Optimization method for brain glioma prognosis prediction model
By constructing a prior knowledge graph for prognosis prediction of brain glioma and a deep learning model based on graph structure, combined with MRI images, the characterization and integration of multimodal and multi-level medical knowledge is solved, and the accuracy and ability of prognosis prediction are improved. Explanatory, achieving more accurate and credible diagnostic decisions.
Patent Information
- Application Number
- CN202410893516.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the auxiliary prognosis prediction process of brain glioma, how to effectively characterize multimodal and multi-level medical prior knowledge and incorporate it into deep learning models to improve prediction accuracy and model performance and avoid useless knowledge pairs Negative effects of performance.
Construct a prior knowledge graph for the prognosis prediction of brain glioma, and build a graph-based structure by analyzing the prior knowledge of doctors in the diagnosis process, using knowledge graphs and deep learning technology, combining the graph structure of MRI images, integrating them into deep neural networks, and building a graph-based structure The prediction model of the model is improved through LRP and deep Bayesian interpretation network, and a hybrid enhanced intelligent prototype verification system is built to realize human-computer communication and decision-making.
It improves the accuracy and credibility of the diagnostic results of the prognostic prediction model of brain glioma, enhances the interpretability and robustness of the model, and makes the system generate more accurate and credible results than a single artificial intelligence system.
Smart Images

Figure CN119092106A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of biological detection, and in particular relates to an optimization method for a prognosis prediction model for brain glioma. Background Art
[0002] Gliomas are the most malignant tumor type and are generally considered a fatal disease among central nervous system cancers. Due to their frequent recurrence, highly diffuse invasion, destruction of normal brain tissue, and resistance to traditional and targeted therapies, the treatment of gliomas remains one of the biggest challenges in cancer treatment.
[0003] At present, prior knowledge in the medical field is usually described in the form of natural language, which has a certain degree of multimodality, multi-level, subjectivity and ambiguity. Doctors often do not make diagnostic decisions based on a single dimension, but rather a combination of multiple empirical knowledge, including age, weight, family / personal medical history, lifestyle habits and other dimensions. In addition, differences in the way doctors read MRIs and their different perceptions of the extent of tumor invasion will also affect the diagnostic results to a certain extent. Therefore, in the process of auxiliary prognosis prediction for brain gliomas, the scientific issue of how to represent multimodal and multi-level medical prior knowledge is crucial to subsequently improving the accuracy of prognosis prediction;
[0004] Secondly, in the field of medical auxiliary diagnosis, effectively integrating the medical prior knowledge graph into the deep learning model can significantly improve the accuracy of the auxiliary diagnosis model. The design of the integration method is crucial to the performance of the model. However, it should be noted that in the hybrid enhanced intelligent system, the more prior knowledge used, the better. Some useless knowledge has almost no help in improving the performance of the model, and may even have a counterproductive effect. The medical prior knowledge graph contains a large amount of useful and useless knowledge, and it is necessary to screen out the knowledge that is valuable to the model. Therefore, how to effectively integrate the medical prior knowledge graph into the deep learning model so that the knowledge of the prior knowledge graph and the knowledge learned by the deep learning model itself can complement each other and automatically adjust to improve the accuracy of auxiliary diagnosis, and can provide valuable reference for subsequent auxiliary diagnosis decisions. For this reason, we propose an optimization method for the prognosis prediction model of brain glioma. Summary of the invention
[0005] The purpose of the present invention is to provide an optimization method for a glioma prognosis prediction model to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a method for optimizing a prognosis prediction model for brain glioma, comprising the following steps:
[0007] S1. Construction of a prior knowledge graph for glioma prognosis prediction: Analyze and extract the prior knowledge used by professional doctors in the actual diagnosis and treatment of glioma, and explore the correlation between different prior knowledge, characterize multimodal and multi-level medical prior knowledge, and construct a prior knowledge graph for glioma prognosis prediction;
[0008] S2. Establishment of a prognostic prediction model for glioma based on prior knowledge graph: Using the constructed medical prior knowledge graph and the existing patient imaging database, study the mechanism of integrating the medical prior knowledge graph into the deep learning model, and establish a prognostic prediction model for glioma based on prior knowledge graph;
[0009] S3. Study on the interpretability of the prognostic prediction model for glioma: Integrate LRP with deep neural network to construct a deep Bayesian interpretation network to achieve a complete explanation of the auxiliary prognostic prediction process and improve the interpretability of the diagnostic model;
[0010] S4. Build a hybrid enhanced intelligent prototype verification system for brain glioma prognosis prediction: Integrate prior knowledge in the medical field into a deep neural network model with strong learning ability to build a hybrid enhanced intelligent prototype verification system, realize human-computer interaction and communication through the hybrid enhanced intelligent prototype verification system, and use the decision-making ability of the hybrid enhanced intelligent prototype verification system to improve adaptability to complex situations and the ability to handle complex tasks, so that the hybrid enhanced intelligent prototype verification system can generate more accurate and reliable results than a single artificial intelligence system.
