis technology to generate the corresponding channel importance ranking for each microstate;

[0057] (2.6) Based on the channel importance ranking, sparse channels are selected for each microstate, resulting in non-uniformly distributed few-channel discovery.

[0058] As a preferred embodiment of the present invention, the step (2.1) specifically includes the following steps:

[0059] (2.1.1) The preprocessed resting-state EEG data were used as input, and the robust K-means MOD clustering algorithm was used to identify four basic topological patterns and obtain four microstate templates;

[0060] (2.1.2) Based on the microstate template, the EEG data is annotated with microstates, and the sliding window size is specified to segment the EEG data to construct EEG time series data.

[0061] As a preferred embodiment of the present invention, the microstate monitoring module of step (2.3) includes a projection layer, a Transformer encoder layer and a classifier in sequence, and the projection layer, the Transformer encoder layer and the classifier are connected in sequence, the projection layer is a layer of linear transformation, which is used to project the input time series into the latent variable space; the Transformer encoder layer includes a multi-layer encoder, each layer includes a multi-head self-attention mechanism and a feedforward neural network, which is used to capture long-distance dependencies in the sequence data; the classifier is used to map the output of the Transformer to the final microstate category space.

[0062] As a preferred embodiment of the present invention, in the step (2.4), the loss function is used to update the micro-state monitoring module parameters, wherein the expression of the loss function is specifically:

[0063] The loss function L of the entire time series is calculated according to the following formula total :

[0064]

[0065] Among them, L total Represents the loss function of the entire time series, representing the average cross entropy loss of all time steps T. is the cross entropy loss for each time step t;

[0066] The cross entropy loss is calculated at each time step t according to the following formula

[0067]

[0068] Among them, {1,2,…,C} represents the microstate category, represents the true category at the t-th time step, Indicates indicator function, indicates condition If true, it takes 1, otherwise it takes 0, y (t,c) represents the output of the classifier for category s at the tth time step, y (t,c′) Represents the output of the classifier for category c′ at the tth time step.

[0069] As a preferred embodiment of the present invention, the step (2.5) specifically includes the following steps:

[0070] (2.5.1) Using the trained microstate monitoring module, input all EEG time series data and infer the monitoring results;

[0071] (2.5.2) Based on the attribution analysis technology of input gradient, the gradient map of each microstate monitoring result to the input feature map in each EEG time series data is calculated respectively;

[0072] (2.5.3) The gradient map of the input feature map is averaged in the EEG channel dimension to obtain the gradient value of each EEG channel. Under each microstate monitoring result, the sum of the EEG channel gradient values ​​of all EEG time series data is counted to obtain the channel importance corresponding to each microstate and sort them.

[0073] As a preferred embodiment of the present invention, the step (3) specifically comprises the following steps:

[0074] (3.1) Construct a microstate test-retest reliability verification module;

[0075] (3.2) The EEG data of all channels, the empirically selected symmetric few-channel EEG data, and the few-channel EEG data found in (2) are input into the microstate retest reliability verification module, and the intraclass correlation coefficient ICC is output to evaluate the microstate retest reliability.

[0076] As a preferred embodiment of the present invention, the microstate retest reliability verification module includes a microstate indicator calculation unit and an indicator consistency evaluation unit; the microstate indicator calculation unit is used to calculate microstate indicators such as duration, frequency, proportion, and transition probability; the indicator consistency evaluation unit is used to use the microstate indicators to calculate the intra-group correlation coefficient ICC to evaluate the consistency of the microstate indicators.

[0077] The system of the present invention for realizing EEG few-channel discovery processing for microstate analysis, wherein the system includes a preprocessing module, an expert annotation module, a microstate monitoring module, a channel importance generation and ranking module, and a microstate retest reliability verification module, for realizing the above-mentioned EEG few-channel discovery method for microstate analysis.

[0078] The device for realizing microstate analysis-oriented EEG few-channel discovery processing of the present invention comprises:

[0079] a processor configured to execute computer executable instructions;

[0080] The memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the method for realizing EEG few-channel discovery processing oriented to microstate analysis are implemented.

