Method, device and storage medium for constructing dynamic causal model for fMRI data analysis
Through the GreedyEC and GreedyROI algorithm combined with Bayesian model, a dynamic causal model was constructed, which solved the problem that a large amount of prior knowledge was required to construct a dynamic causal model, and the optimal model construction under a small amount of prior knowledge was realized, which reduced the calculation. Cost and improve analysis efficiency.
Patent Information
- Application Number
- CN202111238951.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-25
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2041-10-25
AI Technical Summary
Existing dynamic causal models (DCMs) require a lot of prior knowledge when constructing causal analysis of brain networks, and hypotheses-driven methods are difficult to find the optimal model, resulting in high computational costs and limited widespread application.
The GreedyEC algorithm with gradually decreasing effect connections and the GreedyROI algorithm with gradually increasing effect connections combined with Bayesian model to construct a dynamic causal model for fMRI data analysis. Through model space search and Bayesian model comparison, we can find the global and overall optimal model, and reduce it. Reliance on prior knowledge.
It realizes the construction of the optimal dynamic causal model for fMRI data analysis with a small amount of prior knowledge, which lowers the threshold for DCM analysis, and improves the computing efficiency and model reliability.
Smart Images
Figure CN114119478B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical image processing technology, and in particular to a method, device and storage medium for constructing a dynamic causal model for fMRI data analysis. Background Art
[0002] The study of brain activity mechanisms is of great significance for understanding the pathogenesis of various mental illnesses and exploring the cognitive processes of the human brain. Currently, there are two commonly used hypotheses in studying brain function: the functional separation hypothesis and the functional integration hypothesis.
[0003] Among them, the functional integration hypothesis focuses more on the interaction between different brain regions. It believes that human cognitive processes involve interactions between different brain regions, emphasizing cooperation rather than independence. The functional integration hypothesis divides the connections between brain regions into structural connections, functional connections, and effector connections. Effector connections can reflect the direct causal relationship between brain regions and are of great significance for brain function analysis.
[0004] The mainstream effective connectivity analysis methods are Granger causality analysis (GCA) and dynamic causal modeling (DCM). The DCM method simulates neural activity and hemodynamic state at two levels, respectively, and converts neural activity in brain regions into observed functional magnetic resonance imaging (fMRI) signals (Friston et al., 2003). DCM restores the entire process from neural activity to brain BOLD signals, so it has high reliability. However, due to the complexity of physiological process modeling, the computational cost is greatly increased, and it is difficult to generalize to the causal analysis of large-scale brain networks. In addition, DCM is a commonly used effective connectivity (EC) estimation tool in neural image analysis. It is a model-driven analysis method that first needs to propose a hypothetical causal network model, and then obtain the optimal model by comparing different hypothetical models. However, in many cases, our prior knowledge is limited, or there is no clear hypothesis. Therefore, it is difficult to propose a set of hypothetical models. If all possible models are listed, then as the number of nodes increases, the number of competing models will grow exponentially, resulting in an excessively large model space and high computational cost.
[0005] Therefore, DCM has an important advantage over other analytical methods: it can simulate physiological processes with higher reliability. However, early methods for analyzing brain networks using DCM were mainly hypothesis-driven. Proposing hypotheses requires a lot of prior knowledge, and alternative models may not be included in the model space. These two characteristics limit the widespread application of DCM. Summary of the invention
[0006] The present invention provides a method, device and storage medium for constructing a dynamic causal model for fMRI data analysis, so as to solve the problem that the existing hypothesis-based dynamic causal model applied to fMRI data requires a lot of prior knowledge and is difficult to find an optimal model.
[0007] In a first aspect, a method for constructing a dynamic causal model for fMRI data analysis is provided, comprising:
[0008] Acquire brain fMRI image data and perform preprocessing;
[0009] Selecting regions of interest from preprocessed brain fMRI image data;
[0010] The GreedyEC algorithm with gradually decreasing effect connections is used to search the model space of the constructed dynamic causal model for fMRI data analysis, and the first global optimal dynamic causal model for fMRI data analysis is obtained;
[0011] The GreedyROI algorithm with gradually increasing effect connections was used to search the model space of the constructed dynamic causal model for fMRI data analysis, and the second global optimal dynamic causal model for fMRI data analysis was obtained.
