An auxiliary analysis method based on dynamic brain network spatial reconstruction model
Through the auxiliary analysis method of dynamic brain network space reconstruction mode, Hilbert transform and GenLouvain community detection algorithm are used to analyze the dynamic functional connection of the Alzheimer's brain, solving the problem of parameter sensitivity and network attribute simplification of traditional methods , provides detailed abnormal pattern analysis to support early diagnosis and early intervention.
Patent Information
- Application Number
- CN202410549023.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-06
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-05-06
AI Technical Summary
The traditional sliding time window method is sensitive to the selection of window length parameters, and the traditional network attributes oversimplify brain activity information, making it difficult to effectively reveal abnormal patterns of Alzheimer's brain spatial dynamics.
Using an auxiliary analysis method based on the dynamic brain network space reconstruction mode, the instantaneous phase information of the BOLD signal is extracted through Hilbert transform, and a dynamic phase locked network is constructed. The multi-layer brain network community is identified using the GenLouvain community detection algorithm to calculate the brain area. Flexibility and module loyalty, analyzing network role conversion entropy, and combining one-way analysis of variance to perform significant difference tests.
It improves the ability to mine dynamic functional connections in the Alzheimer's brain, provides more detailed analysis of abnormal patterns, supports early diagnosis and early intervention, and overcomes the limitations of parameter selection sensitivity and network attribute simplification of traditional methods.
Smart Images

Figure CN118230945B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of image processing, and in particular relates to an auxiliary analysis method based on a dynamic brain network spatial reconstruction model. Background Art
[0002] Alzheimer's disease is a multifactorial neurodegenerative disease, whose clinical features are progressive cognitive and behavioral dysfunctions. The pathogenesis of Alzheimer's disease has not been fully elucidated, and late intervention is the key reason for treatment failure. Early diagnosis and early intervention are effective measures to delay the progression of the disease. Currently, magnetic resonance imaging is used for the diagnosis and evaluation of Alzheimer's disease with its advantages of being non-invasive and non-invasive. Its high spatial resolution helps to study the spatial characteristics of the brain. Therefore, it is urgent to explore sensitive imaging indicators to assist in the diagnosis of Alzheimer's disease.
[0003] The brain is a complex dynamic system that changes over time. The interactions between different brain regions are characterized by dynamic self-organization characteristics at different spatiotemporal scales. Based on functional magnetic resonance imaging data, a dynamic functional connectivity method was proposed to estimate time-varying functional connectivity on a time scale of seconds to minutes, depicting the working mechanism of the brain from the perspective of dynamic connectivity, and providing a new research method for explaining the pathogenesis of Alzheimer's disease.
[0004] Sliding time window technology is the most commonly used method for constructing dynamic functional connectivity, but it has certain methodological limitations. The choice of window length parameters affects the sensitivity and robustness of the sliding time window method. If the window length is too short, the correlation cannot be reliably calculated, and if the window is too long, meaningful dynamic change information will be lost. In order to make up for the shortcomings of traditional sliding time window technology, frame-by-frame nonlinear phase analysis has been applied to research in this field. Different from the sliding time window technology, frame-by-frame nonlinear phase analysis can capture the coherent relationship between neural oscillations between different brain regions from the perspective of a single time frame. It is not limited by window length and has high temporal resolution and sensitivity. It can be used to quantify the spatial dynamic functional connectivity of the brain in patients with Alzheimer's disease.
[0005] In the past, MRI-assisted diagnosis of Alzheimer's disease was mostly based on graph theory to analyze the topological properties of the network. However, the community structure in the brain will adaptively reconfigure over time. Traditional methods have the problem of over-simplifying brain activity and still have certain limitations in revealing abnormal patterns of spatial dynamics in the Alzheimer's brain. Summary of the invention
[0006] In order to mine dynamic functional connectivity imaging indicators to assist in the diagnosis of Alzheimer's disease and to solve the problems of the sliding time window method being sensitive to parameter selection and the traditional network attributes over-simplifying brain activity information, the present invention provides an auxiliary analysis method based on the dynamic brain network spatial reconstruction model.