[0081] The processor of the present invention is used to implement EEG few-channel discovery processing for microstate analysis, wherein the processor is configured to execute computer-executable instructions, and when the computer-executable instructions are executed by the processor, the various steps of the above-mentioned method for implementing EEG few-channel discovery processing for microstate analysis are implemented.

[0082] The computer-readable storage medium of the present invention stores a computer program thereon, and the computer program can be executed by a processor to implement the various steps of the above-mentioned method for realizing EEG few-channel discovery processing oriented to microstate analysis.

[0083] In a specific implementation of the present invention, first, an EEG few-channel discovery framework for microstate analysis (referred to as the EEG few-channel discovery framework) is generated. After the EEG few-channel discovery framework preprocesses the resting-state EEG data, an expert annotation module is introduced to construct the EEG time series data, and then a learnable microstate monitoring module is trained. The channel importance generation and sorting module is used, combined with the attribution analysis technology based on the input gradient, to discover non-uniformly distributed few channels, and a microstate retest reliability verification module is introduced to verify the retest reliability of the few-channel microstate. In one embodiment of the present invention, the EEG few-channel discovery framework includes five parts: a preprocessing module, an expert annotation module, a microstate monitoring module, a channel importance generation and sorting module, and a microstate retest reliability verification module.

[0084] like Figure 1 As shown, the present invention provides a method for discovering few EEG channels for microstate analysis, comprising the following steps:

[0085] S1. Preprocess the collected resting-state EEG data;

[0086] S2. Using the preprocessed EEG data, a micro-state monitoring module is constructed, which is trained according to the sequence samples, and a channel importance generation and ranking module is constructed, which is combined with the attribution analysis technology based on the input gradient to discover the non-uniformly distributed minority channels;

[0087] S3. Use the few-channel discovery results to verify the retest reliability of the few-channel microstate.

[0088] Furthermore, the step S2 comprises the following steps:

[0089] S201, constructing an expert annotation module to construct EEG time series data based on the preprocessed EEG data;

[0090] S202, randomly sampling at least one time series from a training data set consisting of EEG time series data;

[0091] S203, constructing a learnable micro-state monitoring module, and performing training according to the time series;

[0092] S204, using the loss function to update the parameters of the microstate monitoring module until the microstate monitoring module converges, completing the training of the microstate monitoring module, wherein Adam is used as the optimizer during the training process, and cosine annealing is used as the learning rate scheduling strategy, and an early stopping strategy is used to prevent overfitting;

[0093] S205, constructing a channel importance generation and ranking module, using the trained microstate monitoring module, combined with the attribution analysis technology based on input gradient, to generate a corresponding channel importance ranking for each microstate;

[0094] S206. Based on the channel importance ranking, select sparse channels for each microstate to obtain a non-uniformly distributed few-channel discovery.

[0095] The beneficial effects of the above further scheme are: the expert annotation module can provide microstate labels for each moment of the EEG time series, and construct a time series data set to train the microstate monitoring module; through end-to-end constraints, the microstate monitoring module can learn the mapping of EEG channel feature space to category space, and mine EEG key channels; using the trained microstate monitoring module, the channel importance generation and ranking module can make full use of the gradient information between the monitoring results and the input feature map, apply attribution analysis technology, count and sort the channel importance corresponding to each microstate, and discover non-uniformly distributed few channels that are different from the symmetric few channels selected empirically.

[0096] Furthermore, the step S201 includes the following steps:

[0097] A1. Using the preprocessed resting-state EEG data as input, the robust K-means MOD clustering algorithm is used to identify four basic topological patterns and obtain four microstate templates;

[0098] A2. Based on the microstate template, the EEG data is annotated with microstates, and the sliding window size is specified to segment the EEG data to construct EEG time series data.

[0099] The beneficial effects of the above further scheme are: using the robust K-means MOD clustering algorithm to generate personalized microstate templates on the preprocessed EEG data, and annotating the microstates of the EEG data, which can better reflect the topological structure of the EEG data used; specifying the sliding window size to split the EEG data can construct a dynamic switching sequence of microstates with a suitable time step, thereby reducing the complexity and computational cost of subsequent monitoring modules.

