Network graph generation method based on brain image data

Abstract

The invention provides a network graph generation method based on brain image data, and relates to the technical field of brain connection.The method comprises the steps that brain image data (including fMRI data and sMRI data) of a target object are obtained, the brain image data are preprocessed and registered, and registered brain graph data are obtained; dividing the brain map data into a plurality of brain regions based on a set brain map, and determining time sequence data of each brain region; based on the time sequence data of each brain region, a function connection matrix is determined, and non-diagonal elements in the function connection matrix reveal the connection strength between the brain regions; and based on the brain region and the functional connection matrix of the brain map data, generating a brain network map (which can be visualized) of the target object. According to the scheme, the sMRI data (the precision is millimeter level) is introduced as a registration basis, so that the registration precision can be improved, and a brain network diagram with higher precision can be generated subsequently.

Description

Technical Field

[0001] The present application relates to the field of brain connectivity technology, and in particular to a method for generating a network graph based on brain imaging data. Background Art

[0002] In recent years, fMRI-based brain network construction technology has become an important means to study the mechanisms of brain diseases and cognitive functions.

[0003] Existing schemes are usually based on fMRI data (using low-resolution structural information, such as average functional images for registration), directly dividing brain regions through standardized templates (such as MNI space), ignoring individual anatomical differences, resulting in deviations in the localization of functional activities. In the absence of sMRI data support, the resolution of fMRI directly registered to the standard space is low (usually 3–4 mm). 3 ), making it difficult to accurately align functional activity with anatomical structure, especially in limbic brain regions (such as the hippocampus and amygdala). Existing techniques for representing time series data of brain regions ignore functional heterogeneity within brain regions (such as activation gradients between core and limbic regions), resulting in reduced sensitivity of functional connectivity features and insufficient resolution of functional features.

[0004] Therefore, how to provide a brain network map generation solution with higher accuracy and better ability to reflect brain region functions is a direction that needs further research in this field. Summary of the Invention

[0005] The purpose of the embodiments of the present application is to provide a method for generating a network diagram based on brain imaging data, so that the brain network diagram has higher precision and more accurately reflects the function of brain regions.

[0006] In order to achieve the above objectives, the embodiments of the present application are implemented in the following manner:

[0007] An embodiment of the present application provides a method for generating a network graph based on brain imaging data, comprising: acquiring brain imaging data of a target object, wherein the brain imaging data includes fMRI data and sMRI data; preprocessing the brain imaging data and performing registration to obtain registered brain map data; dividing the brain map data into multiple brain regions based on a set brain map, and determining time series data for each brain region; determining a functional connectivity matrix based on the time series data for each brain region, wherein the non-diagonal elements in the functional connectivity matrix reveal the connection strength between brain regions; and generating a brain network graph of the target object based on the brain regions and the functional connectivity matrix of the brain map data.

[0008] Furthermore, the brain imaging data is preprocessed and registered to obtain registered brain map data, including: performing head motion correction, temporal layer correction, denoising and spatial smoothing on the fMRI data in the brain imaging data, and denoising and brain tissue segmentation on the sMRI data in the brain imaging data; registering the preprocessed fMRI data to the individual structural image space formed based on the preprocessed sMRI data; registering the preprocessed sMRI data to the standard space; and mapping the fMRI data to the standard space through a joint transformation to obtain registered brain map data.

[0009] Furthermore, based on the set brain map, the brain map data is divided into multiple brain regions, and the time series data of each brain region is determined, including: loading the set brain map so that the set brain map and the fMRI data are in the same space; extracting the binary mask corresponding to each set brain region in the set brain map, and determining each brain region in the brain map data accordingly; for each brain region in the brain map data, calculating the time series data based on the time series of each voxel in this brain region.

[0010] Furthermore, for each brain region in the brain map data, time series data is calculated based on the time series of each voxel in this brain region, including: for each brain region in the brain map data: extracting the time series of each voxel in this brain region; using any of the following strategies to calculate the time series data of this brain region: Strategy 1: calculating the average value of the time series of each voxel in this brain region as the time series data of this brain region; Strategy 2: calculating the weighted average value of the time series of each voxel in this brain region based on spatial distance as the time series data of this brain region; Strategy 3: calculating the weighted average value of the time series of each voxel in this brain region based on functional connectivity as the time series data of this brain region; Strategy 4: calculating the weighted average value of the time series of each voxel in this brain region based on activation intensity as the time series data of this brain region; Strategy 5: clustering analysis of functional sub-regions in this brain region, assigning different weights to each functional sub-region, and calculating the average value of the time series in each functional sub-region in this brain region, and then calculating the weighted average value based on the weight and average value of each functional sub-region as the time series data of this brain region.