[0007] The technical solution adopted by the present invention is an auxiliary analysis method based on the dynamic brain network spatial reconstruction model, comprising the following steps:
[0008] Step S1: Preprocess the resting-state functional magnetic resonance imaging data, extract the BOLD time series, i.e., BOLD signal, of each brain region based on the brain map template, evaluate the overlapping area of the cortical surface, and divide the empirical resting subnetwork;
[0009] Step S2: After the BOLD signal is bandpass filtered, the instantaneous phase information between different brain regions is extracted frame by frame using Hilbert transform, and a dynamic phase-locked network is constructed through the phase alignment information between each pair of brain regions; the connection between the same brain regions between adjacent time slices represents the temporal dependency between layers, and a dynamic multi-layer brain network is constructed using a super-adjacency matrix;
[0010] Step S3: using the GenLouvain multi-layer community detection algorithm to identify brain communities in dynamic multi-layer brain networks;
[0011] Step S4: Calculate the dynamic properties of the dynamic multi-layer brain network, including indicators that quantify the stability of cross-layer brain communities: flexibility and temporal core-periphery structure; calculate the module loyalty between each pair of brain regions and construct a module loyalty matrix; use the module loyalty matrix to calculate the recruitment coefficient and integration coefficient of each brain region and analyze its dispersion;
[0012] Step S5: construct a random null model, identify brain regions with lower, higher or indistinguishable dynamic multi-layer brain network integration and recruitment, divide brain region roles, and calculate the brain region state indicator network role transition entropy;
[0013] Step S6: One-way ANOVA was used to test for significant differences.
[0014] Furthermore, in step S1, the preprocessing steps include generating high-precision anatomical masks in the functional original space, skull removal, removal of the first 10 time points, time layer correction, head motion correction, spatial smoothing, noise reduction, filtering, and covariate regression.
[0015] Furthermore, in step S1, the empirical resting subnetwork is divided, specifically including:
[0016] Generate brain tissue boundary template and brain region segmentation template based on T1w-MRI data;
[0017] After data preprocessing, the Desikan-Killiany atlas was used to extract the BOLD time series of each brain region;
[0018] By assessing the overlapping areas of the cortical surface, 68 cortical areas were assigned to seven resting-state networks: the visual network, the sensorimotor network, the frontoparietal network, the dorsal attention network, the limbic subnetwork, the ventral attention network, and the default mode network.
[0019] Furthermore, in step S2, a dynamic multi-layer brain network is constructed, which specifically includes:
[0020] First, the BOLD signal is bandpass filtered at the frequency of interest, and then the Hilbert transform is applied to obtain the instantaneous phase. The Hilbert transform formula is as follows:
[0021]
[0022] Where H[B(t)] is the value of the real signal after Hilbert transform, B(t) is the BOLD signal after bandpass filtering, θ(t) and A(t) represent the instantaneous phase and instantaneous amplitude of the BOLD analytical signal, respectively. The instantaneous phase value of the signal is obtained by calculating the arc tangent of the ratio of the imaginary time series and the real time series. Its value range is [-π, π], and the formula is as follows:
[0023]
[0024] The instantaneous phase information of the BOLD signal was extracted through Hilbert transform, generating the phase evolution of the BOLD time course in each brain region;
[0025] To obtain the whole-brain pattern of BOLD phase coherence at each single time point t, the dynamic phase-locking matrix dPC(i,j,t) is calculated, which estimates the phase alignment between each pair of brain regions i and j at each time t, as shown in the following formula:
[0026] dPC(i,j,t)=cos(θ(i,t)-θ(j,t))
[0027] Among them, θ(i,t) is the instantaneous phase of the BOLD signal of region i at time t. If the two brain regions are temporarily aligned, dPC will be close to 1; if the two brain regions are orthogonal to each other, dPC will be close to 0;
[0028] A dynamic multi-layer brain network based on time reconstruction. The order of layers represents the flow of time. The data structure of the dynamic multi-layer brain network is represented by a hyper-adjacency matrix. The layers of the network and the connection methods between layers appear in the form of blocks in this structure.
[0029] Interlayer coupling only considers the coupling of the same brain regions in adjacent time slices, and the connection weight is set according to the common value of 1, thus forming a dynamic multi-layer brain network. The complete hyper-adjacency matrix of a dynamic multi-layer brain network is expressed as:
[0030]
[0031] This is a [N×T]×[N×T] matrix, where N is the number of brain regions in each layer of the network. In this matrix, the matrix dPC on the diagonal t Represents the intra-layer phase-locking matrix at time t, 1≤t≤T, and the matrix on the off-diagonal line represents the connection matrix between layers.
[0032] Furthermore, in step S3, the brain communities in the dynamic multi-layer brain network are identified, specifically including:
[0033] Consider each brain region in the network as an independent community;
[0034] For each brain region, try to join the community where its neighbors are located, compare the Q values of the modularity quality function before and after joining, and randomly move the brain region under consideration to a community that increases the modularity quality function. The probability of community selection is proportional to the increase in the modularity quality function. Otherwise, the community division will not be changed until the communities to which all brain regions belong no longer change.
