Intelligent ward round making method and system based on ssvep electroencephalogram signals

Abstract

Disclosed in the present invention are an intelligent ward round making method and system based on SSVEP electroencephalogram signals. The method comprises: acquiring a first electroencephalogram signal of a user under a visual stimulus, and defining reference signals; decomposing the first electroencephalogram signal into sub-band components, calculating the differences between the sub-band components, between the reference signals, and between the sub-band components and the reference signals, acquiring a total spatial filter for each sub-band component, filtering the sub-band components, and rearranging the sub-band components to obtain a second electroencephalogram signal; using variational mode decomposition to acquire a plurality of variational mode components, and optimizing weights under each channel to perform reconstruction so as to obtain a third electroencephalogram signal; and acquiring an evoked stimulus frequency on the basis of intrinsic features of the third electroencephalogram signal, and the reference signals, determining a user intent, and completing intelligent ward round making. The present method enhances effective parts by means of multiple signal decompositions and reconstructions, resulting in a higher signal-to-noise ratio. Moreover, the present method uses average features, thus further ensuring the accuracy of frequency identification; and the present method greatly improves the user's ability to communicate with medical staff.

Description

An intelligent ward rounds method and system based on SSVEP electroencephalogram signals Technical Field

[0001] The present invention relates to the field of brain-computer interface technology, and in particular to an intelligent ward rounds method and system based on SSVEP electroencephalogram (EEG) signals. Background Art

[0002] Research shows that one-third of stroke patients develop aphasia. Along with losing the ability to move their limbs, they also lose the ability to communicate, making it impossible for them to effectively communicate their needs with medical staff. Currently, brain-computer interface technology continues to advance, and its application in smart healthcare has become a trend. Steady-state visual evoked potentials (SSVEPs) offer advantages such as simple design, stable signals, a small number of leads, and high transmission rates. Using SSVEP EEG signals, they can identify patients' intentions, helping them communicate with medical staff and significantly improving their ability to express their needs.

[0003] However, due to the complex characteristics of EEG signals, there are a large number of noise signals of non-correlated brain activities and artifacts in the collected SSVEP EEG signals. Some of these noises are also correlated with the stimulation frequency recognition, such as electrooculographic artifacts. This makes the signal-to-noise ratio of SSVEP EEG signals very low, which seriously hinders the recognition of frequency.

[0004] While existing technologies employ methods such as empirical mode decomposition (EMD) and wavelet transforms to decompose and denoise EEG signals, these methods all have limitations. For example, EMD is prone to modal aliasing. Furthermore, these decomposition methods focus solely on the noise signal, without considering the boundary and relationship between the noise signal and the valid signal. By simply filtering out the calculated "noise," they can cause varying degrees of feature loss in the original EEG signal.

[0005] Summary of the Invention

[0006] In response to the problems existing in the above-mentioned prior art, the present invention provides an intelligent ward round method and system based on SSVEP EEG signals, with the aim of ensuring a high signal-to-noise ratio and high recognition rate of SSVEP EEG signals while taking into account real-time performance, so as to ensure that users can communicate with medical staff without obstacles to complete intelligent ward round needs.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides an intelligent ward rounds method based on SSVEP electroencephalogram (EEG) signals, comprising:

[0009] Obtaining a preprocessed EEG signal of the user under the stimulation of a current visual stimulus source, recording it as a first EEG signal, and defining a reference signal according to the visual stimulus frequency that induces the first EEG signal;

[0010] Decomposing the first EEG signal into sub-band components distributed in different frequency ranges, calculating the sub-band components themselves, the reference signal itself, and the differences between the sub-band components and the reference signal, and obtaining a total spatial filter for each sub-band component;

[0011] The sub-band components are respectively processed by the total spatial filter, and rearranged into a new multi-channel signal recorded as a second EEG signal;

[0012] Obtaining multiple variational modal components of the second EEG signal using variational modal decomposition, and optimizing the weights of the variational modal components in each channel to reconstruct the EEG signal to obtain a third EEG signal;

[0013] The induced stimulation frequency of the third EEG signal is obtained based on the maximum correlation coefficient between the third EEG signal and itself and the reference signal, and the user intention is determined based on the induced stimulation frequency to complete the intelligent ward rounds.

[0014] In a second aspect, the present invention provides an intelligent ward rounds system based on SSVEP electroencephalogram (EEG) signals, comprising:

[0015] SSVEP signal acquisition module, used to collect the user's first EEG signal under the stimulation of the current visual stimulus source;

[0016] a reference signal definition module, configured to define a reference signal according to the visual stimulus frequency that induces the first EEG signal;

[0017] a sub-band decomposition module, configured to decompose the first EEG signal into sub-band components distributed in different frequency ranges;

[0018] a subband filtering module, configured to calculate the subband component itself, the reference signal itself, and the difference between the subband component and the reference signal, obtain a total spatial filter for each subband component, and filter the subband component;

[0019] a signal arrangement module, configured to rearrange the sub-band components processed by the total spatial filter into a second EEG signal;

[0020] a signal decomposition and reconstruction module, configured to obtain a plurality of variational modal components of the second EEG signal according to variational modal decomposition, and optimize the weights of the variational modal components in each channel to reconstruct the EEG signal to obtain a third EEG signal;

[0021] The stimulation frequency determination module obtains the induced stimulation frequency of the third EEG signal for determining the user's intention during intelligent ward rounds based on the maximum correlation coefficient between the third EEG signal and itself and the reference signal.

