Big data-based accurate diagnosis and treatment system for liver cancer patient diagnosis data retrieval

Abstract

The invention provides a liver cancer patient diagnosis data retrieval precise diagnosis and treatment system based on big data, and relates to the technical field of medical diagnosis. The liver cancer patient diagnosis data retrieval precise diagnosis and treatment system based on big data comprises a case database module, a clinical decision support system, a deep learning module, an early warning visualization module, a data analysis and processing module, a multi-modal data fusion module, a data collection and storage module, a feature selection and selection module and a data acquisition module. Image data, genome data, pathological data and electronic medical record data are integrated through multi-modal data fusion, comprehensive analysis is carried out, the accuracy of risk prediction is improved, treatment schemes are dynamically adjusted according to specific conditions and treatment responses of patients, personalized risk prediction and treatment scheme recommendation are achieved, and the risk prediction and treatment scheme recommendation efficiency is improved by combining gene data. Medicine selection and dosage are optimized, the treatment risk is reduced, the treatment effect is improved, doctors can understand the basis of system decision making, and the credibility of clinical application is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis technology, and in particular to a liver cancer patient diagnostic data retrieval and precision diagnosis and treatment system based on big data. Background Art

[0002] Big data, or massive data, refers to the amount of data involved that is so large that it cannot be captured, managed, processed, and organized into information that helps companies make more proactive business decisions within a reasonable time through mainstream software tools. Liver cancer, also known as liver malignant tumors, can be divided into two categories: primary and secondary. Primary liver malignant tumors originate from the epithelial or mesenchymal tissue of the liver. The former is called primary liver cancer, which is a high-incidence and extremely harmful malignant tumor. The latter is called sarcoma, which is relatively rare compared with primary liver cancer. Secondary or metastatic liver cancer refers to malignant tumors originating from multiple organs throughout the body that invade the liver. In order to better treat liver cancer, a liver cancer big data platform has been designed. The platform usually integrates multiple data sources such as electronic medical records, imaging examinations, pathological image libraries, bioinformatics data, etc. to form a comprehensive liver cancer clinical and scientific research big data platform. Through big data analysis, these platforms can support early screening, diagnosis, formulation of treatment plans and prognosis evaluation of liver cancer.

[0003] There are still some problems in the use of existing big data platforms. Existing systems often have single functions and it is difficult to fully cover the entire process from data collection to clinical decision-making. Existing deep learning systems may lack model tuning mechanisms, resulting in unstable model performance. Existing CDSS systems may only focus on diagnosis and treatment recommendations and lack comprehensive risk prediction and assessment capabilities. Therefore, technical personnel in this field provide a liver cancer patient diagnostic data retrieval and precision diagnosis and treatment system based on big data to solve the problems raised in the above background technology. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the existing technology, the present invention provides a liver cancer patient diagnostic data retrieval and precision diagnosis and treatment system based on big data, which solves the problems that the existing systems often have single functions and are difficult to comprehensively cover the entire process from data collection to clinical decision-making; the existing deep learning systems may lack model tuning mechanisms, resulting in unstable model performance; and the existing CDSS systems may only focus on diagnosis and treatment recommendations and lack comprehensive risk prediction and assessment capabilities.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a liver cancer patient diagnostic data retrieval and accurate diagnosis and treatment system based on big data, comprising: a case database module, a clinical decision support system, a deep learning module, an early warning visualization module, a data analysis and processing module, a multimodal data fusion module, a data collection and storage module, a feature selection and selection module, and a data acquisition module;

[0008] The clinical decision support system also includes risk prediction and assessment, using a deep reinforcement learning algorithm to dynamically adjust the treatment plan according to the patient's specific situation and treatment response, using a gene variation detection algorithm and a gene expression analysis algorithm to optimize drug selection and dosage, using a multimodal deep learning model for comprehensive analysis, using a LIME algorithm to explain the prediction results of the deep learning model, and using a SHAP algorithm to provide global and local explanations of the model prediction. The risk prediction and assessment predicts the risk of occurrence and recurrence of liver cancer and evaluates the prognosis after treatment, including the following steps:

[0009] A1. Data collection and preprocessing: collect patients' electronic medical records, imaging data, pathological data and genetic data, and then clean, standardize and normalize the data;

[0010] A2. Feature engineering: extract risk-related features such as age, gender, gene mutation and imaging features, and use feature selection algorithms to screen important features;

[0011] A3. Model training: Use machine learning models or deep learning models to predict risks, and use cross-validation and hyperparameter optimization methods to optimize model performance;

[0012] A4. Model evaluation, using evaluation indicators to evaluate model performance;

[0013] A5. Personalized treatment plan adjustment: dynamically adjust the treatment plan according to the patient's specific situation and treatment response, and recommend the best treatment plan based on the current status and model output;

[0014] A6. Genetic variation detection and drug selection, combining genetic data to optimize drug selection and dosage;

[0015] A6.1. Gene data collection: collect the patient's gene sequencing data;

[0016] A6.2. Gene mutation detection: using a gene mutation detection algorithm to detect gene mutations;

[0017] A6.3. Gene expression analysis, using gene expression analysis algorithms to analyze gene expression levels;

[0018] A6.4. Pharmacogenomic analysis, combining gene variation and gene expression data, using the pharmacogenomics database to select appropriate drugs and doses;

[0019] The deep learning module also includes model diagnosis and tuning. Generative adversarial networks use GAN to enhance and generate data. Graph neural networks use GNN to analyze and predict graph structure data to achieve model diversity. Bayesian optimization algorithm is used to automatically optimize model hyperparameters. Reinforcement learning algorithm is used to tune the model. The detailed steps of model tuning by the reinforcement learning algorithm are as follows:

[0020] B1. Generate adversarial networks, use GAN for data enhancement and generation to generate more training data;

[0021] B2. Graph neural network, using GNN to analyze and predict graph structure data and gene network analysis;

[0022] B3. Bayesian optimization algorithm, using Bayesian optimization algorithm to automatically optimize model hyperparameters and improve model performance;

[0023] B4. Reinforcement learning algorithm, use reinforcement learning algorithm to tune the model, and use DQN algorithm to optimize model parameters;

[0024] The early warning visualization uses D3.js for interactive data visualization, WebGL for high-performance graphics rendering, isolation forest, autoencoder and other algorithms to automatically identify potential disease changes and risks, and ARI MA and LSTM algorithms to predict disease trends. The detailed steps of the early warning visualization are as follows:

[0025] C1. Data visualization, using D3.js for interactive data visualization and WebGL for high-performance graphics rendering;

[0026] C2. Identification of disease changes, using algorithms such as isolation forests and autoencoders to automatically identify potential disease changes and risks;

[0027] C3. Disease trend prediction, using ARI MA and LSTM algorithm to predict disease trend;

[0028] The multimodal data fusion module also includes advanced fusion technology and data alignment and matching. The advanced fusion technology uses a multimodal deep learning model for comprehensive analysis and uses an attention mechanism to improve the accuracy and efficiency of data fusion. The multimodal data fusion module integrates imaging data, genomic data, pathological data, electronic medical record data, etc. The detailed steps for comprehensive analysis are as follows:

[0029] D1. Data preprocessing: cleaning, standardization and normalization of data of different modalities;

[0030] D2. Feature extraction: use convolutional neural networks to extract features for image data, use gene expression analysis algorithms to extract features for gene data, and use image processing algorithms to extract features for pathological data;

[0031] D3. Data fusion, using multimodal deep learning models for data fusion;

[0032] D4. Model training and evaluation: Use the fused features for model training and evaluation.

[0033] Preferably, the data analysis and processing module uses the AutoML tool to automatically select and optimize machine learning models, feature selection methods and feature generation algorithms to automatically generate and optimize features, and uses a streaming computing framework for real-time data processing.

