A real-time brain electrical source imaging system

Abstract

The application discloses a real-time electroencephalography source imaging system, comprising: a data input module for continuously and real-timely acquiring brain source activity sampling data and lead field matrix; a sampling data generation module for constructing an electroencephalography sampling data generation model according to the brain source activity sampling data; a covariance matrix calculation module for real-timely calculating the covariance of all brain source activity sampling data; a real-time weight matrix estimation module for real-timely calculating the weight vector of a new beamformer; and a real-time imaging module for real-timely calculating and solving the amplitude of a new brain source activity source for real-time imaging. The application can reconstruct the space-time characteristics of brain source activity in real time under Gaussian and real brain noise conditions, can efficiently process electroencephalography data with an ultra-long sampling time, and the result is robust.

Description

Technical Field

[0001] The present invention belongs to the technical field of brain power imaging, and more specifically, relates to a real-time brain power imaging system. Background Art

[0002] The brain, as the core of the human central nervous system, not only controls our thoughts, emotions, and behaviors but is also closely related to cognitive functions such as perception, memory, and learning. Its importance is self-evident, and it is key to understanding human behavior and psychological processes. With the advancement of science and technology, functional brain imaging techniques have emerged as a vital tool for studying brain structure and function. These techniques can non-invasively observe brain activity patterns in different states, revealing how the brain responds to external stimuli and internal thought processes.

[0003] Functional brain imaging technologies provide an intuitive and dynamic way to observe how the brain works. Through these technologies, scientists can observe changes in brain activity when performing specific tasks, thereby better understanding how the brain coordinates different areas to complete complex cognitive tasks. Functional brain imaging technologies play an important role in the medical field. They help doctors diagnose and treat various brain diseases such as Alzheimer's disease, Parkinson's disease, depression, and schizophrenia. They also provide valuable data and insights for fields such as education, psychology, and artificial intelligence. Through these technologies, researchers can gain a deeper understanding of the brain's plasticity, explore how the brain adapts and learns new skills, and how to improve brain function through training and intervention. As technology continues to advance, functional brain imaging technologies will continue to expand our understanding of the brain and contribute to the development of human health and cognitive science.

[0004] The inverse problem of EEG is the process of retrieving information about neuronal activity in the brain from sampled brain electromagnetic data. It is the core of EEG imaging. A persistent challenge in EEG is the real-time and robust recovery of brain signals, which requires a computationally efficient and accurate method to achieve real-time source imaging. In recent years, new EEG methods have been continuously proposed, including some that are computationally fast and highly accurate, but none of them meet real-time requirements. Furthermore, the computational complexity of the inverse problem and the high amount of data processing required pose significant challenges to the proposed methods and the supporting equipment. Finally, the varying quality of the acquired data inevitably includes some noise and artifacts, which reduce imaging accuracy. Therefore, accurately and in real time using EEG data in the presence of noise has become a challenging problem. Summary of the Invention

[0005] In response to the above-mentioned defects or improvement needs of the existing technology, the present invention provides a real-time brain source imaging system that can realize real-time reconstruction of the spatiotemporal characteristics of brain source activity under Gaussian and real brain noise conditions, and at the same time can efficiently process EEG data with ultra-long sampling time, and the results are robust.

[0006] To achieve the above object, according to one aspect of the present invention, a real-time brain power imaging system is provided, comprising:

[0007] Data input module, used to continuously and in real time acquire brain source activity sampling data and lead field matrix;

[0008] A sampling data generation module is used to build an EEG sampling data generation model based on brain source activity sampling data;

[0009] The covariance matrix calculation module is used to calculate the covariance of all brain source activity sampling data in real time after obtaining new brain source activity sampling data each time, recording the current moment as K and the covariance of all brain source activity sampling data calculated at the previous moment as R(K-1);

[0010] The real-time weight matrix estimation module is used to calculate the weight vector of the beamformer at the current moment K based on the covariance R(K-1) of the brain source activity sampling data calculated at the previous moment and the lead field matrix. Each time the updated value of the covariance of all brain source activity sampling data is obtained, the new beamformer weight vector is calculated in real time. The weight vector of the beamformer at the current moment K is recorded as

[0011] Real-time imaging module, used to calculate the beamformer weight vector according to the current time K The brain source activity source amplitude at the current moment K is calculated based on the brain source activity sampling data at the current moment K, and each time new brain source activity sampling data is obtained, the new brain source activity source amplitude is calculated in real time for real-time imaging.

