A tumor tissue molecular intelligent analysis and recognition system based on data fusion analysis

Abstract

The present invention discloses a tumor tissue molecular intelligent analysis and identification system based on data fusion analysis, which relates to the technical field of tumor tissue molecular intelligent analysis, in order to solve the problem of inaccurate patient tumor data acquisition in the prior art. The present invention extracts biomarkers related to tumor markers from key biomarkers, realizes accurate identification of tumor-related markers, helps to deeply understand the molecular mechanism and characteristics of tumors, and uses molecular feature recognition models to feature-associate tumor-related markers with key biomarkers. It reveals the intrinsic connection between biomarkers and tumor molecules, provides more comprehensive information for tumor diagnosis and prognosis assessment, and uses a hypergraph neural network model to perform data fusion analysis, which can process node data and hyperedge data containing complex relationships.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent analysis of tumor tissue molecules, and in particular to an intelligent analysis and identification system of tumor tissue molecules based on data fusion analysis. Background Art

[0002] Molecular intelligent analysis of tumor tissue is a comprehensive analysis method based on molecular diagnostic technology and artificial intelligence technology. It aims to provide accurate and reliable basis for tumor diagnosis, treatment and prognosis evaluation by in-depth study of the molecular characteristics in tumor tissue.

[0003] The patent application with announcement number CN117831624A discloses a tumor mutation analysis method based on a tumor molecular diagnosis knowledge base. It mainly extracts information data about miRNA and methylation in the tumor genome map, obtains the miRNA expression spectrum of the tumor group and the normal control group, and performs differential expression and correlation analysis, so that it can accurately reveal the gene expression differences and correlations between the tumor group and the normal control group, and then establishes a tumor state judgment model based on the results of differential expression and correlation analysis. At the same time, by obtaining methylation chip data to estimate tumor tissue decomposition, combined with a tumor mutation prediction model constructed by a prediction algorithm, tumor mutations are efficiently predicted, so that tumor mutation analysis can be performed intuitively, greatly improving the efficiency of tumor mutation diagnosis. Although the above patent solves the problem of tumor molecular analysis, the following problems still exist in actual operation:

[0004] 1. After obtaining the patient's medical data, the data is not processed initially, resulting in poor data quality and inability to be integrated.

[0005] 2. Failure to conduct more detailed tumor analysis on the patient’s data resulted in inaccurate diagnosis of the patient’s medical data, leading to poor treatment outcomes.

[0006] 3. The patient’s tumor information is not used for more intelligent molecular judgment and the tumor is not risk assessed, resulting in the patient’s inability to understand the disease in detail. Summary of the invention

[0007] The purpose of the present invention is to provide a tumor tissue molecular intelligent analysis and identification system based on data fusion analysis, extract biomarkers related to tumor markers from key biomarkers, achieve accurate identification of tumor-related markers, help to deeply understand the molecular mechanism and characteristics of tumors, and use molecular feature recognition models to feature-associate tumor-related markers with key biomarkers. The intrinsic connection between biomarkers and tumor molecules is revealed, providing more comprehensive information for tumor diagnosis and prognosis assessment. Doctors can formulate more accurate treatment plans based on patients' molecular feature data and risk assessment results. The hypergraph neural network model is used for data fusion analysis, which can process node data and hyperedge data containing complex relationships, and can solve problems in the prior art.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A tumor tissue molecular intelligent analysis and recognition system based on data fusion analysis, comprising:

[0010] Multi-source heterogeneous data access unit, used for:

[0011] Confirm the patient's medical data from the database, manage the acquired medical data, and mark the medical data as pending medical data after data management;

[0012] Intelligent data pre-processing unit for:

[0013] Performing data quality assessment and data preprocessing on the medical data to be processed, and marking the medical data to be processed after the data quality assessment and data preprocessing are completed as target medical data;

[0014] Deep Fusion Analysis Unit for:

[0015] The target medical data is fused using a hypergraph neural network model, data representation is generated for the fused target medical data, and the generated data representation is annotated as comprehensive representation data;

[0016] Molecular Characterization Unit for:

[0017] Extract key biomarkers from the comprehensive characterization data, perform molecular feature identification on the extracted key biomarkers and tumor molecular features, and mark the molecular features identified in the key biomarkers as molecular feature data to be processed;

[0018] Intelligent diagnostic classification unit for:

[0019] The molecular feature data to be processed are classified into molecular feature categories, risk analysis is performed on the classified molecular features respectively, and the risk analysis results of the molecular features are integrated to obtain standard tumor molecular data after the data integration is completed;

[0020] Interactive result presentation unit for:

[0021] The standard tumor molecular data is converted into a visualization and then transmitted to the mobile terminal for display. At the same time, the patient can view and interact on the mobile terminal.

[0022] Preferably, the multi-source heterogeneous data access unit is further used for:

[0023] Patients’ medical data include clinical medical record data, gene sequencing data, proteomics data, and imaging medical data;

[0024] Among them, clinical medical record data are patients’ medical history, diagnosis, treatment plan and efficacy evaluation data; gene sequencing data are patients’ tumor gene sequencing data; proteomics data are protein expression data of tumor tissue samples; and imaging medical data are patients’ tumor imaging data;

[0025] The acquired clinical medical record data, gene sequencing data, proteomics data and imaging medical data were standardized using the OMOP data model;

[0026] After data standardization is completed, use the data to write scripts for data integration;

[0027] After data integration is completed, the data is converted into a unified data format;

[0028] Finally, clinical medical record data, gene sequencing data, proteomics data and imaging medical data with unified data format are obtained and marked as medical data to be processed.

[0029] Preferably, the intelligent data preprocessing unit is further used for:

[0030] Conduct data quality assessment on the medical data to be processed first;

[0031] The data quality assessment process is to verify the clinical medical record data in the medical data to be processed, and check the patient's diagnosis results and medication records; use FastQC to check the original data of the gene sequencing data in the medical data to be processed, mainly checking the sequencing depth, coverage and sequencing error rate, then remove the connector sequences in the data, filter low-quality sequences and trim the terminal bases, and finally eliminate the contaminated sequences in the data; analyze the distribution of peptide matching errors, peptide number, peptide length and protein molecular weight of the proteomics data in the medical data to be processed; evaluate the image clarity, noise level and no reference image of the imaging data in the medical data to be processed;

[0032] Perform data preprocessing on the pending medical data that has completed data quality assessment;

[0033] The data preprocessing process is to sequentially perform data cleaning, data integration, data transformation and data standardization on the text data in the medical data to be processed; and sequentially perform grayscale adjustment, geometric transformation, image enhancement, filtering, denoising and image normalization on the image data in the medical data to be processed;

[0034] Finally, the unprocessed medical data that has completed data quality assessment and data preprocessing are labeled as target medical data.

