Causal network construction method and system for human brain effective connectivity estimation

Abstract

The application provides a causal network construction method and system for human brain effective connection estimation, comprising: obtaining brain fNIRS time series data to be analyzed, and dividing the time series data into several subsets; for each subset obtained, generating directed edges based on a plurality of preset basic causal discovery algorithms respectively; wherein the directed edges are used to represent the causal relationship between different brain regions; integrating the directed edges corresponding to each subset generated by the same basic causal discovery algorithm, and calculating the directed edge number distribution curve of each basic causal discovery algorithm and the corresponding coefficient of variation; integrating the generation results of each basic causal discovery algorithm by taking the coefficient of variation as the weight, and obtaining an initial causal graph with weights for each directed edge; based on the weight of each directed edge in the initial causal graph and a preset confidence threshold, selecting confident and effective edges to construct the causal graph, and obtaining a final causal network.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of electroencephalogram information processing, and particularly relates to a causal network construction method and system for human brain effective connection estimation. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] With the aging of the world, the global prevalence of neurodegenerative diseases such as Alzheimer's disease (AD), Huntington's disease (HD), mild cognitive impairment (MCI) and Parkinson's disease (PD) is increasing. Neurodegenerative diseases can cause gradual loss of brain cell function, leading to various cognitive and motor disorders, and seriously affect people's quality of life. Unfortunately, most neurodegenerative diseases currently have no effective treatment, which makes early diagnosis particularly critical.

[0004] The development of artificial intelligence methods and theories has prompted people to use advanced technologies such as transfer learning, incremental learning and federated learning when building auxiliary diagnostic models for neurodegenerative diseases. However, these end-to-end models can only make black box predictions; at the same time, causal discovery can infer the causal relationship between different observation variables and may reveal the underlying mechanisms behind neurodegenerative diseases and determine effective digital biomarkers for early disease detection. Researchers at the University of Liege used causal discovery to identify subcortical and frontotemporal-parietal regions as the most important networks for distinguishing between conscious disorders. Similarly, researchers at Beijing Information Science and Technology University found specific changes in the dynamic brain functional connectivity patterns of AD patients and applied their findings to the transition stage of AD to study the performance of these patterns in MCI patients. However, the inventors found that the results obtained by different existing causal discovery methods are not consistent and each has its own advantages and disadvantages, and even when calculating the same data, there can be large differences, making it impossible to quickly and accurately obtain the optimal causal network relying on existing methods. SUMMARY

[0005] In order to overcome the above deficiencies of the prior art, the application provides a causal network construction method and system for human brain effective connection estimation, which adopts a weighted integrated different causal discovery method, measures the weight of different basic methods through a distribution metric function, and obtains a confidence edge set according to a generated directed edge confidence metric curve; secondly, a critical threshold is adaptively selected to obtain the confidence edge set; finally, the condition of a directed acyclic graph is used to redefine the confident edge set and obtain a final causal graph; through the above scheme, the high-confidence inference of different algorithms is integrated, thereby generating a high-reliability and high-precision weighted integrated causal discovery scheme, effectively ensuring the accuracy of the determination of the causal relationship between different brain regions, and laying a foundation for subsequent human brain effective connection estimation.

[0006] According to a first aspect of an embodiment of the application, a causal network construction method for human brain effective connection estimation is provided, comprising:

[0007] obtaining brain fNIRS time series data to be analyzed, and dividing the time series data into a plurality of subsets;

[0008] for each obtained subset, generating a directed edge based on a plurality of preset basic causal discovery algorithms respectively; wherein the directed edge is used to represent the causal relationship between different brain regions;

[0009] integrating the directed edges corresponding to each subset generated by the same basic causal discovery algorithm, and calculating the directed edge number distribution curve of each basic causal discovery algorithm and the corresponding coefficient of variation;

[0010] integrating the generation results of each basic causal discovery algorithm by taking the coefficient of variation as a weight, to obtain an initial causal graph with weights for each directed edge;

[0011] based on the weight of each directed edge in the initial causal graph and a preset confident threshold, selecting confident effective edges for the construction of the causal graph, to obtain a final causal network.

