Brain dynamic analysis method, device, electronic device and storage medium

Abstract

The present invention provides a brain dynamic analysis method, device, electronic device and storage medium, which relate to the field of image processing technology. The brain dynamic analysis method includes obtaining multiple regions of interest and their resting-state networks of a first image; determining the average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is the average of the time series of all regions of interest corresponding to the resting-state network; performing a Hilbert transform on the average time series of each resting-state network to obtain the instantaneous phase of each resting-state network; generating a phase synchronization matrix based on the instantaneous phase of each resting-state network, and obtaining multiple dominant eigenvectors from the phase synchronization matrix; fitting the binarized dominant eigenvectors to a maximum entropy model, determining the energy value of each brain activity pattern, and constructing an energy landscape. The present invention can achieve a comprehensive analysis of brain dynamics.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a brain dynamic analysis method, device, electronic device and storage medium. Background Art

[0002] With the rapid development of neuroscience and computing technology, the study of brain dynamics has gradually become an important research direction in cognitive science and clinical neuroscience. Energy Landscape Analysis (ELA) is a method based on statistical physics. By mapping brain states to energy landscapes, it can reveal the stability of brain states and the energy barriers between brain states. Leading Eigenvector Dynamics Analysis (LEiDA) is a method based on graph theory and time series correlation. By extracting the leading eigenvectors of the instantaneous phase synchronization matrix between brain regions, it can capture the synchronization patterns between brain regions, thereby revealing the brain's transition process between different states.

[0003] In related technologies, the ELA method does not take into account phase synchronization between brain regions, which is an important feature of brain dynamics and reflects the coordination of neuronal activity in different brain regions. For example, in cognitive tasks, neurons in multiple brain regions may collaborate to complete information processing through phase synchronization. If this point is ignored, it will lead to an incomplete analysis of brain dynamics. The LEiDA method cannot directly quantify the energy barriers between brain states, and cannot accurately explain why certain brain state transitions are more likely to occur, and how the energy changes during the brain state transition. Summary of the Invention

[0004] The problem solved by the present invention is how to achieve a comprehensive analysis of brain dynamics.

[0005] To solve the above problems, the present invention provides a brain dynamic analysis method, device, electronic device and storage medium.

[0006] In a first aspect, the present invention provides a method for analyzing brain dynamics, comprising:

[0007] Acquire multiple regions of interest and resting-state networks thereof from a first image, wherein the first image comprises an fMRI image;

[0008] determining an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network;

[0009] performing a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks;

[0010] generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and obtaining a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks;

[0011] The binarized dominant eigenvector is fitted to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape.

[0012] Optionally, generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks includes:

[0013] determining an instantaneous phase difference between the resting-state networks based on the instantaneous phase of each of the resting-state networks;

[0014] According to the instantaneous phase difference between the resting-state networks, an instantaneous phase synchronization value between the resting-state networks is determined to generate the phase synchronization matrix.

[0015] Optionally, fitting the binarized dominant eigenvector to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape includes:

[0016] Binarizing the dominant eigenvector to determine the empirical frequency of each of the brain activity patterns, wherein a preset threshold for the binarization is zero;

[0017] determining a first average activity state of each of the resting-state networks and a first average interaction between the resting-state networks based on the empirical frequency of each of the brain activity patterns;

[0018] adjusting the first parameter and the second parameter of the maximum entropy model with the optimization objectives that the second average activity state of each of the resting-state networks in the maximum entropy model is equal to the first average activity state of each of the resting-state networks, and the second average interaction between the resting-state networks is equal to the first average interaction between the resting-state networks;

[0019] Based on the first parameter and the second parameter, the energy value of each brain activity pattern is determined to construct the energy landscape, wherein the first parameter is the basic activity of all the resting-state networks and the second parameter is the interaction between all the resting-state networks.

[0020] Optionally, adjusting the first parameter and the second parameter of the maximum entropy model with the second average activity state of each of the resting-state networks in the maximum entropy model being equal to the first average activity state of each of the resting-state networks, and the second average interaction between the resting-state networks being equal to the first average interaction between the resting-state networks as the optimization goal, includes:

[0021] Initializing the first parameter and the second parameter;

[0022] determining the second average activity state of each of the resting-state networks and the second average interaction between the resting-state networks based on the first parameter and the second parameter;

[0023] The first parameter and the second parameter are updated according to the first difference between the second average activity state of each resting-state network and the first average activity state of each resting-state network, and the second difference between the second average interaction between the resting-state networks and the first average interaction between the resting-state networks, until the first difference and the second difference both tend to zero.

[0024] Optionally, fitting the binarized dominant eigenvector to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape includes:

[0025] determining a local minimum value among all the brain activity patterns according to the energy value of each of the brain activity patterns;

[0026] constructing a disconnection map based on the local minima in all of the brain activity patterns;

[0027] Based on the disconnection graph, the energy landscape is constructed.