[0022] The intelligent ward rounds method and system based on SSVEP electroencephalogram signals of the present invention have the following beneficial effects:

[0023] 1. The present invention rearranges the original SSVEP EEG signal into a new multi-channel signal using TRCA filtering features, then uses variational mode decomposition to decompose the signal into multiple variational mode components. The parameters of the weights of each variational mode component are optimized using the sparrow search algorithm. The weighted reconstructed EEG signal is subjected to canonical correlation analysis with the average signal and the sine and cosine reference signal, different correlation coefficients are calculated, and the correlation coefficient with the highest correlation coefficient is selected as the final recognition target to obtain the SSVEP EEG signal frequency recognition result. By decomposing and reconstructing the signal multiple times, the effective part of the signal is enhanced, the influence of the noise artifact part in the SSVEP EEG signal is minimized to the greatest extent, and the extracted SSVEP EEG signal has a higher signal-to-noise ratio. In the correlation coefficient calculation process, the correlation between the average feature and the reference signal is calculated separately, taking into account the influence of different individuals and trials, and further ensuring the accuracy of frequency recognition under the premise of a high signal-to-noise ratio. This enables the recognition of the user's intention through the SSVEP EEG signal, helping users communicate with medical staff and greatly improving the user's ability to express their needs.

[0024] 2. Considering that the SSVEP EEG signal is weak and easily affected by some other electrical signals that are correlated with the stimulation frequency recognition, the present invention solves the problem in the prior art that the task-related components extracted from the SSVEP EEG signal by TRCA may be mixed with task-related component noise by maximizing the differences between the sub-band components, the reference signal itself, and the differences between the sub-band components and the reference signal. The noise is filtered out in each sub-band component, reducing the influence between each sub-band component and not easily introducing unwanted frequency information, rather than filtering out the overall noise of the first EEG signal. This more meticulously maximizes the relevance of the visual stimulation task and further improves the signal-to-noise ratio.

[0025] 3. The present invention decomposes the signal into multiple variational modal components through variational modal decomposition for processing, thereby minimizing the influence of artifacts. First, the number of variational modal components is determined by singular values ​​to separate the noise and effective signal in the variational modal components after decomposition as much as possible. The variational modal components obtained by decomposition are weighted by using a sparrow search algorithm, and a larger weight is given to the effective component to effectively filter out the noise signal and improve the frequency recognition accuracy. The number of better variational modal components is preliminarily determined to ensure the real-time frequency recognition of the SSVEP EEG signal. The difference between the variational modal component and the original signal is measured again, and the weight of each variational modal component is automatically optimized using a sparrow search algorithm. This avoids the problem in the prior art that the residual components of the variational modal decomposition are directly removed by using a predetermined standard, which may affect the effective EEG signal to varying degrees. While reducing the influence of the noise on the SSVEP EEG signal, the relationship between different channels and different frequency signals in the original signal is reproduced as much as possible, thereby improving the accuracy of frequency recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] FIG1 is a flow chart of an intelligent ward rounds method based on SSVEP EEG signals according to the present invention;

[0027] FIG2 is a schematic diagram of a process of decomposing and optimizing the weights of each channel to reconstruct a third EEG signal according to the present invention;

[0028] FIG3 is a schematic diagram of the overall framework of the intelligent ward rounds process based on SSVEP EEG signals of the present invention;

[0029] FIG4 is a structural block diagram of an intelligent ward rounds system based on SSVEP electroencephalogram signals according to the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0031] To facilitate understanding of this embodiment, an intelligent ward rounds method based on SSVEP electroencephalogram signals disclosed in an embodiment of the present invention is first introduced in detail.

[0032] An embodiment of the present invention provides an intelligent ward rounds method based on SSVEP electroencephalogram (EEG) signals, as shown in FIG1 , including:

[0033] S1, obtaining a pre-processed EEG signal of the user under the stimulation of a current visual stimulus source and recording it as a first EEG signal, and defining a reference signal according to the visual stimulus frequency that induces the first EEG signal;

[0034] S2, decomposing the first EEG signal into sub-band components distributed in different frequency ranges, calculating the difference between the sub-band components themselves, the reference signal itself, and the difference between the sub-band components and the reference signal, and obtaining a total spatial filter of each sub-band component;

[0035] S3, processing the sub-band components respectively through the total spatial filter, and rearranging them into new multi-channel signals recorded as the second EEG signal;

[0036] S4, using variational mode decomposition to obtain multiple variational mode components of the second EEG signal, and optimizing the weights of the variational mode components in each channel to reconstruct the EEG signal to obtain a third EEG signal;

[0037] S5, determining a correlation coefficient based on the self-features of the third EEG signal and the reference signal to obtain an induced stimulation frequency of the third EEG signal, determining the user intention based on the induced stimulation frequency, and completing intelligent ward rounds.

[0038] The EEG signal under the stimulation of the visual stimulus source is the EEG signal of the user under the rhythmic visual stimulation of the SSVEP paradigm. For the sake of convenience, the EEG signal of the user under the rhythmic visual stimulation of the SSVEP paradigm initially obtained is hereinafter referred to as the SSVEP signal.