[0012] The Bayesian model was used to compare the first global optimal fMRI data analysis dynamic causal model and the second global optimal fMRI data analysis dynamic causal model to obtain the overall optimal fMRI data analysis dynamic causal model.
[0013] The above scheme builds the corresponding global optimal fMRI data analysis dynamic causal model based on two algorithms. The GreedyEC algorithm searches the model space based on the strategy of gradually decreasing effect connections to find the first global optimal fMRI data analysis dynamic causal model; the GreedyROI algorithm searches the model space based on the strategy of gradually increasing effect connections to find the second global optimal fMRI data analysis dynamic causal model; finally, the Bayesian model comparison is used to select the overall optimal fMRI data analysis dynamic causal model. This scheme can build the optimal fMRI data analysis dynamic causal model under the premise of only a small amount of prior knowledge.
[0014] Furthermore, preprocessing of brain fMRI image data includes:
[0015] Head motion correction: Using the 3D brain function image at the first moment as a reference, perform rigid body transformation on the 3D brain function images at other moments to align them;
[0016] Spatial normalization: align the brain structural image to the brain functional image obtained after head motion correction, and use the template to segment the brain structural image; according to the gray matter cortex in the segmentation result, align the brain functional image and brain structural image to the standard brain template;
[0017] Spatial smoothing: Gaussian smoothing is performed on the spatially normalized image.
[0018] Furthermore, the head motion correction process also includes: only retaining the three-dimensional brain function image whose translation movement does not exceed a preset distance and whose rotation angle does not exceed a preset angle.
[0019] Furthermore, the number of regions of interest selected from the preprocessed brain fMRI image data does not exceed 8.
[0020] Furthermore, the GreedyEC algorithm with gradually decreasing effect connections is used to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the first global optimal fMRI data analysis dynamic causal model, which specifically includes:
[0021] A1: Number the bidirectional effect connections between all regions of interest ROIs, and define the fMRI data analysis dynamic causal model based on the existence of bidirectional effect connections between all regions of interest ROIs as the current optimal model;
[0022] A2: Consider the effect connection No. 1 and set i = 1;
[0023] A3: Close the i-th effect connection based on the current optimal model and construct the current candidate model;
[0024] A4: Use Bayesian model selection methods to compare the model evidence between the current best model and the current candidate models;
[0025] A5: If the model evidence of the current optimal model is greater than that of the current candidate model, the i-th effect connection is enabled in all subsequent candidate models; otherwise, the current candidate model is used as the current optimal model, and the i-th effect connection is disabled in all subsequent candidate models.
[0026] A6: Let i=i+1, and repeat steps A3-A6 until the n*(n-1)th efficient connection is considered, where n is the number of selected regions of interest;
[0027] A7: The current optimal model finally obtained is the first global optimal fMRI data analysis dynamic causal model.
[0028] GreedyEC algorithm starts from the perspective of effect connection, and the initial model is set as a fully connected model. The effect connections are numbered, and the contribution of each effect connection to the model is considered in turn; if closing the effect connection reduces the model quality, it will be kept open in the subsequent steps of the algorithm; otherwise, the effect connection will be kept closed in the subsequent steps of the algorithm.
[0029] Furthermore, the GreedyROI algorithm with gradually increasing effect connections is used to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the second global optimal fMRI data analysis dynamic causal model, which specifically includes:
[0030] B1: Number a total of n regions of interest ROIs, k represents the number of regions of interest ROIs considered in the current stage;
[0031] B2: Define the model space composed of the effect connection between the first region of interest ROI and the second region of interest ROI, and construct 2 2 Each fMRI data analysis dynamic causal model was used as an alternative model;
[0032] B3: Use the Bayesian model selection method to compare all candidate models in the model space, take the optimal model as the 2ROI optimal model, and set k = 2;
[0033] B4: Fix the kROI optimal model obtained in the previous step, add the k+1th ROI as a new node, and there are 2*k possible effect connections between the new node and the original ROI;
[0034] B5: Define the model space composed of the effect connections between the ROIs 1 to k+1, keep the original kROI optimal model structure unchanged, and only change the state of the effect connection with the new node as the vertex, and construct a total of 2 2*k Each fMRI data analysis dynamic causal model was used as an alternative model;
[0035] B6: Use the Bayesian model selection method to compare all candidate models in the model space and take the optimal model as the (k+1)ROI optimal model;
[0036] B7: Let k=k+1, and repeat steps B4 to B7 until the nth region of interest ROI is considered. The final nROI optimal model is the second global optimal fMRI data analysis dynamic causal model.