[0100] Furthermore, the micro-state monitoring module in step S203 includes a projection layer, a Transformer encoder layer, and a classifier in sequence;

[0101] The projection layer is a linear transformation layer, which is intended to project the input time series into the latent variable space;

[0102] The Transformer encoder layer contains multiple layers of encoders, each of which includes a multi-head self-attention mechanism and a feed-forward neural network, aiming to efficiently capture long-range dependencies in sequence data;

[0103] The classifier maps the output of the Transformer to the final microstate category space.

[0104] The beneficial effect of the above further scheme is that the microstate monitoring module first projects the input time series into the latent variable space through the projection layer, and then models the strong correlation between the channel space topology and the microstate category from a global perspective of all time steps through the Transformer encoder layer, and reduces the influence of the time order on the microstate classification by not introducing position encoding. Finally, the output of the Transformer is mapped to the final microstate category space through the classifier.

[0105] Furthermore, the loss function in step S204 is expressed as follows:

[0106]

[0107] Among them, L total Represents the loss function of the entire time series, representing the average cross entropy loss of all time steps T, and the cross entropy loss of each time step t The expression is:

[0108]

[0109] Among them, {1,2,…,C} represents the microstate category, represents the true category at the t-th time step, Indicates indicator function, indicates condition If true, it takes 1, otherwise it takes 0, y (t,c) represents the output of the classifier for category c at the tth time step, y (t,c′) Represents the output of the classifier for category c′ at the tth time step.

[0110] Furthermore, the step S205 includes the following steps:

[0111] B1. Using the micro-state monitoring module trained in step S204, input all EEG time series data and infer the monitoring results;

[0112] B2. Based on the attribution analysis technology of input gradient, the gradient map of each microstate monitoring result to the input feature map in each EEG time series data is calculated respectively;

[0113] B3. The gradient map of the input feature map is averaged in the EEG channel dimension to obtain the gradient value of each EEG channel. Under each microstate monitoring result, the sum of the EEG channel gradient values ​​of all EEG time series data is counted to obtain the channel importance corresponding to each microstate and sort them.

[0114] The beneficial effect of the above further scheme is: all EEG time series are input into the trained microstate monitoring module, and the gradient map between the monitoring results and the input feature map is calculated to reflect the contribution of the input to the result, and then averaged in the EEG channel dimension, which can further explore the statistical correlation between the channel space topology and the microstate category, so as to obtain the channel importance corresponding to each microstate.

[0115] Furthermore, step S3 includes the following steps:

[0116] S301, constructing a microstate retest reliability verification module;

[0117] S302, using the EEG data of all channels, the empirically selected symmetric few-channel EEG data, and the few-channel EEG data found in step S2, input them into the microstate retest reliability verification module, and output the intra-group correlation coefficient ICC to evaluate the microstate retest reliability.

[0118] The beneficial effect of the above further scheme is that the microstate retest reliability verification module can provide a standard process for verifying the microstate retest reliability, and realize the evaluation of the microstate retest reliability by calculating the intraclass correlation coefficient ICC for evaluating the consistency of microstate indicators.

[0119] Further, the microstate retest reliability verification module includes a microstate index calculation unit and an index consistency evaluation unit;

[0120] The microstate index calculation unit is intended to calculate microstate indexes such as duration, frequency, proportion, and transition probability;

[0121] The indicator consistency evaluation unit is intended to use the microstate indicator to calculate the intraclass correlation coefficient ICC to evaluate the consistency of the microstate indicator.

[0122] The beneficial effect of the above further scheme is that it stipulates that the microstate indicators included in the evaluation include duration, frequency, proportion, transition probability, etc., and can provide a standard calculation method for a microstate indicator and an intraclass correlation coefficient ICC.

