Alzheimer's disease diagnosis method based on adaptive fine-grained causal brain network

Abstract

The invention discloses an Alzheimer's disease diagnosis method based on a self-adaptive fine-grained causal brain network, and relates to the field of bioinformatics. The method mainly comprises four parts of adaptive fine-grained division, causal brain network construction, causal network fusion based on intervention and disease diagnosis based on the causal brain network. The method comprises the following specific steps: firstly, finely dividing a brain network according to tense period difference of a brain region by adopting a self-adaptive fine-grained division method, and capturing spatiotemporal characteristics of different scales; then, spatial-temporal features are extracted from the fine-grained partitions, feature learning is performed by using deep learning and a frequency domain algorithm, and multi-granularity feature representation of the brain network is obtained; and then, constructing a causal brain network by using a causal inference technology, identifying a causal relationship between brain regions, and analyzing correlation between different brain regions. Furthermore, in combination with a causal intervention strategy, information of the brain-level causal network is integrated into the fine-grained causal network. And finally, by constructing a diagnosis model based on a fine-grained causal brain network, brain network features are classified by using a graph neural network, and early diagnosis of the Alzheimer's disease is realized. The process allows a causal brain network to be constructed based on brain region division, causal changes between brain regions are revealed, and the Alzheimer's disease diagnosis result is improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and more specifically to an Alzheimer's disease diagnosis method based on an adaptive fine-grained causal brain network. Background Art

[0002] Alzheimer's disease (AD) is one of the most prevalent neurodegenerative diseases, and recent advances in deep learning have demonstrated that neuroimaging data analysis has great potential in diagnosing this type of disease. Traditional methods usually rely on static and linear models, such as convolutional neural networks or Transformer, but they may not be able to effectively capture the complex, dynamic and nonlinear nature of brain connections. Therefore, it is natural for researchers to introduce graph structures to build brain network models. Among the methods based on graph neural networks (GNNs), Shimizu applied LiNGAM, a method that focuses on estimating non-Gaussian causal structures, to provide a framework for revealing causal relationships in brain networks. However, while these methods are sophisticated, they often ignore cyclic temporal dynamics and causal dependencies, which are crucial for a comprehensive understanding of the neurodegenerative processes behind attention deficit disorder. Li et al. proposed BrainGNN, an interpretable brain graph neural network designed for fMRI analysis, to address some of these challenges. BrainGNN has played an important role in identifying linear causal structures in brain networks, but it still has difficulties in dealing with the complexity of nonlinear interactions and the high-dimensional data of typical fMRI studies. Therefore, there is still a great need for comprehensive methods that combine deep learning with causal reasoning to more accurately model the dynamic nonlinear interactions in neurodegenerative diseases such as AD. Summary of the invention

[0003] In view of this, the present invention provides an Alzheimer's disease diagnosis method based on an adaptive fine-grained causal brain network to solve the problem of combining temporal dynamics and causal dependencies for analyzing and diagnosing Alzheimer's disease, which is helpful for the study of brain network construction methods, so as to better discover and diagnose Alzheimer's disease, which is related to the early diagnosis and intervention of Alzheimer's disease, and is of great significance for slowing the progression of the disease and alleviating the aggravation of symptoms.

[0004] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network, the specific steps of which include the following:

[0005] Diagram of the process of diagnosing Alzheimer's disease using an adaptive fine-grained causal brain network-based approach Data collection and preprocessing: First, we collected a real sample dataset of functional magnetic resonance imaging (fMRI) of Alzheimer's disease; data preprocessing included denoising, removing irrelevant signals, etc., and processing the image data into time series data for subsequent analysis and modeling; Adaptive fine-grained division: In brain network analysis, the time series periodic decoupling method is used to adaptively divide different brain regions into fine-grained time series; Feature extraction at different granularities: Based on adaptive fine-grained division, information at different granularities is extracted. This includes extracting data features of neural activity at multiple levels, such as brain region level and time slice level: using deep learning and frequency domain methods to perform multi-level feature learning on high-dimensional data to capture the dynamic changes and structural characteristics of brain networks; Causal brain network construction: Using the extracted multi-granularity data features, we establish multi-granularity causal variables, use the Granger causal discovery method to identify the causal relationship between different brain regions, and analyze the causal direction and degree of influence between brain regions or time slices; Intervention-based causal network fusion: Based on the construction of causal brain networks of different granularities, combined with the basic principles of causal graph intervention, the coarse-grained causal brain network information at the brain region level is integrated into the fine-grained causal brain network at the time slice level to complete the final fine-grained causal brain network construction; Disease diagnosis based on causal brain network: Using the constructed fine-grained causal brain network and multi-level features, Alzheimer's disease is diagnosed based on graph neural network.

