Mild cognitive impairment detection system, electronic device, and storage medium

Abstract

The application provides a mild cognitive impairment detection system, an electronic device and a storage medium, and relates to the technical field of brain network. The application can identify the heterogeneous characteristics of a functional structural brain network based on DTI and fMRI, obtain effective features of a homogeneous network region according to different types of nodes and edges, and simultaneously obtain biomarkers of a heterogeneous region network, so as to comprehensively and effectively integrate the features of different regions in the brain, form a whole-brain level feature representation that can accurately express the brain network characteristics of a person to be detected, and improve the detection accuracy of the mild cognitive impairment detection system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of brain network, and particularly relates to a mild cognitive impairment detection system, an electronic device and a storage medium. BACKGROUND

[0002] Mild cognitive impairment (MCI) is a clinical symptom characterized by cognitive decline and is considered a precursor to Alzheimer's disease. Early detection of MCI is crucial for timely intervention and prevention of further cognitive decline. In the prior art, deep learning methods such as graph convolutional neural networks are generally used to analyze and detect brain diseases, and the construction of brain networks using neuroimaging is a prerequisite for such methods.

[0003] However, many graph neural network studies of brain networks fail to analyze the heterogeneity of brain networks, and the established graph network methods cannot capture real heterogeneity connection information and simulate real biological information flow of the brain, resulting in low accuracy of detection results output by existing mild cognitive impairment detection systems. SUMMARY

[0004] (I) Technical problems solved

[0005] To overcome the deficiencies of the prior art, the present application provides a mild cognitive impairment detection system, an electronic device and a storage medium, which solve the technical problem of low accuracy of detection results output by existing mild cognitive impairment detection systems.

[0006] (II) Technical solutions

[0007] To achieve the above object, the present application is implemented by the following technical solutions:

[0008] In a first aspect, the present application provides a mild cognitive impairment detection system, comprising a brain network processing module and a brain network detection module;

[0009] The brain network processing module is configured to obtain diffusion tensor imaging neuroimaging and magnetic resonance functional imaging neuroimaging of a person to be detected, and to construct a whole brain network based on the diffusion tensor imaging neuroimaging and the magnetic resonance functional imaging neuroimaging of the person to be detected;

[0010] The brain network detection module is configured to process the whole brain network of the person to be detected by a pre-trained mild cognitive impairment detection model based on a heterogeneous graph attribute segmentation and integration network, and to output a mild cognitive impairment detection result;

[0011] The mild cognitive impairment detection model comprises a brain network division layer, a graph diffusion layer, an encoder layer, an adaptive feature fusion layer and a classification prediction layer.

[0012] The brain network segmentation layer is used to divide the whole brain network into the left hemisphere brain network, the hemisphere diencephalon network, and the right hemisphere brain network; the left hemisphere brain network and the right hemisphere brain network are both homogeneous networks, while the hemisphere diencephalon network is a heterogeneous network.

[0013] The graph diffusion layer is used to perform diffusion processing on the left hemisphere brain network and the right hemisphere brain network respectively, to obtain a left hemisphere brain network and a right hemisphere brain network containing potential connectivity relationships.

[0014] The encoder layer includes a first homogeneous encoder, a second homogeneous encoder, and a heterogeneous encoder. The first homogeneous encoder and the second homogeneous encoder are used to extract network features with homogeneous properties from the left hemisphere brain network and the right hemisphere brain network containing potential connections, respectively. The heterogeneous encoder is used to extract network features with heterogeneous properties from the interhemispheric network.

[0015] Adaptive Feature Fusion Layer: Used to fuse network features extracted by homogeneous encoders and heterogeneous encoders. For the same nodes that coexist in different networks, they are merged according to their contribution to their respective networks to obtain a whole-brain level brain network representation.

[0016] The classification prediction layer is used to process the whole-brain level brain network representation and output the detection results of mild cognitive impairment.

[0017] Preferably, the diffusion process of the diffusion layer includes:

[0018] The left and right hemisphere brain networks were diffused separately using a graph diffusion technique based on thermonuclear spectroscopy.

