Multiple epileptic seizure analysis method based on dynamic high-order directed network information flow

Abstract

The present invention discloses a method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow, which belongs to the field of image analysis. It solves the problems that simple brain network analysis technology does not take time factors into account, existing brain network technology fails to capture high-order electrode propagation patterns, and the complexity of high-order matrices greatly affects the performance of similarity indicators. It mainly captures the interaction of information flows between electrodes at different hierarchical levels by fusing high-order information, effectively eliminates matrix redundant information based on the non-similarity index of singular value decomposition, and quantitatively compares the differences in seizure paths. Ultimately, it successfully tracks the different patterns of propagation paths of multiple epileptic seizures, providing a reference value for surgical resection of patients with focal epilepsy.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow. Background Art

[0002] Focal epilepsy originates from a specific seizure zone and spreads to adjacent areas of the brain until the entire cortex is affected. Previous studies have found that some patients have similar seizure characteristics, while some patients have multiple types of seizure evolution. However, the treatment given clinically may only consider some seizures in patients with epilepsy, and seizures may still not be completely controlled after surgery. How do multiple epileptic seizures vary in individual patients? Existing network analysis technologies have not yet been able to accurately and effectively distinguish the differences between multiple seizures. Therefore, based on existing network analysis technologies, accurately and reliably observing the variability of epileptic seizures and objectively quantifying the similarity of seizure pathways are issues that urgently need to be addressed in brain and cognitive science research for patients with epilepsy.

[0003] Existing network analysis techniques have certain methodological limitations due to their inherent principles. For example, epileptic seizures have a temporal effect, and simple brain network analysis techniques fail to account for this factor. Existing brain network techniques can only capture low-level electrode propagation patterns, but epileptic seizures involve multiple layers and involve high-order information flow. Seizures last for varying lengths, and matrices containing both temporal and high-order information are complex and contain a certain amount of redundant information, which affects the performance of similarity metrics. Summary of the Invention

[0004] The present invention provides a method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow, aiming to solve the above-mentioned problems.

[0005] The present invention is implemented as follows: a method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow comprises the following steps:

[0006] Step S1: acquiring iEEG data of a patient with focal epilepsy, and preprocessing the iEEG data to obtain preprocessed iEEG data;

[0007] Step S2: Based on the preprocessed iEEG data, the time series of epileptic seizure stages is extracted, and each seizure is divided using a 1s sliding window;

[0008] Step S3: Under each window, construct an effective connection matrix through the extracted time series and add high-order information to obtain the corresponding dynamic high-order effective connection matrix;

[0009] Step S4: Analyze the evolution of multiple seizures in epileptic patients using a dynamic high-order effective connection matrix, and use a non-similarity index based on singular value decomposition to eliminate redundant matrix information and quantitatively compare the differences in seizure paths.

[0010] Furthermore, in step S1, the iEEG data were preprocessed using Brainstorm, specifically including: removing bad channels, removing spurious signals using ICA, band-pass filtering 0.16-97 Hz, and downsampling to 500 Hz.

[0011] Furthermore, in step S3, the corresponding dynamic high-order effective connection matrix is ​​obtained, which specifically includes:

[0012] By calculating the transfer entropy, the effective connection matrix between electrodes is obtained: given system X and system Y, and For the corresponding time series, p represents the probability of information transmission, and the degree of information flow transmission between variables is obtained by calculating the information entropy:

[0013]

[0014] The degree to which system X is affected by system Y is measured by the probability difference between the systems, namely the transfer entropy:

[0015]

[0016] in, represents the k-order X system, Represents the l-order Y system. On this basis, the K-order matrix constructed is as follows:

[0017] STE K =STE 1 *STE K-1 , k≥2

[0018] The false connections with p = 0.001 in the matrix were removed by FDR correction, and then the high-order effective connection matrix was calculated in each time window. The obtained matrices were combined according to the order of the time windows to obtain the dynamic high-order effective connection matrix H containing time information.

[0019] Furthermore, in step S4, the evolution process of multiple seizures of epileptic patients is analyzed, including reducing the dimensionality of the original data through multidimensional scaling analysis, keeping the relative relationship of the samples unchanged, and visualizing the path evolution.