[0035] Preferably, the in-depth fusion analysis unit is further used for:

[0036] Convert clinical medical record data, gene sequencing data, proteomics data and imaging medical data in the target medical data into a unified representation form;

[0037] Among them, clinical medical record data is converted into structured or semi-structured form;

[0038] Gene sequencing data is represented as a graph structure;

[0039] Proteomic data are represented as functional networks of interactions between proteins;

[0040] Key features are extracted from medical imaging data through a deep learning model and represented as a graph structure, where the key features are characteristic data of tumor location and size.

[0041] Preferably, the in-depth fusion analysis unit is further used for:

[0042] After the representation of the target medical data is unified, the target medical data is imported into the hypergraph neural network model, wherein the hypergraph neural network model is retrieved from the database;

[0043] Confirm the data points in the hypergraph neural network model according to the target medical data;

[0044] The data points in the hypergraph neural network model include node data and hyperedge data. The node data are patient data points, gene data points, protein data and image feature data points in the target medical data, and the hyperedge data are data points of the relationship between the data in the target medical data.

[0045] Adjusting model parameters in the hypergraph neural network model according to data points in the hypergraph neural network model, wherein the model parameters include the number of layers, node embedding dimension, hyperedge embedding dimension, and activation function of the hypergraph neural network model;

[0046] Perform model training on the Hypergraph Neural Network model after adjusting the model parameters, and determine whether to perform model tuning on the Hypergraph Neural Network model based on the training results;

[0047] Among them, when the evaluation indicators in the training results are within the qualified range, the model tuning is not performed; when the evaluation indicators in the training results are not within the qualified range, the model tuning is performed;

[0048] Generate data representation for the trained or optimized hypergraph neural network model;

[0049] Among them, the data representation generation process is to pool the node data and hyperedge data in the trained hypergraph neural network model or the tuned hypergraph neural network model, and after the pooling operation, obtain the comprehensive representation data of the trained hypergraph neural network model or the tuned hypergraph neural network model.

[0050] Preferably, when the evaluation index in the training result is not within the qualified range, the in-depth fusion analysis unit performs model tuning, including:

[0051] The data points in the hypergraph neural network model are randomly divided into a preset number of data point groups in equal proportion, and the data point groups are randomly sorted to obtain a data point group sequence;

[0052] Based on the data point group sequence, the first data point group is input into the hypergraph neural network model to obtain a first result, and based on a preset model tuning rule, an optimized value of the first result for the model parameter is obtained, and the optimized value is added to the judgment process of the second data point group to obtain a second result;

[0053] According to the above-mentioned method for obtaining the second result, the data point group sequence is judged based on the hypergraph neural network model cycle to obtain a result sequence, and the result change law of the result sequence is obtained, and the numerical sequence is optimized;

[0054] Based on the result variation rule, each value in the optimized value sequence is evaluated, the optimization effect value of each value is determined, and an optimization effect sequence is obtained;

[0055] The data point groups are randomly sorted multiple times without repetition to obtain the optimization effect sequence after each sorting, forming an optimization effect matrix, and the optimization value matrix is ​​obtained based on the data point group sequence sorted each time;

[0056] Matching the optimization value matrix with the optimization effect matrix to obtain a matrix matching result, and determining the optimization effects of different optimization values ​​under different model parameters according to the matrix matching result;

[0057] Based on the optimization effects of different optimization values ​​under different model parameters, the optimization values ​​corresponding to the best optimization effect under the model parameters closest to the model parameters of the hypergraph neural network model are selected to tune the hypergraph neural network model.

[0058] Preferably, the molecular feature recognition unit is further used for:

[0059] The comprehensive representation data is used to select feature data highly related to the disease from the fused data using the feature selection mechanism in the hypergraph neural network, where the feature selection mechanism is a ranking aggregation method;

[0060] Determine key biomarkers based on the results of feature selection, where key biomarkers are specific gene variations, protein expression patterns, and imaging features;

[0061] The confirmed key biomarkers are identified as tumor-related markers, wherein the tumor-related markers are identified as biomarkers that are related to tumor markers among the key biomarkers;

[0062] The tumor-related markers and key biomarkers are associated with the characteristics of biomarkers and tumor molecules through molecular feature recognition models;

[0063] After feature association, the molecular features identified in the key biomarkers are obtained and marked as molecular feature data to be processed.

[0064] Preferably, the molecular feature recognition unit associates the tumor-related markers with the key biomarkers through a molecular feature recognition model to perform feature association between the biomarkers and tumor molecules, including:

[0065] Inputting the tumor-related marker into the molecular feature recognition model to obtain a first tumor molecule corresponding to the tumor-related marker, and determining a first region of the first tumor molecule;

[0066] Inputting the key biomarker into the molecular feature recognition model, obtaining a second tumor molecule corresponding to the key biomarker, and determining a second region of the second tumor molecule;

[0067] obtaining the same tumor molecule between the first tumor molecule and the second tumor molecule based on the first unique tumor molecule in the first tumor molecule and the second unique tumor molecule in the second tumor molecule;

[0068] Acquire a first region feature and a second region feature of the same tumor molecule, acquire the same features and different features of the first region feature and the second region feature, re-input the markers corresponding to the different features into the molecular feature recognition model, and determine whether the different features are the final tumor molecules according to the output results;

[0069] Based on the same features and the judgment results of different features, the first target tumor molecule finally left by the same tumor molecule is determined;

[0070] intercepting a first marker corresponding to a first specific tumor molecule from the tumor-related marker, re-inputting the first marker into a molecular feature recognition model, and determining whether the first specific tumor molecule is a final tumor molecule;

[0071] Extracting a second marker corresponding to a second specific tumor molecule from the key biomarker, re-inputting the second marker into the molecular feature recognition model, and determining whether the second specific tumor molecule is the final tumor molecule;

[0072] Determining the second target tumor molecule that is finally left based on the judgment results of the first specific tumor molecule and the second specific tumor molecule;

[0073] The first target tumor molecule and the second target tumor molecule are characteristically associated with biomarkers.