[0039] Specifically, step S1 includes two aspects:

[0040] S11, pre-processing the acquired SSVEP signal, connecting the electrode to the input of the signal acquisition module to collect the SSVEP signal, amplifying, filtering and analog-to-digital conversion, and outputting the first EEG signal. Specifically,

[0041] S111, amplify the SSVEP signal.

[0042] Since the SSVEP signal amplitude is very weak, generally not exceeding 100 μV, it first needs to pass through a signal amplification circuit to amplify the SSVEP signal.

[0043] When collecting SSVEP signals, power supply noise interference is inevitable, and the SSVEP signal of the required specific frequency will be submerged, so a 50Hz notch filter is essential.

[0044] S112, power frequency filtering: filtering the SSVEP signal through a 50 Hz notch filter to remove the power frequency.

[0045] Where H(z) is the transfer function of the notch filter, z0 and is the conjugate pole in the complex plane, associated with the desired frequency of 50 Hz.

[0046] S113, frequency domain filtering: the SSVEP signal is passed through a Butterworth filter to obtain an SSVEP signal of a specific frequency (3-40 Hz) required by the present invention.

[0047] S114, analog-to-digital conversion: converting the collected SSVEP signal from an analog signal to a digital signal for subsequent analysis and processing.

[0048] S12, based on the visual stimulation frequency f that induces the first EEG signal i , i=1,2,…,N f , define the sine and cosine periodic signals as the reference signal Y f , the formula is as follows:

[0049] Among them, N h N represents the number of harmonics of the stimulation frequency that induces the first EEG signal. s Indicates the number of channels, f s Indicates the sampling frequency, N f is the total number of stimuli.

[0050] Obtaining the subband components of the first EEG signal in step S2 includes the following steps:

[0051] S21, determining the number K of sub-band components based on the frequency range of the first EEG signal, wherein the analysis filter bank includes a plurality of band-pass filters, each band-pass filter is used to extract information of a specific frequency range of the first EEG signal to obtain a sub-band component;

[0052] Specifically, the frequency range [f1, f2] of the first EEG signal is obtained, where f1 and f2 represent the low-frequency and high-frequency cutoff frequencies of the first EEG signal, respectively. In this embodiment, f1 = 3 Hz and f2 = 40 Hz. The number of subband components, i.e., the number of bandpass filters, is determined based on the frequency range and frequency step size of the first EEG signal. In this embodiment, a frequency step size of 4 Hz is used, i.e., a subband component is extracted every 4 Hz.

[0053] S22, calculating the center frequency of each bandpass filter, wherein the center frequency is selected based on equally spaced frequencies within the frequency range of the first EEG signal;

[0054] The center frequency in this embodiment is denoted as f c , f c Take frequencies that are equally spaced between f1 and f2 to ensure that the entire frequency range of the first EEG signal is covered.

[0055] S23, establishing a bandpass filter for each center frequency, wherein the transfer function of the bandpass filter is:

[0056] Among them, f c represents the center frequency, and z represents a complex variable.

[0057] S24, decomposing the first EEG signal into K segments in the frequency domain through the analysis filter bank, where each segment is a subband component, to obtain K subband components of the first EEG signal, each subband component being distributed in a different frequency range.

[0058] By establishing multiple discrete bandpass filters, the first EEG signal passes through each bandpass filter, retaining only the frequencies near the specific center frequency of the bandpass filter. This results in signal features within each frequency range (i.e., subband components). Each subband component contains only the features of a small segment of the first EEG signal, making the effective features within each segment more similar and the noise more pronounced. By using each segment separately for subsequent difference index extraction and finding the correlation between signals within each small frequency range, noise segments unrelated to the SSVEP stimulus can be removed more accurately and meticulously.

[0059] Step S2 calculates the subband component itself, the reference signal itself, and the difference between the subband component and the reference signal to obtain the total spatial filter of each subband component:

[0060] The TRCA method is used to maximize the trial difference index of the same subband component in different trials under the same stimulus, the reference difference index between the same subband component in different trials under the same stimulus and the reference signal, and the template difference index between the reference signals to obtain the spatial filter of the subband component under the corresponding stimulus. All spatial filters under each stimulus are connected to obtain the total spatial filter of each subband.

[0061] The trial difference index of the same subband component in different trials under the same stimulation is the difference between the subband component itself under the same stimulation. The reference difference index between the same subband component in each trial under the same stimulation and the reference signal is the difference between the subband component and the reference signal under the same stimulation. The template difference index between the reference signals is the difference between the sine and cosine of the reference signal.

[0062] The trial difference index, reference difference index, and template difference index are measured by covariance, specifically:

[0063] The trial difference index is the sum of the covariances of the same sub-band components in different trials under the same stimulus, denoted as

[0064] Where i and j represent different trials under the same stimulus, and i≠j, K represents the Kth subband component, N t represents the number of trials, Cov() represents the covariance;

[0065] The reference difference index includes the sum of the covariances S between the same sub-band component in each trial under the same stimulation and the sinusoidal periodic signal in the reference signal. 21 And the sum of the covariances S between the same sub-band component in each trial under the same stimulation and the cosine periodic signal in the reference signal 22 ;

[0066] in is the sinusoidal periodic signal in the reference signal;

[0067] is the sinusoidal periodic signal in the reference signal;

[0068] represents the sub-component of the K-th sub-band component in the i-th trial;

[0069] The template difference index is the sum of the covariances between the sine and cosine periodic signals.

