Auxiliary identification method, device and equipment for early schizophrenia and storage medium

Abstract

The invention relates to the technical field of data analysis, and discloses an auxiliary identification method and device for early schizophrenia, equipment and a storage medium, which are used for improving the auxiliary identification accuracy of schizophrenia. The auxiliary identification method for early schizophrenia comprises the following steps: classifying multi-dimensional data of a target individual, including a topological structure class, a sequence class and a non-special structure class, extracting feature information of the three classes of data by using a graph neural network, a recurrent neural network and a convolutional neural network, fusing the feature information to form a target vector, and identifying the target vector according to the feature information. And processing the target vector through a preset schizophrenia early stage identification model to obtain an identification result so as to judge the risk degree of the target individual in the schizophrenia early stage.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular, to an auxiliary recognition method, device, equipment and storage medium for early schizophrenia. Background Art

[0002] Currently, the auxiliary recognition method for early schizophrenia usually relies on a single or limited data source, such as only based on brain imaging data or gene data, and uses feature extraction and classification algorithms to identify the risk degree of an individual being in the early stage of schizophrenia.

[0003] However, as a complex mental illness, schizophrenia shows a high degree of heterogeneity in its pathogenesis and clinical manifestations, involving multiple factors at the genetic, environmental, neurobiological and psychosocial levels. Therefore, relying solely on a single data source for recognition may ignore other important factors closely related to the onset of schizophrenia, resulting in limitations in the generalization ability of the model and making it difficult to achieve accurate recognition. Summary of the Invention

[0004] The present invention provides an auxiliary recognition method, device, equipment and storage medium for early schizophrenia, so as to solve the technical problem in the prior art that due to the single data source, the generalization ability of the model is limited and it is difficult to achieve accurate recognition.

[0005] In a first aspect of the present invention, an auxiliary recognition method for early schizophrenia is provided, including: classifying the multi-dimensional comprehensive data collected for a target individual to obtain topological structure class information, sequence class information and non-special structure class information, where the multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex network, gut microbiota omics and cognitive function test levels; respectively extracting features from the topological structure class information, the sequence class information and the non-special structure class information based on a preset target graph neural network, a target recurrent neural network and a target convolutional neural network to obtain first feature information, second feature information and third feature information respectively; fusing the first feature information, the second feature information and the third feature information to obtain a target vector; and processing the target vector based on a preset early schizophrenia recognition model to obtain a recognition result, where the recognition result is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

[0006] In a feasible implementation manner, classifying the multi-dimensional comprehensive data collected for the target individual to obtain topological structure class information, sequence class information, and non-special structure class information includes: classifying the data with node and edge connection patterns in the multi-dimensional comprehensive data as topological structure class information; identifying sequence class information based on the sequentiality and time-dependence of the multi-dimensional comprehensive data; and classifying the data that does not belong to the topological structure class information and the sequence class information as non-special structure class information.

[0007] In a feasible implementation manner, before respectively extracting first feature information, second feature information, and third feature information from the topological structure class information, the sequence class information, and the non-special structure class information based on a preset target graph neural network, target recurrent neural network, and target convolutional neural network, it further includes: grouping a large amount of pre-collected data to be processed to obtain data to be processed of topological structure class, data to be processed of sequence class, and data to be processed of non-special structure class; constructing a target graph neural network according to the data to be processed of topological structure class; constructing a target recurrent neural network according to the data to be processed of sequence class; and constructing a target convolutional neural network according to the data to be processed of non-special structure class.

[0008] In a feasible implementation manner, constructing a target graph neural network according to the data to be processed of topological structure class includes: preprocessing the data to be processed of topological structure class; selecting the architecture of the graph neural network and configuring the corresponding number of graph convolutional layers, activation function, and output layer configuration; based on the configured graph neural network, using the preprocessed data to be processed of topological structure for model training, and adjusting the network parameters through the backpropagation algorithm until the preset convergence condition is reached, thereby constructing a target graph neural network capable of extracting topological structure features.

[0009] In a feasible implementation manner, constructing a target recurrent neural network according to the data to be processed of sequence class includes: analyzing the temporal characteristics of the data to be processed of sequence class to determine the sequence length, time step, and potential periodic or trend components; selecting the architecture of the recurrent neural network and configuring the corresponding number of hidden layers, unit size, and parameters of the forget gate, input gate, and output gate; based on the configured recurrent neural network, segmenting and batch-processing the analysis structure of the data to be processed of sequence class, and adopting a preset training strategy, and optimizing the network weights in combination with the cross-entropy loss function to construct a target recurrent neural network capable of capturing sequence dynamic features.

[0010] In a feasible implementation manner, constructing the target convolutional neural network according to the to-be-processed data of non-special structure class includes: preprocessing the to-be-processed data of non-special structure class; selecting the architecture of the convolutional neural network, and configuring the corresponding number of convolutional layers, convolutional kernel size, stride, pooling method, and the configuration of the fully connected layer; based on the configured convolutional neural network, using the preprocessed to-be-processed data of non-special structure class for model training, and adjusting the network weights through the backpropagation algorithm and optimizer until the model reaches a stable accuracy rate or loss value on the validation set, so as to construct a target convolutional neural network capable of extracting the features of the data of non-special structure.

[0011] In a feasible implementation manner, fusing the first feature information, the second feature information, and the third feature information to obtain a target vector includes: calculating the contribution degrees of all features in the first feature information, the second feature information, and the third feature information; removing the features with contribution degrees less than the preset threshold to obtain the processed first feature information, second feature information, and third feature information; splicing the processed first feature information, second feature information, and third feature information to obtain a target vector.