[0028] Optionally, after fitting the binarized dominant eigenvector to a maximum entropy model and determining the energy value of each brain activity pattern to construct an energy landscape, the brain dynamic analysis method further includes:

[0029] Based on the energy landscape, dynamic indicators of brain states are determined, wherein the dynamic indicators include fractional occupancy, dwell time, and transition probability.

[0030] Optionally, before acquiring the multiple regions of interest and their resting-state networks of the first image, the brain dynamic analysis method further includes:

[0031] The first image is preprocessed, wherein the preprocessing includes at least one of time correction, motion correction, scrubbing, co-registration, spatial normalization, bandpass filtering, spatial smoothing, and time normalization.

[0032] In a second aspect, the present invention provides a brain dynamic analysis device, comprising:

[0033] A first acquisition module is configured to acquire a plurality of regions of interest and resting-state networks thereof from a first image, wherein the first image comprises an fMRI image;

[0034] an averaging module, configured to determine an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network;

[0035] a transformation module, configured to perform a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks;

[0036] a second acquisition module, configured to generate a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and acquire a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks;

[0037] A construction module is used to fit the binarized dominant eigenvector to a maximum entropy model, determine the energy value of each brain activity pattern, and construct an energy landscape.

[0038] In a third aspect, the present invention provides an electronic device comprising a memory and a processor;

[0039] The memory is used to store computer programs;

[0040] The processor is used to implement the brain dynamic analysis method as described in the first aspect when executing the computer program.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the brain dynamic analysis method as described in the first aspect is implemented.

[0042] The beneficial effects of the brain dynamic analysis method, device, electronic device and storage medium of the present invention are as follows: a first image is obtained to provide activity information of the brain in a resting state. The first image is divided into multiple ROIs, and the resting-state network composed of these ROIs is determined, so as to facilitate subsequent processing in units of resting-state networks. The average value of the time series of all ROIs in each resting-state network is calculated to obtain the average time series of each resting-state network, which can reduce the influence of noise and data dimension and accurately reflect the overall activity pattern of each resting-state network. The average time series of each resting-state network is Hilbert transformed to obtain the instantaneous phase of each resting-state network, which can reflect the phase change of the BOLD signal in time and capture the instantaneous phase synchronization degree between resting-state networks, thereby generating a phase synchronization matrix. Extracting the dominant eigenvector from the phase synchronization matrix can not only reduce the data dimension and improve the signal-to-noise ratio, but also capture the main phase synchronization pattern at each time point. The binarized dominant eigenvector is fitted to the maximum entropy model to determine the energy value of each brain activity pattern and complete the construction of the energy landscape. Because the constructed energy landscape quantifies the stability of brain states and the energy barriers between brain states while taking into account the phase synchronization between resting-state networks, it can achieve a comprehensive analysis of brain dynamics. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the flow of a brain dynamic analysis method according to an embodiment of the present invention;

[0044] Figure 2 A disconnection diagram of the brain dynamic analysis method according to an embodiment of the present invention;

[0045] Figure 3 This is a system architecture diagram of a brain dynamic analysis device according to an embodiment of the present invention;

[0046] Figure 4 2 is a system architecture diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0048] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0049] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0050] It should be noted that the modifications of "one" and "multiple" mentioned in the present invention are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0051] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0052] like Figure 1 As shown, an embodiment of the present invention provides a brain dynamic analysis method, including:

[0053] S100: Acquire multiple regions of interest and resting-state networks of a first image, wherein the first image includes an fMRI image.

[0054] Functional Magnetic Resonance Imaging (fMRI) is a non-invasive brain imaging technique that reflects neural activity by measuring changes in the brain's blood oxygen level-dependent (BOLD) signal. The subject's brain is scanned using an fMRI device to obtain a primary image. Before scanning, the fMRI device parameters (such as magnetic field strength and scan time) should be appropriately set to ensure high-quality primary images that accurately reflect neural activity. During the scan, the subject must remain in a resting state to avoid interference from external stimuli. The primary image is then input into specialized image processing software (such as SPM or FSL) to identify multiple regions of interest (ROIs) and their resting-state networks. Professional image processing software can segment the primary image into multiple ROIs using standard templates (such as the Schaefer 100 atlas and the AAL atlas) or data-driven methods (such as independent component analysis, principal component analysis, and cluster analysis). Professional image processing software can then assign each ROI to a corresponding resting-state network based on the Schaefer 100 atlas or the functional connectivity patterns between ROIs. In some embodiments, the resting-state network includes the visual network (VIS), sensorimotor network (SMN), dorsal attention network (DAN), ventral attention network (SAN), limbic network (LMN), control network (CON), and default mode network (DMN).