[0011] Furthermore, when using Strategy 1 to calculate the time series data of this brain region, it includes:

[0012] The time series data of this brain region is calculated using the following formula:

[0013]

[0014] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i(t) represents the signal value of voxel i in brain area M at time node t, and T is the total number of time nodes; then the time series data of brain area M is: M = {M(1),…,M(t),…,M(T)}.

[0015] Furthermore, when using Strategy 2 to calculate the time series data of this brain region, it includes:

[0016] The time series data of this brain region is calculated using the following formula:

[0017]

[0018] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, δ i is the weight corresponding to voxel i in brain region M, weight δ i satisfy:

[0019]

[0020] Among them, d ij represents the relative distance between voxel i and voxel j in brain region M.

[0021] Furthermore, when using strategy three to calculate the time series data of this brain region, it includes:

[0022] The time series data of this brain region is calculated using the following formula:

[0023]

[0024] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, r i is the weight corresponding to voxel i in brain area M, and weight r i satisfy:

[0025] r i =max{r ic (k),r' ic (k)},k∈[0,K],K∈Z + ,K <T,

[0026]

[0027] Among them, max{r ic (k),r' ic (k)} means taking ric (k) and r' ic (k), k is the number of lagging nodes, K is the maximum number of lagging nodes, and K is a positive integer. <T,r ic (k) represents the correlation coefficient between voxel i in brain region M and the target brain region, is the average signal value of voxel i in brain area M at each time node, x c (t+k) is the signal value of the reference time series of the target brain area at the time node (t+k), is the average signal value of the reference time series of the target brain area at each time node, r' ic (k) also represents the correlation coefficient between voxel i in brain region M and the target brain region, x i (t+k) is the signal value of voxel i in brain area M at time node (t+k), x c (t) is the signal value of the reference time series of the target brain area at time node (t).

[0028] Furthermore, when using strategy 4 to calculate the time series data of this brain region, the method includes: processing the fMRI data using a generalized linear model to obtain a task-state activation map for a specific task, where the specific task includes one of a memory task, a decision-making task, a language task, an attention task, a perception task, an emotion task, and an executive function task; extracting the activation intensity of each voxel in this brain region based on the task-state activation map; and calculating the time series data of this brain region using the following formula:

[0029]

[0030] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, w i is the activation intensity corresponding to voxel i in brain region M, and

[0031] Furthermore, when strategy five is used to calculate the time series data of this brain region, it includes: using the nearest neighbor propagation algorithm or the K-means clustering algorithm to perform cluster analysis on the time series of voxels in this brain region, and determining S clusters as the functional sub-regions in this brain region; based on the number of voxels contained in the S clusters, determining the weight of the functional sub-region corresponding to each cluster; based on the time series of the voxels in each functional sub-region in this brain region, calculating the average value of the time series in each functional sub-region; and calculating the weighted average value based on the weight and average value of each functional sub-region as the time series data of this brain region.

[0032] Furthermore, based on the time series data of each brain region, a functional connectivity matrix was determined, including: numbering all brain regions from 1 to m; calculating the correlation coefficient between the time series of every two brain regions, where the correlation coefficient between the i-th brain region and the j-th brain region is recorded as P ij , i, j∈[1,m], and i≠j; based on all correlation coefficients, construct the functional connectivity matrix P with the main diagonal as 1:

[0033]

[0034] Among them, P m1 is the correlation coefficient between the mth brain region and the first brain region, P 1m is the correlation coefficient between the first brain region and the mth brain region.

[0035] Furthermore, based on the brain regions and functional connectivity matrix of the brain map data, a brain network diagram of the target object is generated, including: taking each brain region of the brain map data as a node in the network diagram; determining the non-zero elements of the off-diagonal line from the upper triangular part of the functional connectivity matrix; for each determined non-zero element: determining the corresponding two nodes based on the position of the non-zero element, connecting the two nodes with an edge, and using the value of the non-zero element as the connection strength of the two nodes, thereby constructing a brain network diagram; and visualizing the brain network diagram.

[0036] Beneficial effects:

[0037] The target subject's brain imaging data (including fMRI data and sMRI data) is acquired, preprocessed, and registered to obtain registered brain map data. Based on a set brain atlas, the brain map data is divided into multiple brain regions, and the time series data for each brain region is determined. Based on the time series data for each brain region, a functional connectivity matrix is determined, where the off-diagonal elements in the functional connectivity matrix reveal the strength of connections between brain regions. Based on the brain regions and functional connectivity matrix of the brain map data, a brain network diagram of the target subject (which can be visualized) is generated. This solution introduces sMRI data (with millimeter-level accuracy) as the basis for registration, which can improve registration accuracy and facilitate the subsequent generation of higher-precision brain network diagrams.