[0123] On the other hand, the EEG few-channel discovery system for microstate analysis includes a preprocessing module, an expert annotation module, a microstate monitoring module, a channel importance generation and sorting module, and a microstate retest reliability verification module, which are used to execute the EEG few-channel discovery method for microstate analysis.

[0124] In this embodiment, the present invention constructs a preprocessing module to preprocess the collected resting-state EEG data; constructs an expert annotation module to construct EEG time series data using the preprocessed EEG data; randomly samples at least one time series from a training data set composed of EEG time series data, constructs a microstate monitoring module, trains according to the sequence samples, and constructs a channel importance generation and ranking module, uses the trained microstate monitoring module, combined with the attribution analysis technology based on the input gradient, generates a corresponding channel importance ranking for each microstate, selects sparse channels for each microstate, and obtains non-uniformly distributed few-channel discovery; constructs a microstate retest reliability verification module, and uses the few-channel discovery results to verify the retest reliability of the few-channel microstate.

[0125] In this embodiment, a preprocessing module is constructed to preprocess the collected resting-state EEG data, and the EEG data is preprocessed and analyzed offline using EEGLAB (version R2018b) in MATLAB; a second-order Butterworth filter is used for bandpass filtering from 1 to 40 Hz, and the data is re-referenced to the whole-brain average; and artifacts such as blinking are removed using independent component analysis.

[0126] In this embodiment, an expert annotation module is constructed, and the preprocessed resting-state EEG data is used as input. The robust K-means MOD clustering algorithm is used to identify four basic topological graph patterns, and four microstate templates A, B, C, and D are obtained; based on the microstate templates, the EEG data is microstate labeled, and the sliding window size T is specified to segment the EEG data to construct the EEG time series data; a time series is randomly sampled from the training data set composed of the EEG time series data, which is recorded as {x[1:T], y true [1:T]}, where x[1:T] represents an EEG time series with time step T, y true [1:T] represents the real microstate category label of the EEG time series at time step T, using represents the true category at the tth time step, and {1,2,…,C} represents the microstate category.

[0127] In this embodiment, a micro-state monitoring module is constructed, which includes a projection layer, a Transformer encoder layer, and a classifier in sequence. The projection layer is a linear transformation layer, which aims to project the input EEG time series x[1:T] into the latent variable space. The linear transformation can be expressed as:

[0128] z=wp x+b p

[0129] Among them, z is a hidden variable, w p and b p are the linear transformation matrix and the bias term respectively. The Transformer encoder layer contains multiple layers of encoders, each of which includes a multi-head self-attention mechanism and a feedforward neural network, aiming to efficiently capture long-distance dependencies in sequence data; assuming there are l-head self-attention mechanisms, in each self-attention mechanism, a non-shared weight matrix w is used q 、w k 、w v , generate query q, key k, value v:

[0130] q=w q z,k=w k z,v=w v z

[0131] in T is the time step, and h is the dimension of the hidden variable z. In order to consider all time steps at the same time, the strong correlation between the channel space topology and the microstate category is modeled from a global perspective, and the similarity between the query q, key k, and value v is calculated:

[0132]

[0133] Among them, the softmax function is used to normalize the score, and the results of all l heads are connected, and a learnable linear transformation matrix is ​​used to generate the output of the multi-head self-attention mechanism, which is then input into the feedforward neural network to obtain the result of the encoder of this layer. Each layer of encoder does not introduce position encoding to reduce the influence of time order on microstate classification. Then, the output of the multi-layer Transformer encoder is used as the input of the classifier, and is mapped to the final microstate category space through a simple fully connected layer to obtain the category result y of the microstate. Training is performed according to the time series, and the parameters of the microstate monitoring module are updated using the loss function.

[0134] In one embodiment of the present invention, the loss function is expressed as follows:

[0135]

[0136] Among them, L total Represents the loss function of the entire time series, representing the average cross entropy loss of all time steps T, and the cross entropy loss of each time step t The expression is:

[0137]

[0138] Among them, {1,2,…,C} represents the microstate category, represents the true category at the t-th time step, Indicates indicator function, indicates condition If true, it takes 1, otherwise it takes 0, y (t,c) represents the output of the classifier for category c at the tth time step, y (t,c′) Represents the output of the classifier for category c′ at the tth time step.