[0035] Simplify the brain regions of the obtained communities and repeat the operation until the Q value no longer increases;
[0036] The modularity quality function Q is used to evaluate the quality of the community in a multi-layer network, and its formula is as follows:
[0037]
[0038] Among them, A ijl is the correlation matrix between brain regions i and j in layer l, μ represents the total weight of the edges in the network, and P ijl is the correlation matrix in the Newman-Girvan null model, δ(g il ,g jr ) is equal to 1 when brain regions i and j belong to the same module, otherwise it is equal to 0; γ l is the topological resolution parameter of layer l, which affects the connection strength of brain regions within the layer; ω jlr is the interlayer coupling strength between the l layer and the r layer.
[0039] Furthermore, in step S4, it specifically includes:
[0040] Flexibility indicates the frequency of switching communities in a brain region over a period of time. It is an indicator for quantifying cross-layer community stability. It is defined as the ratio of the number of times a brain region changes its community assignment across a time window to the total number of all possible changes. It quantifies the complexity of spatial reconstruction of the brain region. The calculation formula for brain region flexibility is as follows:
[0041]
[0042] Among them, f i It is a value between 0 and 1. The closer the value is to 0, the more stable the brain area is; the closer the value is to 1, the more flexible the brain area is.
[0043] The temporal core-periphery structure is a mesoscale structure that exists simultaneously with the community. The brain regions with lower flexibility are responsible for performing relatively stable functions and belong to the temporal core regions; the regions with higher flexibility are believed to make adaptive adjustments and belong to the temporal periphery regions.
[0044] The module loyalty matrix reflects the consistency of community allocation between two brain regions during the entire scanning process. It is an N×N square matrix. Each element in the matrix represents the proportion of the number of times one brain region shares the same community with another brain region during the scanning process. The module loyalty matrix is used to calculate the recruitment coefficient and integration coefficient of each brain region. By comparing the relationship between the empirical functional network and the actual divided community, the alliance preference of different brain regions during dynamic community reconstruction is quantified.
[0045] The recruitment coefficient of a brain region refers to the probability that the brain region shares the same community with other brain regions in its own experience functional network during community reconstruction. Specifically, the recruitment coefficient of region i in the functional network S is defined as follows:
[0046]
[0047] The integration coefficient of a brain region refers to the probability that the brain region shares the same community with other brain regions in the empirical functional network during the community reconstruction process; for region i in the functional network S, its integration coefficient can be defined as:
[0048]
[0049] Among them, n S is the number of brain regions in the network S, O is the number of repetitions of the experiment, and T is the number of layers; if brain regions i and j are in the same community, then the element If brain regions i and j are not in the same community,
[0050] The dispersion distributions of integration coefficients and recruitment coefficients were analyzed to assess the unique contribution of functional network dispersion in the integration and segregation of brain functions. The intra-network dispersion was quantified as the sum of the squares of the Euclidean distances from individual-level network nodes to the network centroid, and the inter-network dispersion was calculated as the Euclidean distances between network centroids.
[0051] Furthermore, in step S5, it specifically includes:
[0052] The random null model is a random permutation of the assignment of brain regions to subnetworks. Each value observed in the actual data was compared with the 95% confidence interval of the random null model to identify brain regions with lower, higher, or indistinguishable dynamic network integration and recruitment, thereby dividing into 9 different role states: 01, -11, 11, 00, -10, 10, 0-1, -1-1, and 1-1;
[0053] A two-dimensional state sequence describing the role transition is obtained, and the occurrence probability and duration of each state can be calculated;
[0054] The probability of occurrence is the proportion of time in a specific state to all time, and the duration is the continuous length of time between the current state and the next state.
[0055] The brain region state indicator network role transition entropy measures the complexity of the brain region cross-frame role state transition and is defined as follows:
[0056]
[0057] Among them, M = 9 is the number of states, p i is the probability of occurrence of state i, that is, the ratio of the time slice occupied by state i to the total time slice.
[0058] Furthermore, in step S6, it specifically includes:
[0059] The Spearman correlation method was used to evaluate the correlation between the indicators and cognitive performance;
[0060] All results were corrected for multiple comparisons using the Bonferroni correction method;
[0061] The threshold for significant difference was P<0.05.