[0037] The GreedyROI algorithm starts from the perspective of the region of interest and first searches for the optimal model of two regions of interest. Then, based on the optimal model, it gradually increases the region of interest to search a larger model space until the optimal model that takes all regions of interest into consideration is found.
[0038] Furthermore, it also includes:
[0039] Acquire brain fMRI image data of multiple subjects, and use multiple overall optimal fMRI data analysis dynamic causal models obtained based on the acquired data as candidate models;
[0040] The group Bayes factor was used to perform group-level analysis on multiple alternative models and select the optimal model for the group.
[0041] In order to be suitable for the exploratory analysis of group brain fMRI image data, multiple overall optimal fMRI data analysis dynamic causal models corresponding to multiple subjects were constructed, and the group Bayes factor was used to select the overall optimal fMRI data analysis dynamic causal model with the highest total fit as the group optimal model, so as to achieve exploratory analysis suitable for group brain fMRI image data.
[0042] Furthermore, the use of the group Bayesian factor to perform group-level analysis on multiple candidate models to select the optimal group model specifically includes:
[0043] Calculate the group Bayes factor when each candidate model matches the brain fMRI image data of all subjects, and select the candidate model with the highest total score as the group optimal model;
[0044] The optimal group model is matched to the brain fMRI image data of all subjects, and the Bayesian parameters of the model are averaged to obtain the dynamic causal model for fMRI data analysis that ultimately represents the group effect.
[0045] In a second aspect, a device for constructing a dynamic causal model for fMRI data analysis is provided, comprising:
[0046] A preprocessing module, used to obtain brain fMRI image data and perform preprocessing;
[0047] A region of interest selection module is used to select a region of interest from the preprocessed brain fMRI image data;
[0048] The first model building module is used to use the GreedyEC algorithm with gradually decreasing effect connections to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the first global optimal fMRI data analysis dynamic causal model;
[0049] The second model building module is used to use the GreedyROI algorithm with gradually increasing effect connections to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the second global optimal fMRI data analysis dynamic causal model;
[0050] The overall optimal model building module is used to use the Bayesian model to compare the first global optimal fMRI data analysis dynamic causal model and the second global optimal fMRI data analysis dynamic causal model to obtain the overall optimal fMRI data analysis dynamic causal model.
[0051] In a third aspect, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the method for constructing a dynamic causal model for fMRI data analysis as described above is implemented.
[0052] Beneficial effects:
[0053] The present invention proposes a method, device and storage medium for constructing a dynamic causal model for fMRI data analysis, which has the following advantages:
[0054] (1) The present invention proposes two algorithms for constructing a dynamic causal model for fMRI data analysis based on reducing the model space. Only a small amount of prior knowledge is needed to construct the most suitable dynamic causal model for fMRI data analysis, so as to mine useful information from the data with less prior knowledge and lower the threshold of DCM analysis.
[0055] (2) The GreedyEC algorithm takes into account the numbering problem of effect connections in the greedy strategy, providing greater freedom for analysis. The numbering can be adjusted according to the research purpose to adapt to different data. At the same time, it takes into account all effect connections, which reduces the estimation error caused by the numbering problem to a certain extent.