[0012] Further, the directed edges corresponding to each subset generated by the same basic causal discovery algorithm are integrated, specifically: the directed edges are generated based on the data in each subset by using the same basic causal discovery algorithm, and the number of directed edges is counted based on the directed edges generated in each subset.

[0013] Further, the calculation of the directed edge number distribution curve of each basic causal discovery algorithm and the corresponding coefficient of variation is specifically:

[0014] The abscissa of the directed edge number distribution curve represents the number of times that all different sub-data can obtain a certain directed edge, and the ordinate represents the total number of all different directed edges that can reach a certain number, so as to realize the construction of the directed edge number distribution curve.

[0015] Further, the generation results of each basic causal discovery algorithm are integrated by taking the variation coefficient as the weight, specifically: the weight of different algorithms is obtained by the variation coefficient value of different algorithms, the number of directed edges of each algorithm is multiplied by the weight corresponding to the algorithm, then the number of the same directed edges in all algorithms is added and averaged, and then integrated into a total causal graph, and the weight of each edge is the integer of the average number of directed edges.

[0016] Further, the confident and effective edges are selected based on the weight of each directed edge in the initial causal graph and the preset confidence threshold to construct the causal graph, specifically: a first threshold and a second threshold are preset, wherein the first threshold is less than the second threshold; the causal graph is constructed based on the directed edges greater than the second threshold in the initial causal graph, and the directed edges less than the first threshold in the initial causal graph are removed; at the same time, the remaining directed edges are used as candidate edges, and the candidate edges are added according to whether the directed acyclic graph condition is met after the candidate edges are added to the causal graph, until all candidate edges are traversed.

[0017] Further, the basic causal discovery algorithm includes but is not limited to Peter-Clark algorithm, non-temporal linear non-Gaussian model and GES causal discovery algorithm.

[0018] Further, the different brain regions include left prefrontal cortex, right prefrontal cortex, left temporal cortex, right temporal cortex, left occipital cortex and right occipital cortex.

[0019] According to a second aspect of the embodiment of the present application, a causal network construction system for human brain effective connection estimation is provided, comprising:

[0020] A data acquisition unit is configured to acquire brain fNIRS time series data to be analyzed, and divide the time series data into a plurality of subsets;

[0021] A directed edge generation unit is configured to generate directed edges based on a plurality of preset basic causal discovery algorithms for each obtained subset; wherein the directed edges represent the causal relationship between different brain regions.

[0022] A weight calculation unit is configured to integrate the directed edges corresponding to each subset generated by the same basic causal discovery algorithm, and calculate the directed edge number distribution curve of each basic causal discovery algorithm and the corresponding variation coefficient.

[0023] an integration unit configured to integrate the generated results of each basic causal discovery algorithm with the coefficient of variation as a weight to obtain an initial causal graph with weights of each directed edge;

[0024] a causal network generation unit configured to select confident and effective edges based on the weights of each directed edge in the initial causal graph and a preset confidence threshold to construct a final causal network.

[0025] According to a third aspect of the embodiments of the present disclosure, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and run on the memory, and the processor implements the method for causal network construction for brain effective connectivity estimation when executing the program.

[0026] According to a fourth aspect of the embodiments of the present disclosure, a non-transitory computer readable storage medium is provided, which stores a computer program, and the program is executed by a processor to implement the method for causal network construction for brain effective connectivity estimation.

[0027] The above one or more technical solutions have the following beneficial effects:

[0028] (1) The present application provides a method and system for causal network construction for brain effective connectivity estimation, which adopts a weighted integration of different causal discovery methods, measures the weights of different basic methods through a distribution metric function, and obtains a confidence edge set by adaptively selecting a critical threshold according to the generated directed edge confidence metric curve. Finally, the condition of directed acyclic graph is used to redefine the confident edge set and obtain the final causal graph. Through the above scheme, the high-confidence inference of different algorithms is integrated, thereby producing a high-reliability and high-precision weighted collective causal discovery scheme, which effectively ensures the accuracy of the causal relationship determination between brain regions and lays a foundation for subsequent brain effective connectivity estimation.