[0062] The present invention discloses an auxiliary analysis method based on a dynamic brain network spatial reconstruction pattern, which solves the problems of sensitive parameter selection of traditional dynamic brain network construction methods and over-simplification of brain activity information by traditional network attributes, and improves the analysis framework of brain spatial reorganization patterns from the perspective of dynamic analysis; the present invention first uses Hilbert transform to extract the BOLD signal phase frame by frame and constructs a dynamic phase-locked brain network, then uses the GenLouvain community detection algorithm to complete multi-layer community detection of the dynamic brain network, and then calculates frequency and probability quantitative indicators to analyze the dynamic community reconstruction pattern of the brain region, uses discreteness to evaluate the unique contribution of functional subnetworks in the dynamic integration and separation of functions, and proposes network role transition entropy to quantitatively analyze the temporal changes of brain states; the present invention applies the proposed method to Alzheimer's disease, which can mine the abnormal functional spatial dynamics caused by the disease, discovers a large number of abnormal features related to cognitive scales through statistical analysis, and provides a disease auxiliary diagnosis method, which has important clinical significance for the early diagnosis of AD. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 It is a flow chart of a method for analyzing a dynamic brain network spatial reconstruction pattern based on frame-by-frame nonlinear phase analysis according to an embodiment of the present invention.
[0064] Figure 2 It is a schematic diagram comparing the results of the present invention and the traditional dynamic network construction method.
[0065] Figure 3 The discreteness and network role conversion entropy characteristics proposed by the present invention are related to the cognitive scale. DETAILED DESCRIPTION
[0066] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0067] An auxiliary analysis method based on the dynamic brain network spatial reconstruction model is specifically carried out in the following steps:
[0068] Step S1: preprocess the resting-state functional magnetic resonance imaging data, extract the time series according to the brain map template, and evaluate the overlapping area of the cortical surface to divide the empirical resting subnetwork;
[0069] Step S2: After bandpass filtering the BOLD signal, the Hilbert transform is used to extract the instantaneous phase information between different brain regions frame by frame, and a dynamic phase-locked network is constructed through the phase alignment information between each pair of brain regions. For inter-layer connections, only the connections between the same brain regions between adjacent time slices are considered to represent the temporal dependence between layers, and a dynamic time-varying multi-layer functional brain network is constructed using a super-adjacency matrix;
[0070] Step S3: using the GenLouvain multi-layer community detection algorithm to identify brain communities in the multi-layer network;
[0071] Step S4: Calculate four dynamic properties of the dynamic network: including two frequency-based brain region dynamic indicators flexibility and temporal core-periphery structure, construct a module loyalty matrix, and calculate two probability-based brain region dynamic indicators integration coefficient and recruitment coefficient;
[0072] Step S5: construct a random null model and divide the brain region roles according to the null model, and calculate the brain region state index network role conversion entropy;
[0073] Step S6: One-way ANOVA was used to test for significant differences, Bonferroni correction was used (the significance level was set to 0.05) to evaluate the feature validity, and Spearman correlation test (the significance level was set to 0.05) was used to analyze the relationship between the indicators and cognitive performance.
[0074] Furthermore, in step S1, the resting-state functional magnetic resonance imaging data is preprocessed, and the preprocessing step at least includes generating a high-precision anatomical mask of the functional original space, removing the skull, removing the first 10 time points, time layer correction, head motion correction, spatial smoothing, noise reduction, filtering, and covariate regression.
[0075] The time series extraction and resting subnetwork division specifically include the following steps: generating a brain tissue boundary template and a brain region segmentation template based on a high-resolution mask obtained from T1w-MRI data to ensure extraction of neural signals with a high signal-to-noise ratio; extracting the BOLD time series of each brain region using the Desikan–Killiany atlas after data preprocessing; and assigning the 68 cortical regions of Desikan-Killiany to seven resting-state networks by evaluating the overlapping area of the cortical surface: a visual network, a sensorimotor network, a frontoparietal network, a dorsal attention network, a limbic subnetwork, a ventral attention network, and a default mode network.
[0076] Furthermore, in step S2, constructing a frame-by-frame nonlinear phase-coherent dynamic brain network specifically includes:
[0077]
[0078] Where H[B(t)] is the value of the real signal after Hilbert transform, B(t) is the BOLD signal after bandpass filtering, θ(t) and A(t) represent the instantaneous phase and instantaneous amplitude of the BOLD analytical signal, respectively. The instantaneous phase value of the signal is obtained by calculating the arc tangent of the ratio of the imaginary time series and the real time series. Its value range is [-π, π], and the formula is as follows:
[0079]
[0080] The Hilbert transform generates the time evolution of the BOLD instantaneous phase information of each brain region. In order to obtain the whole-brain pattern of BOLD phase coherence at each single time point t, characterizing the phase alignment between each pair of brain regions i and j at each time t, a dynamic phase-locking matrix dPC(i,j,t) is calculated, as shown in the following formula:
[0081] dPC(i,j,t)=cos(θ(i,t)-θ(j,t))
[0082] Wherein, θ(i,t) is the instantaneous phase of the BOLD signal of region i at time t, and dPC. When dPC is close to 0, it means that the BOLD signals of the two brain regions are orthogonal to each other and have zero phase coherence (cos(90°)=0); when dPC is close to 1, it means that the BOLD signals of the two brain regions are temporarily aligned, that is, they have similar phase angles (cos(0°)=1).