[0012] Furthermore, each time new brain source activity sampling data is obtained, the covariance of all brain source activity sampling data is calculated in real time using the following formula:

[0013]

[0014] in, represents the brain source activity sampling data at time t, R(K) represents the covariance of all brain source activity sampling data at the current time K, and T is the matrix transpose symbol;

[0015] The calculation formula for R(K-1) is:

[0016] Furthermore, R is initialized using the earliest P data points.

[0017] Further, P=10 or P=20.

[0018] Furthermore, the calculation formula for calculating the weight vector of the beamformer at the current moment K based on the covariance R(K-1) of the brain source activity sampling data calculated at the previous moment and the lead field matrix is:

[0019]

[0020] in, is the lead field matrix, For spatial location The orientation of the source, For spatial location Lead field vector.

[0021] Furthermore, the calculation formula for the brain source activity amplitude at the current moment K is:

[0022]

[0023] in, For spatial location The amplitude of brain activity at the current moment K, is the brain source activity sampling data at the current moment K.

[0024] Furthermore, the EEG sampling data generation model is:

[0025]

[0026] in, represents the brain source activity sampling data at time t, represents the lead field matrix at the j-th voxel, represents the source vector at the j-th voxel.

[0027] In general, the above technical solutions conceived by the present invention have beneficial effects compared with the existing technology:

[0028] (1) A real-time EEG source imaging solution system was invented, which provides an efficient and accurate source imaging algorithm, achieves almost real-time source imaging, and can efficiently process EEG data with ultra-long sampling time.

[0029] (2) The present invention utilizes a minimum variance beamformer with a unity gain constraint to achieve source imaging. Although the beamformer output power contains not only noise contributions but also some other unwanted contributions, due to the existence of the unity gain constraint, these effects can be minimized by minimizing this unity gain constraint, thereby obtaining a robust source imaging result.

[0030] (3) The algorithm proposed in the present invention simplifies the calculation process of the inverse problem of source imaging, has small computational complexity, high computational efficiency, and low requirements for equipment.

[0031] (4) The simulation results show that detailed experiments were carried out under simulated and real noise conditions to demonstrate the effectiveness of the proposed brain source imaging algorithm, indicating that the method of the present invention achieves almost real-time calculation results under both simulated and real noise conditions, and the source imaging results are very accurate. At the same time, it has efficient and robust performance when processing EEG data with ultra-long sampling time. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a schematic diagram of a real-time brain power imaging system according to an embodiment of the present invention;

[0033] Figure 2 1 is a schematic diagram of experimental results of an embodiment of the present invention under low signal-to-noise ratio conditions (-5dB);

[0034] Figure 3 Schematic diagram showing how various data according to an embodiment of the present invention change as the duration of the simulated EEG signal increases. DETAILED DESCRIPTION

[0035] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0036] The terms "including" and "having" and any variations thereof in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product or device comprising a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products or devices.

[0037] The naming or numbering of the steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0038] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0039] like Figure 1 As shown, the real-time EEG source imaging system of an embodiment of the present invention includes a data input module, a sampled data generation module, a covariance matrix calculation module, a real-time brain source activity reconstruction module, and a real-time imaging module. This system is applicable to various scenarios for diagnosing diseases through brain electromagnetic activity source imaging. It can achieve real-time reconstruction of the spatiotemporal characteristics of brain source activity under Gaussian and real brain noise conditions, and can efficiently process EEG data with extremely long sampling times, and the results are robust.

[0040] The data input module is used to continuously and in real time acquire brain source activity sampling data and lead field matrix.

[0041] Furthermore, relevant preprocessing can be performed on the input data while obtaining it.

[0042] Every time the data input module obtains new data, it sends it to other modules for real-time processing.

[0043] The brain source activity sampling data may be electroencephalogram (EEG) sampling data or magnetoencephalogram (MEG) sampling data.