[0070] Record in are the sine periodic signal and cosine periodic signal in the reference signal respectively.

[0071] The total spatial filter for each subband is specifically obtained as follows:

[0072] Construct a difference matrix based on the trial difference index, reference difference index, and template difference index

[0073] Solution The weight of the spatial filter is obtained to construct the spatial filter corresponding to the stimulus; is the weight of the spatial filter, is the difference matrix of the Kth subband component under the nth stimulus;

[0074] The final spatial filters of K subband components are obtained by concatenating the weights of all spatial filters under each stimulus: where N f represents the total number of stimuli.

[0075] Considering that the SSVEP signal inevitably contains some other task-related electrical signals, such as electrooculogram (EOG) noise, which may overlap in frequency with the SSVEP signal, simply put, due to the normal physiological activities of the human body, the EOG noise generated by the rhythmic visual stimulation of the SSVEP paradigm will also be related to the SSVEP stimulation, and these noises cannot be removed by frequency filtering. The present invention addresses the problem in the prior art that task-related components extracted from SSVEP signals through TRCA may be mixed with task-related component noise by maximizing the differences between sub-band components, the reference signal itself, and the differences between the sub-band components and the reference signal. Noise is filtered out in each sub-band component, rather than filtering out the overall noise of the first EEG signal, thereby more meticulously maximizing the relevance of the visual stimulation task and further improving the signal-to-noise ratio of the first EEG signal.

[0076] Step S3 is specifically as follows:

[0077] The first EEG signal after TRCA spatial filtering is filtered according to N f The sequence of spatial filters is vertically organized into new channels along the y-axis;

[0078] In each new channel, the K sub-band components are reorganized along the z-axis to obtain a new signal formed by rearrangement and recorded as the second EEG signal.

[0079] For N f Stimulus targets can get N f The filtered signals processed by different spatial filters have complementary information of the first EEG signal, just like the signals collected by electrodes at different scalp locations. Therefore, the first EEG signals processed by different spatial filters are first arranged into a vertical column along the y-axis, and each row represents a first EEG signal processed by a spatial filter. Then, in each row, the K sub-band components of the first EEG signal are arranged along the z-axis to obtain the second EEG signal. In this way, each row of the z-axis represents a first EEG signal, and each column of the y-axis represents a sub-band component, which facilitates the combination of the complementary information of the SSVEP signal, and each sub-band component is filtered out as much as possible after the above-mentioned maximization difference method.

[0080] Referring to FIG2 , step S4 specifically includes:

[0081] S41, representing the second EEG signal as the sum of multiple variational modal components and performing singular value decomposition on the second EEG signal;

[0082] Among them, channel N s The i-th modal component of Indicates channel N s The instantaneous amplitude of the i-th modal component, Indicates channel N s The instantaneous phase function of the i-th modal component; then the second EEG signal is expressed as

[0083] S42, determining the number P of variational modal components based on the change trend of the singular value;

[0084] S43, constructing a variational model based on the number P of the variational modal components, and solving the variational model using an alternating direction multiplier algorithm to obtain multiple variational modal components of the second EEG signal;

[0085] The variational mode component contains the local information characteristics of the original signal at different time scales, which can stabilize the non-stationary SSVEP signal data. At the same time, variational mode decomposition avoids the mode aliasing problem existing in empirical mode decomposition.

[0086] S44, weighting the variational modal components of each frequency band using a sparrow search algorithm, wherein the weights of the variational modal components are determined according to the fitness value of the sparrow search algorithm;

[0087] S45 , reconstructing the EEG signal of each channel based on the weight of the variational modal component to obtain a third EEG signal.

[0088] S42 specifically includes:

[0089] S421, constructing an m×n-order Hankel matrix based on the variational modal component and the number of channels of the second EEG signal, where N s is the number of channels, [] means rounding up;

[0090] According to the second EEG signal and the number of channels N s The resulting Hankel matrix is:

[0091] Perform singular value decomposition on the Hankel matrix, H = U∑V T , where U and V are orthogonal matrices, respectively, left and right singular matrices, ∑=diag(σ1,σ2,…,σ r ), σ i are the singular values ​​of the matrix H, and r is the number of singular values.

[0092] S422, performing singular value decomposition on the Hankel matrix, and sorting the obtained singular values ​​in descending order;

[0093] S423, plot i-σ in descending order i Singular value map, obtain the horizontal coordinate I corresponding to the starting point of the line segment with the largest slope in the i-σ singular value map, where σi represents the i-th singular value after descending sorting;

[0094] S424 , determining the number P of variational modal components according to I=2P.

[0095] It should be noted that when the number of decomposed variational modal components is optimal, the residual after variational modal decomposition is a noise signal. Therefore, the choice of the number of variational modal components determines the noise representation ability of the decomposed signal. While existing optimization algorithms have strong global search capabilities, they suffer from long iterations and high computational complexity, making them incapable of meeting the real-time requirements of SSVEP EEG signal recognition. Furthermore, they are prone to falling into local optima. Therefore, this embodiment proposes using the singular value size to determine the number of variational modal components.