[0055] S200: Determine an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network.

[0056] Specifically, each resting-state network is composed of multiple ROIs. The BOLD signal of each ROI will change over time to form a time series. By calculating the average of the time series of all ROIs in each resting-state network, that is, adding the values ​​of the BOLD signals of all ROIs in the same resting-state network at corresponding positions in the time series, and then dividing by the number of ROIs in the resting-state network, the average time series of each resting-state network is obtained. In some embodiments, assuming that the VIS consists of ROI A, ROI B, and ROI C, the time series of ROI A is (0.2, 0.5, 0.8, 0.6, 0.3), the time series of ROI B is (0.4, 0.6, 0.9, 0.5, 0.2), and the time series of ROI C is (0.3, 0.7, 0.7, 0.4, 0.1). Then, the average time series of the VIS is (0.3, 0.6, 0.8, 0.5, 0.2).

[0057] S300: Performing Hilbert transform on the average time series of each of the resting-state networks to obtain the instantaneous phase of each of the resting-state networks.

[0058] Specifically, in order to study the synchronization between different resting-state networks in the time dimension, the average time series of each resting-state network is Hilbert transformed to obtain the phase of each resting-state network at all time points, that is, all instantaneous phases. The Hilbert transform satisfies the following formula:

[0059]

[0060] in, is applied to real signals The Hilbert transform of is the Cauchy principal value, Is a real signal Time shift , It's time, It's about time Perform integration.

[0061] According to the real signal and applied to real signals Hilbert transform , we can get the complex signal , complex signal Determined by the following formula:

[0062]

[0063] According to the real signal and applied to real signals Hilbert transform , the instantaneous phase of each resting-state network can be obtained , instantaneous phase Determined by the following formula:

[0064]

[0065] S400: Based on the instantaneous phase of each of the resting-state networks, a phase synchronization matrix is ​​generated, and a plurality of dominant eigenvectors are obtained from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks.

[0066] Specifically, an instantaneous phase synchronization matrix can be generated based on the phases of all resting-state networks at the same time point. Multiple instantaneous phase synchronization matrices can be generated based on all instantaneous phases of each resting-state network. These multiple instantaneous phase synchronization matrices form a three-dimensional phase synchronization matrix, with the number of instantaneous phase synchronization matrices equal to the number of time points. Each instantaneous phase synchronization matrix is ​​then subjected to eigendecomposition to determine its eigenvalues ​​and eigenvectors. The eigenvector corresponding to the largest eigenvalue, known as the dominant eigenvector, is selected. Each element in the dominant eigenvector represents the projection of the instantaneous phase of each resting-state network onto the dominant eigenvector. The direction of the element represents the direction of the projection of the instantaneous phase of the resting-state network relative to the dominant eigenvector. When the element is positive, the instantaneous phase of the resting-state network is in the same direction as the dominant eigenvector; when the element is negative, the instantaneous phase of the resting-state network is in the opposite direction to the dominant eigenvector. The magnitude of the element represents the strength of the projection of the instantaneous phase of the resting-state network onto the dominant eigenvector, reflecting the contribution of the resting-state network to the current brain activity pattern.

[0067] S500: Fitting the binarized dominant eigenvector to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape.

[0068] Specifically, the maximum entropy model is a statistical modeling method based on information theory. Its core concept is to select a probability distribution that maximizes its entropy, subject to given constraints. This probability distribution is considered the least biased because it makes the fewest assumptions about unknown information among all possible probability distributions, thus avoiding overfitting. The present invention fits the binarized dominant eigenvectors to the maximum entropy model, determining the energy value of each brain activity pattern and thus constructing an energy landscape.

[0069] In this embodiment, obtaining a first image can provide information about the activity of the brain in a resting state. The first image is divided into multiple ROIs, and the resting-state network composed of these ROIs is determined, so as to facilitate subsequent processing in units of resting-state networks. The average value of the time series of all ROIs in each resting-state network is calculated to obtain the average time series of each resting-state network, which can reduce the influence of noise and data dimension and accurately reflect the overall activity pattern of each resting-state network. The average time series of each resting-state network is subjected to Hilbert transform to obtain the instantaneous phase of each resting-state network, which can reflect the phase change of the BOLD signal in time and capture the instantaneous phase synchronization degree between resting-state networks, thereby generating a phase synchronization matrix. Extracting the dominant eigenvector from the phase synchronization matrix can not only reduce the data dimension and improve the signal-to-noise ratio, but also capture the main phase synchronization pattern at each time point. Fitting the binarized dominant eigenvector to the maximum entropy model can determine the energy value of each brain activity pattern and complete the construction of the energy landscape. Because the constructed energy landscape quantifies the stability of brain states and the energy barriers between brain states while taking into account the phase synchronization between resting-state networks, it can achieve a comprehensive analysis of brain dynamics.