[0034] Preferably, the detailed steps of performing comprehensive analysis by integrating imaging data, genomic data, pathological data, electronic medical record data, etc. through the multimodal data fusion are as follows:

[0035] S1. Data preprocessing: cleaning, standardization and normalization of data of different modalities;

[0036] S2. Feature extraction: use convolutional neural network to extract features for image data, use gene expression analysis algorithm to extract features for gene data, and use image processing algorithm to extract features for pathological data;

[0037] S3. Data fusion, using multimodal deep learning models for data fusion;

[0038] S4. Model training and evaluation, use the fused features for model training and evaluation.

[0039] Preferably, the clinical decision support system also includes model interpretability, and the model interpretability provides the interpretability of model prediction and also includes LIME algorithm and SHAP algorithm. The LIME algorithm selects instances, perturbs data, generates new data points, performs model prediction on the perturbed data points, uses a linear model to fit the prediction results of the perturbed data points, and generates explanation results. The SHAP algorithm calculates the Shap l ey value of each feature, uses the global Shap l ey value to analyze the overall contribution of each feature to the model prediction, and uses the local Shap l ey value to analyze the contribution of each feature to the prediction of a single instance.

[0040] (III) Beneficial effects

[0041] The present invention provides a liver cancer patient diagnostic data retrieval and precision diagnosis and treatment system based on big data.

[0042] It has the following beneficial effects:

[0043] 1. In the present invention, multimodal data fusion integrates imaging data, genomic data, pathological data and electronic medical record data for comprehensive analysis, improves the accuracy of risk prediction, dynamically adjusts the treatment plan according to the patient's specific situation and treatment response, realizes personalized risk prediction and treatment plan recommendation, combines genetic data, optimizes drug selection and dosage, reduces treatment risks, improves treatment effects, provides interpretability of model predictions, enables doctors to understand the basis of system decisions, and improves the credibility of clinical applications.

[0044] 2. In the present invention, a generative adversarial network is used to perform data enhancement, generate more training data, and improve the generalization ability of the model. GNN is used to analyze and predict graph structure data and gene network analysis to improve the model's ability to process complex data, automatically optimize model hyperparameters, and improve model performance. A reinforcement learning algorithm is used to tune the model to further improve the accuracy and stability of the model.

[0045] 3. In the present invention, D3.js and WebGL provide interactive and efficient early warning information visualization to help doctors and patients understand the changes and risks of the disease in a timely manner. Isolation forest and autoencoder automatically identify potential changes and risks of the disease and provide timely warnings. ARI MA and LSTM predict the disease trend and predict the development of the disease in advance to provide support for clinical decision-making.

[0046] 4. In the present invention, the system integrates multiple modules, including data acquisition, storage, analysis, deep learning, early warning and multimodal data fusion, forming a complete closed-loop system. This modular design makes the system highly flexible and scalable. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0048] 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.

[0049] Embodiment 1:

[0050] like Figure 1As shown, an embodiment of the present invention provides a liver cancer patient diagnostic data retrieval and accurate diagnosis and treatment system based on big data, comprising: a case database module, a clinical decision support system, a deep learning module, an early warning visualization module, a data analysis and processing module, a multimodal data fusion module, a data collection and storage module, a feature selection and selection module, and a data acquisition module;

[0051] The clinical decision support system also includes risk prediction and assessment, using deep reinforcement learning algorithms to dynamically adjust treatment plans based on the patient's specific conditions and treatment responses, using gene variation detection algorithms and gene expression analysis algorithms to optimize drug selection and dosage, using multimodal deep learning models for comprehensive analysis, using the LIME algorithm to explain the prediction results of the deep learning model, and using the SHAP algorithm to provide global and local explanations of model predictions. Risk prediction and assessment predicts the risk of occurrence and recurrence of liver cancer and evaluates the prognosis after treatment, including the following steps:

[0052] A1. Data collection and preprocessing: collect patients' electronic medical records, imaging data, pathological data and genetic data, and then clean, standardize and normalize the data;