[0083] The order of layers in the present invention represents the flow of time. A dynamic multi-layer brain network is constructed based on time reconstruction. For inter-layer coupling, only the coupling of the same brain regions in adjacent time slices is considered, and the connection weight is set according to the common value of 1.
[0084] Using the hyper-adjacency matrix to represent the data structure of a multi-layer network can retain all the information of the network. The layers of the network and the connection methods between layers appear in this structure in the form of blocks. A complete hyper-adjacency matrix of a multi-layer network can be expressed as:
[0085]
[0086] This is a [N×T]×[N×T] matrix, where N is the number of brain regions in each layer of the network. In this matrix, the matrix dPC on the diagonal t Represents the intra-layer network adjacency matrix of the Sth layer, 1≤t≤T. The matrix I on the off-diagonal line kl represents the connection matrix between the kth layer and the lth layer, 1≤k≤T, 1≤l≤T.
[0087] Furthermore, in step S3, the specific steps of detecting brain communities in the multilayer network include: initially treating each brain region in the network as an independent community; for each brain region, trying to join the community where the neighbor is located, and comparing the gain of the modularity Q value before and after joining. Randomly move the brain region under consideration to a community that makes the quality function, and the probability of randomly selecting to move to a specific community is proportional to its increase in the quality function; if the gain is less than or equal to 0, the community division is not changed until the communities to which all brain regions belong no longer change; simplify each obtained community into a brain region, the weight of the new brain region ring is the weight between brain regions in the community, and the edge weight between new brain regions is the weight of the edge between communities; repeat the operation until the modularity Q value no longer increases.
[0088] The modularity quality function Q is used to evaluate the quality of the community in a multi-layer network, and its formula is as follows:
[0089]
[0090] Among them, A ijl is the correlation matrix between brain regions i and j in layer l, μ represents the total weight of the edges in the network, and P ijl is the correlation matrix in the Newman-Girvan null model. When brain regions i and j belong to the same module, δ(g il ,g jr )=1, otherwise it is equal to 0. jlr is the interlayer coupling parameter between layer l and layer r, which determines the coupling strength between the interlayer brain areas; l is the topological resolution parameter of layer l, which affects the connection strength of brain regions within the layer. Q ranges from 0 to 1. The closer the Q value is to 0, the lower the separation of the network and the worse the modularity; conversely, the closer the Q value is to 1, the higher the separation of the network and the better the modularity.
[0091] Further, the step S4 is performed according to the following steps:
[0092] First, we use a combination of flexibility and temporal core-periphery structure to mine the dynamic patterns of different brain regions from a frequency perspective. Flexibility indicates the frequency of brain regions switching communities over a period of time and is an indicator for quantifying cross-layer community stability. It is defined as the ratio of the number of times a brain region changes its community assignment across a time window to the total number of all possible changes, quantifying the complexity of spatial reconstruction of the brain region. The formula for calculating the flexibility of a brain region is as follows:
[0093]
[0094] If the brain area belongs to the same community in layer S and layer S+1, then δ(g i,S ,g i,S+1 )=0; otherwise δ(gi,S ,g i,S+1 )=1. i It is a value between 0 and 1. The closer the value is to 0, the more stable the brain area is, and the smaller the change in community distribution in the entire time-varying multi-layer network; the closer the value is to 1, the more flexible the brain area is, and the community it belongs to changes in almost every time slice.
[0095] The temporal core-periphery structure is a mesoscale structure that exists simultaneously with the community. The brain regions with lower flexibility are responsible for performing relatively stable functions and belong to the temporal core region, while the regions with higher flexibility are believed to make adaptive adjustments and belong to the temporal periphery region.
[0096] Then, different brain regions can participate in communities within their own resting-state networks or across resting-state networks, and the tendency of brain region reconstruction is called "community reconstruction preference". The present invention analyzes the community reconstruction pattern of each brain region by using some probabilistic quantitative indicators, specifically from two perspectives: the overlap probability of the dynamic community to which each brain region belongs and the brain regions within the same empirical resting-state network and between different empirical resting-state networks.