[0038] A key improvement of this approach lies in the fact that after dividing the brain map data into multiple brain regions, multiple extraction strategies are designed to meet the differentiated needs of different scenarios when extracting time series data from brain regions. These strategies fully account for the functional heterogeneity of brain regions, optimize the calculation of brain region time series, and highlight the core functional characteristics of brain regions. For conventional application scenarios, strategy one (calculating the average of the time series of each voxel in this brain region as the time series data for this brain region) is adopted. This approach reduces the amount of computation through mean calculation and improves the efficiency of brain network map generation. It is suitable for scenarios that do not require in-depth analysis of specific brain regions or research under specific tasks, and can quickly generate brain network maps with high efficiency. In many brain regions, functional activity exhibits a gradient change from core regions to peripheral regions, such as in the primary sensory cortex or certain areas of the default network. Traditional methods may ignore this gradient feature through simple averaging, resulting in blurred functional connectivity features. Strategy 2 (calculating the weighted average of the time series of each voxel within the brain region based on spatial distance as the time series data for this brain region) introduces spatial distance weighting to incorporate the spatial location information of the voxels into the time series calculation. This can more accurately reflect the gradient of functional changes within the brain region, thereby better reflecting the functional characteristics of the brain region and taking into account features that may be omitted when calculating the average. Based on this, the corresponding time series data is extracted as the basis for generating a brain network map, which makes the generated brain network map more reflective of the functional characteristics of the brain region. For example, when focusing on analyzing the association of a certain brain region with other brain regions (core nodes of specific networks such as the default network and the salience network), strategy 3 is designed (calculating the weighted average of the time series of each voxel within the brain region based on functional connectivity as the time series data for this brain region. This weighted average based on functional connectivity specifies the target brain region, considers the correlation information of each brain region with the target brain region, and assigns corresponding weights). Through the weighted average based on functional connectivity, the strength of the association between these core nodes in the network is highlighted. By calculating the correlation weights between voxels and target brain regions, the functional roles of core nodes can be more accurately reflected, allowing for better consideration of associations with target brain regions during the extraction of time-series data, thereby generating brain network diagrams more suitable for this application scenario. For example, when focusing on analyzing brain region functions during specific tasks (including memory tasks, decision-making tasks, language tasks, attention tasks, perception tasks, emotion tasks, and executive function tasks), the activity characteristics of these regions are often closely related to the task demands.Therefore, design strategy four (calculating the time series of each voxel in this brain region based on the weighted average of the activation intensity as the time series data of this brain region) extracts the activation intensity of each voxel in this brain region based on the task-state activation map, uses it as the corresponding weight, calculates the weighted average of the voxels in the brain region, and extracts the time series data of the primary brain region based on this. By using the weighted average based on the activation intensity, the generalized linear model is used to extract the task-related activation intensity as the weight, ensuring that the network diagram can accurately capture the characteristics of brain region activity driven by the task. This method avoids the loss of information in traditional methods and provides higher sensitivity and specificity for task-state brain function research. With the continuous improvement of computing power, the requirements for precision in brain science research are also constantly increasing. Design strategy five (this strategy currently has a large amount of computation. Taking the division of 379 brain regions as an example, the amount of voxel data for each brain region ranges from thousands to tens of thousands or even more than 100,000. Such a huge amount of computation is currently too costly to calculate, but in theory it may be possible to implement it at low cost in the future). This strategy uses a clustering algorithm to identify functional sub-regions within the brain region, and assigns weights based on the characteristics of the sub-regions to further determine the time series data of the brain region, providing a technical reserve for future high-precision brain network research. This method can break through the limitations of traditional methods that treat brain regions as homogeneous units, and can more realistically reflect the complex functional organization within the brain region. Although the current computational workload is large, with the improvement of hardware performance and algorithm optimization, this strategy is expected to become the mainstream method for future brain network research.

[0039] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. The following drawings only show certain embodiments of the present application and are intended to assist in understanding. Therefore, they should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 This is a flowchart of a method for generating a network diagram based on brain imaging data provided in an embodiment of the present application.

[0042] Figure 2 A schematic diagram of the software interface.

[0043] Figure 3 Schematic diagram of application operations after generating the brain network map. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0045] See also Figure 1 , Figure 1 Flowchart of a method for generating a network graph based on brain imaging data provided in an embodiment of the present application. In this embodiment, the method for generating a network graph based on brain imaging data may include steps S10, S20, S30, S40, and S50.