[0056] (3) GreedyROI algorithm is also a greedy strategy from the perspective of region of interest. It provides a new perspective for model space search. It starts from a single node model and adds nodes of the region of interest, gradually increasing the effect connections, and gradually growing a complex model from a simple model. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0058] Figure 1It is a flow chart of a method for constructing a dynamic causal model for fMRI data analysis provided by an embodiment of the present invention;
[0059] Figure 2 is a flowchart of GreedyEC taking 5 ROIs as an example provided by an embodiment of the present invention;
[0060] Figure 3 is a schematic diagram of a GreedyROI process using five ROIs as an example provided by an embodiment of the present invention;
[0061] Figure 4 It is a schematic diagram of an effect connection network provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0062] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.
[0063] Early methods of using dynamic causal models (DCM) to analyze brain networks were mainly hypothesis-driven. Proposing hypotheses requires a lot of prior knowledge, and alternative models may not be included in the model space. That is, under the premise of clarifying the effect connection relationship between the region of interest ROI and each region of interest ROI, constructing a corresponding dynamic causal model is a prior art. The difficulty lies in how to construct the optimal dynamic causal model with a small amount of prior knowledge. Based on this, the present invention provides the following embodiments.
[0064] Example 1
[0065] like Figure 1 As shown, this embodiment provides a method for constructing a dynamic causal model for fMRI data analysis, including:
[0066] S1: Acquire brain fMRI image data and perform preprocessing. Specifically, the preprocessing process includes:
[0067] S11: Head motion correction: Taking the three-dimensional brain function image at the first moment as the reference, perform rigid body transformation on the three-dimensional brain function images at other moments to align them; only retain the three-dimensional brain function images whose translation motion does not exceed the preset distance and whose rotation angle does not exceed the preset angle. The preset distance and preset angle can be selected according to actual conditions. For example, in this embodiment, the preset distance is set to 2.0 mm and the preset angle is set to 2.0 degrees.
[0068] S12: Spatial normalization: align the brain structure image to the brain function image obtained after head motion correction, and use the template to segment the brain structure image; according to the gray matter cortex in the segmentation result, align the brain function image and the brain structure image to the standard brain template. Commonly used brain templates include MNI template and Brodmann template. In this embodiment, the brain template uses the MNI template.
[0069] S13: Spatial smoothing: Gaussian smoothing is performed on the spatially normalized image. The Gaussian kernel size during the Gaussian smoothing process is twice the size of the voxel of the brain functional image. In this embodiment, a Gaussian kernel of 6×6×6 mm is used for smoothing.
[0070] S2: Select regions of interest (ROIs) from the preprocessed brain fMRI image data. ROIs are all the brain sub-regions of interest in the brain network being studied. The selection of ROIs has a lot of freedom and should be carefully selected according to different data types and research purposes. The statistical parameter mapping software package SPM can be used to obtain spheres and cubes of ROI centered on custom coordinates. You can also directly use activation clusters and custom templates. At the same time, the number of ROIs should not exceed 8, which is an inherent limitation of the DCM model itself.
[0071] like Figure 4 As shown, in this embodiment, ROIs are defined by combining brain activation areas and prior visual pathway areas with specific coordinates as the center of the sphere. A total of five ROIs are defined, including the primary visual cortex (V1 area), visual area 4 (V4 area), fusiform gyrus (FG area), intraparietal sulcus (IPS area) and inferior temporal gyrus (IT area). Figure 4 Shown is the final effector connection network.
[0072] S3: Use the GreedyEC algorithm with gradually decreasing effect connections to search the model space of the constructed dynamic causal model for fMRI data analysis, and obtain the first global optimal dynamic causal model for fMRI data analysis. Specifically, it includes the following steps:
[0073] Defining "open" an effect connection means that the value of the effect connection is significantly different from zero after data fitting in the DCM model. "Close" an effect connection means that the effect connection does not exist in the DCM model regardless of the result of data fitting; define a fully connected model, that is, a model in which bidirectional effect connections exist between all regions of interest ROIs.
[0074] S31: Numbering the bidirectional effect connections between all regions of interest ROIs can be done by constructing an effect connection matrix, with each row and column representing a different region of interest ROI, and the elements on the diagonal representing the self-connections of the region of interest, and numbering all the effect connections according to the effect connection matrix. In this embodiment, a 5×5 effect connection matrix is numbered, and the fully connected model is a DCM model with 25 effect connections. In this embodiment, it is assumed that all self-connections exist, and only non-self-connected effect connections are considered. First, the fully connected model is defined as the current optimal model, that is, the fMRI data analysis dynamic causal model constructed based on the existence of bidirectional effect connections between all regions of interest ROIs is defined as the current optimal model.