[0097] The consistency of community assignment between two brain regions during the entire scanning process is represented by an N×N square matrix (N is the number of brain regions) module loyalty matrix, in which each element represents the proportion of the number of times one brain region shares the same community with another brain region during the scanning process. If the module loyalty of two brain regions is 1, it means that they are always assigned to the same community, and the module loyalty will be 0, which means that they are never in the same community.
[0098] Next, the recruitment coefficient and integration coefficient of each brain region are calculated using the module loyalty matrix, and the alliance preference of different brain regions during dynamic community reconstruction is quantified by comparing the relationship between the empirical functional network and the actual divided communities. The recruitment coefficient of a brain region refers to the probability that this brain region shares the same community with other brain regions in its own empirical functional network during community reconstruction. Specifically, the recruitment coefficient of region i in the functional network S is defined as follows:
[0099]
[0100] Among them, n S is the number of brain regions in the network S, O is the number of repetitions of the experiment, and T is the number of layers; if brain regions i and j are in the same community, then the element If brain regions i and j are not in the same community, Regions with higher recruitment coefficients consistently communicated with other regions in the same functional network during the scan.
[0101] The integration coefficient of a brain region refers to the probability that the brain region shares the same community with other brain regions in the empirical functional network during the community reconstruction process. For region i in the functional network S, its integration coefficient can be defined as:
[0102]
[0103] Regions with higher integration coefficients are more likely to be found in communities dominated by brain regions in other functional networks during scanning. The integration and recruitment coefficients capture the spatial reorganization patterns during the dynamic changes of brain networks, which can better reflect that the brain is a dynamic and complex nonlinear system compared with the traditional single-layer network properties.
[0104] The dispersion distributions of integration coefficients and recruitment coefficients were analyzed to assess the unique contribution of functional network dispersion in the integration and segregation of brain functions. The intra-network dispersion was quantified as the sum of the squares of the Euclidean distances from individual-level network nodes to the network centroid, and the inter-network dispersion was calculated as the Euclidean distances between network centroids.
[0105] Furthermore, in step S5, the steps of solving the network status change are as follows:
[0106] The null model is a random permutation of the assignment of brain regions to subnetworks. This process ensures that the number and size of subnetworks in the null model are the same as those observed in the real data. Comparing each value observed in the real data with the 95% confidence interval obtained from the null model allows the identification of systems with low (-1), high (1), or indistinguishable (0) dynamic network integration and recruitment. Based on this comparison, nine different role states are identified: 01, -11, 11, 00, -10, 10, 0-1, -1-1, and 1-1. We thus obtain a two-dimensional sequence of states that describes the fluctuations between states, and can calculate the probability and duration of each state. The probability of occurrence is the proportion of time in a particular state out of all time, and the duration is the length of time that the current state remains stable from the next state.
[0107] The network role transition entropy measures the complexity of the role state transitions of brain regions across frames and is defined as follows:
[0108]
[0109] Among them, M = 9 is the number of states, p i is the probability of occurrence of state i. The network role transition entropy measures the complexity of flexible changes in brain states by comparing it with a random system based on the integration and separation of dynamic network functions. A higher value of the network role transition entropy indicates that the state of the brain region changes flexibly during the temporal reconstruction process, while a lower value indicates that the state of the brain region is stable and has low temporal complexity.
[0110] Furthermore, in step S6, one-way ANOVA was used to evaluate the differences between the groups. Spearman correlation method was used to evaluate the correlation between the index and cognitive performance. All results were corrected for multiple comparisons using the Bonferroni correction method, and the threshold for significant differences was P<0.05.
[0111] Beneficial effects of the present invention: First, compared with the traditional dynamic brain network construction method, the present invention uses Hilbert transform to quantify the phase coherence between different brain regions at the same time to establish a dynamic multi-layer brain network, which has high temporal resolution and sensitivity and solves the parameter selection problem. Then, the analysis framework of the phase coherent dynamic brain network spatial reconstruction pattern is improved through frequency and probability quantitative indicators, and the important role played by different functional subnetworks in brain function is evaluated according to the discreteness analysis, and the network role conversion entropy index is proposed to measure the complexity of brain region state changes and community reconstruction from the perspective of time and space respectively. It overcomes the limitation that the analysis of traditional network topology attributes cannot reflect the reconfiguration of community structure in the complex system of human brain network, and can comprehensively analyze the research on brain spatial reorganization pattern. In addition, the method proposed by the present invention is applied to Alzheimer's disease for the first time, which can mine the abnormal functional spatial dynamics caused by the disease and provide a disease auxiliary diagnosis method.