[0139] In this embodiment, the loss function L total The role of is to ensure the accuracy of the microstate classification output by the microstate monitoring module, and to constrain the microstate classification of the entire time series obtained by each microstate monitoring module to make it close to the true value.

[0140] In this embodiment, the micro-state monitoring module uses the loss function L during the training process. total Update the network parameters until the micro-state monitoring module finally converges completely. In one embodiment of the present invention, Adam is used as the optimizer during the training process, and cosine annealing is used as the learning rate scheduling strategy, and an early stopping strategy is used to prevent overfitting.

[0141] In this embodiment, a channel importance generation and ranking module is constructed, and a trained microstate monitoring module is used to input all EEG time series data and infer the monitoring results. In combination with the input gradient-based attribution analysis technology, the gradient map of the input feature map for each microstate monitoring result in each EEG time series data is calculated respectively. The gradient map of the input feature map is averaged in the EEG channel dimension to obtain the gradient value of each EEG channel. Under each microstate monitoring result, the sum of the EEG channel gradient values ​​of all EEG time series data is counted to obtain the channel importance corresponding to each microstate and sort them. Sparse channels are selected for each microstate to obtain unevenly distributed few-channel discovery.

[0142] In this embodiment, a microstate retest reliability verification module is constructed, including a microstate indicator calculation unit and an indicator consistency evaluation unit. The EEG data of all channels, the empirically selected symmetric few-channel EEG data, and the discovered few-channel EEG data are input into the microstate retest reliability verification module. The microstate indicator calculation unit is used to calculate microstate indicators such as duration, frequency, proportion, and transition probability. The indicator consistency evaluation unit is used to calculate the intra-group correlation coefficient ICC of the microstate indicators to evaluate the consistency of the microstate indicators. Finally, the ICC is output as the evaluation result of the microstate retest reliability.

[0143] In this embodiment, in order to verify the actual effect of the EEG few-channel discovery method for microstate analysis provided by the embodiment of the present invention, it was implemented on real resting EEG data. Specifically, the inventor selected the resting EEG of 534 subjects, including 240 clinical high-risk people for mental illness (127 females and 113 males; age: M = 18.83 years, SD = 5.17), 111 first-episode schizophrenia patients (46 females and 65 males; age: M = 24.21 years, SD = 7.36), and 183 healthy controls (46 females and 92 males; age: M = 22.79 years, SD = 3.81). Clinical high-risk population met the prodromal state criteria of the Structured Interview for Prodromal Symptoms / Prodromal Symptom Scale (SIPS / SOPS); first-episode schizophrenia patients were diagnosed with schizophrenia or schizophreniform disorder by the Structured Clinical Interview for DSM-IV; healthy controls excluded those who currently or previously had DSM-IV Axis I disorders or whose first-degree relatives had mental disorders. This study was approved by the Ethics Committee of Shanghai Mental Health Center. All subjects signed written informed consent before enrollment. Participants sat in an electrically shielded room with reduced sound and dimmed light. They were asked to close their eyes and sit in a relaxed position for 5 minutes. EEG data were obtained from a 64-lead scalp electrode elastic cap with a sampling rate of 1000 Hz. The EEG signal was referenced online to the tip of the nose and filtered online with a bandpass filter of 0.016-200 Hz. The impedance of the electrode was kept below 5 kilo-ohms.

[0144] In this embodiment, the EEG few-channel discovery method for microstate analysis is verified based on real resting-state EEG data. The trained microstate monitoring module achieves efficient classification of EEG microstates, with a classification accuracy of 0.87 and an AUC of up to 0.95; then the channel importance generation and sorting module is used to discover 8 non-uniformly distributed EEG channels; based on the final microstate retest reliability verification module of the method, the EEG data of the discovered 8 channels and the empirically selected symmetrical 8-channel EEG data are used, and the intra-group correlation coefficient ICC of the microstate indicators (duration, frequency, proportion, and transition probability) of the full-channel EEG data is used as the evaluation indicator. The results are shown in Table 1:

[0145] Table 1

[0146]

[0147] As can be seen from Table 1, the present invention can achieve relatively excellent microstate retest reliability in the EEG few-channel discovery task, and the intra-group correlation coefficient ICC is better than the empirically selected symmetric few-channel scheme, which has good practical value.