[0019] Preferably, the diffusion process of the diffusion layer includes:

[0020] PageRank graph diffusion technology was used to perform diffusion processing on the left and right hemisphere brain networks respectively.

[0021] Preferably, the homogeneous encoder employs a deep graph convolutional network that can mitigate oversmoothing, and its l-th convolutional layer is represented as:

[0022]

[0023] Where σ is the activation function, α l ∈(0,1] is the preset probability activation parameter, β l To adjust the learnable weight matrix W l hyperparameters, For left or right hemisphere brain networks containing potential connectivity relationships, H (l) This is the output of the hidden layer of the convolution.

[0024] Preferably, the expression for the encoding embedding process of the heterogeneous encoder is:

[0025]

[0026] Where R represents the edge type, A r This represents the adjacency structure under the characteristic relation type, where W represents the learnable parameter matrix, X is the feature matrix, and H... I This indicates the encoding of heterogeneous regions.

[0027] Preferably, the adaptive feature fusion layer is specifically used for:

[0028] The learned representations of the encoder layer output network are merged, and the network features H with homogeneous attributes of the right hemisphere brain network are used. R Network features H with homogeneous properties to the left hemisphere brain network L By directly splicing these together, homogeneous regions of the brain network can be represented:

[0029] H homo =CONCAT(H R H L )

[0030] For heterogeneous region coding H I All its nodes are also in H homo The node v has been encoded once; for a node v that exists simultaneously with both encoded representations... i ∈N m and corresponding to t∈{H homo H hetero The representation of h in} i t A node-level adaptive fusion mechanism is designed to merge nodes, and its corresponding adaptive contribution coefficient is calculated as follows:

[0031]

[0032] Where q, W, and b are all learnable parameterized vectors, tanh is the activation function, and β is the activation vector. i t For node h i In t∈{H homo H hetero The contribution coefficient of} This is the result after softmax normalization;

[0033] Based on the contribution coefficients of these nodes within their respective encodings, they are fused to obtain a whole-brain-level brain network representation Z = {z1, z2, ..., z...}. n}, where z i Represented as:

[0034]

[0035] Where, N m This represents the set of nodes in the whole-brain network.

[0036] Preferably, the classification prediction layer uses a multilayer perceptron to read out the whole-brain level brain network representation and output the mild cognitive impairment detection result.

[0037] Preferably, the loss function during the training process of the pre-trained mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks includes:

[0038] For a sample of T individuals, the predicted output of the mild cognitive impairment detection model is: If the actual label of the sample is y, then the cross-entropy function is used to optimize the model, defined as follows:

[0039]

[0040] Secondly, the present invention provides a computer-readable storage medium for storing computer instructions, programs, code sets, or instruction sets, which, when run on a computer, cause the computer to perform the functions corresponding to the brain network processing module and brain network detection module in the mild cognitive impairment system described above.

[0041] Thirdly, the present invention provides an electronic device, comprising:

[0042] One or more processors; and a computer-readable storage medium for storing computer instructions, programs, code sets, or instruction sets that, when executed on a computer, cause the one or more processors to perform the functions corresponding to the brain network processing module and brain network detection module in the mild cognitive impairment system described above.

[0043] (III) Beneficial Effects

[0044] This invention provides a mild cognitive impairment detection system, electronic device, and storage medium. Compared with the prior art, it has the following advantages:

[0045] This invention can identify the heterogeneity features of functional structural brain networks based on DTI and fMRI. According to different types of nodes and edges, it can obtain effective features of homogeneous network regions and obtain biomarkers for heterogeneous network regions. This comprehensively and effectively integrates the features of different regions in the brain to form a whole-brain feature representation that can accurately express the brain network features of the tested person, thereby improving the detection accuracy of mild cognitive impairment detection systems. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a general framework diagram of a mild cognitive impairment detection system according to an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the division of the whole-brain network in an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] This application provides a mild cognitive impairment detection system, electronic device, and storage medium based on heterogeneous graph attribute segmentation and integration networks, which solves the technical problem of low accuracy of detection results output by existing mild cognitive impairment detection systems, and achieves accurate detection of mild cognitive impairment.