[0044] The brain source activity sampling data can be real-time input data, and then it will be pre-processed into valid data.

[0045] For the input data, a series of sensors are used to obtain bioelectromagnetic measurements, and the output of the mth sensor at time t is defined as y m (t), and the column vector contains the outputs of all sensors, we get:

[0046]

[0047] In formula (1), M is the total number of sensors, and the column vector represents the output of the sensor array at time t, which can be called a data vector or array measurement value.

[0048] The spatial position is represented by a three-dimensional vector express, The source vector is defined as a three-dimensional column vector

[0049]

[0050] In formula (2), Indicates spatial location The source vector at time t, and are the x, y and z components of the source vector respectively. The physical property of the source vector is the electromotive force generated by neuronal activity in the brain. The magnitude of the source vector is recorded as a scalar The orientation of the source is denoted by the three-dimensional unit vector The superscript T represents the transpose of the matrix. From this we can get:

[0051]

[0052] For the lead field matrix, assuming that There is a unit magnitude source at , when the unit magnitude source points to the x, y and z directions respectively, the output of the mth sensor due to the unit magnitude source is expressed as and Define column vector and for:

[0053]

[0054] These column vectors represent the sensitivity of the sensor array to the source located at r and pointing in the x, y and z directions respectively. Using these column vectors, the sensitivity of the entire sensor array to the source at r is expressed as an M×3 matrix Expressed as:

[0055]

[0056] The matrix in formula (5) It is called the lead field matrix, and the lead field vector is defined It indicates that the sensor array is in a specific source direction The sensitivity on , is expressed as:

[0057]

[0058] Using the lead field matrix in formula (5) Sensor data and the source vector The relationship can be expressed as:

[0059]

[0060] In formula (7) Denote the volume element, and the integral is performed over the volume assuming the existence of a source, which is called the source space and denoted as Ω.

[0061] The inverse bioelectromagnetic problem is based on the measured values Estimated source vector Here we assume that the sensor lead field matrix is known.

[0062] The sampling data generation module is used to build an EEG sampling data generation model based on brain source activity sampling data.

[0063] from estimate hour, It is continuous in space, and It is discrete in space. A common strategy is to introduce voxel discretization in the source space. We define the number of voxels as N and the position of the voxel as Then the discrete form of formula (7) is expressed as:

[0064]

[0065] In formula (8), the source vector at the jth voxel is For simplicity, let’s call it We introduce the augmented lead field matrix F at all voxel locations:

[0066]

[0067] The augmented lead field matrix F in formula (9) is a matrix of size M×3N. We define a 3N×1 column vector It contains the source vectors for all voxel positions, expressed as:

[0068]

[0069] The column vector Substitute into formula (8), then It can be expressed as:

[0070]

[0071] In formula (11), since the augmented lead field matrix F is a known quantity, the only unknown quantity is the 3N×1 column vector This column vector is called the voxel source vector. Here, assuming that the source directions of all voxel positions are predetermined, the direction of a source at the jth voxel is defined as Then the lead field vector at the jth voxel is expressed as a column vector According to formula (6), we can get Therefore, the augmented lead field matrix is ​​represented as a matrix H of size M×N, defined as follows:

[0072]

[0073] Therefore, formula (11) can be simplified as:

[0074]

[0075] In this case, the voxel source vector is an N×1 column vector:

[0076]

[0077] in The component of the j-th voxel is s j (t), is the scalar intensity at the jth voxel. In summary, ∈ represents the additive noise in the sensor data, and the sensor data With the voxel source vector The relationship between them is expressed as:

[0078]

[0079] When the voxel has a predetermined direction, the augmented lead field matrix H in formula (12) is used. and The relationship between them is expressed as:

[0080]

[0081] Depend on Figure 2 As shown in (a), there are three randomly simulated source activity signals with a duration of 5 seconds. Figure 2 (b) shows the addition of random noise to the randomly simulated source activity signal. Here, the signal-to-noise ratio is set to -5dB to simulate an EEG signal with a low signal-to-noise ratio. The sampling rate is set to 100 Hz. The process of adding noise is as follows: first, the initial signal-to-noise ratio of the generated simulated signal is calculated; then, the gain factor of the noise is adjusted so that the simulated EEG signal meets the set target signal-to-noise ratio; then, the current signal-to-noise ratio is recalculated and the noise is added to the signal; finally, the noisy signal is normalized. This results in a simulated EEG signal with a low signal-to-noise ratio.