[0006] 2. Furthermore, the main period of the time series is generated by using periodic pattern decoupling and TopM method. For example, the value of the main period is: , .

[0007] 3. Furthermore, the segmentation point selection of the fine-grained division is that for each brain region, the main period of the time series and its multiples are the segmentation points, and the value of the main period is: .

[0008] 4. Furthermore, the fusion of the causal network uses the causal graph intervention method to intervene in each causal variable of the coarse-grained brain network, remove all its incoming edges, and then calculate the difference between the causal effects before and after the intervention: ; Then the intervention results Add to the fine-grained causal brain network to enhance the fine-grained causal brain network.

[0009] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network, and the beneficial effects of the present invention are:

[0010] 1. The present invention proposes an Alzheimer's disease diagnosis method based on an adaptive fine-grained causal brain network, which can utilize graph deep learning and causal discovery capabilities at the brain region level and time segment level to comprehensively improve the performance of fMRI in diagnosing Alzheimer's disease.

[0011] 2. The present invention is based on the periodic decoupling method of time series, and performs adaptive division and embedded representation of fine-grained causal variables, which is conducive to a more thorough understanding of the causal dynamics of brain neuropathies at the time segment level in the time dimension. Compared with the existing Alzheimer's disease diagnosis method, it has a simpler and more interpretable embedded representation.

[0012] 3. The causal graph fusion model proposed in the present invention integrates the coarse-grained causal graph into the fine-grained brain network through intervention operations on the causal graph, thereby effectively enhancing the brain network structure representation of macroscopic features in an interpretable way. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 Diagram of the Alzheimer's disease diagnosis process based on an adaptive fine-grained causal brain network;

[0015] Figure 2 It is the overall flow chart of the present invention; DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0017] The present invention discloses a method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network, and the specific steps include the following:

[0018] Step 1: Data collection and preprocessing. Collect real fMRI data of Alzheimer's disease. First, preprocess the original fMRI data set of N subjects into a time series. , represents the coarse-grained time series at the brain region level. Each subject is represented by K brain regions across time T, expressed as ,in is the time point t of the kth brain region of the ith subject. Different from the coarse-grained representation, the fine-grained representation of time series involves segmenting time segments in the same brain region, which contain different lengths of time according to the period. In the effective functional connection network of the brain, these fine-grained time segments are regarded as causal variables and are represented as ,in represents the a-th time segment corresponding to the feature k of the ith subject obtained from the time series.

[0019] Step 2: Adaptive fine-grained division. For the time series of each brain region , are projected into the Ramanujan subspace to achieve periodic pattern decomposition. Then, the top M main cycles with the highest occurrence rate in each brain region sample are selected as the segmentation points of the cycle. This process can adaptively segment fine-grained time segments without manual parameter setting. The value of M is determined based on the observed data set S as follows: , .in, represents the number of main periodic components of the time series in brain region k extracted by the APMD method for sample i, represents the average number of all sample periodic components in brain region k. Once the main periodic components of sequence S are determined, each extracted period is systematically stored in a designed three-dimensional tensor , where represents the mth main cycle extracted from the time series of feature k in sample i. The second dimension 𝑘 maps to the various features under consideration, capturing the periodic behavior of each specific brain region in the dataset, and the third dimension 𝑚 encompasses the different periods determined for each feature to reflect the multi-scale periodic dynamics in the time series. Specifically, for each brain region, the period and its multiples are determined as the split point of the time series in that specific brain region: , Among them, C k is the main cycle set of brain area k.