[0096] Since large singular values ​​correspond to the signal's primary information, they represent valid signals with large spectra and high energy, while small singular values ​​represent noise signals with small spectra and low energy, or invalid signals. Therefore, the singular values ​​are arranged from large to small, and the dividing point between "large" and "small" singular values ​​(i.e., the point where the slope is maximum) is found based on their changing trends. This dividing point can separate valid and invalid signals and is the point where the number of variational modal components is optimal.

[0097] S43 specifically includes:

[0098] S431, calculating the analytical signal of each variational modal component using Hilbert transform;

[0099] Among them, u i (t) is the i-th variational mode component, represents the analytical signal of the i-th variational mode component, δ(t) represents the impulse function, j is the imaginary unit, and * represents the convolution operation.

[0100] S432, estimating the center frequency of each analytical signal, and shifting the analytical signal to baseband using a frequency shift operation;

[0101] where ω i is the center frequency of the analytical signal of the i-th variational mode component, is the index correction term.

[0102] S433, calculating the gradient square norm of the analytical signal after being shifted to the baseband and performing Gaussian smoothing to estimate the bandwidth of each variational modal component and construct a constrained variational model;

[0103] Where X(t) is the second EEG signal, P is the total number of variational modal components, It means partial derivative with respect to t.

[0104] S434, converting the constrained variational model into an unconstrained variational model using a quadratic penalty factor and a Lagrangian operator;

[0105] Where α is the quadratic penalty factor, λ(t) is the Lagrangian operator, and <> represents the inner product operation.

[0106] S435, using an alternating direction multiplier algorithm to solve the unconstrained variational model, alternately updating the variational modal components, the center frequency, and the Lagrangian operator in the unconstrained variational model, and optimizing the variational modal components to obtain optimized variational modal components;

[0107] S436: When the optimized variational modal components satisfy the discrimination accuracy condition, the iteration is completed to output each variational modal component.

[0108] The discrimination accuracy condition is set to meet When , output each variational modal component, otherwise return to continue iteration. is the nth iteration value of the ith variational mode component, is the (n+1)th iteration value of the ith variational mode component, and ε>0 is the preset discrimination accuracy.

[0109] By extracting effective variational modal components from multi-channel SSVEP EEG signals, the influence of irrelevant brain activity and artifacts (noise, i.e., invalid signals) is reduced. At the end of each iteration, new variational modal components are obtained based on the optimized variational modal components. These variational modal components have different frequencies and amplitudes to capture different components of the signal. When the variational modal components reach a certain stability (satisfying the discrimination accuracy), the iteration is terminated.

[0110] It should be noted that when the number of decomposed variational modal components is optimal, the remainder after variational modal decomposition is noise. Therefore, the existing technologies all determine the number of variational modal decompositions of the variational modal decomposition algorithm to decompose the signal as optimally as possible. For example, the number of optimal variational modal components is determined by the correlation coefficient between the remainder and the superposition signal of the variational modal components. However, this only achieves qualitative analysis. The remainder after decomposition cannot be all of the noise. Other variational modal components also contain a small amount of noise signals. At the same time, after removing the variational modal residual components through artificially set standards, the denoised signal is reconstructed through the inverse operation of the algorithm. The removed residual components may contain valid signals, so that directly deleting the component segment will cause "deformation" of the EEG signal to varying degrees. The unremoved part may also contain noise parts, so that noise still exists in the signal obtained. Considering that it is impossible to quantitatively focus on the noise signal to minimize its impact, the present invention adopts the method of assigning different weights to different variational modal components to increase the proportion of valid components and reduce the proportion of invalid components. Specifically including:

[0111] The relative entropy between the variational modal component and the second EEG signal is used as the fitness function of the sparrow search algorithm; the larger the fitness function value of the variational modal component, the greater the weight.

[0112] Relative entropy is a measure of the asymmetry of the difference between two probability distributions. The present invention uses relative entropy to measure the difference between each variational modal component and the original undecomposed signal. It should be noted that the smaller the relative entropy between the component and the original signal, the closer the two are. In the collected SSVEP signal, the noise component is much smaller than the real signal part. Therefore, the component with a small difference from the original signal is recorded as a valid component and is given a larger weight; the component with a large difference from the original signal is recorded as an invalid component and its weight is reduced. After weighting the weights through the sparrow search algorithm, the reconstructed signal can increase the proportion of valid components in the signal and reduce the proportion of invalid components, thereby achieving the effect of removing artifacts.

[0113] Specifically, the fitness function is

[0114] in represents relative entropy, X(t) is the second EEG signal, u i (t) is the i-th variational mode component.

[0115] Specifically, the steps of the above sparrow search algorithm are as follows:

[0116] S441, based on the weight range of the variational modal component, initialize the population, the number of iterations, and the ratio of discoverers, followers, and guardians, and randomly generate initial positions of sparrows, where each sparrow represents a weight of a variational modal component;

[0117] S442, each sparrow randomly selects a target point at its current position and moves toward the target point, and a fitness function value of each sparrow is calculated based on the relative entropy of the variational modal component and the second EEG signal;

[0118] S443, updating the location information of the discoverer, follower, and guardian;

[0119] S444, determine whether the iteration stop condition is met, if so, output the fitness function value of all current sparrows to obtain the weight of each variational mode component; otherwise, return to step S442.