[0074] Preferably, the intelligent diagnosis classification unit is further used for:

[0075] The molecular feature data to be processed are subjected to data dimension reduction using t-SNE technology;

[0076] The molecular feature data to be processed after dimension reduction is clustered into different clusters using a hierarchical clustering method, wherein different clusters are different molecular subtypes.

[0077] After the tumor samples are clustered into different clusters, the key feature samples in the clustered tumor samples are screened according to the coefficient of variation;

[0078] The screened samples with key characteristics are molecularly typed, and the classified molecular characteristic data are obtained after molecular typing.

[0079] Preferably, the intelligent diagnosis classification unit processes the molecular feature data using t-SNE technology to perform data dimension reduction as follows:

[0080] The molecular characteristic data to be processed are divided based on data variable units to obtain n initial data variables. The initial data variables are standardized based on the mean and variance of the data variables to obtain standard data variables. Based on the standard data variables, a standard variable matrix is ​​obtained. ;

[0081] The central function value of the standard data variable and the initial data variable is calculated according to the following formula ;

[0082] ;

[0083] in, represents the variable value of the i-th initial data variable, represents the variable value of the jth standard data variable, The data processing coefficient represents the molecular characteristic data to be processed, and its value is (0,1);

[0084] Based on the central function value of all standard data variables and initial data variables , and get the center matrix ;

[0085] The t-SNE technique is used to obtain the weight matrix of the molecular feature data to be processed, and based on the weight matrix and the center matrix , the target two-dimensional feature Y of the molecular feature data to be processed is calculated using the following formula;

[0086] ;

[0087] in, Represents the standard variable matrix The transposed matrix of express The inverse matrix of , P represents the loading matrix of the molecular feature data to be processed;

[0088] Based on the target two-dimensional features, the molecular feature data to be processed after dimension reduction is obtained.

[0089] Preferably, the intelligent diagnosis classification unit is further used for:

[0090] Performing risk assessment on the classified molecular feature data, wherein the classified molecular feature data is converted into a sample curve, and the converted sample curve is overlapped and compared with the clinical data curve;

[0091] After the curves are overlapped and compared, the risk range of the sample curve is determined based on the area of ​​the non-overlapping region when the two curves are superimposed. The larger the area of ​​the non-overlapping region, the higher the risk value.

[0092] Among them, the risk range is divided into low risk, medium risk and high risk;

[0093] Confirm the risk analysis results of the molecular signature data classified according to the risk range;

[0094] The risk analysis results of the classified molecular feature data are integrated to obtain standard tumor molecular data.

[0095] Compared with the prior art, the present invention has the following beneficial effects:

[0096] 1. The tumor tissue molecular intelligent analysis and recognition system based on data fusion analysis provided by the present invention uses standardized data models such as OMOP to process data, which can eliminate the format and semantic differences between different data sources, improve the interoperability and comparability of data, and through data quality assessment and preprocessing, timely discover and correct errors and anomalies in the data, reduce the interference of invalid data, and improve the overall quality of the data.

[0097] 2. The tumor tissue molecular intelligent analysis and identification system based on data fusion analysis provided by the present invention uses a hypergraph neural network model to perform data fusion analysis, and can process node data and hyperedge data containing complex relationships. By adjusting model parameters and performing model training / tuning, a model that can accurately characterize the molecular characteristics of tumor tissue can be obtained, and whether model tuning is needed can be determined by evaluating the indicators in the training results. This iterative optimization process helps to improve the accuracy and generalization ability of the model, so that the final model can better adapt to actual application scenarios.

[0098] 3. The tumor tissue molecular intelligent analysis and identification system based on data fusion analysis provided by the present invention extracts biomarkers related to tumor markers from key biomarkers, realizes accurate identification of tumor-related markers, helps to deeply understand the molecular mechanism and characteristics of tumors, and uses molecular feature recognition models to feature-associate tumor-related markers with key biomarkers. It reveals the intrinsic connection between biomarkers and tumor molecules, provides more comprehensive information for tumor diagnosis and prognosis assessment, and doctors can formulate more accurate and personalized treatment plans based on patients' molecular feature data and risk assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 This is a module schematic diagram of the tumor tissue molecular intelligent analysis and identification system of the present invention;

[0100] Figure 2 It is a schematic diagram of the process of the intelligent analysis and identification method of tumor tissue molecules of the present invention. DETAILED DESCRIPTION

[0101] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0102] In order to solve the problem in the prior art that after obtaining the patient's medical data, the data is not processed initially, resulting in poor data quality and inability to be integrated, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0103] A tumor tissue molecular intelligent analysis and recognition system based on data fusion analysis, comprising:

[0104] Multi-source heterogeneous data access unit, used for:

[0105] Confirm the patient's medical data from the database, manage the acquired medical data, and mark the medical data as pending medical data after data management;

[0106] Intelligent data pre-processing unit for:

[0107] Performing data quality assessment and data preprocessing on the medical data to be processed, and marking the medical data to be processed after the data quality assessment and data preprocessing are completed as target medical data;

[0108] Deep Fusion Analysis Unit for:

[0109] The target medical data is fused using a hypergraph neural network model, data representation is generated for the fused target medical data, and the generated data representation is annotated as comprehensive representation data;

[0110] Molecular Characterization Unit for:

[0111] Extract key biomarkers from the comprehensive characterization data, perform molecular feature identification on the extracted key biomarkers and tumor molecular features, and mark the molecular features identified in the key biomarkers as molecular feature data to be processed;

[0112] Intelligent diagnostic classification unit for:

[0113] The molecular feature data to be processed are classified into molecular feature categories, risk analysis is performed on the classified molecular features respectively, and the risk analysis results of the molecular features are integrated to obtain standard tumor molecular data after the data integration is completed;

[0114] Interactive result presentation unit for:

[0115] The standard tumor molecular data is converted into a visualization and then transmitted to the mobile terminal for display. At the same time, the patient can view and interact on the mobile terminal.