[0053] A2. Feature engineering: extract risk-related features such as age, gender, gene mutation and imaging features, and use feature selection algorithms to screen important features;

[0054] A3. Model training: Use machine learning models or deep learning models to predict risks, and use cross-validation and hyperparameter optimization methods to optimize model performance;

[0055] A4. Model evaluation, using evaluation indicators to evaluate model performance;

[0056] A5. Personalized treatment plan adjustment: dynamically adjust the treatment plan according to the patient's specific situation and treatment response, and recommend the best treatment plan based on the current status and model output;

[0057] A6. Genetic variation detection and drug selection, combining genetic data to optimize drug selection and dosage;

[0058] A6.1. Gene data collection: collect the patient's gene sequencing data;

[0059] A6.2. Gene mutation detection: using a gene mutation detection algorithm to detect gene mutations;

[0060] A6.3. Gene expression analysis, using gene expression analysis algorithms to analyze gene expression levels;

[0061] A6.4. Pharmacogenomic analysis, combining gene variation and gene expression data, using the pharmacogenomics database to select appropriate drugs and doses;

[0062] The deep learning module also includes model diagnosis and tuning. Generative adversarial networks use GAN for data enhancement and generation. Graph neural networks use GNN to analyze and predict graph structure data to achieve model diversity. The Bayesian optimization algorithm is used to automatically optimize model hyperparameters. The reinforcement learning algorithm is used for model tuning. The detailed steps for model tuning through the reinforcement learning algorithm are as follows:

[0063] B1. Generate adversarial networks, use GAN for data enhancement and generation to generate more training data;

[0064] B2. Graph neural network, using GNN to analyze and predict graph structure data and gene network analysis;

[0065] B3. Bayesian optimization algorithm, using Bayesian optimization algorithm to automatically optimize model hyperparameters and improve model performance;

[0066] B4. Reinforcement learning algorithm, use reinforcement learning algorithm to tune the model, use DQN algorithm to optimize model parameters, use generative adversarial network to perform data enhancement, generate more training data, improve the generalization ability of the model, use GNN to analyze and predict graph structure data, gene network analysis, improve the model's ability to handle complex data, automatically optimize model hyperparameters, improve model performance, use reinforcement learning algorithm to tune the model, and further improve the accuracy and stability of the model;

[0067] Early warning visualization uses D3.js for interactive data visualization, WebGL for high-performance graphics rendering, isolation forest, autoencoder and other algorithms to automatically identify potential disease changes and risks, and ARIMA and LSTM algorithms to predict disease trends. The detailed steps of early warning visualization are as follows:

[0068] C1. Data visualization, using D3.js for interactive data visualization and WebGL for high-performance graphics rendering;

[0069] C2. Identification of disease changes, using algorithms such as isolation forests and autoencoders to automatically identify potential disease changes and risks;

[0070] C3. Disease trend prediction: ARIMA and LSTM algorithms are used to predict disease trends. D3.js and WebGL provide interactive and efficient early warning information visualization to help doctors and patients understand disease changes and risks in a timely manner. Isolation forest and autoencoder can automatically identify potential disease changes and risks, and provide timely warnings. ARIMA and LSTM are used to predict disease trends, predict disease development in advance, and provide support for clinical decision-making.

[0071] The multimodal data fusion module also includes advanced fusion technology and data alignment and matching. The advanced fusion technology uses a multimodal deep learning model for comprehensive analysis and uses an attention mechanism to improve the accuracy and efficiency of data fusion. The multimodal data fusion module integrates imaging data, genomic data, pathological data, electronic medical record data, etc. The detailed steps for comprehensive analysis are as follows:

[0072] D1. Data preprocessing: cleaning, standardization and normalization of data of different modalities;

[0073] D2. Feature extraction: use convolutional neural networks to extract features for image data, use gene expression analysis algorithms to extract features for gene data, and use image processing algorithms to extract features for pathological data;

[0074] D3. Data fusion, using multimodal deep learning models for data fusion;

[0075] D4. Model training and evaluation: Use the fused features for model training and evaluation. Multimodal data fusion integrates imaging data, genomic data, pathological data, and electronic medical record data for comprehensive analysis to improve the accuracy of risk prediction. Dynamically adjust treatment plans based on the patient's specific situation and treatment response to achieve personalized risk prediction and treatment plan recommendations. Combined with genetic data, optimize drug selection and dosage, reduce treatment risks, improve treatment effects, provide explainability of model predictions, enable doctors to understand the basis for system decision-making, and improve the credibility of clinical applications.

