Cerebral cortex dynamic function organization quantification method based on edge function network
Through the k-means algorithm based on Euclidean distance, the edge time series is clustered and the edge participation coefficient is calculated, which solves the problem that the existing technology cannot capture the edge characteristics of the brain network, and realizes the quantification of dynamic functional tissues in the cerebral cortex and individual differences identification. , reveals the pattern of brain development and disease mechanisms.
Patent Information
- Application Number
- CN202410819478.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-24
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-06-24
AI Technical Summary
Existing node-centered brain network analysis cannot effectively capture the underlying features and interaction patterns between network edges, and it is difficult to reveal the dynamic functional organization and individual differences in the cerebral cortex.
The k-means algorithm based on Euclidean distance is used to cluster edge time series, divide the modular edge functional network, and calculate the participation coefficient of each edge as a new neuroimaging feature for disease prediction and classification models. Revealing the dynamic development patterns of brain function and their cognitive relationships.
Through an edge-centered perspective, more individual-specific information is captured, new disease biomarkers are identified, brain development and aging patterns are revealed, and brain retest reliability and sensitivity are provided, which is characterized by the brain. Dynamic functional tissue of the cortex.
Smart Images

Figure CN119092092A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of functional networks based on functional magnetic resonance imaging (fMRI) data, and in particular to a method for quantifying dynamic functional organization of cerebral cortex based on edge functional networks. Background Art
[0002] Network science provides a promising framework for representing and modeling the nervous system. From interconnected cells, to clusters of neurons, to large-scale brain regions, network analysis has contributed to the topological principles that govern the organization of the nervous system and shape brain function. At the core of network neuroscience is the simple modeling of the brain as a network, in which neural elements and their interactions are nodes and edges of the network, respectively. This standard model is fundamentally node-centric because it regards neural elements (nodes) as irreducible units of brain structure and function. This node-centric analysis of brain networks only focuses on the topological properties of nodes and cannot capture the potentially meaningful features or interaction patterns between edges. Recent studies have begun to model brain networks from the perspective of interacting edges, revealing the overlapping system-level architecture of the cerebral cortex by constructing edge-centric functional networks.
[0003] Many natural and engineered networks are modular. Highly modular networks can be divided into communities such that the density of connections between nodes within a community is large relative to the density of connections between nodes in different communities. The inter-module connectivity of a node is often quantified using the participation coefficient (PC). The participation coefficient provides insights into how nodes communicate across modules in a wide variety of real networks. Summary of the invention
[0004] The present invention provides a method for quantifying the dynamic functional organization of the cerebral cortex based on an edge functional network. The present invention proposes a method for quantifying the dynamic functional organization of the cerebral cortex based on a modular edge functional network from a new edge-centric perspective to capture the potential meaningful features or interaction patterns between network edges, as described below:
[0005] A method for quantifying dynamic functional organization of cerebral cortex based on edge functional network, the method comprising:
[0006] The edge time series of a group of subjects are merged into a high-dimensional edge time series, and the edge time series is clustered using the k-means algorithm based on Euclidean distance;
[0007] Based on the clustering results, the brain region pairs are divided into k clusters, and the edge functional connection matrix is arranged according to the community to obtain the corresponding modular edge functional network;
[0008] For each subject's edge function network, the participation coefficient of each edge was calculated according to the modular edge function network partitioning;
[0009] The edge participation coefficient is used as a new neuroimaging feature to establish a disease prediction and classification model to explore the functional mechanisms of major multiple diseases and evaluate the prognosis; by establishing a regression model, it is used to reveal the developmental laws of brain function dynamics and its cognitive associations.
[0010] The method of clustering the edge time series using the k-means algorithm based on Euclidean distance is specifically as follows:
[0011] Merge the edge time series of a group of subjects into a high-dimensional cross-subject edge time series, cluster the edge time series to generate representative edge community assignments, and use this to divide the eFC matrix into non-overlapping modules;
[0012] The edge time series are directly clustered using the standard Euclidean distance k-means algorithm.