[0051] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:

[0052] Neuroimaging-based brain network disease detection technologies play a crucial supporting role in doctors' diagnostic work, helping to reduce the influence of subjective factors on diagnosis. Furthermore, the invention and application of advanced technologies can significantly improve the diagnostic capabilities of hospitals. However, existing research neglects the differences in properties between regions of interest (ROIs) and the different connection patterns between them, even though heterogeneity is prevalent in brain networks. For example, in terms of node properties, the left hemisphere is associated with language and logic processing, while the right hemisphere excels at detecting and responding to emotional changes. Moreover, hemispheric specialization, a key feature of the human brain, also known as brain asymmetry, is considered relevant to the early identification of neurological diseases. Currently, there is still a lack of graph-based methods that effectively utilize the heterogeneous structure of brain networks. Although some graph-based brain network analysis methods have recognized the existence of heterogeneity in brain networks, these studies often examine heterogeneity from a graph theory perspective, deviating from the essential characteristics of brain network heterogeneity analysis. Furthermore, as one of the most complex structures in living organisms, diffusion tensor imaging (DTI) technology only reflects the objective structural connectivity between brain regions through water molecule diffusion, revealing less about their potential connectivity. In fact, brain regions exhibit a variety of complex connectivity patterns, and revealing these patterns is key to accurately characterizing the state of brain network activity. Therefore, brain network analysis methods based on graph neural networks need to consider the following two issues:

[0053] (1) Establishing heterogeneous brain networks and analysis methods: Current brain network research for MCI detection is mainly based on the homogeneity assumption, lacking heterogeneity analysis of brain networks. As a result, the established graph network methods cannot capture real heterogeneous connection information and simulate the real biological information flow of the brain.

[0054] (2) Discovering potential biological connectivity patterns in brain networks: Current research on the structure of brain networks mainly relies on the objective structural connectivity of the brain revealed by DTI, while ignoring the potential connectivity patterns in brain networks, resulting in incomplete and subjective modeling of brain networks.

[0055] To address the aforementioned issues, this invention proposes a mild cognitive impairment (MCI) detection system based on heterogeneous graph attribute segmentation and integration networks. By mining potential connections between brain regions within the brain network, it better captures heterogeneous information and the true information flow within the network, thereby improving the accuracy of MCI detection. The invention first constructs a structural-functional brain network using digital brain imaging (DTI) and fMRI. DTI describes the brain network structure, and fMRI serves as the network nodes. Based on node types and DTI connection strength, the whole-brain network is automatically divided into three regions: a homogeneous network in the left hemisphere, a homogeneous network in the right hemisphere, and a heterogeneous network between hemispheres. Then, for the homogeneous network, which occupies the majority of the network, graph diffusion techniques are used to mine potential connection patterns. The three network regions are then used to learn subgraph embedding representations using their respective optimal graph convolutional neural network methods. Finally, a node-level adaptive fusion mechanism is used to fuse the brain region node representations and integrate all subgraph representations to form a whole-brain embedding, thereby predicting the MCI detection result.

[0056] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0057] This invention provides a mild cognitive impairment detection system, including a brain network processing module and a brain network detection module;

[0058] The brain network processing module is used to acquire the neural images of diffusion tensor imaging (DTI) and functional magnetic resonance imaging (fMRI) of the subject (i.e., both the neural images of diffusion tensor imaging and functional magnetic resonance imaging are input into the computer), and to construct a whole-brain network based on the neural images of diffusion tensor imaging and functional magnetic resonance imaging of the subject.

[0059] The brain network detection module is used to process the whole brain network of the tested person through a pre-trained mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble network, and output mild cognitive impairment detection results.

[0060] The mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks includes a brain network partitioning layer, a graph diffusion layer, an encoder layer, an adaptive feature fusion layer, and a classification prediction layer. The processing procedure of the mild cognitive impairment detection model on the whole-brain network is as follows: Figure 1 As shown.