[0116] Specifically, the multi-source heterogeneous data access unit can eliminate the format and semantic differences between different data sources, improve the interoperability and comparability of data, and perform data quality assessment and preprocessing through the intelligent data preprocessing unit, timely discover and correct errors and anomalies in the data, reduce the interference of invalid data, and improve the overall quality of the data. Through the deep fusion analysis unit, the molecular characteristics of tumor tissue can be deeply analyzed to provide doctors with more accurate diagnosis and treatment suggestions, thereby promoting the development of precision medicine. The molecular feature recognition unit provides more comprehensive information for tumor diagnosis and prognosis evaluation, which not only improves the accuracy and personalization of tumor diagnosis, but also provides new ideas and methods for tumor research. The in-depth molecular feature analysis and risk assessment of the intelligent diagnostic classification unit provide an important basis for personalized medicine.

[0117] The multi-source heterogeneous data access unit is also used for:

[0118] Patients’ medical data include clinical medical record data, gene sequencing data, proteomics data, and imaging medical data;

[0119] Among them, clinical medical record data are patients’ medical history, diagnosis, treatment plan and efficacy evaluation data; gene sequencing data are patients’ tumor gene sequencing data; proteomics data are protein expression data of tumor tissue samples; and imaging medical data are patients’ tumor imaging data;

[0120] The acquired clinical medical record data, gene sequencing data, proteomics data and imaging medical data were standardized using the OMOP data model;

[0121] After the data standardization process is completed, use the data to write scripts for data integration;

[0122] After data integration is completed, the data is converted into a unified data format;

[0123] Finally, clinical medical record data, gene sequencing data, proteomics data and imaging medical data with unified data format are obtained and marked as medical data to be processed.

[0124] Specifically, it accesses and integrates heterogeneous data from different sources (such as clinical medical records, gene sequencing, proteomics, and imaging medicine). This capability is crucial for a comprehensive understanding of the patient's tumor characteristics, because tumors are multi-factorial, multi-level complex diseases that require comprehensive consideration of a variety of molecular and clinical information. Using standardized data models such as OMOP (Observational Medical Outcomes Data Model) to process data can eliminate format and semantic differences between different data sources, improve data interoperability and comparability, and automate data processing by integrating and converting data formats through data scripting. This not only improves data processing efficiency, but also reduces the possibility of human errors, ensures data consistency and accuracy, and ultimately generates medical data with a unified data format, providing doctors with comprehensive patient information and helping to develop personalized treatment plans.

[0125] Intelligent data pre-processing unit is also used for:

[0126] Conduct data quality assessment on the medical data to be processed first;

[0127] The data quality assessment process is to verify the clinical medical record data in the medical data to be processed, and check the patient's diagnosis results and medication records; use FastQC to check the original data of the gene sequencing data in the medical data to be processed, mainly checking the sequencing depth, coverage and sequencing error rate, then remove the connector sequences in the data, filter low-quality sequences and trim the terminal bases, and finally eliminate the contaminated sequences in the data; analyze the distribution of peptide matching errors, peptide number, peptide length and protein molecular weight of the proteomics data in the medical data to be processed; evaluate the image clarity, noise level and no reference image of the imaging data in the medical data to be processed;

[0128] Perform data preprocessing on the pending medical data that has completed data quality assessment;

[0129] The data preprocessing process is to sequentially perform data cleaning, data integration, data transformation and data standardization on the text data in the medical data to be processed; and sequentially perform grayscale adjustment, geometric transformation, image enhancement, filtering, denoising and image normalization on the image data in the medical data to be processed;

[0130] Finally, the unprocessed medical data that has completed data quality assessment and data preprocessing are labeled as target medical data.

[0131] Specifically, through a detailed data quality assessment process, the accuracy and reliability of the medical data to be processed are ensured. Whether it is clinical medical record data, gene sequencing data, proteomics data or imaging medical data, they have all undergone strict quality checks, including data verification, error rate checks, contamination sequence elimination, image clarity assessment, etc., thereby improving the accuracy and credibility of subsequent analysis results, and fine processing has been carried out for different types of data (text data and image data). After cleaning, integration, transformation and standardization, the text data has been removed from redundant and erroneous information and the data format has been unified; the image data has been improved through grayscale adjustment, geometric transformation, image enhancement and other steps. The quality and analyzability of the image. This refined processing helps to extract more valuable information and provide support for subsequent data fusion analysis. Through data quality assessment and preprocessing, errors and anomalies in the data can be discovered and corrected in a timely manner, the interference of invalid data can be reduced, and the overall quality of the data can be improved.

[0132] In order to solve the problem that the existing technology does not conduct more detailed tumor analysis on the patient's data, resulting in inaccurate diagnosis of the patient's medical data and poor treatment effect, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0133] The Deep Fusion Analysis Unit is also used to:

[0134] Convert clinical medical record data, gene sequencing data, proteomics data and imaging medical data in the target medical data into a unified representation form;

[0135] Among them, clinical medical record data is converted into structured or semi-structured form;

[0136] Gene sequencing data is represented as a graph structure;

[0137] Proteomic data are represented as functional networks of interactions between proteins;

[0138] Key features are extracted from medical imaging data through a deep learning model and represented as a graph structure, where the key features are characteristic data of tumor location and size.

[0139] After the representation of the target medical data is unified, the target medical data is imported into the hypergraph neural network model, wherein the hypergraph neural network model is retrieved from the database;

[0140] Confirm the data points in the hypergraph neural network model according to the target medical data;

[0141] The data points in the hypergraph neural network model include node data and hyperedge data. The node data are patient data points, gene data points, protein data and image feature data points in the target medical data, and the hyperedge data are data points of the relationship between the data in the target medical data.