[0120] The weight W of each variational mode component (IMF) is obtained through the above i,P =[W i,1 ,W i,2 ,…,W i,P ], and the weighted EEG signal is obtained: X i =W i1 ×IMF i1 +W i2 ×IMF i2 +…+W iP ×IMF iP ,

[0121] Among them, X i represents the EEG signal after weighted reconstruction of the i-th channel, P is the total number of variational modal components, W iP For different weights.

[0122] S5 specifically includes: using a canonical correlation analysis method to calculate the Pearson coefficients of the third EEG signal, the average signal of the third EEG signal, and the reference signal as correlation coefficients, and taking the frequency corresponding to the maximum correlation coefficient as the induced stimulation frequency of the third EEG signal.

[0123] For each pair of two signals, a canonical correlation analysis method is used to create two weight vectors that maximize the correlation between the two weight vectors and the linear combination of the corresponding signals. These weight vectors act as spatial filters to filter the two signals, and the Pearson coefficient between the filtered signals is calculated. The correlation coefficient is obtained by summing the weighted squares of the multiple Pearson coefficients obtained for multiple pairs of signals.

[0124] The Pearson coefficients of the third EEG signal, the average signal of the third EEG signal, and the reference signal calculated by using the canonical correlation analysis method are specifically:

[0125] The spatial filter of the third EEG signal and the reference signal is calculated using a canonical correlation analysis method to obtain a first weight vector The third EEG signal X and the reference signal Y are obtained based on the first weight vector f The first Pearson coefficient ρ1;

[0126] ρ() means calculating the Pearson coefficient.

[0127] The spatial filter of the third EEG signal and the average signal of the third EEG signal is calculated by using the canonical correlation analysis method to obtain the second weight vector Calculate the average signal of the third EEG signal X and the third EEG signal based on the second weight vector The second Pearson coefficient of

[0128] The average signal of the third EEG signal is N for the same stimulus. t The average value of the third EEG signal in each trial.

[0129] Therefore the correlation coefficient is:

[0130] R i represents the correlation coefficient between the ith stimulus and the reference signal, represents the first Pearson coefficient of the third EEG signal under the i-th stimulus, represents the second Pearson coefficient of the third EEG signal under the i-th stimulus, and sign() is a sign function used to characterize the positive or negative correlation.

[0131] Then the induced stimulation frequency of the third EEG signal is where i=1,2,…,N f Indicates the number of stimulus targets.

[0132] The overall framework of the intelligent ward rounds process based on SSVEP EEG signals of the present invention is shown in FIG3 .

[0133] Before determining the correlation coefficient, this embodiment reduces the influence of noise and artifacts by multiple decomposition and reconstruction of the SSVEP signal and assigning greater weights to the effective components therein. Therefore, the signal-to-noise ratio of the obtained signal is high. Therefore, the self-feature information is extracted by averaging the experimental signal itself, reducing the influence of individuals and trials. The average feature has higher accuracy and robustness. Combined with the correlation coefficient calculated with the reference signal, it can effectively ensure the frequency recognition accuracy of the signal under high signal-to-noise ratio.

[0134] Based on the assumption that filters of different targets in the same frequency band are similar to each other, the spatial filters obtained for different stimulus targets are integrated and combined for target recognition, which can significantly improve the performance of spatial filtering.

[0135] The present invention also provides an intelligent ward rounds system based on SSVEP electroencephalogram signals, as shown in FIG4 , comprising:

[0136] SSVEP signal acquisition module, used to collect the user's first EEG signal under the stimulation of the current visual stimulus source;

[0137] a reference signal definition module, configured to define a reference signal according to the visual stimulus frequency that induces the first EEG signal;

[0138] a sub-band decomposition module, configured to decompose the first EEG signal into sub-band components distributed in different frequency ranges;

[0139] a subband filtering module, configured to calculate the subband component itself, the reference signal itself, and the difference between the subband component and the reference signal, obtain a total spatial filter for each subband component, and filter the subband component;

[0140] a signal arrangement module, configured to rearrange the sub-band components processed by the total spatial filter into a second EEG signal;

[0141] a signal decomposition and reconstruction module, configured to obtain a plurality of variational modal components of the second EEG signal according to variational modal decomposition, and optimize the weights of the variational modal components in each channel to reconstruct the EEG signal to obtain a third EEG signal;

[0142] The stimulation frequency determination module determines the correlation coefficient based on the self-features of the third EEG signal and the reference signal to obtain the induced stimulation frequency of the third EEG signal for determining the user's intention during intelligent ward rounds.

[0143] The present invention is not limited to the above-mentioned specific implementation methods. Various changes made by ordinary technicians in this field based on the above-mentioned concept without creative work are all within the scope of protection of the present invention.