[0046] In order to realize the generation of the network diagram, step S10 may be executed first.

[0047] Step S10: Acquire brain imaging data of the target object, wherein the brain imaging data includes fMRI data and sMRI data.

[0048] In this embodiment, brain imaging data of the target subject can be obtained, including fMRI data and sMRI data collected during the same time period. Of course, for some specific scenarios, fMRI data and sMRI data from other time periods may also be required. For example, when performing a specific task (such as a memory task, decision-making task, language task, attention task, perception task, emotion task, and executive function task), sMRI data and resting-state fMRI data are also required, which is not limited here.

[0049] After obtaining the brain image data of the target object, step S20 may be executed.

[0050] Step S20: pre-processing the brain image data and performing registration to obtain registered brain map data.

[0051] In this embodiment, motion correction, temporal slice correction, denoising, and spatial smoothing can be performed on the fMRI data within the brain imaging data. Furthermore, denoising and brain tissue segmentation can be performed on the sMRI data within the brain imaging data. Subsequently, the preprocessed fMRI data is registered to an individual structural image space formed based on the preprocessed sMRI data. The preprocessed sMRI data is then registered to a standard space (e.g., MNI space). A joint transformation is then used to map the fMRI data to the standard space, yielding registered brain map data.

[0052] After obtaining the registered brain map data, step S30 may be executed.

[0053] Step S30: Based on the set brain map, the brain map data is divided into multiple brain regions, and the time series data of each brain region is determined.

[0054] In this embodiment, a predefined brain map can be loaded so that the predefined brain map and fMRI data are in the same space (e.g., MNI space). There are multiple predefined brain maps, such as the AAL and Desikan-Killiany maps, that can be selected as needed. Other brain maps can also be used (this unit uses a brain map that divides 379 brain regions), or they can be replaced if there are subsequent updates to the brain map. This is not limited here.

[0055] Afterwards, the binary mask corresponding to each set brain region in the set brain atlas can be extracted to determine each brain region in the brain map data. For each brain region in the brain map data, the time series data of this brain region can be calculated based on the time series of each voxel in the brain region.

[0056] In order to adapt to various specific application scenarios, this embodiment specifically designs time series data extraction strategies for various brain regions.

[0057] Exemplarily, for each brain region in the brain map data: extract the time series of each voxel in this brain region, and use any of the following strategies to calculate the time series data of this brain region:

[0058] Specifically, for routine scenarios (for example, mainly to observe lesions at the structural level), strategy 1 can be used.

[0059] Strategy 1: Calculate the average value of the time series of each voxel in this brain region as the time series data of this brain region.

[0060] When using strategy 1 to calculate the time series data of this brain region, the following formula can be used to calculate the time series data of this brain region:

[0061]

[0062] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, and T is the total amount of time nodes.

[0063] That is, the time series data of brain region M is: M = {M(1),…,M(t),…,M(T)}.

[0064] For common scenarios, strategy one is adopted (calculating the average value of the time series of each voxel in this brain area as the time series data of this brain area). This scheme reduces the amount of calculation through mean calculation, improves the efficiency of brain network map generation, and adapts to common scenarios. It does not require specific analysis of a certain brain area or research under specific tasks. Therefore, this scheme is used to generate brain network maps with high efficiency.

[0065] Specifically, in many brain regions, functional activity exhibits a gradient from core to peripheral regions (e.g., different activity differences from core to peripheral regions, such as the primary sensory cortex and default mode network). Traditional methods, such as simple averaging, may overlook this gradient feature, resulting in blurred functional connectivity features. Therefore, strategy 2 was designed.

[0066] Strategy 2: Calculate the weighted average of the time series of each voxel in this brain region based on spatial distance as the time series data of this brain region.

[0067] When using strategy 2 to calculate the time series data of this brain region, the following formula can be used to calculate the time series data of this brain region:

[0068]

[0069] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, δ i is the weight corresponding to voxel i in brain region M, weight δ i satisfy:

[0070]

[0071] Among them, d ij represents the relative distance between voxel i and voxel j in brain area M, and σ is used to adjust the decay speed, that is, to adjust the amplitude of the weight change based on the distance.

[0072] Then, the time series data of brain area M is: M = {M(1),…,M(t),…,M(T)}.

[0073] Strategy 2 introduces spatial distance weighting to integrate the spatial position information of voxels into time series calculations, which can more accurately reflect the functional change gradient within the brain region and better reflect the functional characteristics of the brain region. It can take into account the features that may be omitted when calculating the average value, and extract the corresponding time series data as the basis for generating a brain network map, so that the generated brain network map can better reflect the functional characteristics of the brain region.