[0070] Optionally, generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks includes:

[0071] determining an instantaneous phase difference between the resting-state networks based on the instantaneous phase of each of the resting-state networks;

[0072] According to the instantaneous phase difference between the resting-state networks, an instantaneous phase synchronization value between the resting-state networks is determined to generate the phase synchronization matrix.

[0073] Specifically, assuming that the resting-state network The instantaneous phase of , resting-state network The instantaneous phase of , then the instantaneous phase difference between the two resting-state networks is Satisfies the following formula:

[0074]

[0075] In some embodiments, since the cosine function takes a maximum value of 1 when the instantaneous phase difference is 0 (i.e., completely in phase), and takes a minimum value of -1 when the instantaneous phase difference is π (i.e., completely out of phase), it can well reflect the instantaneous phase synchronization degree, and the cosine value of the instantaneous phase difference is used as the instantaneous phase synchronization value. For example, for a resting state network and resting-state networks The instantaneous phase synchronization value of It can be expressed as:

[0076]

[0077] In other embodiments, since the phase locking value is an indicator of the degree of phase synchronization between neural signals, the phase locking value is used as the instantaneous phase synchronization value. and resting-state networks The instantaneous phase synchronization value of It can be expressed as:

[0078]

[0079] Create a matrix with the same number of resting-state networks as the number of networks. The rows and columns of the matrix represent different resting-state networks. Fill the instantaneous phase synchronization value of each pair of resting-state networks into the corresponding position in the matrix. For example, the resting-state network and resting-state networks The instantaneous phase synchronization value of Fill in the matrix Rank Column and Rank The positions of the columns are adjusted to ensure the symmetry of the matrix. In this way, all instantaneous phase synchronization matrices are obtained, and then a three-dimensional phase synchronization matrix is ​​obtained.

[0080] In this optional embodiment, by calculating the instantaneous phase difference between resting-state networks based on the instantaneous phase of each resting-state network, the temporal phase relationship between the resting-state networks can be quantified, providing basic data for subsequent analysis of the degree of instantaneous phase synchronization between the resting-state networks. Subsequently, by calculating the instantaneous phase synchronization value between the resting-state networks based on the instantaneous phase difference between the resting-state networks, the degree of instantaneous phase synchronization between the resting-state networks can be quantified, thereby generating a phase synchronization matrix to facilitate the subsequent acquisition of the dominant eigenvector.

[0081] Optionally, fitting the binarized dominant eigenvector to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape includes:

[0082] Binarizing the dominant eigenvector to determine the empirical frequency of each of the brain activity patterns, wherein a preset threshold for the binarization is zero;

[0083] determining a first average activity state of each of the resting-state networks and a first average interaction between the resting-state networks based on the empirical frequency of each of the brain activity patterns;

[0084] adjusting the first parameter and the second parameter of the maximum entropy model with the optimization objectives that the second average activity state of each of the resting-state networks in the maximum entropy model is equal to the first average activity state of each of the resting-state networks, and the second average interaction between the resting-state networks is equal to the first average interaction between the resting-state networks;

[0085] Based on the first parameter and the second parameter, the energy value of each brain activity pattern is determined to construct the energy landscape, wherein the first parameter is the basic activity of all the resting-state networks and the second parameter is the interaction between all the resting-state networks.

[0086] Specifically, a preset threshold (i.e., zero) is set, and the elements in the dominant eigenvector are classified according to their numerical value. When the element's value is greater than or equal to the preset threshold, the element is assigned a value of 1, indicating that the corresponding resting-state network is active; when the element's value is less than the preset threshold, the element is assigned a value of -1, indicating that the corresponding resting-state network is inactive. This results in a binary dominant eigenvector, i.e., the brain activity pattern. The frequency of each binary dominant eigenvector observed at all time points is calculated, i.e., the empirical frequency of each brain activity pattern. Based on the empirical frequency of each brain activity pattern, the average activity state of each resting-state network observed at all time points, i.e., the first average activity state of each resting-state network, is determined. The average interaction value between resting-state networks observed at all time points, i.e., the first average interaction value between resting-state networks, is also determined. Activity states include active and inactive states.

[0087] The first average activity state is determined by the following formula:

[0088]

[0089] The first average interaction is determined by the following formula:

[0090]

[0091] in, is the observed resting-state network The first average activity state, is the observed resting-state network and resting-state networks The first average interaction between is the number of resting-state networks, It is a resting state network Activity status, It is a resting state network Activity status, It is a resting state network The experienced frequency of the active state.