[0012] The second aspect of the present invention provides an auxiliary recognition device for early schizophrenia, including: a classification module for classifying the multi-dimensional comprehensive data collected from a target individual to obtain topological structure class information, sequence class information, and non-special structure class information, where the multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex network, gut microbiota omics, and cognitive function test levels; a feature extraction module for respectively extracting features from the topological structure class information, the sequence class information, and the non-special structure class information based on a preset target graph neural network, target recurrent neural network, and target convolutional neural network to obtain first feature information, second feature information, and third feature information respectively; a feature fusion module for fusing the first feature information, the second feature information, and the third feature information to obtain a target vector; and an identification module for processing the target vector based on a preset early schizophrenia identification model to obtain an identification result, where the identification result is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

[0013] In a feasible implementation manner, the classification module is specifically configured to: classify the data with node and edge connection patterns in the multi-dimensional comprehensive data as topological structure class information; identify sequence class information according to the sequentiality and time dependence of the multi-dimensional comprehensive data; and classify the data that does not belong to the topological structure class information and the sequence class information as non-special structure class information.

[0014] In a feasible implementation manner, the auxiliary recognition device for early schizophrenia further includes: a processing module, configured to group a large amount of pre-collected data to be processed to obtain data of a to-be-processed topological structure type, data of a to-be-processed sequence type, and data of a to-be-processed non-special-structure type; a first construction module, configured to construct a target graph neural network according to the data of the to-be-processed topological structure type; a second construction module, configured to construct a target recurrent neural network according to the data of the to-be-processed sequence type; and a third construction module, configured to construct a target convolutional neural network according to the data of the to-be-processed non-special-structure type.

[0015] In a feasible implementation manner, the first construction module is specifically configured to: preprocess the data of the to-be-processed topological structure type; select an architecture of the graph neural network, and configure corresponding numbers of graph convolutional layers, activation functions, and output layer configurations; based on the configured graph neural network, use the preprocessed to-be-processed topological structure for model training, and adjust network parameters through the backpropagation algorithm until a preset convergence condition is reached, so as to construct a target graph neural network capable of extracting topological structure features.

[0016] In a feasible implementation manner, the second construction module is specifically configured to: analyze the temporal characteristics of the data of the to-be-processed sequence type, and determine the sequence length, time step, and potential periodic or trend components; select an architecture of the recurrent neural network, and configure corresponding numbers of hidden layers, unit sizes, and parameters of forget gates, input gates, and output gates; based on the configured recurrent neural network, segment and batch process the analysis results of the data of the to-be-processed sequence type, and adopt a preset training strategy to optimize network weights in combination with the cross-entropy loss function, so as to construct a target recurrent neural network capable of capturing sequence dynamic features.

[0017] In a feasible implementation manner, the third construction module is specifically configured to: preprocess the data of the to-be-processed non-special-structure type; select an architecture of the convolutional neural network, and configure corresponding numbers of convolutional layers, convolutional kernel sizes, strides, pooling methods, and configurations of fully connected layers; based on the configured convolutional neural network, use the preprocessed to-be-processed non-special-structure type data for model training, and adjust network weights through the backpropagation algorithm and an optimizer until the model reaches a stable accuracy or loss value on the validation set, so as to construct a target convolutional neural network capable of extracting features of non-special-structure data.

[0018] In a feasible implementation, the feature fusion module is specifically used to: calculate the contribution of all features in the first feature information, the second feature information and the third feature information; eliminate features whose contribution is less than a preset threshold to obtain processed first feature information, second feature information and third feature information; splice the processed first feature information, the second feature information and the third feature information to obtain a target vector.

[0019] The third aspect of the present invention provides an auxiliary identification device for early schizophrenia, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the auxiliary identification device for early schizophrenia executes the above-mentioned auxiliary identification method for early schizophrenia.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned auxiliary identification method for early schizophrenia.

[0021] In the technical solution provided by the present invention, the multi-dimensional comprehensive data collected from the target individual are classified to obtain topological structure information, sequence information and information without special structure, wherein the multi-dimensional comprehensive data include basic information, environmental information, candidate genes, proteomics, brain complex network, intestinal flora and cognitive function test level; based on a preset target graph neural network, a target recurrent neural network and a target convolutional neural network, feature extraction is performed on the topological structure information, the sequence information and the information without special structure, respectively, to obtain first feature information, second feature information and third feature information, respectively; the first feature information, the second feature information and the third feature information are feature fused to obtain a target vector; the target vector is processed based on a preset schizophrenia early recognition model to obtain a recognition result, and the recognition result is used to indicate the risk level of the target individual in the early stage of schizophrenia. In the embodiment of the present invention, the multi-dimensional comprehensive data of the target individual is classified and processed, and feature extraction is performed on topological structure information, sequence information and information without special structure respectively using a target graph neural network, a target recurrent neural network and a target convolutional neural network, thereby achieving accurate feature capture, and then a comprehensive target vector is obtained through feature fusion, and identification is performed based on a preset schizophrenia early identification model, thereby improving the auxiliary identification accuracy of early schizophrenia. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of an embodiment of an auxiliary identification method for early schizophrenia in an embodiment of the present invention;

[0023] Figure 2 Another schematic diagram of an embodiment of the method for assisting in the identification of early schizophrenia in the embodiments of the present invention;

[0024] Figure 3 A schematic diagram of an embodiment of the device for assisting in the identification of early schizophrenia in the embodiments of the present invention;

[0025] Figure 4 Another schematic diagram of an embodiment of the device for assisting in the identification of early schizophrenia in the embodiments of the present invention;

[0026] Figure 5 A schematic diagram of an embodiment of the device for assisting in the identification of early schizophrenia in the embodiments of the present invention. Specific implementation manners

[0027] The embodiments of the present invention provide a method, a device, equipment, and a storage medium for assisting in the identification of early schizophrenia. By classifying and fusing the features of multi-dimensional data, the accuracy of early identification of schizophrenia is improved.