[0075] S32: Consider the effect connection No. 1 and set i=1.
[0076] S33: based on the current optimal model, the i-th effect connection is closed. In this embodiment, the effect connection from V1 to V4 is first closed to construct the current candidate model.
[0077] S34: Use Bayesian Model Selection (BMS) method to compare the model evidence of the current best model and the current alternative models.
[0078] S35: If the model evidence of the current optimal model is greater than that of the current candidate model, the i-th effect connection is enabled in all subsequent candidate models; otherwise, the current candidate model is used as the current optimal model, and the i-th effect connection is disabled in all subsequent candidate models. In this embodiment, the current candidate model constructed when the effect connection from V1 to V4 is disabled is inferior to the current optimal model after Bayesian model comparison, so the effect connection is enabled in all subsequent candidate models.
[0079] S36: Let i=i+1, and repeat steps S33-S36 until the 20th efficiency connection is considered.
[0080] S37: The current optimal model finally obtained is the first global optimal fMRI data analysis dynamic causal model.
[0081] like Figure 2The figure shows the flowchart of GreedyEC algorithm with 5 regions of interest (ROI) as an example, where (A) represents the fully connected model, the edge with a one-way arrow represents the effect connection, and the number next to it represents the corresponding effect connection number. (B) is the flowchart of GreedyEC algorithm. The EC diagram with an arrow at the end represents the effect connection mode of the alternative model when the algorithm proceeds to the current step. The model space is represented by a matrix, with columns representing different effect connection states and rows representing different alternative models. The black cells in the matrix indicate that the effect connection in the model is turned on, and the white cells indicate that the effect connection in the model is turned off.
[0082] GreedyEC algorithm starts from the perspective of effect connection, and the initial model is set as a fully connected model. The effect connections are numbered, and the contribution of each effect connection to the model is considered in turn; if closing the effect connection reduces the model quality, it will be kept open in the subsequent steps of the algorithm; otherwise, the effect connection will be kept closed in the subsequent steps of the algorithm.
[0083] S4: Use the GreedyROI algorithm with gradually increasing effect connections to search the model space of the constructed dynamic causal model for fMRI data analysis, and obtain the second global optimal dynamic causal model for fMRI data analysis. Specifically include:
[0084] S41: numbering a total of n regions of interest ROIs. In this embodiment, n is 5, and k represents the number of regions of interest ROIs considered in the current stage;
[0085] S42: Define the model space composed of the effect connection between the first region of interest ROI and the second region of interest ROI, and construct 2 2 Each fMRI data analysis dynamic causal model was used as an alternative model;
[0086] S43: Use the Bayesian model selection method to compare all candidate models in the model space, take the optimal model as the 2ROI optimal model, and set k=2;
[0087] S44: fix the kROI optimal model obtained in the previous step, add the k+1th region of interest ROI as a new node, and the possible effect connections between the new node and the original region of interest ROI are 2*k in total;
[0088] S45: Define the model space composed of the effect connections between the ROIs 1 to k+1, keep the original kROI optimal model structure unchanged, and only change the state of the effect connection with the new node as the vertex (open or close), and construct a total of 2 2*k Each fMRI data analysis dynamic causal model was used as an alternative model;
[0089] S46: Use the Bayesian model selection method to compare all candidate models in the model space, and take the best model as the (k+1)ROI optimal model;
[0090] S47: Let k=k+1, and repeat steps S4 to S47 until the nth region of interest ROI is considered, and the optimal model of n ROIs finally obtained is the second global optimal dynamic causal model for fMRI data analysis.
[0091] like Figure 3 The figure shows a schematic diagram of the GreedyROI algorithm flow chart using five regions of interest (ROIs) as an example. The dotted directed edges represent effect connections whose states have not yet been determined. The black directed edges represent effect connections whose states are turned on. The model space is represented by a matrix, with columns representing different effect connections and rows representing different competitive models. Black indicates that an effective connection in the model is turned on, and white indicates that it is turned off.