[0061] The brain network segmentation layer is used to divide the whole brain network into the left hemisphere brain network, the hemisphere diencephalon network, and the right hemisphere brain network; the left hemisphere brain network and the right hemisphere brain network are both homogeneous networks, while the hemisphere diencephalon network is a heterogeneous network.

[0062] The graph diffusion layer is used to perform diffusion processing on the left and right hemisphere brain networks respectively, resulting in left and right hemisphere brain networks containing potential connections. In this layer, the transition matrix of the left and right brain networks is reconstructed, potential connections in the network are mined, and interfering connections are removed.

[0063] The encoder layer includes a first homogeneous encoder, a second homogeneous encoder, and a heterogeneous encoder. The first homogeneous encoder and the second homogeneous encoder are used to extract network features with homogeneous properties from the left hemisphere brain network and the right hemisphere brain network containing potential connections, respectively, and learn their network representations. The heterogeneous encoder is used to extract network features with heterogeneous properties from the interhemispheric network and learn its network representation.

[0064] Adaptive Feature Fusion Layer: Used to fuse network features learned by homogeneous encoders and heterogeneous encoders. For the same nodes that coexist in different networks, they are merged according to their contribution to their respective networks to obtain a whole-brain-level brain network representation.

[0065] The classification prediction layer is used to process the whole-brain level brain network representation and output the detection results of mild cognitive impairment.

[0066] The embodiments of the present invention can identify the heterogeneity features of functional structural brain networks based on DTI and fMRI. According to different types of nodes and edges, effective features of homogeneous network regions are obtained respectively. At the same time, biomarkers that are determined to be heterogeneous regional networks are obtained, thereby comprehensively and effectively integrating the features of different regions in the brain to form a whole-brain feature representation that can accurately express the brain network features of the tested person, thereby improving the detection accuracy of the mild cognitive impairment detection system.

[0067] The following section provides a detailed description of the mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks, incorporating technical principles:

[0068] In the construction and segmentation of the brain network, the basic structure of the whole-brain network was determined by the structural connections derived from DTI (diffusion tensor imaging) neuroimages, and the nodes were characterized by temporal signals from fMRI (functional magnetic resonance imaging) for each brain region. By analyzing the inter-brain connectivity strength within different hemispheres and classifying node types according to their location in lateral brain regions of different hemispheres, the brain networks of the subjects were described as heterogeneous networks with multiple types of nodes and edges. Based on this heterogeneity, the whole-brain network was divided into a homogeneous left hemisphere network, a heterogeneous interhemispheric network, and a homogeneous right hemisphere network.

[0069] In the graph diffusion layer, for subnetworks with high homogeneity, such as the left and right hemisphere brain networks, which include large brain regions and connections, two diffusion methods, namely graph diffusion technology based on hot kernels and personalized PageRank, are used to perform structural denoising and obtain high-order connections, respectively, as potential brain network structural relationships that cannot be reflected by DTI.

[0070] In the encoder layer, dedicated homogeneous convolutional encoders and heterogeneous encoders are designed for each of the divided brain network subnets to encode brain network subnets with different attributes.

[0071] In the adaptive feature fusion layer, a node-level adaptive mechanism is used to fuse the embedding representations of corresponding nodes in different subnetworks for the results encoded by each encoder. Finally, all subnetworks are integrated into a whole-brain level brain network embedding representation.

[0072] In the classification prediction layer, the whole-brain level brain network representation is used to obtain the final embedding representation using a graph-level readout function, which is then fed into the classifier to obtain the final MCI recognition result.

[0073] The following section provides a further description of each layer in the mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks, using specific expressions:

[0074] In the brain network segmentation layer, the structural-functional brain network of each MCI subject is first input, denoted as G=(A,X), where A∈R n×d X is the adjacency matrix of a brain network constructed based on DTI neuroimaging, where X = {v1, v2, ..., v...} n}∈R n×d This is a feature matrix of brain network nodes constructed based on fMRI, where each v i This represents a specific brain region node in the brain network. Each subject's brain network was divided into a left hemisphere brain network G. L =(A L ,X L Right hemisphere brain network G R =(A R ,X R ) and the hemispheric diencephalon network G I =(A I ,X I ), the specific network partitioning is as follows Figure 2 As shown.