[0013] The participation coefficient of each edge is calculated according to the modular edge function network partitioning:
[0014] The participation coefficient of each node in the modular eFC network is calculated, and the functional role of each edge is explored from the perspective of interaction edges.
[0015] The edge function connection matrix is obtained as follows:
[0016] Acquire resting-state functional magnetic resonance imaging data; preprocess the acquired magnetic resonance imaging data; extract brain region time series from the preprocessed data, select a suitable cerebral cortex partitioning template, each brain region represents a node, extract the average fMRI time series of each brain region and perform Z-score standardization;
[0017] For any two brain regions, the time series are multiplied time point by time point to obtain the co-fluctuation time series, and the correlation coefficient between the node pairs is calculated to obtain the edge functional connection matrix.
[0018] The beneficial effects of the technical solution provided by the present invention are:
[0019] 1. This invention aims to propose a new method for quantifying the dynamic functional organization of the cerebral cortex. Through the edge participation coefficient index calculated based on the edge functional network, the node-centric network analysis is extended to the perspective of interactive edges. This edge-centric analysis method may help identify new disease biomarkers, characterize individual differences, and reveal the laws of brain development and aging.
[0020] 2. The present invention regards the interaction edge as an irreducible unit and calculates the edge participation coefficient. The high-dimensional information inherent in the edge function network enables the edge participation coefficient to capture more individual specificity, which will be helpful for research related to dynamic brain development. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 A flowchart of a method for quantifying the dynamic functional organization of the cerebral cortex based on edge functional networks;
[0022] Figure 2 Schematic diagram of the inter-group differences in edge participation coefficients;
[0023] Figure 3 It is a comparison diagram of edge participation coefficient and node participation coefficient;
[0024] Figure 4 It is a schematic diagram of the correlation between the edge participation coefficient and age (edge level);
[0025] Figure 5 Schematic diagram of the correlation between edge participation coefficient and age (sub-network level);
[0026] Figure 6 Schematic diagram of cognitive association of edge participation coefficient;
[0027] Figure 7 A schematic diagram showing high test-retest correlation (corr);
[0028] Figure 8 A schematic diagram showing high test-retest stability from the intraclass correlation coefficient (ICC). DETAILED DESCRIPTION
[0029] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.
[0030] Example 1
[0031] A method for quantifying the dynamic functional organization of the cerebral cortex based on edge functional networks, see Figure 1 , the method comprises the following steps:
[0032] 101: Acquire resting-state functional magnetic resonance imaging (fMRI) data;
[0033] 102: preprocessing the acquired magnetic resonance imaging data (ie, raw data);
[0034] 103: Extract brain region time series from preprocessed data, select appropriate cerebral cortex partition template, each brain region represents a node, extract the average fMRI time series of each brain region and perform Z-score standardization (z-score), multiply the time series of any two brain regions time point by time point to obtain the co-fluctuation time series (i.e., edge time series), calculate the correlation coefficient between node pairs, and obtain the edge functional connectivity matrix;
[0035] 104: Merge a group of test subjects' edge time series into a high-dimensional edge time series, and cluster the edge time series using the k-means algorithm based on Euclidean distance;
[0036] The number of communities varies between k=2 and k=20, and the clustering algorithm is iterated 1000 times. Through the processing of step 104, a representative network module division can be obtained.
[0037] 105: The clustering result divides the brain area pairs into k clusters, and the edge function connection matrix is arranged according to the community to obtain the corresponding modular edge function network;
[0038] 106: For each subject's edge functional network, the participation coefficient of each edge (brain region pair) is calculated according to the module division obtained in step 104, that is, the edge participation coefficient.