[0028] The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims, and the above-mentioned drawings of the present invention are used to distinguish similar objects and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or equipment that includes a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or equipment.

[0029] It can be understood that the execution subject of the present invention can be a device for assisting in the identification of early schizophrenia, or a terminal or a server. Specifically, it is not limited here. The embodiments of the present invention are described by taking the server as the execution subject as an example.

[0030] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to Figure 1 , an embodiment of the method for assisting in the identification of early schizophrenia in the embodiments of the present invention includes:

[0031] 101. Classify the multi-dimensional comprehensive data collected from the target individual to obtain topological structure class information, sequence class information, and non-special structure class information. The multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex networks, gut microbiota, and cognitive function test levels;

[0032] Basic information includes age, gender, height, family medical history, etc.; environmental information includes life stress, living environment, social relationships, etc.; candidate genes include genetic variations related to schizophrenia; proteomics includes the expression levels of proteins in the blood; brain complex networks include the connection patterns and activity states among different brain regions; gut microbiota omics includes the structure and function of the gut microbial community; cognitive function test levels include evaluations in aspects such as memory, attention, and thinking flexibility.

[0033] Basic information, environmental information, proteomics, and cognitive function test levels usually present as unstructured text or numerical data. These data lack clear topological structures or sequence characteristics and are thus classified as information of the non-special structure type; candidate gene data exhibits obvious sequence characteristics, that is, the bases in the gene sequence are arranged in a certain order. This linear structure makes candidate gene data suitable for processing methods for sequence-type information; data such as brain complex networks and gut microbiota omics present complex topological structures; brain complex networks describe the connection patterns and activity states among different brain regions. These connections form a huge network, where nodes represent brain regions and edges represent connections between regions. Gut microbiota omics data also has network characteristics and describes the interactions and relationships among gut microbial communities. This type of data is suitable for processing using graph structure methods to capture the relevance and information flow between nodes. During the classification process, specific algorithms or tools are used to identify and label each type of data.

[0034] 102. Feature extraction is respectively performed on topological structure type information, sequence type information, and non-special structure type information based on the preset target graph neural network, target recurrent neural network, and target convolutional neural network, obtaining the first feature information, the second feature information, and the third feature information respectively;

[0035] Convert topological structure type information into graph structure data, where nodes represent different entities, such as brain regions or microbial species, and edges represent connections or relationships between entities. Topological structure type information includes brain complex network and gut microbiota omics data; use convolutional operations, pooling operations, and graph attention mechanisms in the target graph neural network, etc., to learn and extract node features and global structure features in the graph structure data; then, through the transmission and fusion of multi-layer networks, obtain the key features that can represent topological structure type information, that is, the first feature information.

[0036] When processing brain complex network data, it is first constructed into a graph structure, where nodes represent different regions of the brain, such as cortical regions, hippocampus, amygdala, etc., and edges represent the functional or structural connections between these regions, such as functional connection strength, protein fiber bundles, etc.; subsequently, using a pre-set target graph neural network, the information of each node and its neighbor nodes is aggregated through graph convolutional layers. This process not only considers the attributes of the node itself, such as the activity level and metabolic level of local brain regions, but also integrates the information of other connected nodes, thereby learning and extracting the local features of each node. These features may include the activity level of the node, connection strength, information flow direction, etc.; at the same time, the pooling layer in the target graph neural network further reduces the scale of the graph by aggregating the features of adjacent nodes while retaining key information, thereby extracting higher-level global features. These features may reveal the interaction patterns between brain regions, the topological structure of the functional network, and the dynamic process of information processing; this process is iteratively carried out, and each layer further abstracts and refines the information based on the previous layer until finally extracting features that can comprehensively reflect the characteristics of the brain complex network. These features include but are not limited to the functional connection pattern between brain regions, the degree of network modularity, and the information flow efficiency.

[0037] When processing gut microbiome omics data, it is first converted into a graph structure form, where nodes represent different gut microbial species, and edges reflect the interactions or symbiotic relationships between microorganisms. Then, applying the pre-set target graph neural network, weights are dynamically assigned to different neighbor nodes of each node through the graph attention mechanism. This process considers the interaction strength between microorganisms, the dependence between species, and the influence of environmental factors, thereby more precisely capturing the mutual influence between microorganisms. The multi-layer structure of the target graph neural network gradually integrates node features and the overall structure information of the graph on this basis. In each layer, node features not only include the basic information of microbial species, such as abundance, metabolic type, etc., but also integrate the information of connected neighbor nodes and the weight information calculated through the graph attention mechanism. As the network deepens, these features are further abstracted and refined to form higher-level feature representations. These features may include but are not limited to: the relative abundance of microbial species, the structural diversity of the microbial community, the interaction network pattern between microorganisms, community stability indicators, and potential functional associations, such as metabolic pathways, immune regulation, etc.