[0020] Step 3: Feature extraction at different granularities. Wavelet transformers were applied to the fine-grained time segments and coarse-grained time series of each brain region. This step enables us to extract complex temporal patterns into better feature components by projecting the data from the time domain to the frequency domain. The frequency domain representation enhances our ability to detect subtle oscillatory dynamics and temporal changes that may be masked in the time domain. The result of this process is two types of embeddings that can capture causal variables while preserving the original order of the time series. One is a fine-grained time segment embedding that can analyze local adaptive periodic features; the other is a coarse-grained embedding that can analyze the global macroscopic features of the brain region.

[0021] Step 4: Causal brain network construction. In order to independently learn coarse-grained and fine-grained causal brain networks, a graph encoder is used in this stage to derive a dynamic adjacency matrix. The network uses the Granger causal discovery algorithm to analyze time series and discover fine-grained causal graphs and coarse-grained causal graphs to reflect the potential fine-grained causal relationships between brain structures.

[0022] Step 5: Intervention-based causal network fusion. Through the causal information in the coarse-grained network, the graph isomorphism network (GIN) is used to enhance the edge weights of the fine-grained causal brain network. Specifically, we intervene on each causal variable in the coarse-grained graph, remove all incoming edges, and calculate the average causal effect (ACE) using the causal graph after intervention: . This ACE will then enhance the edge weights of the fine-grained brain network by incorporating the intervention enhancement value: ,This integrated approach combines detailed local feature learning with global causal context, improving the overall understanding of brain networks.

[0023] Step 6: Disease diagnosis based on causal brain networks. By inputting the learned causal brain network into GNN, the model can effectively capture and utilize the nonlinear causal relationship between fine-grained brain regions to make accurate diagnoses.

[0024] The Alzheimer's disease diagnosis method based on an adaptive fine-grained causal brain network proposed in the present invention solves the problem of combining temporal dynamics and causal dependencies for analyzing and diagnosing Alzheimer's disease. The method is effective and easy to implement.

[0025] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network, characterized in that: The specific steps include the following: Data collection and preprocessing: First, we collected a real sample dataset of functional magnetic resonance imaging (fMRI) of Alzheimer's disease; data preprocessing included denoising, removing irrelevant signals, etc., and processing the image data into time series data for subsequent analysis and modeling; Adaptive fine-grained division: In brain network analysis, the time series periodic decoupling method is used to adaptively divide different brain regions into fine-grained time series; Feature extraction at different granularities: Based on adaptive fine-grained division, information at different granularities is extracted. This includes extracting data features of neural activity at multiple levels, such as brain region level and time slice level: using deep learning and frequency domain methods to perform multi-level feature learning on high-dimensional data to capture the dynamic changes and structural characteristics of brain networks; Causal brain network construction: Using the extracted multi-granularity data features, we establish multi-granularity causal variables, use the Granger causal discovery method to identify the causal relationship between different brain regions, and analyze the causal direction and degree of influence between brain regions or time slices; Intervention-based causal network fusion: Based on the construction of causal brain networks of different granularities, combined with the basic principles of causal graph intervention, the coarse-grained causal brain network information at the brain region level is integrated into the fine-grained causal brain network at the time slice level to complete the final fine-grained causal brain network construction; Disease diagnosis based on causal brain network: Using the constructed fine-grained causal brain network and multi-level features, Alzheimer's disease is diagnosed based on graph neural network.

2. The method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network according to claim 1, characterized in that: The main cycle of the time series is generated by decoupling the periodic pattern and combining the frequency of the periodic occurrence. For example, the value of the main period is: , .

3. The method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network according to claim 1, characterized in that: The segmentation point selection of the fine-grained division is that for each brain region, the main period of the time series and its multiples are the segmentation points, and the value of the main period is: .

4. The method for diagnosing Alzheimer's disease based on an adaptive fine-grained causal brain network according to claim 1, characterized in that: The fusion of the causal network uses the causal graph intervention method to intervene in each causal variable of the coarse-grained brain network, remove all incoming edges of the intervened nodes in the coarse-grained network, and then calculate the difference in causal effect before and after the intervention: ; Then the intervention results Add to the fine-grained causal brain network to enhance the fine-grained causal brain network.

Images