[0020] Furthermore, the multidimensional scaling analysis specifically includes:

[0021] Calculate the distance matrix of electrode points in the matrix Map the D-dimensional space to the low-dimensional space Z, Z i and Zj Represent two different elements in the matrix Z respectively, and the formula is as follows:

[0022]

[0023] Normalize the vector Z, let

[0024]

[0025] Sum the distance formula, where N represents the number of electrodes:

[0026]

[0027]

[0028]

[0029] Define the inner product matrix B:

[0030] B=Z T Z

[0031] Through the above formula, the element value in B is derived:

[0032]

[0033] Perform eigenvalue decomposition on matrix B to obtain its eigenvalue matrix V and eigenvector matrix V T :

[0034] B=V∧V T

[0035] Take the first Z items with the largest eigenvalue matrix of matrix B and their corresponding eigenvectors to obtain matrix Z:

[0036] Z=V∧ 1 / 2 .

[0037] 6. The method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow according to claim 3, characterized in that in step S4, a non-similarity index based on singular value decomposition is used to eliminate matrix redundant information and quantitatively compare the differences in seizure paths, specifically comprising:

[0038] Perform singular value decomposition on the matrix H to obtain the matrix ∑ containing key information:

[0039] H=U∑V *

[0040] Where U represents the left singular matrix, V represents the right singular matrix, and ∑ is the singular value matrix;

[0041] For each episode, a corresponding matrix ∑ is obtained, and its Euclidean distance is calculated as the non-similarity index;

[0042] Compare the dissimilarity matrix G of multiple episodes of each patient, where n and m represent two episodes respectively. The lower the dissimilarity, the more similar the propagation networks of the two episodes are:

[0043] G(i, j)=|∑n ij -∑m ij |.

[0044] Furthermore, in step S1, the number of seizures of the patient with focal epilepsy is ≥4.

[0045] Compared with the prior art, the beneficial effects of the present invention are: the present invention discloses a method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow, which includes epileptic seizure time information, fuses high-order information to capture the interaction of information flow between electrodes at different hierarchical levels, visualizes path evolution through multidimensional scaling method, and proposes a non-similarity index based on singular value decomposition to effectively eliminate matrix redundant information and quantitatively compare the differences in seizure paths. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of the brain network analysis method of the present invention;

[0047] Figure 2 This is a technical diagram of a dynamic high-order effective network of the present invention;

[0048] Figure 3 This is a result diagram of the present invention for one patient. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0050] Example

[0051] See also Figure 1 The present invention provides a technical solution: a method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow, comprising the following steps:

[0052] Step S1: iEEG data from patients with focal epilepsy were used, requiring each patient to have ≥ 4 seizures; the data were preprocessed, including removing bad channels, removing spurious signals, bandpass filtering 0.16-97 Hz, and downsampling to 500 Hz;

[0053] Step S2: Extract the time series of the seizure phase based on the preprocessed iEEG data, and divide each seizure into 1s sliding windows.

[0054] Step S3: Based on the obtained time series, an effective connection matrix is ​​constructed in each window, and high-order information is added on this basis to obtain the corresponding dynamic high-order effective connection matrix;

[0055] Step S4: Based on the obtained matrix, the original data is reduced in dimension by multidimensional scaling, the relative relationship of the samples is kept unchanged, and the similarities and differences of the paths are visualized; the matrix complexity is high and there is a certain amount of data redundancy, based on which the present invention proposes a non-similarity index based on singular value decomposition;

[0056] In step S3, constructing a dynamic high-order effective connection matrix includes the following steps: (1) obtaining an effective connection matrix between electrodes by calculating transfer entropy; (2) constructing a high-order matrix based on the matrix obtained in (1); (3) removing false connections in the matrix by performing FDR (Benjamini and Hochberg) correction (p = 0.001) on the basis of the matrix obtained in (2); (4) calculating a high-order effective connection matrix in each time window, combining the obtained matrices according to the order of the time windows, and obtaining a high-order matrix containing time information;

[0057] The specific method of step S4 is as follows: (1) calculating the distance matrix of the electrode points in the matrix; (2) obtaining the inner product matrix by normalization and distance summation; (3) obtaining the eigenvalue matrix and eigenvector matrix of the matrix; (4) taking the largest preceding term of the matrix eigenvalue matrix and its corresponding eigenvector to obtain the matrix.

[0058] The specific method of step S4 is as follows: (1) performing singular value decomposition on the matrix to obtain a matrix containing key information; (2) obtaining a corresponding matrix for each attack, and calculating its Euclidean distance as a non-similarity index.