[0142] Adjusting model parameters in the hypergraph neural network model according to data points in the hypergraph neural network model, wherein the model parameters include the number of layers, node embedding dimension, hyperedge embedding dimension, and activation function of the hypergraph neural network model;

[0143] Perform model training on the Hypergraph Neural Network model after adjusting the model parameters, and determine whether to perform model tuning on the Hypergraph Neural Network model based on the training results;

[0144] Among them, when the evaluation indicators in the training results are within the qualified range, the model tuning is not performed; when the evaluation indicators in the training results are not within the qualified range, the model tuning is performed;

[0145] Generate data representation for the trained or optimized hypergraph neural network model;

[0146] Among them, the data representation generation process is to pool the node data and hyperedge data in the trained hypergraph neural network model or the tuned hypergraph neural network model, and after the pooling operation, obtain the comprehensive representation data of the trained hypergraph neural network model or the tuned hypergraph neural network model.

[0147] In one embodiment, when the evaluation index in the training result is not within the qualified range, the deep fusion analysis unit performs model tuning, including:

[0148] The data points in the hypergraph neural network model are randomly divided into a preset number of data point groups in equal proportion, and the data point groups are randomly sorted to obtain a data point group sequence;

[0149] Based on the data point group sequence, the first data point group is input into the hypergraph neural network model to obtain a first result, and based on a preset model tuning rule, an optimized value of the first result for the model parameter is obtained, and the optimized value is added to the judgment process of the second data point group to obtain a second result;

[0150] According to the above-mentioned method for obtaining the second result, the data point group sequence is judged based on the hypergraph neural network model cycle to obtain a result sequence, and the result change law of the result sequence is obtained, and the numerical sequence is optimized;

[0151] Based on the result variation rule, each value in the optimized value sequence is evaluated, the optimization effect value of each value is determined, and an optimization effect sequence is obtained;

[0152] The data point groups are randomly sorted multiple times without repetition to obtain the optimization effect sequence after each sorting, forming an optimization effect matrix, and the optimization value matrix is ​​obtained based on the data point group sequence sorted each time;

[0153] Matching the optimization value matrix with the optimization effect matrix to obtain a matrix matching result, and determining the optimization effects of different optimization values ​​under different model parameters according to the matrix matching result;

[0154] Based on the optimization effects of different optimization values ​​under different model parameters, the optimization values ​​corresponding to the best optimization effect under the model parameters closest to the model parameters of the hypergraph neural network model are selected to tune the hypergraph neural network model.

[0155] In this embodiment, the preset model tuning rules are set in advance according to actual conditions.

[0156] In this embodiment, the optimization effects of different optimization values ​​under different model parameters can be found from the matrix, specifically, the optimization effects of one model parameter under different optimization values.

[0157] The beneficial effect of the above design scheme is: by matching the optimization numerical matrix with the optimization effect matrix, a matrix matching result is obtained, and the optimization effects of different optimization values ​​under different model parameters are determined according to the matrix matching result. Based on the optimization effects of different optimization values ​​under different model parameters, the optimization values ​​corresponding to the best optimization effect under the model parameters closest to the model parameters of the hypergraph neural network model are selected to perform model tuning on the hypergraph neural network model, so as to obtain a model that can accurately characterize the molecular characteristics of tumor tissue, and realize in-depth analysis and identification of the molecular characteristics of tumor tissue.

[0158] Specifically, medical data from different sources and types (clinical medical records, gene sequencing, proteomics, and medical imaging data) are converted into a unified representation so that these data can be analyzed and processed under the same framework. This unified data representation not only simplifies the complexity of data analysis, but also improves the interpretability and comparability of data. Different types of data are given different representations according to their characteristics, such as clinical medical record data is converted into structured or semi-structured forms, gene sequencing data is represented as a graph structure, proteomics data is represented as a functional network, and medical imaging data is also represented as a graph structure after extracting key features through a deep learning model. This flexible data modeling method can better capture the intrinsic characteristics and relationships of the data, and use a hypergraph neural network model for data fusion analysis. The model can handle data sets containing complex relationships, such as node data (patients, genes, proteins, imaging features) and hyperedge data (relationships between data). By adjusting model parameters and performing model training / tuning, a model that can accurately characterize the molecular characteristics of tumor tissue can be obtained, and the need for model tuning can be determined by evaluating the indicators in the training results. This iterative optimization process helps improve the accuracy and generalization ability of the model, so that the final model can better adapt to actual application scenarios. Through the pooling operation, the node data and hyperedge data in the hypergraph neural network model are comprehensively processed to generate comprehensive representation data of the model. This representation data not only contains the information of a single data point, but also integrates the relationship between data points, providing comprehensive data support for subsequent intelligent analysis and identification of tumor tissue molecules, and can achieve in-depth analysis and identification of the molecular characteristics of tumor tissues.

[0159] In order to solve the problem that the existing technology does not use the patient's tumor information for more intelligent molecular judgment and does not conduct risk assessment of the tumor, thus causing the patient to be unable to understand the disease in detail, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0160] The molecular characterization unit is also used for:

[0161] The comprehensive representation data is used to select feature data highly related to the disease from the fused data using the feature selection mechanism in the hypergraph neural network, where the feature selection mechanism is a ranking aggregation method;

[0162] Determine key biomarkers based on the results of feature selection, where key biomarkers are specific gene variations, protein expression patterns, and imaging features;

[0163] The confirmed key biomarkers are identified as tumor-related markers, wherein the tumor-related markers are identified as biomarkers that are related to tumor markers among the key biomarkers;

[0164] The tumor-related markers and key biomarkers are associated with the characteristics of biomarkers and tumor molecules through molecular feature recognition models;

[0165] After feature association, the molecular features identified in the key biomarkers are obtained and marked as molecular feature data to be processed.