[0074] Specifically, when focusing on analyzing the correlation between a certain brain region and other brain regions (core nodes of specific networks such as the default network and the salience network), if only strategy one is adopted, it is inevitable to ignore the correlation information of some brain regions. Therefore, strategy three is designed.

[0075] Strategy 3: Calculate the weighted average of the time series of each voxel in this brain region based on functional connectivity as the time series data of this brain region.

[0076] When using strategy three to calculate the time series data of this brain region, the following formula can be used to calculate the time series data of this brain region:

[0077]

[0078] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, r i is the weight corresponding to voxel i in brain region M, and weight r i satisfy:

[0079] r i =max{r ic (k),r' ic (k)},k∈[0,K],K∈Z + ,K <T, (5)

[0080]

[0081] Among them, max{r ic (k),r' ic (k)} means taking r ic (k) and r' ic The maximum value in (k), k is the number of lagging nodes, K is the maximum number of lagging nodes (for example, 3, then k needs to calculate 4 lag cases from 0 to 3), and K is a positive integer, K <T,r ic (k) represents the correlation coefficient between voxel i in brain region M and the target brain region, is the average signal value of voxel i in brain area M at each time node, x c (t+k) is the signal value of the reference time series of the target brain area at the time node (y+k), is the average signal value of the reference time series of the target brain area at each time node, r' ic (k) also represents the correlation coefficient between voxel i in brain region M and the target brain region, x i (t+k) is the signal value of voxel i in brain area M at time node (t+k), x c (t) is the signal value of the reference time series of the target brain area at time node (t).

[0082] Then, the time series data of brain area M is: M = {M(1),…,M(t),…,M(T)}.

[0083] In this strategy, the target brain region is specified (the reference time series of the target brain region needs to be known), the correlation information between each brain region and the target brain region is considered, and corresponding weights are assigned. In this way, the association with the target brain region can be better considered during the extraction of brain region time series data, thereby generating a brain network diagram more suitable for this scenario. When allocating weights, the time series of the voxels and the reference time series of the target brain region are used to calculate the correlation taking into account the lag. Formula (6) considers the case where the target brain region lags behind brain region M, while formula (7) considers the case where brain region M lags behind the target brain region.

[0084] Specifically, when focusing on analyzing the brain region functions of specific tasks (including memory tasks, decision-making tasks, language tasks, attention tasks, perception tasks, emotion tasks and executive function tasks), the activity characteristics of the brain regions are often closely related to the task requirements. Therefore, strategy four is designed.

[0085] Strategy 4: Calculate the weighted average of the time series of each voxel in this brain region based on the activation intensity as the time series data of this brain region.

[0086] When using strategy 4 to calculate the time series data of this brain region, it is necessary to use a generalized linear model to process the fMRI data to obtain a task-state activation map for a specific task. Then, based on the task-state activation map, the activation intensity of each voxel in this brain region is extracted. Based on this, the time series data of this brain region is calculated using the following formula:

[0087]

[0088] Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, w i is the activation intensity corresponding to voxel i in brain region M, and

[0089] Then, the time series data of brain area M is: M = {M(1),…,M(t),…,M(T)}.

[0090] The fMRI data is processed using a generalized linear model to obtain a task-state activation map for a specific task. Based on the task-state activation map, the activation intensity of each voxel in this brain area is extracted and used as the corresponding weight. The weighted average of the voxels in the brain area is calculated, and the time series data of the primary brain area is extracted based on this. This can be applied to specific tasks to better extract the corresponding brain area features and avoid information loss caused by averaging.

[0091] Specifically, with the continuous improvement of computing power, the requirements for accuracy in brain science research are also constantly increasing. This embodiment also involves strategy five. Currently, this strategy has a large amount of computation. Taking the division of 379 brain regions as an example, the amount of voxel data for each brain region ranges from thousands to tens of thousands or even more than 100,000. Such a huge amount of computation is currently too costly, but in theory it may be achievable at low cost in the future.

[0092] Strategy 5: Through cluster analysis of the functional sub-regions within this brain region, different weights are assigned to each functional sub-region, and the average value of the time series in each functional sub-region within this brain region is calculated. Then, based on the weight and average value of each functional sub-region, a weighted average is calculated as the time series data of this brain region.

[0093] When using strategy five to calculate the time series data of this brain region, a clustering algorithm such as the nearest neighbor propagation algorithm or the K-means clustering algorithm can be used to perform cluster analysis on the time series of voxels in this brain region to determine S clusters as the functional subregions within this brain region. Then, based on the number of voxels contained in the S clusters, the weight of the functional subregion corresponding to each cluster can be determined (for example, the weight is allocated in proportion to the number). Then, based on the time series of the voxels in each functional subregion in this brain region, the average value of the time series of each functional subregion is calculated. Finally, a weighted average value is calculated based on the weight and average value of each functional subregion to serve as the time series data of this brain region.