[0039] 107: Using edge participation coefficient as a new neuroimaging feature, we can establish disease prediction and classification models to explore the functional mechanisms of major multi-communicable diseases and evaluate prognosis. By establishing a regression model, we can reveal the developmental laws of brain function dynamics and its cognitive associations.
[0040] Among them, the edge participation coefficient quantifies the diversity of edge connection patterns in complex functional brain networks. As a new neuroimaging feature, on the one hand, it can be used to explore the functional mechanisms of major multiple diseases (such as stroke, cerebral small vessel disease) and evaluate the prognosis by establishing disease prediction and classification models; on the other hand, by establishing regression models, it can be used to reveal the developmental laws of brain function dynamics and their cognitive associations.
[0041] Example 2
[0042] Combine the following Figure 1-Figure 6 The scheme in Example 1 is further introduced, as described below:
[0043] 1. fMRI Data Acquisition and Preprocessing
[0044] Resting-state fMRI data usually come from open-source neuroimaging databases or clinical data sets. Using a 3T magnetic resonance scanner, scanning parameters such as repetition time (TR), echo time (TE), layer thickness, inter-layer spacing, number of layers, and scanning field of view need to be set before scanning. During the scanning process, the subjects are required to keep their heads still, close their eyes and relax, but stay awake. The preprocessing steps include: head motion correction, inter-layer time correction, spatial normalization, spatial smoothing, time domain filtering (0.01-0.1Hz), and interference signal regression (head motion parameters, white matter, cerebrospinal fluid average signal, and global signal). Subsequently, according to the selected brain atlas, the fMRI time series of all voxels corresponding to each brain region are averaged to obtain the time series of each brain node.
[0045] 2. Constructing edge time series and edge function connection matrix
[0046] For functional connectomes, temporal correlations between brain region activity time series are estimated from fMRI to define functional connectivity (FC) that can be modeled and analyzed using network science. The most common measure of this correlation is to calculate the Pearson correlation coefficient between each pair of node time series. Let x i =[x i (1),…,x i (T)] and x j =[x j (1),…,x j (T)] represents the time series of brain regions i and j. Then each time series is z-score normalized:
[0047]
[0048] in, and represent mean and standard deviation respectively.
[0049] Then the correlation between i and j can be calculated as Repeating this process for all pairs of brain regions, we get a node-by-node correlation matrix, the functional connectivity (FC). If there are N nodes, the dimension of this matrix is [N×N].
[0050] However, Faskowitz et al. recently proposed a new modeling framework by slightly modifying the above method to study functional brain networks from the perspective of edge-centricity. In short, two z-score-normalized node time series z i and z j Element-wise multiplication results in a vector of length T whose elements encode the magnitude of the moment-by-moment co-fluctuation between brain areas i and j. For example, suppose that at time t, brain areas i and j simultaneously increase their activity relative to baseline. These increases are expressed in z i and z j is encoded as a positive value in the t-th position in , so their product is also positive. The same is true if i and j both decrease their activity at the same time. On the other hand, if i increases and j decreases (or i decreases and j increases), this will appear as a negative value. Similarly, if one of them increases or decreases while the other approaches baseline, the corresponding product will be close to zero.
[0051] It is worth noting that the average value of the edge time series is equal to the Pearson correlation of the corresponding two node time series. Repeating this operation for N brain regions will yield edge time series, each of which has a length of T. By calculating the statistical correlation of each pair of edges, the edge functional connectivity (eFC) can be obtained.
[0052] Let c ij =[z i (1) z j (1),…,z i (T)·z j (T)] and c uv =[z u (1) z v (1),…,z u (T)·z v (T)] represent the co-fluctuation time series of edges {i,j} and {u,v} respectively, and eFC can be calculated as:
[0053]
[0054] 3. k-means clustering algorithm
[0055] First, in order to determine the benchmark for subsequent analysis, a group of subject edge time series were merged (concatenated) into a high-dimensional cross-subject edge time series, which were then clustered to generate representative edge community assignments, which were then used to divide the eFC matrix into non-overlapping modules. In terms of the choice of clustering method, the standard Euclidean distance k-means algorithm was used to directly cluster the edge time series. The number of clusters varied from k = 2 to k = 20, and for each k, the clustering algorithm was iterated 1000 times.