[0011] Preferably, the specific steps of constructing the prior knowledge graph for brain glioma prognosis prediction in S1 include:
[0012] S101. Analyze the typical disease characteristics used by doctors in the process of combined radiochemotherapy and targeted therapy for high-grade glioma, and combine the follow-up data of the patients with their own basic information as the medical prior knowledge used in the prognostic evaluation of glioma;
[0013] S102. Using named entity recognition, relationship extraction, and pattern recognition technology, extract the category to which the medical prior knowledge in each sample belongs;
[0014] S103. Using Transformer, encoding and decoding technology, construct a priori knowledge graph for glioma prognosis prediction, and provide effective data support for subsequent treatment and diagnosis decisions of glioma.
[0015] Preferably, the typical disease characteristics include tumor marker data and IDH mutation status; the follow-up data of the patient include the patient's PFS and OS, and the patient's recurrence or metastasis type; the patient's basic information includes age, weight, family, personal medical history and lifestyle habits.
[0016] Preferably, the establishment of the brain glioma prognosis prediction model based on the prior knowledge graph in S2 specifically includes the following steps:
[0017] S201. In-depth analysis of the constructed prior knowledge graph for prognosis prediction of brain glioma, and use of knowledge graph and graph neural network technology to mine the graph structure of the medical prior knowledge graph;
[0018] S202. Analyze the characteristic information of MRI images of patients with brain glioma through autoencoders, convolutional neural networks, and eigenvalue averaging technology, construct the node set and adjacency matrix of MRI, and extract the graph structure on MRI;
[0019] S203. The graph structure of the medical prior knowledge graph is combined with the graph structure of MRI by using feature channel fusion, eigenvalue splicing and adjacency matrix merging technology to achieve graph structure fusion, and a prognosis prediction model for high-grade glioma chemoradiotherapy combined with targeted therapy based on graph structure fusion is constructed, and the response rate of patients in the later stage to treatment is predicted.
[0020] Preferably, the interpretability study in the S3 midbrain glioma prognosis prediction model specifically includes the following steps:
[0021] S301. In-depth analysis of the constructed glioma prognosis prediction model based on prior knowledge graph, and the use of LRP to calculate the contribution of various pathogenic factors and MRI features to the diagnosis results;
[0022] S302. Use encoding and decoding, Bayesian network theory and methods to construct a deep Bayesian interpretation network and summarize the general diagnostic model of brain glioma;
[0023] S303. Learn and infer a decision analysis tree from the deep Bayesian interpretation network to obtain the pathological judgment basis for each feature of glioma, improve the interpretability of the glioma prognosis prediction model, enable doctors to analyze and understand the entire glioma prognosis decision-making process from the pathological and etiological levels, and improve the credibility of doctors and patients in the diagnosis results.
[0024] Preferably, the construction of a hybrid enhanced intelligent prototype verification system for brain glioma prognosis prediction in S4 further includes the following steps:
[0025] S401. Using the built hybrid enhanced intelligent prototype verification system, select different parameters to test each functional module in the diagnostic model;
[0026] S402. Quantify the accuracy, precision, recall, and time and space complexity of the diagnostic results, and conduct rational analysis based on actual clinical diagnosis;
[0027] S403. Use MATLAB to perform statistics and analysis on the test results, and further improve and optimize the designed medical prior knowledge graph, auxiliary diagnosis model, and model interpretability scheme.