[0094] Strategy 5 uses cluster analysis (such as the commonly used nearest neighbor propagation algorithm and K-Means algorithm) on the voxels in each brain region to determine the functional subregions of each brain region. Weights are assigned based on the characteristics of the subregions to further determine the time series data of the brain region, providing a technical reserve for future high-precision brain network research. This method can break through the limitations of traditional methods that treat brain regions as homogeneous units and can more realistically reflect the complex functional organization within the brain region. Although the current computational complexity is large, with the improvement of hardware performance and algorithm optimization, this strategy is expected to become the mainstream method for future brain network research.

[0095] After the time series data of each brain region is determined, step S40 may be executed.

[0096] Step S40: Determine a functional connectivity matrix based on the time series data of each brain region, wherein the off-diagonal elements in the functional connectivity matrix reveal the connection strength between brain regions.

[0097] In this embodiment, all brain regions can be numbered from 1 to m, and then the correlation coefficient between the time series of each two brain regions can be calculated (for example, using the Pearson correlation coefficient), where the correlation coefficient between the i-th brain region and the j-th brain region is denoted as P ij, i,j∈[1,m], and i≠j. Then, based on all the correlation coefficients, construct the functional connectivity matrix P with the main diagonal as 1:

[0098]

[0099] Among them, P m1 is the correlation coefficient between the mth brain region and the first brain region, P 1m is the correlation coefficient between the first brain region and the mth brain region.

[0100] After obtaining the functional connectivity matrix, step S50 may be executed.

[0101] Step S50: Generate a brain network map of the target object based on the brain regions and functional connectivity matrix of the brain map data.

[0102] In this embodiment, each brain region of the brain map data can be used as a node in the network diagram, and the non-zero elements of the off-diagonal line are determined from the upper triangular part of the functional connectivity matrix (or the lower triangular part, because the functional connectivity matrix p is a symmetric matrix with the main diagonal elements being 1). Then, for each non-zero element determined: the corresponding two nodes are determined based on the position of the non-zero element, the two nodes are connected by an edge, and the value of the non-zero element is used as the connection strength of the two nodes. Based on this, a brain network diagram is constructed, and then the brain network diagram is visualized.

[0103] Of course, other network diagram construction methods can also be used. This solution uses graph theory methods to construct brain network diagrams, and then uses tools (such as BrainNetViewer, PyBrainSim, Gephi, etc.) to realize the visualization of brain network diagrams.

[0104] The initial version of the software for the network diagram generation method based on brain imaging data in this embodiment has been developed (e.g. Figure 2 and Figure 3 As shown), some optimization designs in this embodiment may be reflected in subsequent software upgrade versions, but some optimization designs need to wait for the time to be ripe before they can be developed and applied (such as strategy five).

[0105] In summary, the embodiment of the present application provides a method for generating a network diagram based on brain imaging data, by obtaining brain imaging data (including fMRI data and sMRI data) of a target object, preprocessing the brain imaging data, and performing registration to obtain registered brain map data; based on a set brain map, the brain map data is divided into multiple brain regions, and the time series data of each brain region is determined; based on the time series data of each brain region, a functional connection matrix is determined, wherein the non-diagonal elements in the functional connection matrix reveal the connection strength between the brain regions; based on the brain regions and the functional connection matrix of the brain map data, a brain network diagram of the target object is generated (which can be visualized). This solution introduces sMRI data (with an accuracy of millimeters) as the basis for registration, which can improve the registration accuracy and facilitate the subsequent generation of a higher-precision brain network diagram.