[0076] The data analysis and processing module uses the AutoML tool to automatically select and optimize machine learning models, feature selection methods and feature generation algorithms to automatically generate and optimize features, and uses the streaming computing framework for real-time data processing.

[0077] The detailed steps for comprehensive analysis by integrating imaging data, genomic data, pathological data, electronic medical record data, etc. through multimodal data fusion are as follows:

[0078] S1. Data preprocessing: cleaning, standardization and normalization of data of different modalities;

[0079] S2. Feature extraction: use convolutional neural network to extract features for image data, use gene expression analysis algorithm to extract features for gene data, and use image processing algorithm to extract features for pathological data;

[0080] S3. Data fusion, using multimodal deep learning models for data fusion;

[0081] S4. Model training and evaluation, use the fused features for model training and evaluation.

[0082] The clinical decision support system also includes model interpretability. Model interpretability provides interpretability of model predictions and also includes LIME algorithm and SHAP algorithm. The LIME algorithm selects instances, perturbs data, generates new data points, performs model predictions on the perturbed data points, uses a linear model to fit the prediction results of the perturbed data points, and generates explanation results. The SHAP algorithm calculates the Shap l ey value of each feature, uses the global Shap l ey value to analyze the overall contribution of each feature to the model prediction, and uses the local Shap l ey value to analyze the contribution of each feature to the prediction of a single instance. The system integrates multiple modules, including data acquisition, storage, analysis, deep learning, early warning and multimodal data fusion, forming a complete closed-loop system. This modular design makes the system highly flexible and scalable.

[0083] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A liver cancer patient diagnostic data retrieval and accurate diagnosis and treatment system based on big data, characterized by: include: Case database module, clinical decision support system, deep learning module, early warning visualization module, data analysis and processing module, multimodal data fusion module, data collection and storage module, feature selection and selection module, and data acquisition module; The clinical decision support system also includes risk prediction and assessment, using a deep reinforcement learning algorithm to dynamically adjust the treatment plan according to the patient's specific situation and treatment response, using a gene variation detection algorithm and a gene expression analysis algorithm to optimize drug selection and dosage, using a multimodal deep learning model for comprehensive analysis, using a LIME algorithm to interpret the prediction results of the deep learning model, and using a SHAP algorithm to provide global and local explanations of the model prediction. The risk prediction and assessment predicts the risk of occurrence and recurrence of liver cancer and evaluates the prognosis after treatment, including the following steps: A1. Data collection and preprocessing: collect patients' electronic medical records, imaging data, pathological data and genetic data, and then clean, standardize and normalize the data; A2. Feature engineering: extract risk-related features such as age, gender, gene mutation and imaging features, and use feature selection algorithms to screen important features; A3. Model training: using machine learning models or deep learning models for risk prediction, and optimizing model performance using cross-validation and hyperparameter optimization methods; A4. Model evaluation, using evaluation indicators to evaluate model performance; A5. Personalized treatment plan adjustment: dynamically adjust the treatment plan according to the patient's specific situation and treatment response, and recommend the best treatment plan based on the current status and model output; A6. Genetic variation detection and drug selection, combining genetic data to optimize drug selection and dosage; A6.