[0029] (2) On the simulated data set, the method has strong accuracy and strong robustness. On the real fNIRS data set, the method can better reflect the differences between different groups compared with the basic algorithm. The results show that the brain effective connectivity of AD patients and MCI patients is significantly different from that of the HC group under the same state. Through analysis, it is found that the disease may damage some brain regions and affect the effective connectivity between brain regions, which will be of great help to the study of the underlying mechanism of neurodegenerative diseases.

[0030] The advantages of the additional aspects of the present application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The embodiments of the application, and their

[0032] Figure 1 A basic flow chart of the causal network construction method for human brain effective connection estimation described in the embodiments of the application;

[0033] Figure 2 Causal diagrams generated using the WECD method on five simulated data described in the embodiments of the application;

[0034] Figure 3 Causal diagrams generated using the Nopart-WECD method on five simulated data described in the embodiments of the application

[0035] Figure 4 EC situation diagram of the three groups in the Stroop state described in the embodiments of the application;

[0036] Figure 5 EC situation diagram of the three groups in the Rest state described in the embodiments of the application;

[0037] Figure 6 A block diagram of the overall processing process of the causal network construction method for human brain effective connection estimation described in the embodiments of the application. DETAILED DESCRIPTION

[0038] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0039] It should be noted that the terms used herein are only intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the application.

[0040] In the case of no conflict, the embodiments in the application and the features in the embodiments can be combined with each other.

[0041] Embodiment one:

[0042] The purpose of this embodiment is to provide a causal network construction method for human brain effective connection estimation.

[0043] As shown in Figure 1 A causal network construction method for human brain effective connection estimation, comprising:

[0044] Obtaining brain fNIRS time series data to be analyzed, and dividing the time series data into several subsets;

[0045] For each subset obtained, a directed edge is generated based on a preset plurality of basic causal discovery algorithms respectively; wherein the directed edge is used to represent the causal relationship between different brain regions;

[0046] The directed edges corresponding to each subset generated by the same basic causal discovery algorithm are integrated, and the directed edge number distribution curve of each basic causal discovery algorithm and its corresponding coefficient of variation are calculated;

[0047] The generation results of each basic causal discovery algorithm are integrated by taking the coefficient of variation as the weight, and an initial causal graph with weights for each directed edge is obtained;

[0048] Based on the weight of each directed edge in the initial causal graph and a preset confidence threshold, a confident and effective edge is selected for the construction of the causal graph, and a final causal network is obtained.

[0049] In specific implementation, the directed edges corresponding to each subset generated by the same basic causal discovery algorithm are integrated, specifically: using the same basic causal discovery algorithm, the generation of directed edges is based on the data in each subset, and the number of directed edges is counted based on the directed edges generated in each subset.

[0050] In specific implementation, the calculation of the directed edge number distribution curve of each basic causal discovery algorithm and its corresponding coefficient of variation is specifically:

[0051] (1) The horizontal coordinate of the directed edge number distribution curve represents the number of times that the algorithm can obtain a certain directed edge for all different sub-data (a total of 10 times), and the vertical coordinate represents the total number of all different directed edges that can reach a certain number.

[0052] (2) The coefficient of variation is calculated by dividing the standard deviation by the mean to calculate the dispersion degree of the directed edge number distribution curve. The greater the dispersion degree, the greater the coefficient of variation value.

[0053] In specific implementation, the integration of the generation results of each basic causal discovery algorithm by taking the coefficient of variation as the weight is specifically:

[0054] The greater the coefficient of variation value, the more confident the algorithm is in the calculated directed edge. The weights of different algorithms are obtained by the coefficient of variation values of different algorithms. The number of directed edges of each algorithm is multiplied by the weight corresponding to the algorithm, and then the number of the same directed edges in all algorithms is added and averaged to integrate into a total causal graph. The weight of each edge is the integer of the average number of directed edges.