[0038] Input the target sequence data into the target recurrent neural network. This network gradually processes each element in the sequence through its internal recurrent connection mechanism, while retaining and updating the hidden state to capture the temporal dependence and context information in the sequence. As the sequence progresses, the network gradually accumulates and refines key features, namely the second feature information.

[0039] When processing candidate gene data, the sequence data of candidate genes is used as input. The sequence data includes DNA sequences, RNA expression sequences, etc. Each nucleotide or expression value in the sequence is gradually processed through a recurrent neural network. At the same time, hidden states are used to capture the temporal dependencies and structural features in the gene sequence. During this process, the key signs of the gene sequence are learned and extracted. These features include the expression pattern of the gene, the presence and distribution of regulatory elements, the location of mutation sites and their potential effects, etc.

[0040] When extracting features from unstructured information of a specific type based on a pre-set model, first, the original unstructured data is converted into a format suitable for model processing. Then, the model is used to learn the internal representation of the data, and the data features are gradually abstracted and refined through multi-layer non-linear transformations, that is, the third feature information.

[0041] For basic data, the basic data can be converted into numerical feature vectors, and then a pre-set convolutional neural network is used for feature extraction. The network learns the local features and patterns in the basic data through convolutional layers, and reduces the feature dimension and retains key features through pooling layers. These features may include key information related to specific research or prediction tasks such as age, gender, medical history, etc.

[0042] For environmental information, it can be converted into the form of two-dimensional images or sequence data, and then a pre-set convolutional neural network is applied for feature extraction. The network captures the spatial features and temporal dependencies in the environmental information through convolutional operations, and gradually refines the key features through multi-layer convolution and pooling operations. These features may include key information such as the level of life stress, the quality of the living environment, the complexity of social relationships, etc.

[0043] For proteomics, the protein expression data can be converted into the form of two-dimensional matrices or images, and then a pre-set convolutional neural network is used for feature extraction. The network learns the local patterns and differentially expressed features in the protein expression profile through convolutional kernels, and reduces the dimension of the feature map and retains key features through pooling layers. These features may include the expression level of specific proteins, the interaction relationships between proteins, and the overall pattern of the protein expression profile, etc.

[0044] For data on the level of cognitive function tests, it is converted into the form of sequence data or one-dimensional vectors, and a pre-set convolutional neural network is applied for feature extraction. The network extracts the key features related to cognitive function by learning the temporal dependencies and pattern changes in the test data. These features may include the changing trends and patterns of test indicators such as reaction time, accuracy rate, error rate, etc.

[0045] 103. Feature fusion is performed on the first feature information, the second feature information, and the third feature information to obtain a target vector;

[0046] The graph structure features are transformed into low - dimensional vectors using graph embedding techniques, and the sequence features are transformed into fixed - length feature vectors using recurrent neural networks. At the same time, the unstructured data features are transformed into feature vectors according to the data type through numerical conversion, two - dimensional matrix or image representation, and convolutional neural network / recurrent neural network processing respectively. A feature concatenation strategy is adopted, that is, these feature vectors from different sources are directly concatenated end - to - end to form a long vector that synthesizes various types of feature information as the final target vector. The target vector synthesizes the topological structure - type information, sequence - type information, and non - special - structure - type information features extracted from multi - dimensional comprehensive data, and these features together constitute a comprehensive description of the mental health status of the target individual.

[0047] 104. Process the target vector based on a pre - set early schizophrenia recognition model to obtain a recognition result, and the recognition result is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

[0048] Input the target vector into the model. Inside the model, the target vector is non - linearly transformed and feature - extracted through a multi - layer neural network structure, and then a classification algorithm is used to analyze and judge the extracted features. Finally, a recognition result is output, which is usually represented in the form of a probability or a score and is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

[0049] The construction method of the early schizophrenia recognition model can be: collect and pre - process the multi - dimensional comprehensive data of early schizophrenia patients and non - patients, and then use multiple neural network architectures, including graph neural networks, recurrent neural networks, and convolutional neural networks, to process topological structures, sequences, and non - special - structure information respectively for feature extraction and fusion. Then, train a machine learning model through optimization strategies such as backpropagation algorithm and gradient descent, continuously adjust the network weights during the iteration process to minimize the prediction error until the model achieves stable recognition performance on the validation set, and finally output a probability or score that can quantitatively represent the early schizophrenia risk degree of the target individual.

[0050] In the embodiments of the present invention, by classifying and processing the multi - dimensional comprehensive data of the target individual, using the target graph neural network, target recurrent neural network, and target convolutional neural network to extract features for topological structure - type information, sequence - type information, and non - special - structure - type information respectively, accurate feature capture is achieved. Furthermore, a comprehensive target vector is obtained through feature fusion, and based on the pre - set early schizophrenia recognition model, the risk degree of the target individual being in the early stage of schizophrenia is identified, improving the accuracy of the auxiliary recognition of early schizophrenia.

[0051] Please refer to Figure 2 , another embodiment of the auxiliary recognition method for early schizophrenia in the embodiments of the present invention includes:

[0052] 201. Classify the multi-dimensional comprehensive data collected from the target individual to obtain topological structure class information, sequence class information, and non-special structure class information. The multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex network, gut microbiota omics, and cognitive function test levels.