[0056] It is worth noting that, in general, other community detection algorithms can be used instead of k-means. The decision to use the k-means algorithm here was made for practical purposes because k-means exhibits faster running times than other algorithms, such as modularity maximization and Informap, which have been widely used in previous works to generate communities in functional and structural brain networks.
[0057] 4. Calculation of edge participation coefficient
[0058] Node-centric networks can be divided into node communities, also called modules. Some nodes have connections distributed across multiple modules, while others are only connected to other nodes in their own modules. The participation coefficient (PC) provides insights into how nodes communicate between modules and is often used to measure this property in network neuroscience. Similarly, embodiments of the present invention can calculate the participation coefficient of each node (i.e., brain region pair) in a modular eFC network to explore the functional role of each edge from the perspective of the interaction edge.
[0059] The edge participation coefficient measures how evenly the connections of a node in the eFC network are distributed across different communities, i.e., the diversity of functional connections. The closer this value is to 1, the more evenly a given node in the eFC network interacts with all communities, i.e., a greater degree of functional diversity. Formally, the PC of a node s is given by:
[0060]
[0061] Among them, k s (m) is the strength of node s with all nodes in community m, and k s is the total strength of node s. Using the representative community labels generated by the previous clustering, the edge participation coefficient of each subject can be calculated. For comparison, the embodiment of the present invention also calculates the PC of each node in the node-centric functional network.
[0062] 5. Validity Verification Scheme
[0063] Although a range of cluster numbers (k = 2, 4, ..., 20) were considered when performing the clustering operation to investigate the sensitivity of the edge participation coefficient to different cluster numbers k, a more systematic validation scheme is still needed. In order to evaluate the performance of the edge participation coefficient in brain functional network applications, two additional factors are worth considering:
[0064] (1) Test-retest reliability: Test-retest reliability reflects the variability of the edge participation coefficient within the subject and measures the stability and consistency of the edge participation coefficient over time.
[0065] (2) Sensitivity: Statistical analysis was used to examine whether there were significant inter-group differences, age correlations, and cognitive associations in the edge participation coefficients.
[0066] Test-retest reliability refers to the consistency of observations obtained from resting-state fMRI of the same subject collected at different times. In brain functional connectomics, the intraclass correlation coefficient (ICC) is often used to quantitatively evaluate the test-retest reliability of a measure. ICC is defined as the ratio of individual variability to overall variability, so its range is 0 to 1. Higher ICC values reflect higher test-retest reliability or more consistent measurements between different methods. It is generally believed that ICC < 0.4 indicates poor reliability, ICC between 0.41 and 0.59 indicates average reliability, ICC between 0.6 and 0.74 indicates good reliability, and ICC > 0.75 indicates excellent reliability. In the embodiment of the present invention, the ICC (1,1) variant is used, and the calculation formula is shown in formula (3), which is usually used to estimate the consistency of the exact value when the source of error is not specified.
[0067]
[0068] Wherein, MSR and MSW represent the mean square error between groups and within groups respectively, and f represents the number of scans. In this embodiment, f=2.