[0092] The maximum entropy model satisfies the following formula:

[0093]

[0094] in,

[0095]

[0096] in, It is a resting state network Activity status, It is a resting state network The basic activity, It is a resting state network and resting-state networks The interaction between It is a resting state network Activity status, is the number of resting-state networks, It's the pattern of brain activity. is the basic activity of all resting-state networks, is the interaction between all resting-state networks, is the energy value of the brain activity pattern, is the energy value of all possible brain activity patterns, is the model probability of the brain activity pattern.

[0097] By adjusting the first parameter of the maximum entropy model and the second parameter , so that the second average activity state in the maximum entropy model is close to the observed first average activity state, and the second average interaction in the maximum entropy model is close to the observed first average interaction. When the second average activity state in the maximum entropy model is approximately equal to the observed first average activity state, and the second average interaction in the maximum entropy model is approximately equal to the observed first average interaction, according to the first parameter of the maximum entropy model and the second parameter , calculate the energy value of each brain activity pattern and complete the construction of the energy landscape.

[0098] The second average activity state is determined by the following formula:

[0099]

[0100] The second average interaction is determined by the following formula:

[0101]

[0102] in, is the resting state network in the maximum entropy model The second average activity state, is the resting state network in the maximum entropy model and resting-state networks The second average interaction between is the number of resting-state networks, It is a resting state network Activity status, It is a resting state network Activity status, is the resting state network in the maximum entropy model The model probability.

[0103] In this optional embodiment, the dominant eigenvectors are binarized, simplifying the primary phase synchronization pattern at each time point into multiple brain activity patterns. The empirical frequency of each brain activity pattern observed at all time points is then calculated. Using the brain activity patterns and their empirical frequencies, a first average activity state and a first average interaction can be calculated as constraints for the maximum entropy model, facilitating the fitting of the maximum entropy model. Specifically, the first and second parameters of the maximum entropy model are adjusted so that the second average activity state in the maximum entropy model approaches the observed first average activity state, and the second average interaction in the maximum entropy model approaches the observed first average interaction. When the second average activity state in the maximum entropy model is approximately equal to the observed first average activity state, and the second average interaction in the maximum entropy model is approximately equal to the observed first average interaction, the first and second parameters of the maximum entropy model are no longer adjusted. Using the first and second parameters of the maximum entropy model, the energy value of each brain activity pattern is calculated to construct an energy landscape. The brain states in this energy landscape are more biologically interpretable and can comprehensively and accurately reflect the true state of brain dynamics, facilitating subsequent brain research and neurological disease diagnosis.

[0104] Optionally, adjusting the first parameter and the second parameter of the maximum entropy model with the second average activity state of each of the resting-state networks in the maximum entropy model being equal to the first average activity state of each of the resting-state networks, and the second average interaction between the resting-state networks being equal to the first average interaction between the resting-state networks as the optimization goal, includes:

[0105] Initializing the first parameter and the second parameter;

[0106] determining the second average activity state of each of the resting-state networks and the second average interaction between the resting-state networks based on the first parameter and the second parameter;

[0107] The first parameter and the second parameter are updated according to the first difference between the second average activity state of each resting-state network and the first average activity state of each resting-state network, and the second difference between the second average interaction between the resting-state networks and the first average interaction between the resting-state networks, until the first difference and the second difference both tend to zero.

[0108] Specifically, the initial values ​​of the first parameter and the second parameter are randomly set. For example, the initial values ​​of the first parameter and the second parameter are randomly drawn from a normal distribution with a mean of 0 and a standard deviation of 0.1. Alternatively, the initial values ​​of the first parameter and the second parameter are set based on prior knowledge. Alternatively, the initial values ​​of the first parameter and the second parameter are set based on the first average activity state and the first average interaction. Based on the initial values ​​of the first parameter and the second parameter, the model probability of the corresponding brain activity pattern is determined using the maximum entropy model formula. Based on the model probability of the corresponding brain activity pattern, the corresponding second average activity state and second average interaction formula are used to determine the corresponding second average activity state and second average interaction. A first difference between the second average activity state corresponding to the initial value of the first parameter and the observed first average activity state is calculated, as is a second difference between the second average interaction corresponding to the initial value of the second parameter and the observed first average interaction. Based on the first and second differences, the first and second parameters are updated using an optimization algorithm (such as gradient descent).

[0109] Based on the updated first and second parameters, the maximum entropy model formula is used to determine the model probability of the corresponding brain activity pattern. Based on the model probability of the corresponding brain activity pattern, the second average activity state formula and the second average interaction formula are used to determine the corresponding second average activity state and second average interaction. A first difference between the second average activity state corresponding to the updated first parameter and the observed first average activity state is calculated, as is a second difference between the second average interaction corresponding to the updated second parameter and the observed first average interaction. Based on the first and second differences, the first and second parameters are updated using an optimization algorithm (e.g., gradient descent). These steps are repeated until both the first and second differences approach zero, thus achieving the optimization goal.