[0028] Preferably, the prior knowledge graph includes physical indicators, test indicators, clinical characteristics, case characteristics, prognostic data and treatment methods. The physical indicators include height, age, weight and menopausal status, as well as hypertension, diabetes, coronary heart disease, pregnancy and delivery history and HPV infection. The test indicators include blood routine, biochemistry, SCC, keratin and immune factors. The clinical characteristics include tumor size, intracranial pressure symptoms, neurological and functional symptoms. The pathological characteristics include degree of differentiation, MET mutation and genetic testing. The prognostic data include PFS, ORR, OS and recurrence and metastasis locations. The treatment methods include chemotherapy drugs, targeted therapy, immunotherapy and radiotherapy doses.
[0029] Preferably, the construction process of the prior knowledge graph for glioma prognosis prediction in S1 includes weighted gene co-expression network analysis, Lasso-Cox regression analysis, multi-factor Cox regression analysis and risk score calculation.
[0030] Preferably, the weighted gene co-expression network analysis includes the following formula:
[0031] W ij =cor(x i ,x j ) β ;
[0032] Among them, W ij is the network connection strength between gene i and gene j; cor(x i ,x j ) is the correlation coefficient between the expression levels of gene i and gene j; x i ,x j is the expression level of gene i and gene j in different samples; β is an adjustable parameter used to enhance the high correlation connections in the network;
[0033] The Lasso-Cox regression analysis includes the following formula:
[0034]
[0035] Among them, L(β) is the loss function, y i is the response variable for the ith observation, β 0is the intercept term, x ij is the value of the jth explanatory variable for the ith observation, β j is the coefficient of the jth explanatory variable, λ is the regularization parameter used to control the degree of penalty, and p is the number of explanatory variables.
[0036] Preferably, the multivariate Cox regression analysis includes the following formula:
[0037] h(t)=h 0 (t)exp(β 1 x 1 +β 2 x 2 +...+β p x p );
[0038] where h(t) is the instantaneous hazard rate at time t; h 0 (t) is the baseline hazard rate; β 1 ,β 2 ,...,β p is the regression coefficient; x 1 ,x 2 ,...,x p is the explanatory variable; p is the number of explanatory variables;
[0039] The risk score calculation includes the following formula:
[0040]
[0041] Among them, RiskScore is the risk score of the individual; β j is the regression coefficient of the jth variable; x j is the value of the jth variable.
[0042] Compared with the prior art, the present invention has the following beneficial effects:
[0043] (1) The present invention introduces multimodal and multi-level medical prior knowledge used by professional doctors in the process of disease diagnosis, extracts its features by using knowledge graphs, deep learning and other technologies, and constructs a prior knowledge graph for the prognosis prediction of brain glioma, thereby solving the problem of representing medical prior knowledge and providing a theoretical basis and data support for subsequent auxiliary diagnosis decisions, which has great innovative value.
[0044] (2) The present invention proposes a new solution to integrate medical prior knowledge graphs into deep neural networks by mining the graph structure of prior knowledge graphs and MRI images, and constructs a high-level brain glioma cancer prognosis prediction model based on graph structure fusion. The prior knowledge graph and the knowledge learned by the deep neural network can complement each other and automatically adjust, thereby comprehensively improving the performance of the prognosis prediction model for locally advanced brain glioma in terms of the granularity and accuracy of diagnostic results;
[0045] (3) Based on an in-depth analysis of the glioma auxiliary prediction prognosis model based on prior knowledge graphs, this paper proposes a deep Bayesian global interpretation scheme integrating LRP, which can fully explain the glioma prognosis prediction process, improve the reliability and robustness of the glioma model, and enable doctors to analyze and understand the entire diagnostic decision-making process from the perspectives of cases and causes, thereby improving the credibility of doctors and patients in the diagnostic results. This scheme is innovative to a certain extent.
[0046] (4) The present invention integrates prior knowledge in the medical field into a deep neural network model with powerful learning capabilities to construct a hybrid enhanced intelligent prototype verification system. The hybrid enhanced intelligent prototype verification system is used to achieve human-computer interaction and communication. The decision-making ability of the hybrid enhanced intelligent prototype verification system is utilized to improve the adaptability to complex situations and the ability to handle complex tasks, so that the hybrid enhanced intelligent prototype verification system can generate more accurate and reliable results than a single artificial intelligence system. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 is a flow chart of the present invention;
[0048] Figure 2 It is a specific flow chart of S1 of the present invention;
[0049] Figure 3 It is a specific flow chart of S2 of the present invention. DETAILED DESCRIPTION
[0050] 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.