[0075] In the graph diffusion layer, a homogeneous deep graph convolutional network is used to apply G... L =(A L ,X L ) and G R =(A R ,X RGraph diffusion techniques are implemented to uncover potential network connectivity patterns and remove some noisy connections to obtain a new adjacency matrix structure. The general graph diffusion process is defined as follows:

[0076]

[0077] in, Let A represent a general transition matrix, initialized as the adjacency matrix of the brain network. R Or A L Θ k The coefficients are polynomials and satisfy the constraints. And Θ k ∈[0,1]. In particular, a heat kernel and a personalized PageRank (PPR) approach are used to implement a brain network-based diffusion matrix S.

[0078] Based on the graph diffusion process of PPR, the symmetric normalized transition matrix is ​​defined as T. sys =D -1 / 2 AD -1 / 2 and set Where α∈(0,1) represents the transmission probability of a random walk. After simplification, the graph diffusion matrix can be expressed as:

[0079] S PPR =α(I n -(1-α)(I n -T sys )) -1

[0080] Among them, I n ∈R n×n It is an identity matrix.

[0081] Based on the graph diffusion process of the heat core, the normalized transition matrix of the random walk is defined as T. rw =AD -1 and set Where t represents the diffusion time. After simplification, the graph diffusion matrix can be expressed as:

[0082]

[0083] Furthermore, based on the two types of graph diffusion matrices mentioned above, regularization is applied to each to obtain the final diffusion-adjacency matrix representation of the brain network, namely:

[0084]

[0085] Both graph diffusion processes described above, based on PPR and hot kernels, possess the ability to filter out noisy connections and uncover potential correlations within a network. It's important to note that this diffusion method only applies to G.L =(A L ,X L ) and G R =(A R ,X R It was implemented.

[0086] In the encoder layer, targeting the left hemisphere brain network G L =(A L ,X L Right hemisphere brain network G R =(A R ,X R ) and the hemispheric diencephalon network G I =(A I ,X I The three networks are configured with two types of encoders: homogeneous encoders and heterogeneous encoders. The homogeneous encoder is primarily used to encode G. L =(A L ,X L ) and G R =(A R ,X R Heterogeneous encoders are used to encode G. I =(A I ,X I The homogeneous encoder employs a deep graph convolutional network to mitigate oversmoothing, where the l-th convolutional layer is represented as:

[0087]

[0088] Where σ is the activation function, α l ∈(0,1] is the preset probability activation parameter, β l To adjust the learnable weight matrix W l hyperparameters, It offers two selectable formats: PPR and Heat. (l) This is the output of the hidden layer of the convolution. Therefore, for G... L =(A L ,X L ) and G R =(A R ,X R They can be encoded as follows:

[0089]

[0090] For G I =(A I ,X I For this type of heterogeneous network, a heterogeneous encoder was designed for encoding embedding:

[0091]

[0092] Where R represents the type of edge, A r This represents the adjacency structure under the characteristic relation type, where W represents the learnable parameter matrix and X is the feature matrix.

[0093] In the adaptive feature fusion layer, the learned representations of the network obtained by merging the above encodings are used for G. R and G L By directly splicing these together, homogeneous regions of the brain network can be represented:

[0094] H homo =CONCAT(H R H L )

[0095] For heterogeneous region coding H I All its nodes are also in H homo It has been encoded once. For a node v that exists simultaneously in both encoded representations... i ∈N m and corresponding to t∈{H homo H hetero} represents Design a node-level adaptive fusion mechanism for merging nodes, and calculate the corresponding adaptive contribution coefficient as follows:

[0096]

[0097] Where q, W, and b are all learnable parameterized vectors, tanh is the activation function, and β is the activation vector. i t is node h i In t∈{H homo H hetero The contribution coefficient of} This is the result after softmax normalization.

[0098] Based on the contribution coefficients of these nodes within their respective encodings, they are fused to obtain the whole-brain-level brain network representation Z = {z1, z2, ..., z...}. n}, where z i Represented as:

[0099]

[0100] In the classification prediction layer, a multilayer perceptron (MLP) is used to read out the predicted probability for each Z.