[0059] Test example

[0060] like Figure 1 As shown, the method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow of the present invention includes the following steps:

[0061] Step S1: iEEG data of patients with focal epilepsy were used, and each patient was required to have ≥ 4 seizures;

[0062] iEEG (intracranial electroencephalogram) was used in this study. Data from 29 patients with focal epilepsy were used, 11 of which were from the Bern Hospital in Switzerland and 15 from the Hospital of the University of Pennsylvania. Dynamic high-order effective connectivity was used to analyze the evolution of multiple epileptic seizures, and similarity metrics were used to quantitatively analyze the differences between these seizures.

[0063] iEEG data were preprocessed using Brainstorm, including removing bad channels, removing artifacts using ICA, bandpass filtering from 0.16 to 97 Hz, and downsampling to 500 Hz;

[0064] Step S2: Extract the time series of the seizure phase based on the preprocessed iEEG data, and divide each seizure into two parts using a 1-s sliding window;

[0065] Step S3: Based on the obtained time series, an effective connection matrix is ​​constructed in each window, and high-order information is added on this basis to obtain the corresponding dynamic high-order effective connection matrix;

[0066] like Figure 2 As shown, constructing a dynamic high-order effective connection matrix includes the following steps:

[0067] By calculating the transfer entropy, the effective connection matrix between electrodes is obtained: given system X and system Y, and For the corresponding time series, p represents the probability of information transmission, and the degree of information flow transmission between variables is obtained by calculating the information entropy:

[0068]

[0069] The degree to which system X is affected by system Y is measured by the probability difference between the systems, namely the transfer entropy:

[0070]

[0071] in, represents the k-order X system, Represents the l-order Y system. On this basis, the K-order matrix constructed is as follows:

[0072] STE K =STE 1 *STE K-1 , K≥2

[0073] The false connections with p = 0.001 in the matrix were removed by FDR correction, and then the high-order effective connection matrix was calculated in each time window. The obtained matrices were combined according to the order of the time windows to obtain the dynamic high-order effective connection matrix H containing time information.

[0074] Step S4: Based on the obtained matrix, the original data is reduced in dimension through multidimensional scaling analysis, the relative relationship of the samples is kept unchanged, and the path evolution is visualized; the matrix H is relatively complex and has a certain amount of data redundancy. Based on this, the present invention proposes a non-similarity index based on singular value decomposition;

[0075] The specific method of multidimensional scaling analysis is:

[0076] Calculate the distance matrix of electrode points in the matrix Map the D-dimensional space to the low-dimensional space Z, Z i and Z j Represent two different elements in the matrix Z respectively, and the formula is as follows:

[0077]

[0078] Normalize the vector Z, let

[0079]

[0080] Sum the distance formula, where N represents the number of electrodes:

[0081]

[0082]

[0083]

[0084] Define the inner product matrix B:

[0085] B=Z T Z

[0086] Through the above formula, the element value in B is derived:

[0087]

[0088] Perform eigenvalue decomposition on matrix B to obtain its eigenvalue matrix V and eigenvector matrix V T :

[0089] B=V∧V T

[0090] Take the first Z items with the largest eigenvalue matrix of matrix B and their corresponding eigenvectors to obtain matrix Z:

[0091] Z=V∧ 1 / 2 .

[0092] The matrix H is highly complex and has certain data redundancy. Based on this, the present invention proposes a non-similarity index based on singular value decomposition. The specific method is as follows:

[0093] Perform singular value decomposition on the matrix H to obtain the matrix ∑ containing key information:

[0094] H=U∑V *

[0095] Where U represents the left singular matrix, V represents the right singular matrix, and ∑ is the singular value matrix;

[0096] For each episode, a corresponding matrix ∑ is obtained, and its Euclidean distance is calculated as the non-similarity index;

[0097] Compare the dissimilarity matrix G of multiple episodes of each patient, where n and m represent two episodes respectively. The lower the dissimilarity, the more similar the propagation networks of the two episodes are:

[0098] G(i, j)=|∑n ij -∑m ij |.

[0099] Experimental results: The present invention successfully tracked the differences between multiple seizures of patients through a dynamic high-order effective brain network and performed quantitative analysis using similarity indicators.

[0100] like Figure 3 As shown, MDS visualization can accurately see the similarities and differences of the four episodes, and the dissimilarity index can clearly find that the third and fourth episodes are the most similar.