[0082] The covariance matrix calculation module is used to calculate the covariance of all brain source activity sampling data in real time after each new brain source activity sampling data is obtained. The current moment is recorded as K, and the covariance of all brain source activity sampling data calculated at the previous moment is recorded as R(K-1).

[0083] Assume that the direction of the source is predetermined and avoid direction estimation, in which case the data vector Beamformer reconstructs source amplitude The estimated value of can be expressed as:

[0084]

[0085] in It's location and the estimated source amplitude at time t. Where, the column vector represents the beamformer weight, which characterizes the beamformer's characteristics. There are two common types of beamformers: a non-adaptive beamformer, where the weight vector depends only on the sensor array's lead field, and an adaptive beamformer, where the weight vector depends on both the measurement data and the sensor array's lead field. Here, an adaptive beamformer is used.

[0086] The weight vector of the adaptive beamformer is defined as follows, Under the premise of , it can be expressed as:

[0087]

[0088] where R is the data covariance matrix, which is defined as follows, where <·> represents the ensemble mean:

[0089]

[0090] Formula (19) can be expanded into the following form:

[0091]

[0092] In formula (20), R(K) is the value of the data covariance matrix at the current moment, where K is the current moment. According to the definition of the data covariance matrix, is the data covariance matrix R(K-1) at the K-1 moment, which is the R of the previous moment, and Here, since the data is input in real time and continuously, we can assume that K is a time point with an infinitesimal duration. The value is also infinitesimal, so this part can be ignored. Then the data covariance matrix R(K-1) calculated at the previous moment can be directly replaced with the data covariance matrix of the current moment K. Therefore, the real-time estimate of the data covariance matrix at the current moment is for:

[0093]

[0094] Since R(K-1) is known at the current time K, The calculation process will become extremely fast. Here, it is necessary to initialize the data covariance matrix R. The calculation method of R is given by formula (19). The moment when real-time imaging starts is recorded as moment 1, and the previous moment R(0) is used to directly replace the moment 1. Initialize R using P data points before time 1, and the calculated initial covariance matrix R(0) is:

[0095]

[0096] Where N is a value much smaller than the number of samples, such as P = 10 or P = 20. After R is initialized, the subsequent All of them can be estimated by formula (21), thus achieving the effect of real-time calculation of the data covariance matrix.

[0097] The covariance matrix calculation module also calculates R(K) in real time as the data covariance matrix of the next moment K+1 Used for real-time imaging at the next moment K+1.

[0098] The real-time weight matrix estimation module is used to calculate the weight vector of the beamformer at the current moment K based on the covariance R(K-1) of the brain source activity sampling data calculated at the previous moment and the lead field matrix. The weight vector of the new beamformer is calculated in real time each time the covariance of all new brain source activity sampling data is obtained. The weight vector of the beamformer at the current moment K is recorded as

[0099] The following are the technical details of the module:

[0100] In formula (18), the inner product It is called the unity gain constraint, which means that The beamformer output of a unit-level source is, therefore, set This ensures that the beamformer passes the signal.

[0101] This constrained minimization problem can be solved using the Lagrange multiplier method, where the Lagrange multiplier is defined as a scalar ζ and the Lagrange function is You can get:

[0102]

[0103] From formula (23), we can get the Lagrangian function right The derivative of is:

[0104]

[0105] Let the right side of the above equation be 0, and we can get the following relationship:

[0106]

[0107] Substitute formula (25) into the unity gain constraint equation In the equation, we get:

[0108]

[0109] Then substitute this ζ into formula (25) to obtain the weight vector that satisfies formula (18) for:

[0110]

[0111] In summary, can be obtained directly, then the beamformer weight vector It only depends on the data covariance matrix R. If the data covariance matrix R can be calculated in real time, then the weight matrix of the beamformer can be estimated in real time. Since the covariance matrix calculation module can calculate the data covariance matrix of the current input data in real time And because is known, so the beamformer weight vector can be calculated in real time

[0112] Real-time imaging module, used to calculate the beamformer weight vector according to the current time K The brain source activity source amplitude at the current moment K is calculated based on the brain source activity sampling data at the current moment K, and each time new brain source activity sampling data is obtained, the new brain source activity source amplitude is calculated in real time for real-time imaging.