[0106] A key improvement of this approach lies in the fact that after dividing the brain map data into multiple brain regions, multiple extraction strategies are designed to meet the differentiated needs of different scenarios when extracting time series data from brain regions. These strategies fully account for the functional heterogeneity of brain regions, optimize the calculation of brain region time series, and highlight the core functional characteristics of brain regions. For conventional application scenarios, strategy one (calculating the average of the time series of each voxel in this brain region as the time series data for this brain region) is adopted. This approach reduces the amount of computation through mean calculation and improves the efficiency of brain network map generation. It is suitable for scenarios that do not require in-depth analysis of specific brain regions or research under specific tasks, and can quickly generate brain network maps with high efficiency. In many brain regions, functional activity exhibits a gradient change from core regions to peripheral regions, such as in the primary sensory cortex or certain areas of the default network. Traditional methods may ignore this gradient feature through simple averaging, resulting in blurred functional connectivity features. Strategy 2 (calculating the weighted average of the time series of each voxel within the brain region based on spatial distance as the time series data for this brain region) introduces spatial distance weighting to incorporate the spatial location information of the voxels into the time series calculation. This can more accurately reflect the gradient of functional changes within the brain region, thereby better reflecting the functional characteristics of the brain region and taking into account features that may be omitted when calculating the average. Based on this, the corresponding time series data is extracted as the basis for generating a brain network map, which makes the generated brain network map more reflective of the functional characteristics of the brain region. For example, when focusing on analyzing the association of a certain brain region with other brain regions (core nodes of specific networks such as the default network and the salience network), strategy 3 is designed (calculating the weighted average of the time series of each voxel within the brain region based on functional connectivity as the time series data for this brain region. This weighted average based on functional connectivity specifies the target brain region, considers the correlation information of each brain region with the target brain region, and assigns corresponding weights). Through the weighted average based on functional connectivity, the strength of the association between these core nodes in the network is highlighted. By calculating the correlation weights between voxels and target brain regions, the functional roles of core nodes can be more accurately reflected, allowing for better consideration of associations with target brain regions during the extraction of time-series data, thereby generating brain network diagrams more suitable for this application scenario. For example, when focusing on analyzing brain region functions during specific tasks (including memory tasks, decision-making tasks, language tasks, attention tasks, perception tasks, emotion tasks, and executive function tasks), the activity characteristics of these regions are often closely related to the task demands.Therefore, design strategy four (calculating the time series of each voxel in this brain region based on the weighted average of the activation intensity as the time series data of this brain region) extracts the activation intensity of each voxel in this brain region based on the task-state activation map, uses it as the corresponding weight, calculates the weighted average of the voxels in the brain region, and extracts the time series data of the primary brain region based on this. By using the weighted average based on the activation intensity, the generalized linear model is used to extract the task-related activation intensity as the weight, ensuring that the network diagram can accurately capture the characteristics of brain region activity driven by the task. This method avoids the loss of information in traditional methods and provides higher sensitivity and specificity for task-state brain function research. With the continuous improvement of computing power, the requirements for precision in brain science research are also constantly increasing. Design strategy five (this strategy currently has a large amount of computation. Taking the division of 379 brain regions as an example, the amount of voxel data for each brain region ranges from thousands to tens of thousands or even more than 100,000. Such a huge amount of computation is currently too costly to calculate, but in theory it may be possible to implement it at low cost in the future). This strategy uses a clustering algorithm to identify functional sub-regions within the brain region, and assigns weights based on the characteristics of the sub-regions to further determine the time series data of the brain region, providing a technical reserve for future high-precision brain network research. This method can break through the limitations of traditional methods that treat brain regions as homogeneous units, and can more realistically reflect the complex functional organization within the brain region. Although the current computational workload is large, with the improvement of hardware performance and algorithm optimization, this strategy is expected to become the mainstream method for future brain network research.

[0107] The foregoing is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for generating a network graph based on brain imaging data, characterized in that: include: Acquiring brain imaging data of a target object, wherein the brain imaging data includes fMRI data and sMRI data; Preprocessing and registering brain image data to obtain registered brain map data; Based on the set brain map, the brain map data is divided into multiple brain regions, and the time series data of each brain region is determined; Based on the time series data of each brain region, a functional connectivity matrix is determined, where the off-diagonal elements in the functional connectivity matrix reveal the connection strength between brain regions; Based on the brain regions and functional connectivity matrix of the brain map data, a brain network map of the target subject is generated.

2. The method for generating a network graph based on brain imaging data according to claim 1, wherein: Preprocess the brain image data and perform registration to obtain registered brain map data, including: Performing head motion correction, temporal layer correction, denoising and spatial smoothing on fMRI data in brain imaging data, and denoising and brain tissue segmentation on sMRI data in brain imaging data; The preprocessed fMRI data are registered to the individual structural image space formed based on the preprocessed sMRI data; The pre-processed sMRI data were registered to the standard space; The fMRI data are mapped into the standard space through joint transformation to obtain the registered brain map data.

3. The method for generating a network graph based on brain imaging data according to claim 1, wherein: Based on the set brain map, the brain map data is divided into multiple brain regions, and the time series data of each brain region is determined, including: Load the set brain map so that the set brain map and fMRI data are in the same space; Extract the binary mask corresponding to each set brain region in the set brain map, and determine each brain region in the brain map data based on it; For each brain region in the brain map data, time series data is calculated based on the time series of each voxel in this brain region.