[0053] Divide the data with node and edge connection patterns in the multi-dimensional comprehensive data into topological structure class information; identify the sequence class information based on the sequentiality and time-dependence of the multi-dimensional comprehensive data; divide the data that does not belong to the topological structure class information and sequence class information into non-special structure class information.

[0054] First, identify the data that contains clear node and edge connection patterns, such as brain complex network and gut microbiota omics data, which exhibit typical topological structure characteristics and are thus classified as topological structure class information; second, for the data with sequentiality and time-dependence, such as candidate genes, identify them as sequence class information; finally, for the other data that neither belongs to the topological structure class nor conforms to the sequentiality characteristics, such as basic information, environmental information, proteomics, and cognitive function test levels, which have no specific structure or sequence pattern, they are divided into non-special structure class information.

[0055] 202. Group a large amount of pre-collected data to be processed to obtain data to be processed for topological structure class, data to be processed for sequence class, and data to be processed for non-special structure class.

[0056] A large amount of data to be processed has been pre-collected. These data include the to-be-processed multi-dimensional comprehensive information from early-stage schizophrenia patients and non-patients. The to-be-processed multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex network, gut microbiota omics, and cognitive function test levels. These data are sorted and grouped into data to be processed for topological structure class, data to be processed for sequence class, and data to be processed for non-special structure class. Specifically, the brain complex network and gut microbiota omics of early-stage schizophrenia patients and non-patients are divided into data to be processed for topological structure class, the candidate genes of early-stage schizophrenia patients and non-patients are divided into data to be processed for sequence class, and the basic information, environmental information, proteomics, and cognitive function test levels are divided into data to be processed for non-special structure class.

[0057] 203. Construct a target graph neural network based on the data to be processed for topological structure class.

[0058] Preprocess the data of the topological structure class to be processed; select the architecture of the graph neural network, and configure the corresponding number of graph convolutional layers, activation functions, and output layer configurations; based on the configured graph neural network, use the preprocessed topological structure to be processed for model training, and adjust the network parameters through the backpropagation algorithm until the preset convergence condition is reached, thereby constructing a target graph neural network capable of extracting topological structure features.

[0059] Preprocess the data of the topological structure class to be processed, including cleaning the data, removing noise, and structuring the data into a graph form, where nodes represent brain regions or gut microbial species, and edges represent the connections or interactions between them; then, select a suitable graph neural network architecture according to the characteristics of the data and the task requirements, such as the graph convolutional network (GCN) or the graph attention network (GAT). When configuring the graph neural network, adjust the number of graph convolutional layers to capture topological features at different levels, select suitable activation functions such as ReLU or Sigmoid to introduce non-linearity, and design the output layer to match the requirements of classification or regression tasks; then, use the preprocessed data of the topological structure to be processed for model training, calculate the predicted values through forward propagation, and adjust the network parameters through the backpropagation algorithm to minimize the loss function; during the training process, monitor the performance of the model, such as accuracy, recall, or F1 score, and adjust hyperparameters such as the learning rate and batch size in a timely manner until the model reaches the preset convergence condition; finally, construct a target graph neural network capable of accurately extracting topological structure features.

[0060] 204. Construct a target recurrent neural network according to the sequence class data to be processed;

[0061] Analyze the temporal characteristics of the sequence class data to be processed, determine the sequence length, time step, and potential periodic or trend components; select the architecture of the recurrent neural network, and configure the corresponding number of hidden layers, unit sizes, and parameters of the forget gate, input gate, and output gate; based on the configured recurrent neural network, segment and batch process the analysis structure of the sequence class data to be processed, and adopt a preset training strategy to optimize the network weights in combination with the cross-entropy loss function, thereby constructing a target recurrent neural network capable of capturing the dynamic features of the sequence.

[0062] Select the architecture of a recurrent neural network, such as a long short-term memory network or a gated recurrent unit. These network structures are particularly good at handling long-term dependencies in long sequence data. When configuring the recurrent neural network, adjust the number of hidden layers, the size of the units, and the parameters of the forget gate, input gate, and output gate to optimize the memory and expressive capabilities of the model. Then, split the sequence-like data to be processed into a training set, a validation set, and a test set, and perform batch processing to accelerate the training process. During the training process, adopt a preset training strategy, such as gradually reducing the learning rate or using early stopping to avoid overfitting, and combine it with the cross-entropy loss function to optimize the network weights. Through continuous iterative training, finally construct a target recurrent neural network that can accurately capture the dynamic features in the candidate gene sequences.

[0063] 205. Construct a target convolutional neural network based on the data of the class without special structure to be processed;

[0064] Preprocess the data of the class without special structure to be processed; select the architecture of the convolutional neural network and configure the number of convolutional layers, the size of the convolutional kernels, the stride, the pooling method, and the configuration of the fully connected layer; based on the configured convolutional neural network, use the preprocessed data of the class without special structure to be processed for model training, and adjust the network weights through the backpropagation algorithm and an optimizer until the model reaches a stable accuracy or loss value on the validation set, so as to construct a target convolutional neural network that can extract the features of the data without special structure.

[0065] Preprocess the data of the class without special structure to be processed. The data of the class without special structure to be processed includes the basic information, environmental information, proteomics, and cognitive function test levels of early schizophrenia patients and non-patients. The preprocessing includes data cleaning, numerical classification of categorical variables, standardization or normalization of numerical variables, etc. Subsequently, select and configure the architecture of the convolutional neural network, including determining the number and configuration of convolutional layers, selecting the pooling method, and designing the structure of the fully connected layer. Then, use the preprocessed data to train the configured convolutional neural network, and continuously adjust the network weights through the backpropagation algorithm and a suitable optimizer until the accuracy of the model on the validation set tends to be stable or the loss value reaches a preset threshold, thereby constructing a target convolutional neural network that can effectively extract and represent the key features in the data without special structure.