[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.
Claims
1. An auxiliary analysis method based on dynamic brain network spatial reconstruction mode, characterized in that: The following steps are involved: Step S1: Preprocess the resting-state functional magnetic resonance imaging data, extract the BOLD time series, i.e., BOLD signal, of each brain region based on the brain map template, evaluate the overlapping area of the cortical surface, and divide the empirical resting subnetwork; Step S2: After the BOLD signal is bandpass filtered, the instantaneous phase information between different brain regions is extracted frame by frame using Hilbert transform, and a dynamic phase-locked network is constructed through the phase alignment information between each pair of brain regions; the connection between the same brain regions between adjacent time slices represents the temporal dependency between layers, and a dynamic multi-layer brain network is constructed using a super-adjacency matrix; Step S3: using the GenLouvain multi-layer community detection algorithm to identify brain communities in dynamic multi-layer brain networks; Step S4: Calculate dynamic properties of dynamic multilayer brain networks, including metrics that quantify the stability of cross-layer brain communities: flexibility and temporal core-periphery structure; Calculate the module fidelity between each pair of brain regions and construct a module fidelity matrix; The module fidelity matrix was used to calculate the recruitment coefficient and integration coefficient of each brain region and analyze its dispersion; Step S5: construct a random null model, identify brain regions with lower, higher or indistinguishable dynamic multi-layer brain network integration and recruitment, divide brain region roles, and calculate the brain region state indicator network role transition entropy; Step S6: One-way ANOVA was used to test for significant differences; In step S2, a dynamic multi-layer brain network is constructed, which specifically includes: First, the BOLD signal is bandpass filtered at the frequency of interest, and then the Hilbert transform is applied to obtain the instantaneous phase. The Hilbert transform formula is as follows: Where H[B(t)] is the value of the real signal after Hilbert transform, B(t) is the BOLD signal after bandpass filtering, θ(t) and A(t) represent the instantaneous phase and instantaneous amplitude of the BOLD analytical signal, respectively. The instantaneous phase value of the signal is obtained by calculating the arc tangent of the ratio of the imaginary time series and the real time series. Its value range is [-π, π], and the formula is as follows: The instantaneous phase information of the BOLD signal was extracted through Hilbert transform, generating the phase evolution of the BOLD time course in each brain region; To obtain the whole-brain pattern of BOLD phase coherence at each single time point t, the dynamic phase-locking matrix dPC(i,j,t) is calculated, which estimates the phase alignment between each pair of brain regions i and j at each time t, as shown in the following formula: dPC(i,j,t)=cos(θ(i,t)-θ(j,t)) Where θ(i,t) is the instantaneous phase of the BOLD signal of region i at time t. If the two brain regions are temporarily aligned, dPC will be close to 1; if the two brain regions are orthogonal to each other, dPC will be close to 0; A dynamic multi-layer brain network based on time reconstruction. The order of layers represents the flow of time. The data structure of the dynamic multi-layer brain network is represented by a hyper-adjacency matrix. The layers of the network and the connection methods between layers appear in the form of blocks in this structure. Interlayer coupling only considers the coupling of the same brain regions in adjacent time slices, and the connection weight is set according to the common value of 1, thus forming a dynamic multi-layer brain network. The complete hyper-adjacency matrix of a dynamic multi-layer brain network is expressed as: This is a [N×T]×[N×T] matrix, where N is the number of brain regions in each layer of the network. In this matrix, the matrix dPC on the diagonal t represents the phase-locked matrix within a layer at time t, 1≤t≤T, and the matrix on the off-diagonal line represents the connection matrix between layers; In step S4, it specifically includes: Flexibility indicates the frequency of switching communities in a brain region over a period of time. It is an indicator for quantifying cross-layer community stability. It is defined as the ratio of the number of times a brain region changes its community assignment across a time window to the total number of all possible changes. It quantifies the complexity of spatial reconstruction of the brain region. The calculation formula for brain region flexibility is as follows: Among them, f i It is a value between 0 and 1. The closer the value is to 0, the more stable the brain area is; the closer the value is to 1, the more flexible the brain area is. The temporal core-periphery structure is a mesoscale structure that exists simultaneously with the community. The brain regions with lower flexibility are responsible for performing relatively stable functions and belong to the temporal core regions; the regions with higher flexibility are believed to make adaptive adjustments and belong to the temporal periphery regions. The module loyalty matrix reflects the consistency of community allocation between two brain regions during the entire scanning process. It is an N×N square matrix. Each element in the matrix represents the proportion of the number of times one brain region shares the same community with another brain region during the scanning process. The module loyalty matrix is used to calculate the recruitment coefficient and integration coefficient of each brain region. By comparing the relationship between the empirical functional network and the actual divided community, the alliance preference of different brain regions during dynamic community reconstruction is quantified. The recruitment coefficient of a brain region refers to the probability that the brain region shares the same community with other brain regions in its own experience functional network during community reconstruction. Specifically, the recruitment coefficient of region i in the functional network S is defined as follows: The integration coefficient of a brain region refers to the probability that the brain region shares the same community with other brain regions in the empirical functional network during the community reconstruction process; for region i in the functional network S, its integration coefficient can be defined as: Among them, n S is the number of brain regions in the network S, O is the number of repetitions of the experiment, and T is the number of layers; if brain regions i and j are in the same community, then element If brain regions i and j are not in the same community, The dispersion distributions of integration coefficients and recruitment coefficients were analyzed to assess the unique contribution of functional network dispersion in the integration and segregation of brain functions. The intra-network dispersion was quantified as the sum of the squares of the Euclidean distances from individual-level network nodes to the network centroid, and the inter-network dispersion was calculated as the Euclidean distances between network centroids.