[0166] In one embodiment, the molecular feature recognition unit associates the tumor-related markers with the key biomarkers through a molecular feature recognition model to perform feature association between the biomarkers and tumor molecules, including:

[0167] Inputting the tumor-related marker into the molecular feature recognition model to obtain a first tumor molecule corresponding to the tumor-related marker, and determining a first region of the first tumor molecule;

[0168] Inputting the key biomarker into the molecular feature recognition model, obtaining a second tumor molecule corresponding to the key biomarker, and determining a second region of the second tumor molecule;

[0169] obtaining the same tumor molecule between the first tumor molecule and the second tumor molecule based on the first unique tumor molecule in the first tumor molecule and the second unique tumor molecule in the second tumor molecule;

[0170] Acquire a first region feature and a second region feature of the same tumor molecule, acquire the same features and different features of the first region feature and the second region feature, re-input the markers corresponding to the different features into the molecular feature recognition model, and determine whether the different features are the final tumor molecules according to the output results;

[0171] Based on the same features and the judgment results of different features, the first target tumor molecule finally left by the same tumor molecule is determined;

[0172] intercepting a first marker corresponding to a first specific tumor molecule from the tumor-related marker, re-inputting the first marker into a molecular feature recognition model, and determining whether the first specific tumor molecule is a final tumor molecule;

[0173] Extracting a second marker corresponding to a second specific tumor molecule from the key biomarker, re-inputting the second marker into the molecular feature recognition model, and determining whether the second specific tumor molecule is the final tumor molecule;

[0174] Determining the second target tumor molecule that is finally left based on the judgment results of the first specific tumor molecule and the second specific tumor molecule;

[0175] The first target tumor molecule and the second target tumor molecule are characteristically associated with biomarkers.

[0176] In this embodiment, the same tumor molecule is a molecule identified in both the tumor-related marker and the key biomarker, but the identified molecules are within the same region, and the specific regions may be different.

[0177] In this embodiment, the markers are re-input into the molecular feature recognition model for secondary recognition. The difference from the first recognition is that the features of the markers are amplified, the number of markers is reduced, and the recognition efficiency is more accurate.

[0178] The beneficial effect of the above design scheme is: by using a molecular feature recognition model to associate tumor-related markers with key biomarkers, the features of biomarkers and tumor molecules are associated, and multiple analyses and identifications are performed during the model recognition process, the accuracy of the feature association is ensured, so that the feature association better reveals the intrinsic connection between biomarkers and tumor molecules, and provides more comprehensive information for tumor diagnosis and prognosis evaluation, which not only improves the accuracy and personalization of tumor diagnosis, but also provides new ideas and methods for tumor research.

[0179] Specifically, by using the feature selection mechanism in the hypergraph neural network, especially the sorting and aggregation method, it is possible to efficiently screen out feature data that is highly relevant to the disease from the fused data. This method not only takes into account the interaction between features, but also improves the accuracy and robustness of feature selection by sorting and aggregation. Based on the results of feature selection, key biomarkers closely related to tumors, such as specific gene variants, protein expression patterns, and imaging features, can be determined. These markers provide an important molecular basis for subsequent tumor analysis and diagnosis. By further extracting biomarkers related to tumor markers from key biomarkers, accurate identification of tumor-related markers is achieved. This process helps to deeply understand the molecular mechanisms and characteristics of tumors, and provides a basis for formulating personalized treatment plans. Using molecular feature recognition models, tumor-related markers are characteristically associated with key biomarkers. This association not only reveals the intrinsic connection between biomarkers and tumor molecules, but also provides more comprehensive information for tumor diagnosis and prognosis evaluation, which not only improves the accuracy and personalization of tumor diagnosis, but also provides new ideas and methods for tumor research.

[0180] Intelligent diagnostic classification unit is also used for:

[0181] The molecular feature data to be processed are subjected to data dimension reduction using t-SNE technology;

[0182] The molecular feature data to be processed after dimension reduction is clustered into different clusters using a hierarchical clustering method, wherein different clusters are different molecular subtypes.

[0183] After the tumor samples are clustered into different clusters, the key feature samples in the clustered tumor samples are screened according to the coefficient of variation;

[0184] The screened samples with key characteristics are molecularly typed, and the classified molecular characteristic data are obtained after molecular typing.

[0185] In one embodiment, the intelligent diagnosis classification unit processes the molecular feature data using the t-SNE technology to perform data dimension reduction as follows:

[0186] The molecular characteristic data to be processed are divided based on data variable units to obtain n initial data variables. The initial data variables are standardized based on the mean and variance of the data variables to obtain standard data variables. Based on the standard data variables, a standard variable matrix is ​​obtained. ;

[0187] The central function value of the standard data variable and the initial data variable is calculated according to the following formula ;

[0188] ;

[0189] in, represents the variable value of the i-th initial data variable, represents the variable value of the jth standard data variable, The data processing coefficient represents the molecular characteristic data to be processed, and its value is (0,1);

[0190] Based on the central function value of all standard data variables and initial data variables , and get the center matrix ;

[0191] The t-SNE technique is used to obtain the weight matrix of the molecular feature data to be processed, and based on the weight matrix and the center matrix , the target two-dimensional feature Y of the molecular feature data to be processed is calculated using the following formula;

[0192] ;

[0193] in, Represents the standard variable matrix The transposed matrix of express The inverse matrix of , P represents the loading matrix of the molecular feature data to be processed;

[0194] Based on the target two-dimensional features, the molecular feature data to be processed after dimension reduction is obtained.

[0195] In this embodiment, the data variable units are determined according to the t-SNE technique.

[0196] In this embodiment, the purpose of the central matrix is ​​to eliminate matrix offset in the data and provide a data basis for subsequent dimensionality reduction.

[0197] In this embodiment, each element of the weight matrix represents the correlation or importance between the original variable and the factor determined by the t-SNE technique.

[0198] In this embodiment, the loading matrix of the molecular characteristic data to be processed is an important concept in factor analysis. The variables of the molecular characteristic data to be processed are regarded as linear combinations of latent factors, and the loading matrix reflects the strength and direction of the relationship between the original variables and the latent factors. The latent factors are set in advance according to actual conditions.

[0199] The beneficial effects of the above design scheme are: by using t-SNE technology to reduce the dimension of the molecular feature data to be processed, considering the influence of factors such as matrix offset and factor correlation in the dimension reduction process, and using the set algorithm formula to reduce the dimension of the molecular feature data to be processed, the accuracy of the molecular feature data to be processed after the dimension reduction is guaranteed, and an accurate data basis is provided for the screening of key feature samples in tumor samples.

[0200] Performing risk assessment on the classified molecular feature data, wherein the classified molecular feature data is converted into a sample curve, and the converted sample curve is overlapped and compared with the clinical data curve;

[0201] After the curves are overlapped and compared, the risk range of the sample curve is determined based on the area of ​​the non-overlapping region when the two curves are superimposed. The larger the area of ​​the non-overlapping region, the higher the risk value.