4. The method for generating a network graph based on brain image data according to claim 3, wherein: For each brain region in the brain map data, time series data is calculated based on the time series of each voxel in this brain region, including: For each brain region in the brain map data: Extract the time series of each voxel in this brain region; Use any of the following strategies to calculate the time series data for this brain region: Strategy 1: Calculate the average value of the time series of each voxel in this brain region as the time series data of this brain region; Strategy 2: Calculate the weighted average of the time series of each voxel in this brain region based on spatial distance as the time series data of this brain region; Strategy 3: Calculate the weighted average of the time series of each voxel in this brain region based on functional connectivity as the time series data of this brain region; Strategy 4: Calculate the weighted average of the activation intensity of the time series of each voxel in this brain region as the time series data of this brain region; Strategy 5: Through cluster analysis of the functional sub-regions within this brain region, different weights are assigned to each functional sub-region, and the average value of the time series in each functional sub-region within this brain region is calculated. Then, based on the weight and average value of each functional sub-region, a weighted average is calculated as the time series data of this brain region.

5. The method for generating a network graph based on brain image data according to claim 4, characterized in that: When using strategy 1 to calculate the time series data of this brain region, it includes: The time series data of this brain region is calculated using the following formula: Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, and T is the total amount of time nodes; Then the time series data of brain area M is: M = {M(1),…,M(t),…,M(T)}.

6. The method for generating a network graph based on brain image data according to claim 4, characterized in that: When using Strategy 2 to calculate the time series data of this brain region, it includes: The time series data of this brain region is calculated using the following formula: Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, δ i is the weight corresponding to voxel i in brain region M, weight δ i satisfy: Among them, d ij represents the relative distance between voxel i and voxel j in brain region M.

7. The method for generating a network graph based on brain image data according to claim 4, wherein: When using strategy three to calculate the time series data of this brain region, it includes: The time series data of this brain region is calculated using the following formula: Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, r i is the weight corresponding to voxel i in brain area M, and weight r i satisfy: r i =max{r ic (k),r' ic (k)},k∈[0,K],K∈Z + ,K<T, Among them, max{r ic (k),r' ic (k)} means taking r ic (k) and r' ic (k), k is the number of lagging nodes, K is the maximum number of lagging nodes, and K is a positive integer. <T,r ic (k) represents the correlation coefficient between voxel i in brain region M and the target brain region, is the average signal value of voxel i in brain area M at each time node, x c (t+k) is the signal value of the reference time series of the target brain area at the time node (t+k), is the average signal value of the reference time series of the target brain area at each time node, r' ic (k) also represents the correlation coefficient between voxel i in brain region M and the target brain region, x i (t+k) is the signal value of voxel i in brain area M at time node (t+k), x c (t) is the signal value of the reference time series of the target brain area at time node (t).

8. The method for generating a network graph based on brain image data according to claim 4, wherein: When using Strategy 4 to calculate the time series data of this brain region, it includes: The fMRI data were processed using a generalized linear model to obtain task-specific activation maps, where the specific task included one of a memory task, a decision-making task, a language task, an attention task, a perception task, an emotion task, and an executive function task; Extract the activation intensity of each voxel in this brain area based on the task-state activation map; The time series data of this brain region is calculated using the following formula: Where M(t) represents the average signal value of brain region M at time node t, N is the total number of voxels in brain region M, and x i (t) represents the signal value of voxel i in brain area M at time node t, T is the total amount of time nodes, w i is the activation intensity corresponding to voxel i in brain region M, and 9. The method for generating a network graph based on brain image data according to claim 4, wherein: When using strategy 5 to calculate the time series data of this brain region, it includes: A neighbor propagation algorithm or K-means clustering algorithm is used to perform cluster analysis on the time series of voxels in this brain region, and S clusters are determined as functional sub-regions in this brain region; Based on the number of voxels contained in the S clusters, the weight of the functional sub-region corresponding to each cluster is determined; Based on the time series of voxels in each functional subregion within this brain region, the average value of the time series in each functional subregion is calculated; The weighted average was calculated based on the weight and average of each functional subregion as the time series data of this brain region.

10. The method for generating a network graph based on brain image data according to claim 1, wherein: Based on the time series data of each brain region, a functional connectivity matrix was determined, including: All brain regions are numbered from 1 to m; Calculate the correlation coefficient between the time series of each two brain regions, where the correlation coefficient between the i-th brain region and the j-th brain region is recorded as P ij , i, j∈[1,m], and i≠j; Based on all correlation coefficients, construct the functional connectivity matrix P with the main diagonal as 1: Among them, P m1 is the correlation coefficient between the mth brain region and the first brain region, P 1m is the correlation coefficient between the first brain region and the mth brain region.

Images