2. According to claim 1, an auxiliary analysis method based on dynamic brain network spatial reconstruction mode is characterized in that: In step S1, the preprocessing steps include generating high-precision anatomical masks in the functional raw space, skull removal, removal of the first 10 time points, temporal layer correction, head motion correction, spatial smoothing, noise reduction, filtering, and covariate regression.
3. According to claim 1, the auxiliary analysis method based on the dynamic brain network spatial reconstruction model is characterized in that: In step S1, the empirical resting subnetwork is divided, specifically including: Generate brain tissue boundary template and brain region segmentation template based on T1w-MRI data; After data preprocessing, the Desikan-Killiany atlas was used to extract the BOLD time series of each brain region; By assessing the overlapping areas of the cortical surface, 68 cortical areas were assigned to seven resting-state networks: the visual network, the sensorimotor network, the frontoparietal network, the dorsal attention network, the limbic subnetwork, the ventral attention network, and the default mode network.
4. The auxiliary analysis method based on the dynamic brain network spatial reconstruction model according to claim 1 is characterized in that: In step S3, brain communities in a dynamic multi-layer brain network are identified, specifically including: Consider each brain region in the network as an independent community; For each brain region, try to join the community where its neighbors are located, compare the Q values of the modularity quality function before and after joining, and randomly move the brain region under consideration to a community that increases the modularity quality function. The probability of community selection is proportional to the increase in the modularity quality function. Otherwise, the community division will not be changed until the communities to which all brain regions belong no longer change. Simplify the obtained communities into brain regions and repeat the operation until the Q value no longer increases; The modularity quality function Q is used to evaluate the quality of the community in a multi-layer network, and its formula is as follows: Among them, A ijl is the correlation matrix between brain regions i and j in layer l, μ represents the total weight of the edges in the network, and P ijl is the correlation matrix in the Newman-Girvan null model, δ(g il ,g jr ) is equal to 1 when brain regions i and j belong to the same module, otherwise it is equal to 0; γ l is the topological resolution parameter of layer l, which affects the connection strength of brain regions within the layer; ω jlr is the interlayer coupling strength between the l layer and the r layer.
5. The auxiliary analysis method based on the dynamic brain network spatial reconstruction model according to claim 1 is characterized in that: In step S5, it specifically includes: The random null model is a random permutation of the assignment of brain regions to subnetworks. Each value observed in the actual data was compared with the 95% confidence interval of the random null model to identify brain regions with lower, higher, or indistinguishable dynamic network integration and recruitment, thereby dividing into 9 different role states: 01, -11, 11, 00, -10, 10, 0-1, -1-1, and 1-1; A two-dimensional state sequence describing the role transition is obtained, and the occurrence probability and duration of each state can be calculated; The probability of occurrence is the proportion of time in a specific state to all time, and the duration is the continuous length of time between the current state and the next state. The brain region state indicator network role transition entropy measures the complexity of the brain region cross-frame role state transition and is defined as follows: Among them, M = 9 is the number of states, p i is the probability of occurrence of state i, that is, the ratio of the time slice occupied by state i to the total time slice.
6. The auxiliary analysis method based on the dynamic brain network spatial reconstruction model according to claim 1 is characterized in that: In step S6, it specifically includes: The Spearman correlation method was used to evaluate the correlation between the indicators and cognitive performance; All results were corrected for multiple comparisons using the Bonferroni correction method; The threshold for significant difference was P<0.05.