[0202] Among them, the risk range is divided into low risk, medium risk and high risk;

[0203] Confirm the risk analysis results of the molecular signature data classified according to the risk range;

[0204] The risk analysis results of the classified molecular feature data are integrated to obtain standard tumor molecular data.

[0205] Specifically, the t-SNE technology was used for data dimensionality reduction, which effectively reduced the dimension of the molecular feature data and reduced the computational complexity, while retaining the important information in the data. Subsequently, the hierarchical clustering method was used to cluster tumor samples with similar molecular features into different clusters, namely different molecular subtypes, which helped to deeply understand the heterogeneity of tumors and provide a basis for subsequent analysis. The key feature samples were screened according to the coefficient of variation, which could identify the most influential features for tumor classification and diagnosis, and improve the accuracy and sensitivity of diagnosis. This screening method helps to reduce the interference of noise data and improve the reliability of analysis results. The molecular typing of the selected key feature samples further refined the classification of tumors and provided strong support for precision medicine. At the same time, through the conversion of sample curves and the overlapping comparison with clinical data curves, a scientific assessment of the risk range of tumor samples was achieved, and through in-depth molecular feature analysis and risk assessment, an important basis for personalized medicine was provided. Doctors can formulate more accurate and personalized treatment plans based on the patient's molecular feature data and risk assessment results to improve the treatment effect and the patient's quality of life.