[0092] The GreedyROI algorithm starts from the perspective of the region of interest and first searches for the optimal model of two regions of interest. Then, based on the optimal model, it gradually increases the region of interest to search a larger model space until the optimal model that takes all regions of interest into consideration is found.
[0093] S5: Use the Bayesian model to compare the first global optimal fMRI data analysis dynamic causal model and the second global optimal fMRI data analysis dynamic causal model to obtain the overall optimal fMRI data analysis dynamic causal model.
[0094] The above scheme constructs the corresponding global optimal fMRI data analysis dynamic causal model based on two algorithms. The GreedyEC algorithm searches the model space based on the strategy of gradually decreasing effect connections to find the first global optimal fMRI data analysis dynamic causal model; the GreedyROI algorithm searches the model space based on the strategy of gradually increasing effect connections to find the second global optimal fMRI data analysis dynamic causal model; finally, the Bayesian model comparison is used to select the overall optimal fMRI data analysis dynamic causal model. This scheme can construct the optimal fMRI data analysis dynamic causal model under the premise of only a small amount of prior knowledge. It should be noted that the use of the Bayesian model selection method for model comparison is an existing method, and its specific process will not be repeated here.
[0095] Example 2
[0096] In order to be applicable to the exploratory analysis of group brain fMRI image data, this embodiment provides a method for constructing a dynamic causal model for fMRI data analysis, which includes the following steps:
[0097] Acquire brain fMRI image data of multiple subjects, and use the method described in Example 1 based on the acquired data to obtain multiple corresponding overall optimal fMRI data analysis dynamic causal models as candidate models;
[0098] The group Bayes factor was used to perform group-level analysis on multiple alternative models and select the optimal model for the group.
[0099] Specifically, the group Bayesian factor is used to perform group-level analysis on multiple candidate models to select the optimal model for the group, specifically including:
[0100] Apply each candidate model to all subjects, and calculate the group Bayes factor (representing the average score of the candidate model on all subjects) based on the degree of fit of the candidate model on the fMRI image data of the brain of each subject. Select the candidate model with the highest total score as the optimal model of the group;
[0101] The optimal group model is matched to the brain fMRI image data of all subjects, the model parameters of each subject are estimated, and the Bayesian parameter averaging of the model is performed to obtain the final dynamic causal model for fMRI data analysis representing the group effect.
[0102] In order to be applicable to the exploratory analysis of group brain fMRI image data, multiple overall optimal fMRI data analysis dynamic causal models corresponding to multiple subjects are constructed, and the overall optimal fMRI data analysis dynamic causal model with the highest overall fit is selected as the group optimal model using the group Bayes factor to achieve exploratory analysis applicable to group brain fMRI image data. It should be noted that the use of group Bayes factors to perform group-level analysis on multiple models is an existing method, and its specific process will not be repeated here.
[0103] For ease of application, the method of Example 1 or Example 2 is encapsulated into a MATLAB function based on SPM to help neuroscience researchers analyze the brain's causal information flow network.
[0104] Example 3
[0105] This embodiment provides a device for constructing a dynamic causal model for fMRI data analysis, comprising:
[0106] A preprocessing module, used to obtain brain fMRI image data and perform preprocessing;
[0107] A region of interest selection module is used to select a region of interest from the preprocessed brain fMRI image data;
[0108] The first model building module is used to use the GreedyEC algorithm with gradually decreasing effect connections to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the first global optimal fMRI data analysis dynamic causal model;
[0109] The second model building module is used to use the GreedyROI algorithm with gradually increasing effect connections to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the second global optimal fMRI data analysis dynamic causal model;
[0110] The overall optimal model building module is used to use the Bayesian model to compare the first global optimal fMRI data analysis dynamic causal model and the second global optimal fMRI data analysis dynamic causal model to obtain the overall optimal fMRI data analysis dynamic causal model.
[0111] Example 4
[0112] This embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for constructing a dynamic causal model for fMRI data analysis as described above.
[0113] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0114] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0115] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0117] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.
[0118] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.