[0051] See also Figure 1-Figure 3 The present invention provides a technical solution: an optimization method for a prognosis prediction model for brain glioma, comprising the following steps:
[0052] S1. Construction of a prior knowledge graph for glioma prognosis prediction: Analyze and extract the prior knowledge used by professional doctors in the actual diagnosis and treatment of glioma, and explore the correlation between different prior knowledge, characterize multimodal and multi-level medical prior knowledge, and construct a prior knowledge graph for glioma prognosis prediction;
[0053] S2. Establishment of a prognostic prediction model for glioma based on prior knowledge graph: Using the constructed medical prior knowledge graph and the existing patient imaging database, study the mechanism of integrating the medical prior knowledge graph into the deep learning model, and establish a prognostic prediction model for glioma based on prior knowledge graph;
[0054] S3. Study on the interpretability of the prognostic prediction model for glioma: Integrate LRP with deep neural network to construct a deep Bayesian interpretation network to achieve a complete explanation of the auxiliary prognostic prediction process and improve the interpretability of the diagnostic model;
[0055] S4. Build a hybrid enhanced intelligent prototype verification system for brain glioma prognosis prediction: Integrate prior knowledge in the medical field into a deep neural network model with strong learning ability to build a hybrid enhanced intelligent prototype verification system, realize human-computer interaction and communication through the hybrid enhanced intelligent prototype verification system, and use the decision-making ability of the hybrid enhanced intelligent prototype verification system to improve adaptability to complex situations and the ability to handle complex tasks, so that the hybrid enhanced intelligent prototype verification system can generate more accurate and reliable results than a single artificial intelligence system.
[0056] In this embodiment, preferably, the specific steps of constructing the prior knowledge graph for brain glioma prognosis prediction in S1 include:
[0057] S101. Analyze the typical disease characteristics used by doctors in the process of combined radiochemotherapy and targeted therapy for high-grade glioma, and combine the follow-up data of the patients with their own basic information as the medical prior knowledge used in the prognostic evaluation of glioma;
[0058] S102. Using named entity recognition, relationship extraction, and pattern recognition technology, extract the category to which the medical prior knowledge in each sample belongs;
[0059] S103. Using Transformer, encoding and decoding technology, construct a priori knowledge graph for glioma prognosis prediction, and provide effective data support for subsequent treatment and diagnosis decisions of glioma.
[0060] In this embodiment, preferably, the typical disease characteristics include tumor marker data and IDH mutation status; the follow-up data of the patient include the patient's PFS and OS, and the patient's recurrence or metastasis type; and the patient's basic information includes age, weight, family, personal medical history and lifestyle habits.
[0061] In this embodiment, preferably, the establishment of the brain glioma prognosis prediction model based on the prior knowledge graph in S2 specifically includes the following steps:
[0062] S201. In-depth analysis of the constructed prior knowledge graph for prognosis prediction of brain glioma, and use of knowledge graph and graph neural network technology to mine the graph structure of the medical prior knowledge graph;
[0063] S202. Analyze the characteristic information of MRI images of patients with brain glioma through autoencoders, convolutional neural networks, and eigenvalue averaging technology, construct the node set and adjacency matrix of MRI, and extract the graph structure on MRI;
[0064] S203. The graph structure of the medical prior knowledge graph is combined with the graph structure of MRI by using feature channel fusion, eigenvalue splicing and adjacency matrix merging technology to achieve graph structure fusion, and a prognosis prediction model for high-grade glioma chemoradiotherapy combined with targeted therapy based on graph structure fusion is constructed, and the response rate of patients in the later stage to treatment is predicted.
[0065] In this embodiment, preferably, the interpretability study in the S3 midbrain glioma prognosis prediction model specifically includes the following steps:
[0066] S301. In-depth analysis of the constructed glioma prognosis prediction model based on prior knowledge graph, and the use of LRP to calculate the contribution of various pathogenic factors and MRI features to the diagnosis results;
[0067] S302. Use encoding and decoding, Bayesian network theory and methods to construct a deep Bayesian interpretation network and summarize the general diagnostic model of brain glioma;
[0068] S303. Learn and infer a decision analysis tree from the deep Bayesian interpretation network to obtain the pathological judgment basis for each feature of glioma, improve the interpretability of the glioma prognosis prediction model, enable doctors to analyze and understand the entire glioma prognosis decision-making process from the pathological and etiological levels, and improve the credibility of doctors and patients in the diagnosis results.