[0101] It should be noted that the pre-trained mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks needs to be trained and optimized before use to determine the model's parameters. The loss function during the training process includes:

[0102] For a sample of T individuals, the predicted output of the mild cognitive impairment detection model is: If the actual label of the sample is y, then the cross-entropy function is used to optimize the model, defined as follows:

[0103]

[0104] The effectiveness of the mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks proposed in this invention is verified through the following validation experiments:

[0105] The ADNI dataset was used in the validation experiment. A total of 179 participants were collected during the validation process, including 28 healthy controls (NC), 51 participants with mild cognitive impairment (MCI), 64 participants with early mild cognitive impairment (EMCI), and 36 participants with late mild cognitive impairment (LMCI).

[0106] During the verification process, the mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble network of the present invention was compared with existing benchmark methods, including support vector machine (SVM), GCN, GAT, GCNII, GDC, Hebrain GNN, GCNH, and CAGNN.

[0107] Among them, Support Vector Machine (SVM): This supervised learning algorithm solves classification problems by finding the optimal hyperplane with the maximum margin in the feature space.

[0108] GCN: Simplifies multinomial convolution to a first-order form, thereby enabling information aggregation within a one-hop neighborhood.

[0109] GAT: It uses a self-attention mechanism to determine the importance of nodes in the first-order neighborhood, thus promoting local convolution.

[0110] GCNII is an improved version of the GCN model that combines initial residuals and identity mapping techniques to enable the stacking of multiple convolutional layers.

[0111] GDC: A spatial local convolution inspired by spectral methods. It addresses the problems of noisy edges and arbitrary definitions in real graphs by sparsifying the generalized form of graph diffusion.

[0112] HebrainGNN: In this method, the brain network is modeled as a heterogeneous graph, and a white-supervised pre-training strategy is designed to address the overfitting problem caused by the conflict between large-scale parameters and small-scale medical data samples.

[0113] GCNH is a graph learning method applicable to both homogeneous and heterogeneous graphs. It learns and combines individual representations of a node and its neighborhood, and uses learnable coefficients to balance the contributions of the central node and its neighborhood.

[0114] CAGNN is an algorithm specifically designed for heterogeneous networks. It decouples node features into discriminative features for downstream tasks, aggregates features for graph convolutional layers, and uses a shared mixer module to fuse these two features for classification.

[0115] This validation experiment uses four evaluation metrics (accuracy, F1 score, sensitivity, and precision) to demonstrate that the mild cognitive impairment (MCI) detection model based on heterogeneous graph attribute segmentation and ensemble networks outperforms the aforementioned benchmark methods in MCI classification. For fair comparison, all methods were evaluated using five-fold cross-validation, and Bayesian parameter search was performed to find the optimal parameters. It should be noted that graph diffusion includes two diffusion modes: hot core diffusion and personalized PageRank. The MCI detection model based on heterogeneous graph attribute segmentation and ensemble networks has two modes: a PageRank-based mode and a heat-based mode, denoted as HGASIN-PPR and HGASIN-Heat, respectively. Specific results are shown in Tables 1-4.

[0116] Table 1 shows the classification results obtained by comparing NC and MCI using 5-fold cross-validation with competing methods.

[0117]

[0118]

[0119] Table 2 shows the classification results obtained by 5-fold cross-validation with competing methods when comparing NC numbers to LMCI.

[0120]

[0121] Table 3 shows the classification results obtained by 5-fold cross-validation with competing methods when comparing NC and EMICI.

[0122]

[0123]

[0124] Table 4 shows the classification results obtained by performing 5-fold cross-validation with competing methods when MCI, LMCI, and EMCI are considered as a single disease category.