[0110] In this optional embodiment, the first and second parameters are initialized to provide a starting point for subsequent parameter adjustments. The steps of determining a second average activity state and a second average interaction based on the first and second parameters, and updating the first and second parameters based on a first difference between the second average activity state and the first average activity state, and a second difference between the second average interaction and the first average interaction, are repeated to ensure that the second average activity state is as consistent as possible with the first average activity state, and the second average interaction is as consistent as possible with the first average interaction. This allows for the construction of a probability distribution that matches the observed data while maintaining maximum uncertainty, thereby providing more reliable energy values ​​for brain activity patterns to construct an energy landscape.

[0111] Optionally, fitting the binarized dominant eigenvector to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape includes:

[0112] determining a local minimum value among all the brain activity patterns according to the energy value of each of the brain activity patterns;

[0113] constructing a disconnection map based on the local minima in all of the brain activity patterns;

[0114] Based on the disconnection graph, the energy landscape is constructed.

[0115] Specifically, if two brain activity patterns are adjacent, then they differ in the activity state of only one resting-state network. The number of adjacent brain activity patterns for each brain activity pattern is equal to the number of resting-state networks. Traversing all brain activity patterns, when the energy value of a brain activity pattern is less than the energy values ​​of all its adjacent brain activity patterns, the brain activity pattern is a local minimum. A hypercube graph is constructed using the brain activity patterns as nodes. The maximum energy value among all nodes is used as the energy threshold, and nodes with energy values ​​greater than or equal to the energy threshold are deleted. A connectivity check is performed to ensure that each local minimum is connected by reducing the path within the network.

[0116] The maximum energy value among all remaining nodes is used as the energy threshold, and nodes with energy values ​​greater than or equal to the energy threshold are deleted. A connectivity check is performed to ensure that each local minimum is connected by a path within the reduced network. The above steps are repeated until each local minimum is isolated in the reduced network. Based on the above results, a disconnected graph is constructed, such as Figure 2 As shown in Figure 2, the leaf nodes of the disconnection graph represent local minima, while the internal nodes of the disconnection graph represent branching points between different local minima, reflecting the hierarchical relationship between local minima. The disconnection graph intuitively illustrates the energy barriers required to transition from one local minimum to another. Therefore, an energy landscape can be constructed based on the disconnection graph.

[0117] In this optional embodiment, brain activity patterns with energy values ​​less than the energy values ​​of all adjacent brain activity patterns are identified as local minima and used to construct a disconnection map. The disconnection map intuitively displays all local minima and their energy values, as well as the hierarchical relationships between local minima and energy barriers, enabling direct construction of an energy landscape.

[0118] Optionally, after fitting the binarized dominant eigenvector to a maximum entropy model and determining the energy value of each brain activity pattern to construct an energy landscape, the brain dynamic analysis method further includes:

[0119] Based on the energy landscape, dynamic indicators of brain states are determined, wherein the dynamic indicators include fractional occupancy, dwell time, and transition probability.

[0120] Specifically, attractor basins in the energy landscape correspond to brain states, which include multiple brain activity patterns, each of which typically belongs to an attractor basin. Each local minimum in the energy landscape has an attractor basin. Attractor basins are determined by: (i) For brain activity patterns that are not in a local minimum, transition to the adjacent brain activity pattern with the minimum energy value. (ii) Repeat step (i) until a local minimum is reached. When a brain activity pattern finally reaches a local minimum through the above steps, it is considered to belong to the attractor basin corresponding to that local minimum. Brain dynamics can be studied by calculating three key dynamic metrics of brain states (i.e., attractor basins): fractional occupancy, dwell time, and transition probability. Fractional occupancy refers to the proportion of the total time spent in a brain state (i.e., attractor basin) to the total scan time. Dwell time refers to the average continuous time spent in a brain state (i.e., attractor basin). The transition probability refers to the ratio of the number of transitions from a certain brain state (i.e., a certain attractor basin) to the total number of departures from a certain brain state (i.e., a certain attractor basin).

[0121] In this optional embodiment, the attractor basin in the energy landscape corresponds to the brain state. By analyzing the energy landscape, the three main dynamic indicators of the brain state, namely, fractional occupancy, residence time and transition probability, can be determined to further analyze the actual situation of the brain dynamics.

[0122] Optionally, before acquiring the multiple regions of interest and their resting-state networks of the first image, the brain dynamic analysis method further includes:

[0123] The first image is preprocessed, wherein the preprocessing includes at least one of time correction, motion correction, scrubbing, co-registration, spatial normalization, bandpass filtering, spatial smoothing, and time normalization.