[0055] In specific implementation, the confident effective edges are selected based on the weight of each directed edge in the initial causal graph and a preset confidence threshold for the construction of the causal graph, specifically: a first threshold and a second threshold are preset, wherein the first threshold is less than the second threshold; the causal graph is constructed based on the directed edges in the initial causal graph that are greater than the second threshold, and the directed edges in the initial causal graph that are less than the first threshold are removed; meanwhile, the remaining directed edges are used as candidate edges, and the candidate edges are added according to whether the directed acyclic graph condition is met after the candidate edges are added, until all the candidate edges are traversed.

[0056] In specific implementation, the basic causal discovery algorithm includes but is not limited to Peter-Clark algorithm, non-temporal linear non-Gaussian model and GES causal discovery algorithm.

[0057] In specific implementation, the different brain regions include left prefrontal cortex, right prefrontal cortex, left temporal cortex, right temporal cortex, left occipital cortex and right occipital cortex.

[0058] For the convenience of understanding, the scheme described in the embodiment is described in detail as follows with reference to the accompanying drawings:

[0059] Exploring and explaining the EC (Effective Connectivity, effective connectivity) between brain regions can help us explore the mechanism behind neurodegenerative diseases such as Alzheimer's disease, thereby helping to better perform early diagnosis and treatment. The causal discovery method is a tool widely used in this field in recent years, but different methods have inconsistent results due to different internal mechanisms. To solve this problem, the scheme described in the embodiment generates a high-reliability and high-precision weighted ensemble causal discovery method (Weighted Ensemble Causal Discovery, WECD) by integrating high-confidence inferences of different algorithms. For the convenience, the scheme described in the embodiment is referred to as WECD. At the same time, the performance of the WECD method is verified on a simulated data set in the embodiment, and it is found that its ability is superior to existing methods. The WECD method is verified on a real fNIRS data set, which is divided into three groups: AD, MCI and HC (Healthy Control). By analyzing the differences in effective connectivity between groups, the neural mechanisms of each cognitive state at the population level can be better understood.

[0060] In the scheme described in the embodiment, first, a distribution metric function is defined to measure the weight of different basic methods, and the generated directed edge confidence metric curve is used; second, the critical threshold is adaptively selected to obtain the confidence edge set; finally, the condition of directed acyclic graph is used to redefine the confident edge set and obtain the final causal graph. Among them, the coefficient of variation is one of the distribution metric functions, and the distribution metric function here represents the coefficient of variation.

[0061] As Figure 1 shown, the scheme described in the embodiment mainly includes the following processing flow:

[0062] 1) Partition the data according to the original time sequence;

[0063] 2) Put the sub-data into different algorithms respectively, and weight the algorithms by calculating the coefficient of variation of the number of directed edges;

[0064] 3) Integrate all algorithms to obtain a causal graph with each directed edge having a weight;

[0065] 4) Set a confidence threshold and select confident and effective edges to construct a causal graph;

[0066] 5) Determine whether the conditions for a directed acyclic graph are met;

[0067] 6) If the conditions are met, continue to add candidate edges to the causal graph;

[0068] 7) If the conditions are not met, remove the edges that do not meet the conditions;

[0069] 8) Determine whether all candidate edges have been traversed;

[0070] 9) If there are remaining candidate edges, continue to add candidate edges to the causal graph;

[0071] 10) End.

[0072] In specific implementation, the partitioning of the data is specifically: dividing the data into multiple sub-data, increasing the diversity of the results of the same algorithm by using the similarity and difference within the data, thereby improving the stability and certainty of the algorithm. To avoid destroying the time sequence of the original data, the data is divided into N equal parts, rather than randomly sampled. This method ensures that the time sequence of the original data is preserved. When the same algorithm is used to calculate the causal relationship of the N sub-data, the results are slightly different due to the changes within the data. However, since these sub-data come from the same original data, the differences between the sub-data are small.