[0066] 206. Based on the preset target graph neural network, target recurrent neural network, and target convolutional neural network, extract features from the topological structure class information, sequence class information, and non-special structure class information respectively, and obtain the first feature information, the second feature information, and the third feature information;

[0067] The execution process of step 206 is similar to that of the above step 102, and will not be elaborated here.

[0068] 207. Fusing the first feature information, the second feature information, and the third feature information to obtain a target vector;

[0069] Calculate the contribution of all features in the first feature information, the second feature information, and the third feature information; remove features whose contribution is less than a preset threshold to obtain processed first feature information, second feature information, and third feature information; concatenate the processed first feature information, second feature information, and third feature information to obtain a target vector.

[0070] Contribution is a quantitative indicator used to evaluate the importance or effectiveness of each feature in the schizophrenia identification task. The feature importance evaluation method can be used to calculate the contribution of each feature. A statistical feature selection method can be used. Specifically, the GBDT model is used to accumulate the contribution of each feature in reducing the loss function during the iteration process to obtain the importance ranking of the features. At the same time, the chi-square statistic or mutual information value between each feature and the schizophrenia label can also be calculated to quantify the strength of the association between them. Through these methods, the contribution of each feature to schizophrenia identification can be measured.

[0071] 208. Process the target vector based on a preset schizophrenia early stage identification model to obtain an identification result, where the identification result is used to indicate the risk level of the target individual being in the early stage of schizophrenia.

[0072] The method for constructing an early identification model for schizophrenia can be: collecting and processing multi-dimensional comprehensive data of early schizophrenia patients and non-patients, respectively targeting topological structure information, sequence information and information without special structure, using target graph neural network, target recurrent neural network and target convolutional neural network to extract features and then perform feature fusion, using the fused feature vector to train the machine learning model, and through continuous iteration to optimize the model parameters, finally obtaining a prediction model that can accurately assess the risk level of the target individual in the early stage of schizophrenia.

[0073] In an embodiment of the present invention, by classifying multi-dimensional comprehensive data into topological structure type, sequence type and information without special structure type, and constructing target graph neural network, target recurrent neural network and target convolutional neural network for feature extraction, full use is made of multi-source data such as basic information, environmental information, candidate genes, proteomics, brain complex network, intestinal flora group biology and cognitive function test level, and the feature information extracted by different networks is integrated into a comprehensive target vector through feature fusion technology, and then the risk level of the target individual in the early stage of schizophrenia is identified based on the preset schizophrenia early identification model, thereby improving the accuracy and reliability of auxiliary identification of early schizophrenia.

[0074] The above describes the method for assisting in the identification of early schizophrenia in the embodiments of the present invention. Next, the device for assisting in the identification of early schizophrenia in the embodiments of the present invention will be described. Please refer to Figure 3 One embodiment of the device for assisting in the identification of early schizophrenia in the embodiments of the present invention includes:

[0075] A classification module 301, configured to classify the multi-dimensional comprehensive data of a target individual to obtain topological structure class information, sequence class information, and non-special structure class information. The multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex networks, gut microbiota omics, and cognitive function test levels;

[0076] A feature extraction module 302, configured to respectively perform feature extraction on the topological structure class information, the sequence class information, and the non-special structure class information based on a preset target graph neural network, a target recurrent neural network, and a target convolutional neural network, and respectively obtain first feature information, second feature information, and third feature information;

[0077] A feature fusion module 303, configured to perform feature fusion on the first feature information, the second feature information, and the third feature information to obtain a target vector;

[0078] An identification module 304, configured to process the target vector based on a preset early schizophrenia identification model to obtain an identification result, where the identification result is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

[0079] In the embodiments of the present invention, by classifying the multi-dimensional comprehensive data of the target individual, and using the target graph neural network, the target recurrent neural network, and the target convolutional neural network to respectively perform feature extraction on the topological structure class information, the sequence class information, and the non-special structure class information, accurate capture of features is achieved. Furthermore, a comprehensive target vector is obtained through feature fusion, and the risk degree of the target individual being in the early stage of schizophrenia is identified based on the preset early schizophrenia identification model, improving the accuracy of the assisted identification of early schizophrenia.

[0080] Please refer to Figure 4 Another embodiment of the device for assisting in the identification of early schizophrenia in the embodiments of the present invention includes:

[0081] A classification module 301, configured to classify the multi-dimensional comprehensive data of a target individual to obtain topological structure class information, sequence class information, and non-special structure class information. The multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex networks, gut microbiota omics, and cognitive function test levels;

[0082] A feature extraction module 302 is configured to perform feature extraction on the topological structure type information, the sequence type information, and the non-special structure type information respectively based on a preset target graph neural network, a target recurrent neural network, and a target convolutional neural network, and obtain first feature information, second feature information, and third feature information respectively;

[0083] A feature fusion module 303 is configured to perform feature fusion on the first feature information, the second feature information, and the third feature information to obtain a target vector;

[0084] An identification module 304 is configured to process the target vector based on a preset early schizophrenia identification model to obtain an identification result, and the identification result is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

[0085] Optionally, the classification module 301 may be specifically configured to:

[0086] Classify the data with node and edge connection patterns in the multi-dimensional comprehensive data as topological structure type information; identify the sequence type information according to the sequentiality and time-dependency of the multi-dimensional comprehensive data; classify the data that does not belong to the topological structure type information and the sequence type information as non-special structure type information.