[0124] Specifically, functional magnetic resonance imaging (fMRI) acquires the first image slice by slice. Slight differences in acquisition time between slices (e.g., when the TR is 2 seconds, the last slice is 2 seconds behind the first slice), leading to temporal misalignment. Interpolation algorithms (such as linear or spline interpolation) are used to align the BOLD signals of all slices to the same time point (usually with the middle slice as a reference) to eliminate the effects of acquisition time differences. This is known as temporal correction. Head movement during scanning can cause misalignment of the first image. Motion correction is used to align the first images at all time points to a reference image (e.g., the first image at the first time point or the averaged first image) using rigid body transformations (such as translation and rotation). Head movement or noise can cause abnormal BOLD signals in the first image. Abnormal BOLD signals are removed directly, replaced with normal BOLD signals, or marked (a technique known as scrubbing). The first image reflects functional changes in the brain under different states and does not provide detailed anatomical information. Aligning the first image with the T1-weighted anatomical image is known as co-registration. Given the differences between subjects' brains, the first image is deformed to a standard template (a technique known as spatial normalization). The BOLD signal contains high-frequency noise and low-frequency drift. Bandpass filtering is used to retain the target frequency band (e.g., 0.01-0.1 Hz) and filter out irrelevant frequencies. Differences in brain region position and noise across subjects can lead to local discontinuities in the BOLD signal. A Gaussian kernel (e.g., 6-8 mm FWHM) is used to smooth the image, a process known as spatial smoothing. Time series have different means and variances across subjects and within the same subject, so time series are treated with zero mean (decentering) and unit variance (standardization), a process known as temporal normalization.

[0125] In this optional embodiment, temporal correction eliminates the effects of acquisition time differences, motion correction and scrubbing reduce motion artifacts, co-registration provides detailed anatomical information for subsequent processing, spatial normalization eliminates individual differences, bandpass filtering preserves signals related to brain dynamics, spatial smoothing improves signal-to-noise ratio, and temporal normalization eliminates scale differences in the time series. By performing these preprocessing steps on the first image, the quality and reliability of the first image can be significantly improved, facilitating the subsequent construction of the energy landscape.

[0126] like Figure 3 As shown, an embodiment of the present invention provides a brain dynamic analysis device 300, comprising:

[0127] A first acquisition module 310 is configured to acquire a plurality of regions of interest and their resting-state networks from a first image, wherein the first image comprises an fMRI image;

[0128] an averaging module 320, configured to determine an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network;

[0129] a transformation module 330, configured to perform a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks;

[0130] a second acquisition module 340, configured to generate a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and obtain a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks;

[0131] The construction module 350 is used to fit the binarized dominant eigenvector to a maximum entropy model, determine the energy value of each brain activity pattern, and construct an energy landscape.

[0132] The brain dynamic analysis device 300 of this embodiment is used to implement the above-mentioned brain dynamic analysis method. Its advantages over the existing technology are the same as the advantages of the above-mentioned brain dynamic analysis method over the existing technology, and will not be repeated here.

[0133] like Figure 4 As shown, an electronic device 400 provided by an embodiment of the present invention includes a memory 410 and a processor 420; the memory 410 is used to store computer programs; the processor 420 is used to implement the brain dynamic analysis method as described above when executing the computer program.

[0134] In other words, an electronic device 400 includes a memory 410 and a processor 420 coupled to the memory 410; the memory 410 is configured to store a computer program; and the processor 420 is configured to perform the following operations when executing the computer program:

[0135] Acquire multiple regions of interest and resting-state networks thereof from a first image, wherein the first image comprises an fMRI image;

[0136] determining an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network;

[0137] performing a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks;

[0138] generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and obtaining a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks;

[0139] The binarized dominant eigenvector is fitted to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape.

[0140] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the brain dynamic analysis method described above is implemented.

[0141] In other words, a non-volatile computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the following operations:

[0142] Acquire multiple regions of interest and resting-state networks thereof from a first image, wherein the first image comprises an fMRI image;

[0143] determining an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network;

[0144] performing a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks;

[0145] generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and obtaining a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks;

[0146] The binarized dominant eigenvector is fitted to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape.

[0147] An electronic device 400 that can serve as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device 400 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 400 can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0148] Electronic device 400 includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). The RAM can also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. An input / output (I / O) interface is also connected to the bus.

[0149] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM). In this application, the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network elements. Some or all of these units can be selected based on actual needs to achieve the objectives of the embodiments of the present invention. Furthermore, the functional units in the various embodiments of the present invention can be integrated into a single processing unit, each unit can exist physically separately, or two or more units can be integrated into a single unit. These integrated units can be implemented in either hardware or software functional units.