[0073] In a specific implementation, the weighting and integration is specifically: after calculating the sub-data respectively using the base causal discovery algorithm, record the directed edges generated by each algorithm. Integrate the N parts generated by the same algorithm to obtain the distribution curve of the number of directed edges of each algorithm. Estimate the distribution of each algorithm using the coefficient of variation to obtain a weighted causal graph with all directed edges. The distribution curve generated thereby provides self-confidence information about the causal relationship between two nodes for each algorithm. If an algorithm is very confident about the causal relationship between two nodes, its statistical results will appear upstream or downstream of the distribution curve. On the other hand, if an algorithm is uncertain about the causal relationship between two nodes, its statistical results will appear in the middle position of the distribution curve.

[0074] In a specific implementation, the self-confidence threshold setting is specifically: the ideal distribution curve of a stable algorithm is "U" shaped, indicating that all directed edges are confident. However, the base algorithm is often unstable, and the difference in data will lead to uncertain directed edges. In addition, existing directed edges are not necessarily true directed edges, and uncertain directed edges can be true directed edges. In order to solve these problems, the present application introduces two weights, so that the method can adaptively adjust the threshold. The first threshold λ1 is set to remove confident invalid edges, and the second threshold λ2 is set to retain confident valid edges.

[0075] In a specific implementation, the directed acyclic graph condition is specifically: once the confident edges are determined, an initial causal graph is created. The remaining uncertain edges are classified as candidate edges. The candidate edges are evaluated one by one, and if they meet the condition of directed acyclic graph, they are added to the causal graph as new confident edges. However, if a candidate edge cannot meet the condition, it will be excluded from the candidate edge list. Repeat this procedure for all selected edges until the final causal is obtained.

[0076] In order to prove the effectiveness of the scheme described in this embodiment, this embodiment carries out relevant experimental verification:

[0077] This embodiment verifies the effectiveness of WECD on two data sets, including a simulated data set with true connectivity and a real fNIRS data set collected at the Third People's Hospital of Guangxi Zhuang Autonomous Region.

[0078] (1) Simulated data set: Use the data generator of the causal discovery toolbox (CDT) to generate a non-cyclic graph based on the causal mechanism and the corresponding data. Since the real fNIRS data consists of about 5000 sampling points, and mainly analyzes the causal relationship between six brain regions, the simulated data also contains 5000 data points and six nodes. Make it approximate to the properties of real brain data.

[0079] (2) Real fNIRS dataset: fNIRS data was obtained using a desktop near-infrared acquisition instrument with a sampling frequency of 17 Hz and laser light sources with wavelengths of 740 nm and 820 nm. This device recorded continuous wave changes in 61 fNIRS channels, including six brain regions: left prefrontal cortex (LPFC), right prefrontal cortex (RPFC), left temporal cortex (LTC), right temporal cortex (RTC), left occipital cortex (LOC), and right occipital cortex (ROC).

[0080] fNIRS data from 25 subjects in Stroop and Rest states was collected. The Stroop state is a cognitive assessment paradigm that involves subjects making color judgments of text, where the meaning of the text is a certain color but does not necessarily match the color of the text. The Rest state refers to the quiet state of the subject without tasks.

[0081] During these two processes, changes in blood oxygen in the brain were collected, and the fNIRS signal was further processed. First, the light intensity signal was converted to HbO and HbR. Then, physiological noise, conversion drift, high-frequency noise, and motion artifacts were removed. Finally, blood oxygen signals between 0.021 and 2.00 Hz were obtained. HbO was selected from the 61 channels of the six brain regions to estimate the brain EC.

[0082] In the specific experiment of the embodiment, N is the number of subsets of data partitions; λ1 is the threshold value for defining invalid edges, and the directed edges counted as N* λ1 are excluded; λ2 is the threshold value for defining valid edges, and the directed edges counted as N* λ2 are retained. E is defined as the directed edges, where ETrue represents the ground-truth directed edges, and Etarget represents the directed edges of the target causal graph. After multiple repetitions of the data, it was found that setting the number of data partitions N to 10 works best. In the data partitioning method, the relative positions and quantities between the data sampling points are not changed, and the division is carried out in the original order to ensure that the time sequence within the sub-data is unchanged. If WECD is applied to the estimation of the brain EC, the nodes and pos parameters on the simulation data are also set according to the brain. First, referring to the real fNIRS data, the 61 brain channels are re-divided into 6 brain regions as nodes, and the positions of the nodes are determined by the distribution of the 6 brain regions in the brain. We use the original known position coordinates of the 61 channels to get the position of each node. We train the model on the simulation data, adjust the parameters of λ1 and λ2, and set λ1 = 0.2 and λ2 = 0.3 according to the gap between the observed target graph and the ground-truth. Under the premise of eliminating non-existent directed edges, the weights of all directed edges are sorted, λ1 is used to determine the invalid edges in the confident directed edges, and λ2 is used to determine the valid edges. The remaining part is used as a candidate edge to determine whether it needs to be added to the final causal graph.