[0087] Optionally, the early schizophrenia auxiliary identification device further includes:

[0088] A processing module 305 is configured to group a large amount of pre-collected data to be processed to obtain data to be processed of topological structure type, data to be processed of sequence type, and data to be processed of non-special structure type;

[0089] A first construction module 306 is configured to construct a target graph neural network according to the data to be processed of topological structure type;

[0090] A second construction module 307 is configured to construct a target recurrent neural network according to the data to be processed of sequence type;

[0091] A third construction module 308 is configured to construct a target convolutional neural network according to the data to be processed of non-special structure type.

[0092] Optionally, the first construction module 306 may be specifically configured to:

[0093] Preprocess the data to be processed of topological structure type; select the architecture of the graph neural network, and configure the corresponding number of graph convolutional layers, activation function, and output layer configuration; based on the configured graph neural network, use the preprocessed data to be processed of topological structure for model training, and adjust the network parameters through the backpropagation algorithm until the preset convergence condition is reached, so as to construct a target graph neural network capable of extracting topological structure features.

[0094] Optionally, the second construction module 307 can be specifically used for:

[0095] Analyze the temporal characteristics of the sequence-like data to be processed, determine the sequence length, time step, and potential periodic or trend components; select the architecture of the recurrent neural network, and configure the corresponding number of hidden layers, unit size, and parameters of the forget gate, input gate, and output gate; based on the configured recurrent neural network, segment and batch process the analysis structure of the sequence-like data to be processed, and adopt a preset training strategy to optimize the network weights in combination with the cross-entropy loss function, so as to construct a target recurrent neural network capable of capturing the dynamic characteristics of the sequence.

[0096] Optionally, the third construction module 308 can be specifically used for:

[0097] Preprocess the data of the non-special-structure type to be processed; select the architecture of the convolutional neural network, and configure the corresponding number of convolutional layers, convolutional kernel size, stride, pooling method, and the configuration of the fully connected layer; based on the configured convolutional neural network, use the preprocessed data of the non-special-structure type to be processed for model training, and adjust the network weights through the backpropagation algorithm and optimizer until the model reaches a stable accuracy or loss value on the validation set, so as to construct a target convolutional neural network capable of extracting the characteristics of the non-special-structure data.

[0098] Optionally, the feature fusion module 303 can be specifically used for: calculating the contribution degrees of all features in the first feature information, second feature information, and third feature information; removing the features with contribution degrees less than the preset threshold to obtain the processed first feature information, second feature information, and third feature information; splicing the processed first feature information, second feature information, and third feature information to obtain a target vector.

[0099] In the embodiment of the present invention, by classifying the multi-dimensional comprehensive data into topological structure type, sequence type, and non-special-structure type information, and specifically constructing a target graph neural network, a target recurrent neural network, and a target convolutional neural network for feature extraction, multi-source data such as basic materials, environmental information, candidate genes, proteomics, brain complex networks, gut microbiota, and cognitive function test levels are fully utilized, and the feature information extracted by different networks is integrated into a comprehensive target vector through the feature fusion technology. Furthermore, based on the preset early schizophrenia recognition model, the risk degree of the target individual being in the early stage of schizophrenia is recognized, improving the accuracy and reliability of the early schizophrenia assisted recognition.

[0100] Above Figure 3 and Figure 4The auxiliary recognition device for early schizophrenia in the embodiments of the present invention is described in detail from the perspective of modular functional entities. Next, the auxiliary recognition device for early schizophrenia in the embodiments of the present invention is described in detail from the perspective of hardware processing.

[0101] See Figure 5 As shown, the auxiliary recognition device for early schizophrenia includes a processor 500 and a memory 501. The memory 501 stores machine-executable instructions that can be executed by the processor 500, and the processor 500 executes the machine-executable instructions to implement the above-mentioned auxiliary recognition method for early schizophrenia.

[0102] Furthermore, Figure 5 The auxiliary recognition device for early schizophrenia shown further includes a bus 502 and a communication interface 503. The processor 500, the communication interface 503, and the memory 501 are connected through the bus 502.

[0103] Among them, the memory 501 may include a high-speed random access memory (Random Access Memory, RAM), and may also include a non-volatile memory, for example, at least one disk memory. The communication connection between this system network element and at least one other network element is realized through at least one communication interface 503 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 502 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 5 only a single bidirectional arrow is used in

[0104] The processor 500 may be an integrated circuit chip with the ability to process signals. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 500 or the instructions in the form of software. The above-mentioned processor 500 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present disclosure can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by a combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 501, and the processor 500 reads the information in the memory 501 and combines its hardware to complete the method steps of the foregoing embodiments.

[0105] The present invention also provides an auxiliary recognition device for early schizophrenia. The computer device includes a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the processor executes the steps of the auxiliary recognition method for early schizophrenia in the above-mentioned various embodiments.

[0106] The present invention also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium, and the computer-readable storage medium may also be a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer executes the steps of the auxiliary recognition method for early schizophrenia.