[0150] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A brain dynamic analysis method, characterized in that: include: Acquire multiple regions of interest and resting-state networks thereof from a first image, wherein the first image comprises an fMRI image; determining an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network; performing a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks; generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and obtaining a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks; The binarized dominant eigenvector is fitted to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape, including: binarizing the dominant eigenvector to determine the empirical frequency of each brain activity pattern, wherein the preset threshold for binarization is zero; determining the first average activity state of each resting-state network and the first average interaction between the resting-state networks based on the empirical frequency of each brain activity pattern; adjusting the first parameter and second parameter of the maximum entropy model with the optimization goal that the second average activity state of each resting-state network in the maximum entropy model is equal to the first average activity state of each resting-state network, and the second average interaction between the resting-state networks is equal to the first average interaction between the resting-state networks; determining the energy value of each brain activity pattern based on the first parameter and the second parameter to construct the energy landscape, wherein the first parameter is the basal activity of all the resting-state networks, and the second parameter is the interaction between all the resting-state networks.

2. The brain dynamic analysis method according to claim 1, characterized in that: The generating a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks comprises: determining an instantaneous phase difference between the resting-state networks based on the instantaneous phase of each of the resting-state networks; According to the instantaneous phase difference between the resting-state networks, an instantaneous phase synchronization value between the resting-state networks is determined to generate the phase synchronization matrix.

3. The brain dynamic analysis method according to claim 1, characterized in that: The first parameter and the second parameter of the maximum entropy model are adjusted with the second average activity state of each of the resting-state networks in the maximum entropy model being equal to the first average activity state of each of the resting-state networks, and the second average interaction between the resting-state networks being equal to the first average interaction between the resting-state networks as the optimization goal, including: Initializing the first parameter and the second parameter; determining the second average activity state of each of the resting-state networks and the second average interaction between the resting-state networks based on the first parameter and the second parameter; The first parameter and the second parameter are updated according to the first difference between the second average activity state of each resting-state network and the first average activity state of each resting-state network, and the second difference between the second average interaction between the resting-state networks and the first average interaction between the resting-state networks, until the first difference and the second difference both tend to zero.

4. The brain dynamic analysis method according to claim 1, characterized in that: The binarized dominant eigenvector is fitted to a maximum entropy model to determine the energy value of each brain activity pattern to construct an energy landscape, including: determining a local minimum value among all the brain activity patterns according to the energy value of each of the brain activity patterns; constructing a disconnection map based on the local minima in all of the brain activity patterns; Based on the disconnection graph, the energy landscape is constructed.

5. The brain dynamic analysis method according to claim 1, characterized in that: After fitting the binarized dominant eigenvector to a maximum entropy model and determining the energy value of each brain activity pattern to construct an energy landscape, the brain dynamic analysis method further includes: Based on the energy landscape, dynamic indicators of brain states are determined, wherein the dynamic indicators include fractional occupancy, dwell time, and transition probability.

6. The brain dynamic analysis method according to claim 1, characterized in that: Before acquiring the multiple regions of interest and their resting-state networks of the first image, the brain dynamic analysis method further includes: The first image is preprocessed, wherein the preprocessing includes at least one of time correction, motion correction, scrubbing, co-registration, spatial normalization, bandpass filtering, spatial smoothing, and time normalization.

7. A brain dynamic analysis device, characterized in that: include: A first acquisition module is configured to acquire a plurality of regions of interest and resting-state networks thereof from a first image, wherein the first image comprises an fMRI image; an averaging module, configured to determine an average time series of each resting-state network based on the multiple regions of interest and their resting-state networks, wherein the average time series is an average of the time series of all the regions of interest corresponding to the resting-state network; a transformation module, configured to perform a Hilbert transform on the average time series of each of the resting-state networks to obtain an instantaneous phase of each of the resting-state networks; a second acquisition module, configured to generate a phase synchronization matrix based on the instantaneous phase of each of the resting-state networks, and acquire a plurality of dominant eigenvectors from the phase synchronization matrix, wherein the number of the dominant eigenvectors is equal to the number of time points, and the elements of the dominant eigenvectors correspond to the resting-state networks; A construction module is used to fit the binarized dominant eigenvector to the maximum entropy model, determine the energy value of each brain activity pattern, and construct an energy landscape, specifically for binarizing the dominant eigenvector and determining the empirical frequency of each brain activity pattern, wherein the preset threshold for binarization is zero; based on the empirical frequency of each brain activity pattern, determine the first average activity state of each resting-state network and the first average interaction between the resting-state networks; with the second average activity state of each resting-state network in the maximum entropy model being equal to the first average activity state of each resting-state network, and the second average interaction between the resting-state networks being equal to the first average interaction between the resting-state networks as the optimization goal, adjust the first parameter and second parameter of the maximum entropy model; based on the first parameter and the second parameter, determine the energy value of each brain activity pattern to construct the energy landscape, wherein the first parameter is the basal activity of all the resting-state networks, and the second parameter is the interaction between all the resting-state networks.

8. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the brain dynamic analysis method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the brain dynamic analysis method according to any one of claims 1 to 6 is implemented.

Images