[0119] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.
Claims
1. A method for constructing a dynamic causal model for fMRI data analysis, characterized in that: include: Acquire brain fMRI image data and perform preprocessing; Selecting regions of interest from preprocessed brain fMRI image data; The GreedyEC algorithm with gradually decreasing effect connections is used to search the model space of the constructed dynamic causal model for fMRI data analysis, and the first global optimal dynamic causal model for fMRI data analysis is obtained; The GreedyROI algorithm with gradually increasing effect connections was used to search the model space of the constructed dynamic causal model for fMRI data analysis, and the second global optimal dynamic causal model for fMRI data analysis was obtained. Using the Bayesian model to compare the first global optimal fMRI data analysis dynamic causal model and the second global optimal fMRI data analysis dynamic causal model, the overall optimal fMRI data analysis dynamic causal model is obtained; The GreedyEC algorithm with gradually decreasing effect connections is used to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the first global optimal fMRI data analysis dynamic causal model, which specifically includes: A1: Number the bidirectional effect connections between all regions of interest ROIs, and define the fMRI data analysis dynamic causal model based on the existence of bidirectional effect connections between all regions of interest ROIs as the current optimal model; A2: Consider the first effect connection and let i =1; A3: Close the first i The signal effects are connected to construct the current candidate model; A4: Use Bayesian model selection methods to compare the model evidence between the current best model and the current candidate models; A5: If the model evidence of the current optimal model is greater than that of the current candidate model, the first i Otherwise, the current candidate model is used as the current optimal model, and the first effect is closed in the subsequent candidate models. i Number effect connection; A6: Order i = i+ 1, and repeat steps A3-A6 until the n* (n-1)th efficient connection is considered, where n is the number of selected regions of interest; A7: The current optimal model finally obtained is the first global optimal fMRI data analysis dynamic causal model; The GreedyROI algorithm with gradually increasing effect connections is used to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the second global optimal fMRI data analysis dynamic causal model, which specifically includes: B1: Number a total of n regions of interest ROIs, k represents the number of regions of interest ROIs considered in the current stage; B2: Define the model space composed of the effect connection between the first region of interest ROI and the second region of interest ROI, and construct 2 2 Each fMRI data analysis dynamic causal model was used as an alternative model; B3: Use the Bayesian model selection method to compare all candidate models in the model space, take the optimal model as the 2ROI optimal model, and set k=2; B4: Fix the kROI optimal model obtained in the previous step, add the k+1th ROI as a new node, and there are 2*k possible effect connections between the new node and the original ROI; B5: Define the model space composed of the effect connections between the ROIs 1 to k+1, keep the original kROI optimal model structure unchanged, and only change the state of the effect connection with the new node as the vertex, and construct a total of 2 2*k Each fMRI data analysis dynamic causal model was used as an alternative model; B6: Use the Bayesian model selection method to compare all candidate models in the model space and take the optimal model as the (k+1)ROI optimal model; B7: Let k=k+1, repeat steps B4~B7 until the nth region of interest ROI is considered, and the final optimal model of n ROI is the second global optimal dynamic causal model for fMRI data analysis.
2. The method for constructing a dynamic causal model for fMRI data analysis according to claim 1, characterized in that: Preprocessing of brain fMRI image data includes: Head motion correction: Using the 3D brain function image at the first moment as a reference, perform rigid body transformation on the 3D brain function images at other moments to align them; Spatial normalization: align the brain structural image to the brain functional image obtained after head motion correction, and use the template to segment the brain structural image; according to the gray matter cortex in the segmentation result, align the brain functional image and brain structural image to the standard brain template; Spatial smoothing: Gaussian smoothing is performed on the spatially normalized image.
3. The method for constructing a dynamic causal model for fMRI data analysis according to claim 2, characterized in that: The head motion correction process also includes: only retaining the three-dimensional brain function image whose translation movement does not exceed a preset distance and whose rotation angle does not exceed a preset angle.
4. The method for constructing a dynamic causal model for fMRI data analysis according to claim 1, characterized in that: The number of regions of interest selected from the preprocessed brain fMRI image data does not exceed 8.