[0101] The present invention proposes an analysis method for multiple epileptic seizures based on dynamic high-order directed network information flow, which belongs to the field of image analysis. It solves the problems that simple brain network analysis technology does not take time factors into account, existing brain network technology fails to capture high-order electrode propagation patterns, and the complexity of high-order matrices greatly affects the performance of similarity indicators. It mainly captures the interaction of information flows between electrodes at different hierarchical levels by fusing high-order information, and effectively eliminates redundant matrix information based on the non-similarity index of singular value decomposition, and quantitatively compares the differences in seizure paths. Ultimately, it successfully tracks the different patterns of propagation paths of multiple epileptic seizures, providing a reference value for surgical resection of patients with focal epilepsy.

[0102] The above are only preferred embodiments of the present invention and are 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 method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow, characterized in that: The following steps are involved: Step S1: acquiring iEEG data of a patient with focal epilepsy, and preprocessing the iEEG data to obtain preprocessed iEEG data; Step S2: Based on the preprocessed iEEG data, the time series of epileptic seizure stages is extracted, and each seizure is divided using a 1s sliding window; Step S3: Under each window, construct an effective connection matrix through the extracted time series and add high-order information to obtain the corresponding dynamic high-order effective connection matrix; Step S4: Analyze the evolution of multiple seizures in epileptic patients using a dynamic high-order effective connection matrix, and use a non-similarity index based on singular value decomposition to eliminate redundant matrix information and quantitatively compare the differences in seizure paths; In step S3, the corresponding dynamic high-order effective connection matrix is ​​obtained, which specifically includes: By calculating the transfer entropy, the effective connection matrix between electrodes is obtained: given system X and system Y, and For the corresponding time series, p represents the probability of information transmission, and the degree of information flow transmission between variables is obtained by calculating the information entropy: The degree to which system X is affected by Y is measured by the probability difference between the systems, namely the transfer entropy: in, represents the k-order X system, Represents the l-order Y system. On this basis, the K-order matrix constructed is as follows: YOU K =STE 1 *YOU K-1 ,k≥2 The false connections with p = 0.001 in the matrix were removed by FDR correction, and then the high-order effective connection matrix was calculated in each time window. The obtained matrices were combined according to the time window order to obtain the dynamic high-order effective connection matrix H containing time information.

2. The method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow according to claim 1, characterized in that: In step S1, the iEEG data were preprocessed using Brainstorm, including: removing bad channels, removing spurious signals using ICA, band-pass filtering 0.16-97 Hz, and downsampling to 500 Hz.

3. The method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow according to claim 1, characterized in that: In step S4, the evolution process of multiple seizures in epileptic patients is analyzed, including reducing the dimensionality of the original data through multidimensional scaling analysis, keeping the relative relationship of the samples unchanged, and visualizing the path evolution.

4. The method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow according to claim 3, characterized in that: Multidimensional scaling analysis specifically includes: Calculate the distance matrix of electrode points in the matrix Map the D-dimensional space to the low-dimensional space Z, Z i and Z j Represent two different elements in the matrix Z respectively, and the formula is as follows: Normalize the vector Z, let Sum the distance formula, where N represents the number of electrodes: Define the inner product matrix B: B=Z T Z uses the above formula to deduce the element value in B: Perform eigenvalue decomposition on matrix B to obtain its eigenvalue matrix V and eigenvector matrix V T : B=V∧V T Take the first Z items with the largest eigenvalue matrix of matrix B and their corresponding eigenvectors to get matrix Z: Z=V∧ 1 / 2 。 5. The method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow according to claim 1, characterized in that: In step S4, a non-similarity index based on singular value decomposition is used to eliminate matrix redundant information and quantitatively compare the differences in seizure paths, specifically including: Perform singular value decomposition on the matrix H to obtain the matrix ∑ containing key information: H=U∑V * Where U represents the left singular matrix, V represents the right singular matrix, and Σ is the singular value matrix; For each episode, a corresponding matrix Σ is obtained, and its Euclidean distance is calculated as the non-similarity index; Compare the dissimilarity matrix G of multiple episodes of each patient, where n and m represent two episodes respectively. The lower the dissimilarity, the more similar the propagation networks of the two episodes are: G(i,j)=|Σn ij -Σm ij |。 6. The method for analyzing multiple epileptic seizures based on dynamic high-order directed network information flow according to claim 1, characterized in that: In step S1, the number of seizures of patients with focal epilepsy is ≥4.

Images