[0069] Example 3
[0070] The feasibility of the solution in Example 1 is verified by combining a specific example, as described below:
[0071] See also Figure 2 , the edge participation coefficients had significant between-group differences (PT1: preterm infants scanned before term equivalent age; PT 2: preterm infants scanned at term equivalent age; FT: term infants). Figure 3 ,The edge participation coefficient is better than the node participation coefficient in terms of correlation with age, indicating that the edge participation coefficient can capture more complex network topology information related to individual developmental characteristics, and is suitable for brain development-related research. Figure 4 and Figure 5 The correlation between edge participation coefficients and age at the edge level and subnetwork level is shown, indicating the ability of edge participation coefficients in reflecting the age-related dynamic functional organization of the cortex, and also reveals the heterogeneity of the dynamic functional diversity of the infant cerebral cortex, that is, different networks have specific developmental trajectories. Figure 6 , showing the correlation between edge participation coefficient and language and cognitive ability, suggesting the importance of early development of vision and high-order networks for later cognitive abilities. Figure 7 and Figure 8 ,The edge participation coefficient exhibits high test-retest reliability, indicating that the proposed participation coefficient is highly reproducible and stable, and is a reliable indicator that can be applied to functional brain network analysis.
[0072] In summary, the embodiments of the present invention aim to propose a new method for quantifying the dynamic functional organization of the cerebral cortex, which extends the node-centric network analysis to the perspective of interactive edges, and calculates the edge participation coefficient index based on the edge functional network, thereby characterizing the dynamic functional organization of the cerebral cortex. This edge-centric analysis method can characterize individual differences and reveal the laws of brain development such as Figure 2-6 shown.
[0073] The best implementation plan is to adopt patent transfer, technology cooperation or product development. Products developed based on this technology can be applied to statistical analysis, identification, classification of actual populations and pathologies, and scientific research to explore the dynamic functional characteristics of the brain in different diseases or pathologies, such as identifying new disease biomarkers, exploring disease pathogenesis and behavioral associations, etc.
[0074] Those skilled in the art will appreciate that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.
[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for quantifying dynamic functional organization of cerebral cortex based on edge functional network, characterized in that: The method comprises: The edge time series of a group of subjects are merged into a high-dimensional edge time series, and the edge time series is clustered using the k-means algorithm based on Euclidean distance; Based on the clustering results, the brain region pairs are divided into k clusters, and the edge functional connection matrix is arranged according to the community to obtain the corresponding modular edge functional network; For each subject's edge function network, the participation coefficient of each edge was calculated according to the modular edge function network partitioning; The edge participation coefficient is used as a new neuroimaging feature to establish a disease prediction and classification model to explore the functional mechanisms of major multiple diseases and evaluate the prognosis; by establishing a regression model, it is used to reveal the developmental laws of brain function dynamics and its cognitive associations.
2. The method for quantifying dynamic functional organization of cerebral cortex based on edge function network according to claim 1, characterized in that: The k-means algorithm based on Euclidean distance is used to cluster the edge time series as follows: Merge the edge time series of a group of subjects into a high-dimensional cross-subject edge time series, cluster the edge time series to generate representative edge community assignments, and use this to divide the eFC matrix into non-overlapping modules; The edge time series are directly clustered using the standard Euclidean distance k-means algorithm.
3. The method for quantifying dynamic functional organization of cerebral cortex based on edge function network according to claim 1, characterized in that: The participation coefficient of each edge is calculated based on the modular edge function network partitioning: The participation coefficient of each node in the modular eFC network is calculated, and the functional role of each edge is explored from the perspective of interaction edges.
4. The method for quantifying dynamic functional organization of cerebral cortex based on edge functional network according to claim 1, characterized in that: The edge function connection matrix is obtained as follows: Acquire resting-state functional magnetic resonance imaging data; preprocess the acquired magnetic resonance imaging data; extract brain region time series from the preprocessed data, select a suitable cerebral cortex partitioning template, each brain region represents a node, extract the average fMRI time series of each brain region and perform Z-score standardization; For any two brain regions, the time series are multiplied time point by time point to obtain the co-fluctuation time series, and the correlation coefficient between the node pairs is calculated to obtain the edge functional connection matrix.
Citation Information
Patent Citations
Brain network construction method based on brain region weight correlation
CN116561518A
Mental disease brain function activity evaluation device based on brain activation clustering algorithm
CN116597994A
Determining an indication as to whether a patient has a neurocognitive disease
WO2023048557A1