[0113] Since the real-time weight matrix estimation module can calculate the weight vector of the beamformer at the current moment in real time We substitute it into formula (17) to obtain the output of the beamformer in real time, which is the estimated source amplitude for:

[0114]

[0115] in, For spatial location The amplitude of brain activity at the current moment K, is the brain source activity sampling data at the current moment K.

[0116] In order to obtain a real-time estimate of the source amplitude Then we can perform real-time imaging. Figure 2Figure (c) shows the result of real-time source reconstruction. It can be found that this method can robustly estimate source activity under low signal-to-noise ratio conditions. Figure 3 Figure (a) in the middle shows the change in computational time from short-duration signals (1 second) to ultra-long-duration signals (300 seconds). It can be seen that the computational efficiency of this method for ultra-long-duration EEG data is still very high. Although the calculated data increases by two orders of magnitude, the computational time of this method changes slightly, which is sufficient to achieve real-time computing performance. Figure 3 Figure (b) shows a schematic diagram of the performance indicators of this method, such as AP (average precision), A (accuracy), and Correlation, as they change with signal duration. It can be found that the increase in duration does not lead to significant changes in these indicators, and they remain at a high level. This is sufficient to prove that this method can robustly reconstruct source activities.

[0117] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time brain power imaging system, characterized in that include: Data input module, used to continuously and in real time acquire brain source activity sampling data and lead field matrix; A sampling data generation module is used to build an EEG sampling data generation model based on brain source activity sampling data; The covariance matrix calculation module is used to calculate the covariance of all brain source activity sampling data in real time after obtaining new brain source activity sampling data each time, recording the current moment as K and the covariance of all brain source activity sampling data calculated at the previous moment as R(K-1); The real-time weight matrix estimation module is used to calculate the weight vector of the beamformer at the current moment K based on the covariance R(K-1) of the brain source activity sampling data calculated at the previous moment and the lead field matrix. Each time the updated value of the covariance of all brain source activity sampling data is obtained, the new beamformer weight vector is calculated in real time. The weight vector of the beamformer at the current moment K is recorded as Real-time imaging module, used to calculate the beamformer weight vector according to the current time K The brain source activity source amplitude at the current moment K is calculated based on the brain source activity sampling data at the current moment K, and each time new brain source activity sampling data is obtained, the new brain source activity source amplitude is calculated in real time for real-time imaging; The calculation formula for calculating the weight vector of the beamformer at the current moment K based on the covariance R(K-1) of the brain source activity sampling data calculated at the previous moment and the lead field matrix is: in, is the lead field matrix, For spatial location The orientation of the source, For spatial location Lead field vector, T is the matrix transpose sign.

2. The real-time brain power imaging system according to claim 1, characterized in that After obtaining new brain source activity sampling data each time, the calculation formula for calculating the covariance of all brain source activity sampling data in real time is: in, represents the brain source activity sampling data at time t, and R(K) represents the covariance of all brain source activity sampling data at the current time K; The calculation formula for R(K-1) is:

3. The real-time brain power imaging system according to claim 2, characterized in that Initialize R using the earliest P data points, 4. The real-time brain power imaging system according to claim 3, characterized in that P=10 or P=20.

5. The real-time brain power imaging system according to claim 1, wherein: The calculation formula for the brain source activity amplitude at the current moment K is: in, For spatial location The amplitude of brain activity at the current moment K, is the brain source activity sampling data at the current moment K.

6. The real-time brain power imaging system according to claim 1, wherein: The EEG sampling data generation model is: in, represents the brain source activity sampling data at time t, represents the lead field matrix at the j-th voxel, represents the source vector at the jth voxel, and N is the number of voxels.

Images