[0148] In another embodiment, if Figure 2As shown, the present invention provides an EEG few-channel discovery system for microstate analysis, including a preprocessing module, an expert annotation module, a microstate monitoring module, a channel importance generation and sorting module, and a microstate retest reliability verification module, which is used to execute the EEG few-channel discovery method for microstate analysis described in any one of Example 1.

[0149] In another exemplary embodiment, the present invention further provides a computer-readable storage medium including program instructions, which, when executed by a processor, implements the steps of the EEG few-channel discovery method for microstate analysis in any of the above embodiments. For example, the computer-readable storage medium may be the above-mentioned memory including program instructions, and the above-mentioned program instructions may be executed by a processor of the system to complete the above-mentioned EEG few-channel discovery method for microstate analysis, and achieve the same technical effect as the above-mentioned method.

[0150] The specific implementation scheme of this embodiment can refer to the relevant description in the above embodiment, which will not be repeated here.

[0151] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0152] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0153] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0154] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0155] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the corresponding program may be stored in a computer-readable storage medium, which, when executed, includes one of the steps of the method embodiment or a combination thereof.

[0156] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0157] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0158] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0159] The method, system, device, processor and computer-readable storage medium thereof are used to implement EEG few-channel discovery processing for microstate analysis. The provided EEG few-channel discovery method for microstate analysis models the EEG microstate monitoring task from the perspective of explainable deep learning for the first time, extracts the classification representation of the EEG spatial topological structure, and uses the attribution analysis technology based on input gradient to generate a corresponding channel importance ranking for each microstate, thereby obtaining a non-uniformly distributed few-channel discovery. Compared with all channels and empirically selected symmetric few channels, the discovered few-channel EEG microstate retest reliability can reach a high level and is better than the empirically selected symmetric few channels.

[0160] In this specification, the present invention has been described with reference to specific embodiments thereof. However, it is apparent that various modifications and variations may be made without departing from the spirit and scope of the present invention. Therefore, the specification and drawings should be regarded as illustrative rather than restrictive.

Claims

1. A method for realizing EEG few-channel discovery processing for microstate analysis, characterized in that: The method comprises the following steps: (1) Preprocessing the collected resting-state EEG data; (2) Using the preprocessed EEG data, a microstate monitoring module is constructed, which is trained according to the sequence samples, and a channel importance generation and ranking module is constructed. Combined with the attribution analysis technology based on the input gradient, a non-uniformly distributed minority channel is discovered; (3) Use the few-channel discovery results to verify the retest reliability of the few-channel microstates.

2. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 1 is characterized in that: The step (2) specifically comprises the following steps: (2.1) constructing an expert annotation module to construct EEG time series data based on the preprocessed EEG data; (2.2) randomly sampling at least one time series from the training data set consisting of EEG time series data; (2.3) constructing a learnable micro-state monitoring module and performing training according to the time series; (2.4) Using the loss function to update the parameters of the micro-state monitoring module until the micro-state monitoring module converges, the training of the micro-state monitoring module is completed. In the training process, Adam is used as the optimizer, and cosine annealing is used as the learning rate scheduling strategy. The early stopping strategy is used to prevent overfitting. (2.5) Construct a channel importance generation and ranking module, use the trained microstate monitoring module, and combine it with the input gradient-based attribution analysis technology to generate the corresponding channel importance ranking for each microstate; (2.6) Based on the channel importance ranking, sparse channels are selected for each microstate, resulting in non-uniformly distributed few-channel discovery.

3. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 2 is characterized in that: The step (2.1) specifically comprises the following steps: (2.1.1) The preprocessed resting-state EEG data were used as input, and the robust K-means MOD clustering algorithm was used to identify four basic topological patterns and obtain four microstate templates; (2.1.2) Based on the microstate template, the EEG data is annotated with microstates, and the sliding window size is specified to segment the EEG data to construct EEG time series data.

4. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 2, characterized in that: The microstate monitoring module of the step (2.3) includes a projection layer, a Transformer encoder layer and a classifier in sequence, wherein the projection layer, the Transformer encoder layer and the classifier are connected in sequence, wherein the projection layer is a layer of linear transformation, which is used to project the input time series into the latent variable space; the Transformer encoder layer includes a multi-layer encoder, each layer includes a multi-head self-attention mechanism and a feedforward neural network, which is used to capture long-distance dependencies in the sequence data; the classifier is used to map the output of the Transformer to the final microstate category space.

5. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 2, characterized in that: In the step (2.4), the loss function is used to update the microstate monitoring module parameters, wherein the expression of the loss function is specifically: The loss function L of the entire time series is calculated according to the following formula total : Among them, L total Represents the loss function of the entire time series, representing the average cross entropy loss of all time steps T. is the cross entropy loss for each time step t; The cross entropy loss is calculated at each time step t according to the following formula Among them, {1,2,…,C} represents the microstate category, represents the true category at the t-th time step, Indicates indicator function, indicates condition If true, it takes 1, otherwise it takes 0, y (t,c) represents the output of the classifier for category c at the tth time step, y (t,c′) Indicates the classifier for category c at the tth time step ′ Output.

6. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 2, characterized in that: The step (2.5) specifically comprises the following steps: (2.5.1) Using the trained microstate monitoring module, input all EEG time series data and infer the monitoring results; (2.5.2) Based on the attribution analysis technology of input gradient, the gradient map of each microstate monitoring result to the input feature map in each EEG time series data is calculated respectively; (2.5.3) The gradient map of the input feature map is averaged in the EEG channel dimension to obtain the gradient value of each EEG channel. Under each microstate monitoring result, the sum of the EEG channel gradient values ​​of all EEG time series data is counted to obtain the channel importance corresponding to each microstate and sort them.

7. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 1, characterized in that: The step (3) specifically comprises the following steps: (3.1) Construct a microstate test-retest reliability verification module; (3.2) The EEG data of all channels, the empirically selected symmetric few-channel EEG data, and the few-channel EEG data found in (2) are input into the microstate retest reliability verification module, and the intraclass correlation coefficient ICC is output to evaluate the microstate retest reliability.

8. The method for realizing EEG few-channel discovery processing for microstate analysis according to claim 7, characterized in that: The microstate retest reliability verification module includes a microstate indicator calculation unit and an indicator consistency evaluation unit; the microstate indicator calculation unit is used to calculate microstate indicators such as duration, frequency, proportion, and transition probability; the indicator consistency evaluation unit is used to use the microstate indicators to calculate the intra-group correlation coefficient ICC to evaluate the consistency of the microstate indicators.

9. A system for realizing EEG few-channel discovery and processing for microstate analysis, characterized in that: The system includes a preprocessing module, an expert annotation module, a microstate monitoring module, a channel importance generation and ranking module, and a microstate retest reliability verification module, and is used to implement the EEG few-channel discovery method for microstate analysis as described in any one of claims 1 to 8.

10. A device for realizing EEG few-channel discovery processing for microstate analysis, characterized in that: The device comprises: a processor configured to execute computer-executable instructions; A memory storing one or more computer executable instructions, wherein when the computer executable instructions are executed by the processor, the steps of the method for realizing EEG few-channel discovery processing oriented to microstate analysis as described in any one of claims 1 to 8 are implemented.

11. A processor for realizing EEG few-channel discovery processing for microstate analysis, characterized in that: The processor is configured to execute computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the method for realizing EEG few-channel discovery processing for microstate analysis as described in any one of claims 1 to 8 are implemented.

12. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program can be executed by a processor to implement the various steps of the method for realizing EEG few-channel discovery processing oriented to microstate analysis as described in any one of claims 1 to 8.

Images