[0069] In this embodiment, preferably, the construction of a hybrid enhanced intelligent prototype verification system for brain glioma prognosis prediction in S4 further includes the following steps:
[0070] S401. Using the built hybrid enhanced intelligent prototype verification system, select different parameters to test each functional module in the diagnostic model;
[0071] S402. Quantify the accuracy, precision, recall, and time and space complexity of the diagnostic results, and conduct rational analysis based on actual clinical diagnosis;
[0072] S403. Use MATLAB to perform statistics and analysis on the test results, and further improve and optimize the designed medical prior knowledge graph, auxiliary diagnosis model, and model interpretability scheme.
[0073] In this embodiment, preferably, the prior knowledge graph includes physical indicators, test indicators, clinical characteristics, case characteristics, prognostic data and treatment methods, the physical indicators include height, age, weight and menopausal status, as well as hypertension, diabetes, coronary heart disease, pregnancy history and HPV infection, the test indicators include blood routine, biochemistry, SCC, keratin and immune factors, the clinical characteristics include tumor size, intracranial pressure symptoms, neurological and functional symptoms, the pathological characteristics include degree of differentiation, MET mutation and gene detection, the prognostic data include PFS, ORR, OS and recurrence and metastasis location, and the treatment methods include chemotherapy drugs, targeted therapy, immunotherapy and radiotherapy dose.
[0074] In this embodiment, preferably, the construction process of the prior knowledge graph for brain glioma prognosis prediction in S1 includes weighted gene co-expression network analysis, Lasso-Cox regression analysis, multi-factor Cox regression analysis and risk score calculation.
[0075] In this embodiment, preferably, the weighted gene co-expression network analysis includes the following formula:
[0076] W ij =cor(x i ,x j ) β ;
[0077] Among them, W ij is the network connection strength between gene i and gene j; cor(x i ,x j ) is the correlation coefficient between the expression levels of gene i and gene j; x i ,x j is the expression level of gene i and gene j in different samples; β is an adjustable parameter used to enhance the high correlation connections in the network;
[0078] The Lasso-Cox regression analysis includes the following formula:
[0079]
[0080] Among them, L(β) is the loss function, y i is the response variable for the ith observation, β 0 is the intercept term, x ij
[0081] is the value of the jth explanatory variable for the ith observation, β j is the coefficient of the jth explanatory variable, λ is the regularization parameter used to control the degree of penalty, and p is the number of explanatory variables.
[0082] In this embodiment, preferably, the multi-factor Cox regression analysis includes the following formula:
[0083] h(t)=h 0 (t)exp(β 1 x 1 +β 2 x 2 +...+β p x p );
[0084] where h(t) is the instantaneous hazard rate at time t; h 0 (t) is the baseline hazard rate; β 1 ,β 2 ,...,β p is the regression coefficient; x 1 ,x 2 ,...,x p is the explanatory variable; p is the number of explanatory variables;
[0085] The risk score calculation includes the following formula:
[0086]
[0087] Among them, RiskScore is the risk score of the individual; β j is the regression coefficient of the jth variable; x j is the value of the jth variable.
[0088] Principles and advantages of the present invention:
[0089] The present invention introduces multimodal and multi-level medical prior knowledge used by professional doctors in the process of disease diagnosis, and uses knowledge graphs, deep learning and other technologies to extract features, constructs a priori knowledge graph for brain glioma prognosis prediction, solves the problem of representation of medical prior knowledge, and provides a theoretical basis and data support for subsequent auxiliary diagnosis decisions, which has great innovative value; by mining the graph structure of prior knowledge graphs and MRI images, a new scheme of integrating medical prior knowledge graphs into deep neural networks is proposed, and a high-level brain glioma cancer prognosis prediction model based on graph structure fusion is constructed, which can make the prior knowledge graph and the knowledge learned by the deep neural network complement each other and automatically adjust, so as to comprehensively improve the performance of the prognosis prediction model for locally advanced brain glioma in terms of the granularity and accuracy of the diagnostic results; in-depth analysis of the prior knowledge graph based on the prior knowledge graph and the deep neural network can be used to predict the prognosis of local advanced brain glioma. Based on the atlas-assisted prediction model of glioma, a deep Bayesian global interpretation scheme integrating LRP is proposed, which can fully explain the prognosis prediction process of glioma, improve the reliability and robustness of the glioma model, and enable doctors to analyze and understand the entire diagnostic decision-making process from the levels of cases and causes, thereby improving the credibility of doctors and patients in the diagnostic results, which is innovative. Prior knowledge in the medical field is integrated into a deep neural network model with strong learning ability to construct a hybrid enhanced intelligent prototype verification system, and human-computer interaction and communication are achieved through the hybrid enhanced intelligent prototype verification system. The decision-making ability of the hybrid enhanced intelligent prototype verification system is used to improve the adaptability to complex situations and the ability to handle complex tasks, so that the hybrid enhanced intelligent prototype verification system can generate more accurate and reliable results than a single artificial intelligence system.