[0125]

[0126] Tables 1-3 divide the ADMI dataset into three tasks: NC vs. MCI, NC vs. LMCI, and NC vs. EMICI, to confirm that the mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks can distinguish various fine-grained MCI diseases. Experiments show that our two methods achieved optimal results on almost all evaluation metrics in these three discrimination tasks. Table 4 treats MCI, LMCI, and EMCI as a single disease category and classifies them together with NC to verify the competitiveness of the HGASI method in coarse-grained MCI disease detection. Experimental results show that the HGASI-Heat-based process is more competitive in this task and demonstrates overall leading performance. Compared with the suboptimal model GCNH, it improves accuracy by 6.73% (6.12%), F1 score by 2.17%, sensitivity by 6.73%, and precision by 20.48%. In summary, the experimental results in the four tables fully demonstrate the advantages of the mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble network in the present invention for mild cognitive impairment code (MCI) detection tasks. Furthermore, the heat-based heterogeneous graph attribute segmentation and ensemble network-based mild cognitive impairment detection model outperforms the PPR-based heterogeneous graph attribute segmentation and ensemble network-based mild cognitive impairment detection model in most cases.

[0127] This invention also provides a computer-readable storage medium for storing computer instructions, programs, code sets, or instruction sets, which, when run on a computer, enable the computer to perform functions corresponding to the brain network processing module and brain network detection module in the aforementioned mild cognitive impairment system.

[0128] This invention also provides an electronic device, comprising:

[0129] One or more processors; and a computer-readable storage medium for storing computer instructions, programs, code sets, or instruction sets that, when executed on a computer, cause the one or more processors to perform the functions corresponding to the brain network processing module and brain network detection module in the above-described mild cognitive impairment system.

[0130] In summary, compared with existing technologies, it has the following beneficial effects:

[0131] 1. The embodiments of the present invention can identify the heterogeneity features of functional structural brain networks based on DTI and fMRI. According to different types of nodes and edges, effective features of homogeneous network regions are obtained respectively. At the same time, biomarkers of heterogeneous regional networks are obtained, thereby comprehensively and effectively integrating the features of different regions in the brain to form a whole-brain feature representation that can accurately express the brain network features of the tested person, thereby improving the detection accuracy of the mild cognitive impairment detection system.

[0132] 2. The embodiments of the present invention utilize methods based on hot cores and personalized PageRank to denoise the connections between brain region nodes and discover hidden biological connections on the objective physiological connectivity structure based on DTI, thereby making up for the lack of effective patterns in the network and enabling mild cognitive impairment systems to encode brain network features more comprehensively and accurately.

[0133] 3. This invention can adaptively fuse encoded representations of nodes from different brain regions. Nodes from the same brain region may exist in subnetworks of different properties. An adaptive mechanism is used to fuse these node representations and integrate them into a whole-brain-level brain network embedding based on their contributions in different subnetworks. Compared to previous methods, this fully considers the richer properties of some important nodes, thereby further improving the detection accuracy of the mild cognitive impairment detection system.

[0134] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0135] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A mild cognitive impairment detection system, characterized in that, Includes a brain network processing module and a brain network detection module; The brain network processing module is used to acquire diffusion tensor imaging and magnetic resonance functional imaging of the tested person, and to construct a whole brain network based on the diffusion tensor imaging and magnetic resonance functional imaging of the tested person. The brain network detection module is used to process the whole brain network of the tested person through a pre-trained mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble network, and output mild cognitive impairment detection results. The mild cognitive impairment detection model includes a brain network segmentation layer, a graph diffusion layer, an encoder layer, an adaptive feature fusion layer, and a classification prediction layer. The brain network segmentation layer is used to divide the whole brain network into the left hemisphere brain network, the hemisphere diencephalon network, and the right hemisphere brain network; the left hemisphere brain network and the right hemisphere brain network are both homogeneous networks, while the hemisphere diencephalon network is a heterogeneous network. The graph diffusion layer is used to perform diffusion processing on the left hemisphere brain network and the right hemisphere brain network respectively, to obtain a left hemisphere brain network and a right hemisphere brain network containing potential connectivity relationships. The encoder layer includes a first homogeneous encoder, a second homogeneous encoder, and a heterogeneous encoder. The first homogeneous encoder and the second homogeneous encoder are used to extract network features with homogeneous properties from the left hemisphere brain network and the right hemisphere brain network containing potential connections, respectively. The heterogeneous encoder is used to extract network features with heterogeneous properties from the interhemispheric network. Adaptive Feature Fusion Layer: Used to fuse network features extracted by homogeneous encoders and heterogeneous encoders. For the same nodes that coexist in different networks, they are merged according to their contribution to their respective networks to obtain a whole-brain level brain network representation. The classification prediction layer is used to process the whole-brain level brain network representation and output the detection results of mild cognitive impairment.