[0206] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0207] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A tumor tissue molecular intelligent analysis and identification system based on data fusion analysis, characterized in that: include: Multi-source heterogeneous data access unit, used for: Confirm the patient's medical data from the database, manage the acquired medical data, and mark the medical data as pending medical data after data management; Intelligent data pre-processing unit for: Performing data quality assessment and data preprocessing on the medical data to be processed, and marking the medical data to be processed after the data quality assessment and data preprocessing are completed as target medical data; Deep Fusion Analysis Unit for: The target medical data is fused using a hypergraph neural network model, data representation is generated for the fused target medical data, and the generated data representation is annotated as comprehensive representation data; Molecular Characterization Unit for: Extract key biomarkers from the comprehensive characterization data, perform molecular feature identification on the extracted key biomarkers and tumor molecular features, and mark the molecular features identified in the key biomarkers as molecular feature data to be processed; Intelligent diagnostic classification unit for: The molecular feature data to be processed are classified into molecular features, the classified molecular features are respectively subjected to risk analysis, and the risk analysis results of the molecular features are integrated to obtain standard tumor molecular data after the data integration is completed; Interactive result presentation unit for: The standard tumor molecular data is converted into a visual data, and after the conversion is completed, it is transmitted to the mobile terminal for display. At the same time, the patient can view and interact on the mobile terminal; The multi-source heterogeneous data access unit is further used for: Patients’ medical data include clinical medical record data, gene sequencing data, proteomics data, and imaging medical data; Among them, clinical medical record data are patients’ medical history, diagnosis, treatment plan and efficacy evaluation data; gene sequencing data are patients’ tumor gene sequencing data; proteomics data are protein expression data of tumor tissue samples; and imaging medical data are patients’ tumor imaging data; The acquired clinical medical record data, gene sequencing data, proteomics data and imaging medical data were standardized using the OMOP data model; After data standardization is completed, use the data to write scripts for data integration; After data integration is completed, the data is converted into a unified data format; Finally, clinical medical record data, gene sequencing data, proteomics data and imaging medical data with unified data format are obtained and marked as medical data to be processed; The intelligent data preprocessing unit is further used for: Conduct data quality assessment on the medical data to be processed; The data quality assessment process is to verify the clinical medical record data in the medical data to be processed, and check the patient's diagnosis results and medication records; use FastQC to check the original data of the gene sequencing data in the medical data to be processed, including checking the sequencing depth, coverage and sequencing error rate, removing the connector sequences in the data, filtering low-quality sequences and trimming the terminal bases, and finally eliminating the contaminated sequences in the data; analyze the distribution of peptide matching errors, peptide number, peptide length and protein molecular weight of the proteomics data in the medical data to be processed; evaluate the image clarity, noise level and no reference image of the imaging data in the medical data to be processed; Perform data preprocessing on the pending medical data that has completed data quality assessment; The data preprocessing process is to sequentially perform data cleaning, data integration, data transformation and data standardization on the text data in the medical data to be processed; and sequentially perform grayscale adjustment, geometric transformation, image enhancement, filtering, denoising and image normalization on the image data in the medical data to be processed; Finally, the unprocessed medical data that has completed data quality assessment and data preprocessing are labeled as target medical data; The in-depth fusion analysis unit is further used for: Convert clinical medical record data, gene sequencing data, proteomics data and imaging medical data in the target medical data into a unified representation form; Among them, clinical medical record data is converted into structured or semi-structured form; Gene sequencing data is represented as a graph structure; Proteomic data are represented as functional networks of interactions between proteins; The key features of medical imaging data are extracted through a deep learning model and represented as a graph structure, where the key features are the characteristic data of tumor location and size; After the representation of the target medical data is unified, the target medical data is imported into the hypergraph neural network model, wherein the hypergraph neural network model is retrieved from the database; Confirm the data points in the hypergraph neural network model according to the target medical data; The data points in the hypergraph neural network model include node data and hyperedge data. The node data are patient data points, gene data points, protein data and image feature data points in the target medical data, and the hyperedge data are data points of the relationship between the data in the target medical data. Adjusting model parameters in the hypergraph neural network model according to data points in the hypergraph neural network model, wherein the model parameters include the number of layers, node embedding dimension, hyperedge embedding dimension, and activation function of the hypergraph neural network model; Perform model training on the Hypergraph Neural Network model after adjusting the model parameters, and determine whether to perform model tuning on the Hypergraph Neural Network model based on the training results; Among them, when the evaluation indicators in the training results are within the qualified range, the model tuning is not performed; when the evaluation indicators in the training results are not within the qualified range, the model tuning is performed; Generate data representation for the trained or optimized hypergraph neural network model; The data representation generation process is to perform a pooling operation on the node data and hyperedge data in the trained hypergraph neural network model or the tuned hypergraph neural network model, and obtain comprehensive representation data of the trained hypergraph neural network model or the tuned hypergraph neural network model after the pooling operation; When the evaluation index in the training result in the in-depth fusion analysis unit is not within the qualified range, model tuning is performed, including: The data points in the hypergraph neural network model are randomly divided into a preset number of data point groups in equal proportion, and the data point groups are randomly sorted to obtain a data point group sequence; Based on the data point group sequence, the first data point group is input into the hypergraph neural network model to obtain a first result, and based on a preset model tuning rule, an optimized value of the first result for the model parameter is obtained, and the optimized value is added to the judgment process of the second data point group to obtain a second result; According to the above-mentioned method for obtaining the second result, the data point group sequence is judged based on the hypergraph neural network model cycle to obtain a result sequence, and the result change law of the result sequence is obtained, and the numerical sequence is optimized; Based on the result variation rule, each value in the optimized value sequence is evaluated, the optimization effect value of each value is determined, and an optimization effect sequence is obtained; The data point groups are randomly sorted multiple times without repetition to obtain the optimization effect sequence after each sorting, forming an optimization effect matrix, and the optimization value matrix is ​​obtained based on the data point group sequence sorted each time; Matching the optimization value matrix with the optimization effect matrix to obtain a matrix matching result, and determining the optimization effects of different optimization values ​​under different model parameters according to the matrix matching result; Based on the optimization effects of different optimization values ​​under different model parameters, the optimization value corresponding to the best optimization effect under the model parameters that are closest to the model parameters of the hypergraph neural network model is selected to perform model tuning on the hypergraph neural network model; The molecular feature recognition unit is also used for: The comprehensive representation data is used to select feature data highly related to the disease from the fused data using the feature selection mechanism in the hypergraph neural network, where the feature selection mechanism is a ranking aggregation method; Determine key biomarkers based on the results of feature selection, where key biomarkers are specific gene variations, protein expression patterns, and imaging features; The confirmed key biomarkers are identified as tumor-related markers, wherein the tumor-related markers are identified as biomarkers that are related to tumor markers among the key biomarkers; The tumor-related markers and key biomarkers are associated with the characteristics of biomarkers and tumor molecules through molecular feature recognition models; After feature association, the molecular features identified in the key biomarkers are obtained and marked as molecular feature data to be processed; The molecular feature recognition unit associates the tumor-related markers with the key biomarkers through a molecular feature recognition model to perform feature association between the biomarkers and tumor molecules, including: Inputting the tumor-related marker into the molecular feature recognition model to obtain a first tumor molecule corresponding to the tumor-related marker, and determining a first region of the first tumor molecule; Inputting the key biomarker into the molecular feature recognition model, obtaining a second tumor molecule corresponding to the key biomarker, and determining a second region of the second tumor molecule; obtaining the same tumor molecule between the first tumor molecule and the second tumor molecule based on the first unique tumor molecule in the first tumor molecule and the second unique tumor molecule in the second tumor molecule; Acquire a first region feature and a second region feature of the same tumor molecule, acquire the same features and different features of the first region feature and the second region feature, re-input the markers corresponding to the different features into the molecular feature recognition model, and determine whether the different features are the final tumor molecules according to the output results; Based on the same features and the judgment results of different features, the first target tumor molecule finally left by the same tumor molecule is determined; intercepting a first marker corresponding to a first specific tumor molecule from the tumor-related marker, re-inputting the first marker into a molecular feature recognition model, and determining whether the first specific tumor molecule is a final tumor molecule; Extracting a second marker corresponding to a second specific tumor molecule from the key biomarker, re-inputting the second marker into the molecular feature recognition model, and determining whether the second specific tumor molecule is the final tumor molecule; Determining the second target tumor molecule that is finally left based on the judgment results of the first specific tumor molecule and the second specific tumor molecule; Characteristically associating the first target tumor molecule and the second target tumor molecule with a biomarker; The intelligent diagnosis classification unit is also used for: The molecular feature data to be processed are subjected to data dimension reduction using t-SNE technology; The molecular feature data to be processed after dimension reduction is clustered into different clusters using a hierarchical clustering method, wherein different clusters are different molecular subtypes. After the tumor samples are clustered into different clusters, the key feature samples in the clustered tumor samples are screened according to the coefficient of variation; The selected key characteristic samples are molecularly typed, and the molecular characteristic data after molecular typing is obtained; The intelligent diagnosis classification unit processes the molecular feature data using the t-SNE technology to perform data dimension reduction as follows: The molecular characteristic data to be processed are divided based on data variable units to obtain n initial data variables. The initial data variables are standardized based on the mean and variance of the data variables to obtain standard data variables. Based on the standard data variables, a standard variable matrix N is obtained. (1,n) ; The central function value M of the standard data variable and the initial data variable is calculated according to the following formula i ; Among them, X i represents the variable value of the i-th initial data variable, X j represents the variable value of the jth standard data variable, β X The data processing coefficient represents the molecular characteristic data to be processed, and its value is (0,1); Based on the central function value M of all standard data variables and initial data variables i , get the center matrix M (1,n) ; The weight matrix of the molecular feature data to be processed is obtained by using the t-SNE technology, and based on the weight matrix and the center matrix M (1,n) , the target two-dimensional feature Y of the molecular feature data to be processed is calculated using the following formula; in, Represents the standard variable matrix N (1,n) The transposed matrix of express The inverse matrix of , P represents the loading matrix of the molecular feature data to be processed; Obtaining the molecular feature data to be processed after dimension reduction based on the target two-dimensional features; The intelligent diagnosis classification unit is also used for: Performing risk assessment on the classified molecular feature data, wherein the classified molecular feature data is converted into a sample curve, and the converted sample curve is overlapped and compared with the clinical data curve; After the curves are overlapped and compared, the risk range of the sample curve is determined based on the area of ​​the non-overlapping region when the two curves are superimposed. The larger the area of ​​the non-overlapping region, the higher the risk value. Among them, the risk range is divided into low risk, medium risk and high risk; Confirm the risk analysis results of the molecular signature data classified according to the risk range; The risk analysis results of the classified molecular feature data are integrated to obtain standard tumor molecular data.

Images