[0090] 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 method for optimizing a prognosis prediction model for brain glioma, characterized in that: The following steps are involved: S1. Construction of a prior knowledge graph for glioma prognosis prediction: Analyze and extract the prior knowledge used by professional doctors in the actual diagnosis and treatment of glioma, and explore the correlation between different prior knowledge, characterize multimodal and multi-level medical prior knowledge, and construct a prior knowledge graph for glioma prognosis prediction; S2. Establishment of a prognostic prediction model for glioma based on prior knowledge graph: Using the constructed medical prior knowledge graph and the existing patient imaging database, study the mechanism of integrating the medical prior knowledge graph into the deep learning model, and establish a prognostic prediction model for glioma based on prior knowledge graph; S3. Study on the interpretability of the prognostic prediction model for glioma: Integrate LRP with deep neural network to construct a deep Bayesian interpretation network to achieve a complete explanation of the auxiliary prognostic prediction process and improve the interpretability of the diagnostic model; S4. Build a hybrid enhanced intelligent prototype verification system for brain glioma prognosis prediction: Integrate prior knowledge in the medical field into a deep neural network model with strong learning ability to build a hybrid enhanced intelligent prototype verification system, realize human-computer interaction and communication through the hybrid enhanced intelligent prototype verification system, and use the decision-making ability of the hybrid enhanced intelligent prototype verification system to improve adaptability to complex situations and the ability to handle complex tasks, so that the hybrid enhanced intelligent prototype verification system can generate more accurate and reliable results than a single artificial intelligence system.
2. The optimization method for a prognosis prediction model for glioma according to claim 1, characterized in that: The specific steps of constructing the prior knowledge graph for brain glioma prognosis prediction in S1 include: S101. Analyze the typical disease characteristics used by doctors in the process of combined radiochemotherapy and targeted therapy for high-grade glioma, and combine the follow-up data of the patients with their own basic information as the medical prior knowledge used in the prognostic evaluation of glioma; S102. Using named entity recognition, relationship extraction, and pattern recognition technology, extract the category to which the medical prior knowledge in each sample belongs; S103. Using Transformer, encoding and decoding technology, construct a priori knowledge graph for glioma prognosis prediction, and provide effective data support for subsequent treatment and diagnosis decisions of glioma.
3. The optimization method for the prognosis prediction model of brain glioma according to claim 2, characterized in that: The typical disease characteristics include tumor marker data and IDH mutation status; the follow-up data of the patient include the patient's PFS and OS, and the type of recurrence or metastasis of the patient; the basic information of the patient includes age, weight, family, personal medical history and lifestyle habits.
4. The optimization method for a prognosis prediction model for glioma according to claim 3, characterized in that: The establishment of the brain glioma prognosis prediction model based on the prior knowledge graph in S2 specifically includes the following steps: S201. In-depth analysis of the constructed prior knowledge graph for prognosis prediction of brain glioma, and use of knowledge graph and graph neural network technology to mine the graph structure of the medical prior knowledge graph; S202. Analyze the characteristic information of MRI images of patients with brain glioma through autoencoders, convolutional neural networks, and eigenvalue averaging technology, construct the node set and adjacency matrix of MRI, and extract the graph structure on MRI; S203. The graph structure of the medical prior knowledge graph is combined with the graph structure of MRI by using feature channel fusion, eigenvalue splicing and adjacency matrix merging technology to achieve graph structure fusion, and a prognosis prediction model for high-grade glioma chemoradiotherapy combined with targeted therapy based on graph structure fusion is constructed, and the response rate of patients in the later stage to treatment is predicted.