2. The mild cognitive impairment detection system as described in claim 1, characterized in that, The diffusion process of the diffusion layer includes: The left and right hemisphere brain networks were diffused separately using a graph diffusion technique based on thermonuclear spectroscopy.

3. The mild cognitive impairment detection system as described in claim 1, characterized in that, The diffusion process of the diffusion layer includes: PageRank graph diffusion technology was used to perform diffusion processing on the left and right hemisphere brain networks respectively.

4. The mild cognitive impairment detection system as described in any one of claims 1 to 3, characterized in that, The homogeneous encoder employs a deep graph convolutional network that can mitigate oversmoothing, and its l-th convolutional layer is represented as: Where σ is the activation function, α l ∈(0,1] is the preset probability activation parameter, β l To adjust the learnable weight matrix W l hyperparameters, For left or right hemisphere brain networks containing potential connectivity relationships, H (l) This is the output of the hidden layer of the convolution.

5. The mild cognitive impairment detection system as described in any one of claims 1 to 3, characterized in that, The expression for the encoding embedding process of the heterogeneous encoder is: Where R represents the type of edge, A r This represents the adjacency structure under the characteristic relation type, where W represents the learnable parameter matrix, X is the feature matrix, and H... I This indicates the encoding of heterogeneous regions.

6. The mild cognitive impairment detection system as described in any one of claims 1 to 3, characterized in that, The adaptive feature fusion layer is specifically used for: The learned representations of the encoder layer output network are merged, and the network features H with homogeneous attributes of the right hemisphere brain network are used. R Network features H with homogeneous properties to the left hemisphere brain network L By directly splicing these together, homogeneous regions of the brain network can be represented: H homo =CONCAT(H R ,H L ) For heterogeneous region coding H I All its nodes are also in H homo The node v has been encoded once; for a node v that exists simultaneously with both encoded representations... i ∈N m and corresponding to t∈{H homo H hetero } represents Design a node-level adaptive fusion mechanism for merging nodes, and calculate the corresponding adaptive contribution coefficient as follows: Where q, W, and b are all learnable parameterized vectors, and tanh is the activation function. For node h i In t∈{H homo H hetero The contribution coefficient of} This is the result after softmax normalization; Based on the contribution coefficients of these nodes within their respective encodings, they are fused to obtain a whole-brain-level brain network representation Z = {z1, z2, ..., z...}. n }, where z i Represented as: Where, N m This represents the set of nodes in the whole-brain network.

7. The mild cognitive impairment detection system as described in any one of claims 1 to 3, characterized in that, The classification prediction layer uses a multilayer perceptron to read out the whole-brain level brain network representation and outputs the detection result of mild cognitive impairment.

8. The mild cognitive impairment detection system as described in any one of claims 1 to 3, characterized in that, The loss function during the training process of the pre-trained mild cognitive impairment detection model based on heterogeneous graph attribute segmentation and ensemble networks includes: For a sample of T individuals, the predicted output of the mild cognitive impairment detection model is: If the actual label of the sample is y, then the cross-entropy function is used to optimize the model, defined as follows:

9. A computer-readable storage medium, characterized in that, The computer storage medium is used to store computer instructions, programs, code sets, or instruction sets, which, when run on a computer, enable the computer to perform the functions corresponding to the brain network processing module and brain network detection module in the mild cognitive impairment system as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, include: One or more processors; And a computer-readable storage medium for storing computer instructions, programs, code sets or instruction sets, which, when run on a computer, enable the one or more processors to perform the functions corresponding to the brain network processing module and the brain network detection module in any one of claims 1 to 8 of the mild cognitive impairment system.

Images