[0083] To verify the experimental effect of the WECD method, six existing methods are selected as comparison methods, including PC (Peter-Clark algorithm), LiNGAM (non-temporal linear non-Gaussian model), GES (optimal structure identification with greedy search), CAM (causal additive models), CCDr (concave penalized coordinate descent with reparameterization) and GIES (greedy interventional equivalence search). The WECD method is compared with these basic algorithms, and the WECD method integrates the self-confidence part of the basic algorithm, and the result is closer to the real causal graph.

[0084] The experimental results on the simulation data are summarized, and the mean and variance are calculated, as shown in Table 1. From the data in the table, in the results of each evaluation index, except for the recall rate, the WECD method is outstanding in both the mean and the variance among the seven algorithms. This test shows that the basic method is often not accurate and stable when estimating causal relationships, and the WECD method just makes up for this deficiency.

[0085] Table 1 Comparison of experimental results

[0086]

[0087]

[0088] To better reflect the superiority of the WECD method, the following ablation experiments are performed: the method with missing data partition, and the method based on the majority voting integration strategy.

[0089] The result is to compare the real causal graph of the simulation data and the causal graph after adding the candidate edge. For convenience of observation, the target causal graph is fused with the ground-truth, and the correct edge obtained from the target causal graph, the correct edge not obtained and the error edge appearing are represented by different gray scales, and the color gray scale gradually increases in order, as shown in Figure 2 .

[0090] Putting all the data into the algorithm to calculate the causal relationship cannot make the algorithm have diversified results, and the most certain result is selected from it. According to this idea, we removed the data segmentation step in the method, and instead put the whole data repeatedly into it to get the causal graph of the method on the five simulation data. From Figure 3It can be seen that the target causal graph obtained by the method without data segmentation has a large gap with the ground-truth, and the number of wrong directed edges and missed directed edges is increased compared with WECD.

[0091] The most common integration strategy is majority voting (MV), that is, voting on the results of all base algorithms, and if more than half of the votes, it will be added to the integrated result. In this method, the data integration stage and the algorithm integration stage vote on the number of directed edges, and if the number of votes is more than half, it means that this directed edge exists, and finally the final integrated result is obtained.

[0092] The final ablation experiment results are shown in Table 2.

[0093] Table 2 Ablation experiment results

[0094]

[0095]

[0096] The WECD method achieved good results on the simulation data, and it was applied to fNIRS data to estimate the effective connection on the brain. The fNIRS data of 25 subjects were preprocessed, including two states: Stroop state and Root state. They were divided into AD group, MCI group and control group, and the WECD method was applied at the group level to observe the differences between groups.

[0097] As shown in Figure 4 , the effective connection of the three groups in the Stroop state, from the comparison, it can be seen that the AD patients only have LPFC→RTC and LPFC→RPFC connections; MCI patients have connections between LPFC and other brain regions in addition to the connections of AD, and also increase LTC→RTC and LOC→ROC, which are left brain connections to right brain; and HC group increases LTC→LOC and LOC→RTC, which are temporal lobe and left occipital lobe connections, and RPFC→RTC and RPFC→LTC, which are right frontal lobe connections to temporal lobe. The above results show that as the disease worsens, the temporal lobe, frontal lobe and occipital lobe of the brain are damaged to varying degrees, among which the right frontal lobe, left temporal lobe and left occipital lobe are most severely damaged in terms of signal transmission.