5. The optimization method for the prognosis prediction model of brain glioma according to claim 4, characterized in that: The interpretability study of the S3 midbrain glioma prognosis prediction model specifically includes the following steps: S301. In-depth analysis of the constructed glioma prognosis prediction model based on prior knowledge graph, and the use of LRP to calculate the contribution of various pathogenic factors and MRI features to the diagnosis results; S302. Use encoding and decoding, Bayesian network theory and methods to construct a deep Bayesian interpretation network and summarize the general diagnostic model of brain glioma; S303. Learn and infer a decision analysis tree from the deep Bayesian interpretation network to obtain the pathological judgment basis for each feature of glioma, improve the interpretability of the glioma prognosis prediction model, enable doctors to analyze and understand the entire glioma prognosis decision-making process from the pathological and etiological levels, and improve the credibility of doctors and patients in the diagnosis results.
6. The optimization method for the prognosis prediction model of brain glioma according to claim 5, characterized in that: The hybrid enhanced intelligent prototype verification system for brain glioma prognosis prediction in S4 also includes the following steps: S401. Using the built hybrid enhanced intelligent prototype verification system, select different parameters to test each functional module in the diagnostic model; S402. Quantify the accuracy, precision, recall, and time and space complexity of the diagnostic results, and conduct rational analysis based on actual clinical diagnosis; S403. Use MATLAB to perform statistics and analysis on the test results, and further improve and optimize the designed medical prior knowledge graph, auxiliary diagnosis model, and model interpretability scheme.
7. The optimization method for a prognosis prediction model for glioma according to claim 1, characterized in that: The prior knowledge graph includes physical indicators, test indicators, clinical characteristics, case characteristics, prognostic data and treatment methods. The physical indicators include height, age, weight and menopausal status, as well as hypertension, diabetes, coronary heart disease, pregnancy and delivery history and HPV infection. The test indicators include blood routine, biochemistry, SCC, keratin and immune factors. The clinical characteristics include tumor size, intracranial pressure symptoms, neurological and functional symptoms. The pathological characteristics include degree of differentiation, MET mutation and genetic testing. The prognostic data include PFS, ORR, OS and recurrence and metastasis locations. The treatment methods include chemotherapy drugs, targeted therapy, immunotherapy and radiotherapy doses.
8. The optimization method for a prognosis prediction model for glioma according to claim 1, characterized in that: The construction process of the prior knowledge graph for brain glioma prognosis prediction in S1 includes weighted gene co-expression network analysis, Lasso-Cox regression analysis, multifactor Cox regression analysis and risk score calculation.
9. The optimization method for the prognosis prediction model of brain glioma according to claim 8, characterized in that: The weighted gene co-expression network analysis includes the following formula: W ij =cor(x i ,x j ) β ; Among them, W ij is the network connection strength between gene i and gene j; cor(x i ,x j ) is the correlation coefficient between the expression levels of gene i and gene j; x i ,x j is the expression level of gene i and gene j in different samples; β is an adjustable parameter used to enhance the high correlation connections in the network; The Lasso-Cox regression analysis includes the following formula: Among them, L(β) is the loss function, y i is the response variable of the ith observation, β0 is the intercept term, and x ij is the value of the jth explanatory variable for the ith observation, β j is the coefficient of the jth explanatory variable, λ is the regularization parameter used to control the degree of penalty, and p is the number of explanatory variables.
10. The optimization method for the prognosis prediction model of brain glioma according to claim 9, characterized in that: The multivariate Cox regression analysis includes the following formula: h(t)=h0(t)exp(β1x1+β2x2+...+β p x p ); where h(t) is the instantaneous hazard rate at time t; h0(t) is the baseline hazard rate; β1,β2,...,β p are the regression coefficients; x1,x2,...,x p is the explanatory variable; p is the number of explanatory variables; The risk score calculation includes the following formula: Among them, RiskScore is the risk score of the individual; β j is the regression coefficient of the jth variable; x j is the value of the jth variable.
Citation Information
Cited By
Interpretable method and system for ventricular premature beat auxiliary diagnosis model
CN120131035A
An interpretable method and system for auxiliary diagnostic models of ventricular premature beats
CN120131035B
Brain tumor survival prediction method and system based on multi-modal medical knowledge graph
CN121565447A