Claims

1. An intelligent ward-round method based on SSVEP electroencephalogram signals, characterized in that, it includes the following steps: S1. Obtain the electroencephalogram signal of the user under the current visual stimulus source after preprocessing, denoted as the first electroencephalogram signal, and define a reference signal according to the visual stimulus frequency that induces the first electroencephalogram signal; S2. Decompose the first electroencephalogram signal into sub-band components distributed in different frequency ranges, calculate the differences between the sub-band components themselves, the reference signal itself, and between the sub-band components and the reference signal, and obtain the total spatial filter for each sub-band component; S3. Process the sub-band components respectively through the total spatial filter and rearrange them into a new multi-channel signal, denoted as the second electroencephalogram signal; S4. Use variational mode decomposition to obtain multiple variational mode components of the second electroencephalogram signal, and optimize the weights of the variational mode components in each channel to reconstruct the electroencephalogram signal, obtaining the third electroencephalogram signal; S5. Based on the self-characteristics of the third electroencephalogram signal and the reference signal, determine the correlation coefficient to obtain the induced stimulus frequency of the third electroencephalogram signal, and determine the user's intention based on the induced stimulus frequency to complete the intelligent ward-round.

2. The intelligent ward-round method based on SSVEP electroencephalogram signals according to claim 1, characterized in that, the preprocessing in step S1 includes: amplifying, filtering, and analog-to-digital converting the obtained electroencephalogram signal of the user under the rhythmic visual stimulus of the SSVEP paradigm; wherein the filtering process includes passing the electroencephalogram signal through a 50Hz notch filter and a 3 - 40Hz Butterworth filter in sequence to obtain the first electroencephalogram signal.

3. The intelligent ward-round method based on SSVEP electroencephalogram signals according to claim 1, characterized in that, defining the reference signal according to the visual stimulus frequency that induces the first electroencephalogram signal in step S1 includes: Based on the visual stimulus frequency that induces the first electroencephalogram signal, construct a sine-cosine periodic signal as the reference signal, and the frequency of the reference signal is the same as or a multiple frequency of the visual stimulus frequency that induces the first electroencephalogram signal.

4. The intelligent ward-round method based on SSVEP electroencephalogram signals according to claim 3, characterized in that, specifically decomposing the first electroencephalogram signal into sub-band components distributed in different frequency ranges in step S2 is to construct a decomposition filter bank, and decompose the first electroencephalogram signal into multiple frequency bands using the decomposition filter bank to obtain the sub-band components of the first electroencephalogram signal, specifically including: S21. Determine the number K of sub-band components based on the frequency range of the first electroencephalogram signal. The decomposition filter bank includes multiple band-pass filters, and each band-pass filter is used to extract the information of a specific frequency range of the first electroencephalogram signal to obtain a sub-band component; S22. Calculate the center frequency of each band-pass filter, and the center frequency is selected based on the equally spaced frequencies between the frequency ranges of the first electroencephalogram signal; S23. Establish a band-pass filter for each center frequency, and the transfer function of the band-pass filter is as follows: where f c represents the center frequency and z represents a complex variable. S24. Decompose the first EEG signal into K segments in the frequency domain through the decomposition filter bank. Each segment is a sub-band component, and K sub-band components of the first EEG signal are obtained. Each sub-band component is distributed in a different frequency range.

5. The intelligent ward round method based on SSVEP EEG signals according to claim 4, wherein, in step S2, calculating the differences between the sub-band component itself, the reference signal itself, and between the sub-band component and the reference signal to obtain the total spatial filter for each sub-band component specifically includes: Using the TRCA method to maximize the trial difference index of the same sub-band component in different trials under the same stimulus, the reference difference index between the same sub-band component and the reference signal in different trials under the same stimulus, and the template difference index between the reference signals, to obtain the spatial filter of the sub-band component under the corresponding stimulus, and connecting all the spatial filters under each stimulus to obtain the total spatial filter for each sub-band.

6. The intelligent ward round method based on SSVEP EEG signals according to claim 5, wherein, the trial difference index, the reference difference index, and the template difference index are measured by covariance, specifically: The trial difference index is the sum of the covariances of the same sub-band component in different trials under the same stimulus, denoted as where i and j represent different trials under the same stimulus, and i ≠ j, K represents the Kth subband component, N t represents the number of trials, and Cov() represents covariance; The reference difference index includes the sum S of the covariances between the same sub-band component in each trial under the same stimulus and the sine periodic signal in the reference signal 21 and the sum S of the covariances between the same sub-band component in each trial under the same stimulus and the cosine periodic signal in the reference signal 22 ; Among them is the sinusoidal periodic signal in the reference signal; is the sinusoidal periodic signal in the reference signal; represents the sub-component of the Kth sub-band component in the ith trial; the template difference index is the sum of the covariances between the sine and cosine periodic signals, Denoted as Among them are the sinusoidal periodic signal and the cosine periodic signal in the reference signal respectively.

7. The intelligent ward round method based on SSVEP EEG signals according to claim 6, wherein, the obtaining of the total spatial filter for each sub-band specifically includes: Construct a difference matrix based on the trial difference index, reference difference index, and template difference index Solve Obtain the weights of the spatial filter to construct the spatial filter corresponding to the stimulus; among which is the weight of the spatial filter, is the difference matrix of the Kth sub-band component under the nth stimulus; The final spatial filters for the K sub-band components are obtained by concatenating the weights of all the spatial filters under each stimulus: where N f represents the total number of stimuli.