[0098] As shown in Figure 5As shown, three groups of effect connection in Root state, compared with Stroop state, AD patients increased LPFC→LTC connection in Root state; MCI patients lack of LTC→RTC, LOC→ROC left brain to right brain connection; HC group lack of RPFC→PTC connection. The above results show that Stroop task mainly requires deployment of frontal lobe, but needs the assistance of left temporal lobe and left occipital lobe, and because AD has serious loss in right frontal lobe, there is also a compensatory effect of left frontal lobe.

[0099] Embodiment Two:

[0100] The purpose of the embodiment is to provide a causal network construction system for human brain effect connection estimation.

[0101] A causal network construction system for human brain effect connection estimation, comprising:

[0102] A data acquisition unit for acquiring brain fNIRS time series data to be analyzed and dividing the time series data into several subsets;

[0103] A directed edge generation unit for generating directed edges based on a plurality of pre-set basic causal discovery algorithms for each obtained subset; wherein the directed edges represent the causal relationship between different brain regions;

[0104] A weight calculation unit for integrating the directed edges corresponding to each subset generated by the same basic causal discovery algorithm and calculating the directed edge number distribution curve of each basic causal discovery algorithm and its corresponding coefficient of variation;

[0105] An integration unit for integrating the generation results of each basic causal discovery algorithm with the coefficient of variation as the weight to obtain an initial causal graph with weights for each directed edge;

[0106] A causal network generation unit for selecting confident and effective edges based on the weight of each directed edge in the initial causal graph and a pre-set confidence threshold to construct a causal graph and obtain a final causal network.

[0107] Further, the system of the embodiment corresponds to the method of embodiment one, and the technical details have been described in detail in embodiment one, so here is not repeated.

[0108] In more embodiments, there is also provided:

[0109] An electronic device comprising a memory and a processor, and computer instructions stored on the memory and running on the processor, when the computer instructions are run by the processor, the method described in embodiment one is completed. For brevity, this is not repeated here.

[0110] It should be understood that the processor in the embodiments can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0111] The memory can include read-only memory and random access memory, and provide instructions and data to the processor, and a portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0112] A computer readable storage medium is used to store computer instructions, which are executed by a processor to complete the method described in Embodiment I.

[0113] The method in Embodiment I can be directly embodied as hardware processor execution completion, or executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. The storage medium is located in the memory, and the processor reads the information in the memory, and combines the hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0114] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software mode depends on the specific application of the technical solution and the design constraints. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present disclosure.

[0115] The causal network construction method and system for human brain effective connection estimation provided by the above embodiments can be realized, and has a broad application prospect.

[0116] The above only describes the preferred embodiments of the present disclosure and is not intended to limit the present disclosure. Those skilled in the art can make various modifications and changes to the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for constructing causal networks for estimating human brain effect connectivity, characterized in that, include: Obtain the brain fNIRS time series data to be analyzed, and divide the time series data into several subsets; For each obtained subset, directed edges are generated based on multiple pre-defined basic causal discovery algorithms; wherein, the directed edges are used to represent causal relationships between different brain regions; The directed edges corresponding to each subset generated by the same basic causal discovery algorithm are integrated, and the distribution curve of the number of directed edges and its corresponding coefficient of variation for each basic causal discovery algorithm are calculated. The generation results of each basic causal discovery algorithm are integrated using the coefficient of variation as the weight to obtain an initial causal graph with weights for each directed edge. Based on the weight of each directed edge in the initial causal graph and a preset confidence threshold, the edges with valid confidence are selected to construct the causal graph and obtain the final causal network. The integration of the generation results of each basic causal discovery algorithm using the coefficient of variation as the weight is specifically as follows: the weight of each algorithm is obtained by using the coefficient of variation of different algorithms, the number of directed edges of each algorithm is multiplied by the weight of the corresponding algorithm, and then the number of the same directed edges in all algorithms is added together and averaged to form a total causal graph, and the weight of each edge is the integer part of the average number of directed edges. The method of selecting confident and valid edges to construct the causal graph based on the weight of each directed edge in the initial causal graph and a preset confidence threshold is as follows: a first threshold and a second threshold are preset, wherein the first threshold is less than the second threshold; the causal graph is constructed based on the directed edges in the initial causal graph that are greater than the second threshold, and the directed edges in the initial causal graph that are less than the first threshold are removed; at the same time, the remaining directed edges are used as candidate edges, and candidate edges are added according to whether the causal graph satisfies the condition of a directed acyclic graph after adding candidate edges, until all candidate edges have been traversed. The basic causal discovery algorithms are specifically: the Peter-Clark algorithm, the non-temporal linear non-Gaussian model, and the GES causal discovery algorithm.