5. The method for constructing a dynamic causal model for fMRI data analysis according to any one of claims 1 to 4, characterized in that: Also includes: Acquire brain fMRI image data of multiple subjects, and use multiple overall optimal fMRI data analysis dynamic causal models obtained based on the acquired data as candidate models; The group Bayes factor was used to perform group-level analysis on multiple alternative models and select the optimal model for the group.
6. The method for constructing a dynamic causal model for fMRI data analysis according to claim 5, characterized in that: The method uses the group Bayesian factor to perform group-level analysis on multiple candidate models to select the optimal model for the group, specifically including: Calculate the group Bayes factor when each candidate model matches the brain fMRI image data of all subjects, and select the candidate model with the highest total score as the group optimal model; The optimal group model is matched to the brain fMRI image data of all subjects, and the Bayesian parameters of the model are averaged to obtain the dynamic causal model for fMRI data analysis that ultimately represents the group effect.
7. A device for constructing a dynamic causal model for fMRI data analysis, characterized in that: include: A preprocessing module, used to obtain brain fMRI image data and perform preprocessing; A region of interest selection module is used to select a region of interest from the preprocessed brain fMRI image data; The first model building module is used to use the GreedyEC algorithm with gradually decreasing effect connections to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the first global optimal fMRI data analysis dynamic causal model; The second model building module is used to use the GreedyROI algorithm with gradually increasing effect connections to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the second global optimal fMRI data analysis dynamic causal model; An overall optimal model building module is used to compare the first global optimal fMRI data analysis dynamic causal model and the second global optimal fMRI data analysis dynamic causal model using a Bayesian model to obtain an overall optimal fMRI data analysis dynamic causal model; The GreedyEC algorithm with gradually decreasing effect connections is used to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the first global optimal fMRI data analysis dynamic causal model, which specifically includes: A1: Number the bidirectional effect connections between all regions of interest ROIs, and define the fMRI data analysis dynamic causal model based on the existence of bidirectional effect connections between all regions of interest ROIs as the current optimal model; A2: Consider the first effect connection and let i =1; A3: Close the first i The signal effects are connected to construct the current candidate model; A4: Use Bayesian model selection methods to compare the model evidence between the current best model and the current candidate models; A5: If the model evidence of the current optimal model is greater than that of the current candidate model, the first i Otherwise, the current candidate model is used as the current optimal model, and the first effect is closed in the subsequent candidate models. i Number effect connection; A6: Order i = i+ 1, and repeat steps A3-A6 until the n* (n-1)th efficient connection is considered, where n is the number of selected regions of interest; A7: The current optimal model finally obtained is the first global optimal fMRI data analysis dynamic causal model; The GreedyROI algorithm with gradually increasing effect connections is used to search the model space of the constructed fMRI data analysis dynamic causal model to obtain the second global optimal fMRI data analysis dynamic causal model, which specifically includes: B1: Number a total of n regions of interest ROIs, k represents the number of regions of interest ROIs considered in the current stage; B2: Define the model space composed of the effect connection between the first region of interest ROI and the second region of interest ROI, and construct 2 2 Each fMRI data analysis dynamic causal model was used as an alternative model; B3: Use the Bayesian model selection method to compare all candidate models in the model space, take the optimal model as the 2ROI optimal model, and set k=2; B4: Fix the kROI optimal model obtained in the previous step, add the k+1th ROI as a new node, and there are 2*k possible effect connections between the new node and the original ROI; B5: Define the model space composed of the effect connections between the ROIs 1 to k+1, keep the original kROI optimal model structure unchanged, and only change the state of the effect connection with the new node as the vertex, and construct a total of 2 2*k Each fMRI data analysis dynamic causal model was used as an alternative model; B6: Use the Bayesian model selection method to compare all candidate models in the model space and take the optimal model as the (k+1)ROI optimal model; B7: Let k=k+1, repeat steps B4~B7 until the nth region of interest ROI is considered, and the final optimal model of n ROI is the second global optimal dynamic causal model for fMRI data analysis.
8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing a dynamic causal model for fMRI data analysis as described in any one of claims 1 to 6 is implemented.