8. The intelligent ward round method based on SSVEP EEG signals according to claim 1, wherein, in S3, processing the sub-band components through the total spatial filter respectively and rearranging them into a new multi-channel signal denoted as the second EEG signal specifically includes: Organize the first EEG signal after TRCA spatial filtering vertically in the order of N f spatial filters into new channels along the y-axis; Recombining the K sub-band components along the z-axis in each new channel to obtain a newly rearranged signal denoted as the second EEG signal.

9. The intelligent ward round method based on SSVEP EEG signals according to claim 1, wherein, in S4, using variational mode decomposition to obtain multiple variational mode components of the second EEG signal, and optimizing the weights of the variational mode components in each channel to reconstruct the EEG signal, to obtain the third EEG signal including: S41. Represent the second EEG signal as the sum of multiple variational mode components and perform singular value decomposition on the second EEG signal; S42. Determine the number P of variational mode components based on the changing trend of the singular values; S43. Construct a variational model based on the number P of variational mode components and use the alternating direction multiplier algorithm to solve the variational model to obtain multiple variational mode components of the second EEG signal; S44. Use the sparrow search algorithm to weight the variational mode components in each frequency band, where the weight of the variational mode component is determined according to the fitness value of the sparrow search algorithm; S45, Reconstruct the EEG signals under each channel based on the weights of the variational mode components to obtain the third EEG signal.

10. The intelligent ward-round method based on SSVEP EEG signals according to claim 9, wherein, S42 determining the number P of variational mode components based on the variation trend of the singular values includes: S421, construct an m×n order Hankel matrix based on the variational mode components and the number of channels of the second electroencephalogram signal, where n = N s -m + 1, N s is the number of channels, [] represents rounding up; S422, Perform singular value decomposition on the Hankel matrix and sort the obtained singular values in descending order; S423, Plot the sorted i-σ in descending order i Singular value graph, obtain the abscissa I corresponding to the starting point of the line segment with the largest slope in the i-σ singular value graph, where σ i represents the i-th singular value after sorting in descending order; S424, Determine the number P of variational mode components according to I = 2P.

11. The intelligent ward-round method based on SSVEP EEG signals according to claim 10, wherein, S43 constructing a variational model based on the number P of variational mode components and using the alternating direction multiplier algorithm to solve the variational model to obtain multiple variational mode components of the second EEG signal includes: S431, Calculate the analytic signal of each variational mode component using the Hilbert transform; S433, Calculate the gradient squared norm of the analytic signal after moving to the baseband for Gaussian smoothing to estimate the bandwidth of each variational mode component and construct a constrained variational model; S434, Use the quadratic penalty factor and the Lagrangian operator to convert the constrained variational model into an unconstrained variational model; S435, Use the alternating direction multiplier algorithm to solve the unconstrained variational model, alternately update the variational mode components, the center frequencies, and the Lagrangian operator in the unconstrained variational model, and optimize the variational mode components to obtain optimized variational mode components; S436, When the optimized variational mode components satisfy the discrimination accuracy condition, complete the iteration and output each variational mode component.

12. The intelligent ward-round method based on SSVEP EEG signals according to claim 9, wherein, S44 using the sparrow search algorithm to weight each frequency band variational mode component, and the weight of the variational mode component is determined according to the fitness function value of the sparrow search algorithm includes: Using the relative entropy between the variational mode component and the second EEG signal as the fitness function of the sparrow search algorithm; The larger the fitness function value of the variational mode component, the larger the weight.

13. The intelligent ward-round method based on SSVEP EEG signals according to claim 1, wherein, S5 uses the canonical correlation analysis method to calculate the Pearson correlation coefficients between the third EEG signal and the average signal of the third EEG signal, and with the reference signal as the correlation coefficients, and takes the frequency corresponding to the maximum correlation coefficient as the evoked stimulus frequency of the third EEG signal.

14. The intelligent ward-round method based on SSVEP EEG signals according to claim 13, wherein, The specific method of using the canonical correlation analysis method to calculate the Pearson correlation coefficients between the third EEG signal and the average signal of the third EEG signal, and with the reference signal is: Calculate the spatial filter of the third electroencephalogram signal and the reference signal by using the canonical correlation analysis method to obtain the first weight vector Calculating a first Pearson coefficient ρ of the third electroencephalogram signal X and the reference signal Y based on the first weight vector f ; 1 ; Calculate the spatial filter of the third EEG signal and the average signal of the third EEG signal using the canonical correlation analysis method to obtain the second weight vector Obtaining an average signal of the third electroencephalogram signal X and the third electroencephalogram signal based on the second weight vector The second Pearson correlation coefficient; 15. The intelligent ward-round method based on SSVEP EEG signals according to claim 14, wherein, The specific correlation coefficient is as follows: where R i represents the correlation coefficient between the i-th stimulus and the reference signal, The first Pearson coefficient of the third electroencephalogram signal under the i-th stimulus, Represents the second Pearson coefficient of the third electroencephalogram signal under the i-th stimulus, and sign() is the sign function used to characterize the positivity and negativity of the correlation. Then the induced stimulation frequency of the third electroencephalogram signal is where i = 1, 2, …, N f indicates the number of stimulation targets.

Images