1. Gene data collection: collect the patient's gene sequencing data; A6.

2. Gene mutation detection: using a gene mutation detection algorithm to detect gene mutations; A6.

3. Gene expression analysis, using gene expression analysis algorithms to analyze gene expression levels; A6.

4. Pharmacogenomic analysis, combining gene variation and gene expression data, using the pharmacogenomics database to select appropriate drugs and doses; The deep learning module also includes model diagnosis and tuning. Generative adversarial networks use GAN to enhance and generate data. Graph neural networks use GNN to analyze and predict graph structure data to achieve model diversity. Bayesian optimization algorithm is used to automatically optimize model hyperparameters. Reinforcement learning algorithm is used to tune the model. The detailed steps of model tuning by the reinforcement learning algorithm are as follows: B1. Generate adversarial networks, use GAN for data enhancement and generation to generate more training data; B2. Graph neural network, using GNN to analyze and predict graph structure data and gene network analysis; B3. Bayesian optimization algorithm, using Bayesian optimization algorithm to automatically optimize model hyperparameters and improve model performance; B4. Reinforcement learning algorithm, use reinforcement learning algorithm to tune the model, and use DQN algorithm to optimize model parameters; The early warning visualization uses D3.js for interactive data visualization, WebGL for high-performance graphics rendering, isolation forest, autoencoder and other algorithms to automatically identify potential disease changes and risks, and ARIMA and LSTM algorithms to predict disease trends. The detailed steps of the early warning visualization are as follows: C1. Data visualization, using D3.js for interactive data visualization and WebGL for high-performance graphics rendering; C2. Identification of disease changes, using algorithms such as isolation forests and autoencoders to automatically identify potential disease changes and risks; C3. Disease trend prediction, using ARIMA and LSTM algorithms to predict disease trends; The multimodal data fusion module also includes advanced fusion technology and data alignment and matching. The advanced fusion technology uses a multimodal deep learning model for comprehensive analysis and uses an attention mechanism to improve the accuracy and efficiency of data fusion. The multimodal data fusion module integrates imaging data, genomic data, pathological data, electronic medical record data, etc. The detailed steps for comprehensive analysis are as follows: D1. Data preprocessing: cleaning, standardization and normalization of data of different modalities; D2. Feature extraction: use convolutional neural networks to extract features from image data, use gene expression analysis algorithms to extract features from gene data, and use image processing algorithms to extract features from pathological data; D3. Data fusion, using multimodal deep learning models for data fusion; D4. Model training and evaluation: Use the fused features for model training and evaluation.

2. According to claim 1, a liver cancer patient diagnostic data retrieval and accurate diagnosis and treatment system based on big data is characterized by: The data analysis and processing module uses the AutoML tool to automatically select and optimize machine learning models, feature selection methods and feature generation algorithms to automatically generate and optimize features, and uses a streaming computing framework for real-time data processing.

3. According to claim 1, a liver cancer patient diagnostic data retrieval and accurate diagnosis and treatment system based on big data is characterized by: The detailed steps of comprehensive analysis by integrating imaging data, genome data, pathological data, electronic medical record data, etc. through multimodal data fusion are as follows: S1. Data preprocessing: cleaning, standardization and normalization of data of different modalities; S2. Feature extraction: use convolutional neural network to extract features for image data, use gene expression analysis algorithm to extract features for gene data, and use image processing algorithm to extract features for pathological data; S3. Data fusion, using multimodal deep learning models for data fusion; S4. Model training and evaluation, use the fused features for model training and evaluation.

4. According to claim 1, a liver cancer patient diagnostic data retrieval and accurate diagnosis and treatment system based on big data is characterized by: The clinical decision support system also includes model interpretability, and the model interpretability provides the interpretability of model prediction and also includes LIME algorithm and SHAP algorithm. The LIME algorithm selects instances, perturbs data, generates new data points, performs model prediction on the perturbed data points, uses a linear model to fit the prediction results of the perturbed data points, and generates explanation results. The SHAP algorithm calculates the Shapley value of each feature, uses the global Shapley value to analyze the overall contribution of each feature to the model prediction, and uses the local Shapley value to analyze the contribution of each feature to the prediction of a single instance.

Images