[0107] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0108] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0109] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An auxiliary recognition method for early schizophrenia, characterized in that, The auxiliary recognition method for early schizophrenia includes: Classify the multi-dimensional comprehensive data collected from the target individual to obtain topological structure class information, sequence class information, and non-special structure class information. The multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex network, gut microbiota omics, and cognitive function test levels; Based on the preset target graph neural network, target recurrent neural network, and target convolutional neural network, respectively extract features from the topological structure class information, the sequence class information, and the non-special structure class information, and obtain the first feature information, the second feature information, and the third feature information respectively; Fuse the first feature information, the second feature information, and the third feature information to obtain a target vector; Based on the preset early schizophrenia recognition model, process the target vector to obtain a recognition result, and the recognition result is used to indicate the risk degree of the target individual being in the early stage of schizophrenia.

2. The auxiliary recognition method for early schizophrenia according to claim 1, characterized in that, The classifying the multi-dimensional comprehensive data collected from the target individual to obtain topological structure class information, sequence class information, and non-special structure class information includes: Classify the data with node and edge connection patterns in the multi-dimensional comprehensive data collected from the target individual into topological structure class information; Identify the sequence class information according to the sequentiality and time dependence of the multi-dimensional comprehensive data; Classify the data that does not belong to the topological structure class information and the sequence class information into non-special structure class information.

3. The auxiliary recognition method for early schizophrenia according to claim 1, characterized in that, Before respectively extracting features from the topological structure class information, the sequence class information, and the non-special structure class information based on the preset target graph neural network, target recurrent neural network, and target convolutional neural network, and obtaining the first feature information, the second feature information, and the third feature information respectively, it further includes: Group a large amount of pre-collected data to be processed to obtain data to be processed for topological structure class, data to be processed for sequence class, and data to be processed for non-special structure class; Construct a target graph neural network according to the data to be processed for topological structure class; Construct a target recurrent neural network according to the data to be processed for sequence class; Construct a target convolutional neural network according to the data to be processed for non-special structure class.

4. The auxiliary recognition method for early schizophrenia according to claim 3, characterized in that, The constructing a target graph neural network according to the data to be processed for topological structure class includes: Preprocess the data to be processed for topological structure class; Select the architecture of the graph neural network, and configure the corresponding number of graph convolutional layers, activation function, and output layer configuration; Based on the configured graph neural network, use the preprocessed data to be processed for topological structure for model training, and adjust the network parameters through the backpropagation algorithm until the preset convergence condition is reached, so as to construct a target graph neural network capable of extracting topological structure features.

5. The auxiliary recognition method for early schizophrenia according to claim 3, wherein, The constructing a target recurrent neural network according to the data to be processed for sequence class includes: Analyze the temporal characteristics of the data to be processed for sequence class, and determine the sequence length, time step, and potential periodic or trend components; Select the architecture of the recurrent neural network, and configure the corresponding number of hidden layers, unit size, and parameters of the forget gate, input gate, and output gate; Based on the configured recurrent neural network, the analysis structure of the sequence data to be processed is segmented and batched, and the preset training strategy is adopted, combined with the cross entropy loss function to optimize the network weights, to construct a target recurrent neural network that can capture the dynamic characteristics of the sequence.

6. The auxiliary recognition method for early schizophrenia according to claim 3, wherein, The step of constructing a target convolutional neural network according to the to-be-processed data without special structure includes: Preprocessing the to-be-processed data without special structure; Select the architecture of the convolutional neural network and configure the corresponding number of convolutional layers, convolution kernel size, step size, pooling method, and configuration of the fully connected layer; Based on the configured convolutional neural network, the preprocessed data without special structure is used for model training. The network weights are adjusted through the back propagation algorithm and optimizer until the model reaches a stable accuracy or loss value on the validation set, so as to construct a target convolutional neural network that can extract features of data without special structure.

7. The method for auxiliary identification of early schizophrenia according to any one of claims 1-6, characterized in that, The step of fusing the first feature information, the second feature information, and the third feature information to obtain a target vector includes: Calculating the contribution of all features in the first feature information, the second feature information, and the third feature information; Eliminate features whose contribution is less than a preset threshold to obtain processed first feature information, second feature information, and third feature information; The processed first feature information, the second feature information and the third feature information are concatenated to obtain a target vector.

8. An auxiliary recognition device for early schizophrenia, characterized in that, The auxiliary identification device for early schizophrenia comprises: A classification module is used to classify the multi-dimensional comprehensive data collected from the target individual to obtain topological structure information, sequence information and information without special structure. The multi-dimensional comprehensive data includes basic information, environmental information, candidate genes, proteomics, brain complex network, intestinal flora and cognitive function test level; A feature extraction module is used to extract features of the topological structure information, the sequence information and the information without special structure based on a preset target graph neural network, a target recurrent neural network and a target convolutional neural network, respectively, to obtain first feature information, second feature information and third feature information respectively; A feature fusion module, used for fusing the first feature information, the second feature information and the third feature information to obtain a target vector; The recognition module is used to process the target vector based on a preset schizophrenia early recognition model to obtain a recognition result, wherein the recognition result is used to indicate the risk level of the target individual in the early stage of schizophrenia.

9. An auxiliary recognition device for early schizophrenia, characterized in that, The auxiliary identification device for early schizophrenia includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the device for assisting in identifying early-stage schizophrenia to execute the method for assisting in identifying early-stage schizophrenia as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by a processor, the auxiliary identification method for early schizophrenia according to any one of claims 1 to 7 is implemented.

Images