2. The method for constructing a causal network for estimating human brain effect connections as described in claim 1, characterized in that, The directed edges corresponding to each subset generated by the same basic causal discovery algorithm are integrated. Specifically, the same basic causal discovery algorithm is used to generate directed edges based on the data in each subset, and the number of directed edges generated in each subset is counted.

3. The method for constructing a causal network for estimating human brain effect connections as described in claim 1, characterized in that, The calculation of the directed edge number distribution curve and its corresponding coefficient of variation for each basic causal discovery algorithm is specifically as follows: The x-axis of the directed edge number distribution curve represents the number of times a certain directed edge can be obtained by all different sub-data, and the y-axis represents the total number of times that a certain directed edge can be obtained by all different directed edges, thus realizing the construction of the directed edge number distribution curve; the dispersion of the directed edge number distribution curve is calculated by dividing the standard deviation by the mean, which is used as the coefficient of variation.

4. The method for constructing a causal network for estimating human brain effect connections as described in claim 1, characterized in that, The different brain regions include the left prefrontal cortex, right prefrontal cortex, left temporal cortex, right temporal cortex, left occipital cortex, and right occipital cortex.

5. A causal network construction system for estimating human brain effect connectivity, characterized in that, include: The data acquisition unit is used to acquire brain fNIRS time series data to be analyzed and divide the time series data into several subsets; A directed edge generation unit is used to generate directed edges for each obtained subset based on multiple pre-defined basic causal discovery algorithms; wherein the directed edges are used to represent causal relationships between different brain regions; The weight calculation unit is used to integrate the directed edges corresponding to each subset generated by the same basic causal discovery algorithm, and to calculate the distribution curve of the number of directed edges and its corresponding coefficient of variation for each basic causal discovery algorithm. An integration unit is used to integrate the generation results of each basic causal discovery algorithm with the coefficient of variation as the weight to obtain an initial causal graph with weights for each directed edge. The causal network generation unit is used to construct the causal graph by selecting confident and valid edges based on the weight of each directed edge in the initial causal graph and a preset confidence threshold, thereby obtaining the final causal network. The integration of the generation results of each basic causal discovery algorithm using the coefficient of variation as the weight is specifically as follows: the weight of each algorithm is obtained by using the coefficient of variation of different algorithms, the number of directed edges of each algorithm is multiplied by the weight of the corresponding algorithm, and then the number of the same directed edges in all algorithms is added together and averaged to form a total causal graph, and the weight of each edge is the integer part of the average number of directed edges. The method of selecting confident and valid edges to construct the causal graph based on the weight of each directed edge in the initial causal graph and a preset confidence threshold is as follows: a first threshold and a second threshold are preset, wherein the first threshold is less than the second threshold; the causal graph is constructed based on the directed edges in the initial causal graph that are greater than the second threshold, and the directed edges in the initial causal graph that are less than the first threshold are removed; at the same time, the remaining directed edges are used as candidate edges, and candidate edges are added according to whether the causal graph satisfies the condition of a directed acyclic graph after adding candidate edges, until all candidate edges have been traversed. The basic causal discovery algorithms are specifically: the Peter-Clark algorithm, the non-temporal linear non-Gaussian model, and the GES causal discovery algorithm.

6. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and running thereon, characterized in that, When the processor executes the program, it implements a causal network construction method for estimating human brain effect connectivity as described in any one of claims 1-4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements a causal network construction method for estimating human brain effect connectivity as described in any one of claims 1-4.

Images