Data analysis method in high-value medical data asset construction process

Abstract

The invention provides a data analysis method in a high-value medical data asset construction process, and relates to the technical field of medical data analysis, and the method comprises the following steps: generating a gene feature subset, a clinical feature subset and an image feature subset; performing time dimension coding on the gene feature subset to generate a gene expression time sequence vector; marking timestamps for the clinical feature subsets, and extracting clinical event time sequence features; carrying out three-dimensional space registration on the image feature subset, and extracting space-time dynamic features of the focus area; based on the spatio-temporal dynamic features, image focus coordinates are obtained, and the image focus coordinates are mapped to a gene-clinical united space; constructing a heterogeneous graph network comprising gene nodes, clinical nodes and image nodes, and calculating edge weights among the nodes; in the heterogeneous graph network, redundant edges are cut off, a space-time correlation graph is generated, and then a correlation mapping matrix is obtained. According to the method, heterogeneous data are integrated, so that data originally dispersed in different systems can establish an effective relation.

Description

Technical Field

[0001] The present application relates to the technical field of medical data analysis, and in particular to a data analysis method in the process of constructing high-value medical data assets. Background Art

[0002] With the rapid development of medical information technology, medical data has shown explosive growth. These data contain huge value and can provide strong support for disease diagnosis, treatment plan formulation, etc. However, since medical data often comes from heterogeneous data sources, when doctors or hospitals want to study the relationship between certain disease types and one or several medical data, it is difficult to quickly and accurately extract data related to the disease from different systems and establish connections, resulting in low analysis efficiency. Summary of the invention

[0003] In view of the above-mentioned defects or deficiencies in the prior art, the present application aims to provide a data analysis method in the process of building high-value medical data assets to establish a connection between multimodal medical data. The method includes the following steps: Extract effective gene loci related to the target disease and generate a gene feature subset; extract clinical indicators associated with the target disease and generate a clinical feature subset; extract quantitative features of the lesion area and generate an image feature subset; The gene feature subset is encoded in the time dimension to generate a gene expression time series vector; the clinical feature subset is marked with a timestamp to extract the clinical event time series features; the image feature subset is registered in three-dimensional space to extract the spatiotemporal dynamic features of the lesion area; Aligning the gene expression time series of the gene expression time series vector with the clinical event time series of the clinical event time series feature, and calculating the time alignment similarity; Based on the spatiotemporal dynamic characteristics, image lesion coordinates are obtained, the image lesion coordinates are mapped to the gene-clinical joint space, and the spatial correlation is calculated; Constructing a heterogeneous graph network including gene nodes, clinical nodes, and imaging nodes, and calculating edge weights between nodes based on the time alignment similarity and the spatial correlation; In the heterogeneous graph network, redundant edges corresponding to the edge weights below a first preset threshold are pruned to generate a spatiotemporal association graph, thereby obtaining an association mapping matrix.

[0004] According to the technical solution provided by the present application, the generation of a spatiotemporal correlation map and then obtaining a correlation mapping matrix includes the following steps: Converting the spatiotemporal association graph into a weighted adjacency matrix, wherein the matrix element value of the weighted adjacency matrix is ​​the weight of the corresponding edge; Performing symmetric normalization processing on the weighted adjacency matrix to obtain a normalized matrix; Perform singular value decomposition on the normalized matrix, retain the first k principal components, and generate a low-dimensional mapping matrix as the association mapping matrix.

[0005] According to the technical solution provided by the present application, the singular value decomposition of the normalized matrix is ​​performed, the first k principal components are retained, and a low-dimensional mapping matrix is ​​generated as the association mapping matrix, comprising the following steps: Performing singular value decomposition on the normalized matrix, retaining the first k principal components, generating a low-dimensional mapping matrix, and generating feature clusters based on the low-dimensional mapping matrix; The cluster labels of the feature clusters are fed back to the singular value decomposition as auxiliary information, and the truncation rank k value in the singular value decomposition process is iteratively optimized to obtain the association mapping matrix.

[0006] According to the technical solution provided in this application, the extraction of effective gene loci related to the target disease includes the following steps: Through the chi-square test, the SNP sites whose occurrence frequency in the population exceeds the second preset threshold are screened to obtain a candidate SNP list; Obtain the GWAS database, compare the candidate SNP list with the GWAS database, and retain the SNP sites with p-values ​​less than a third preset threshold as valid gene sites; wherein the third preset threshold is less than 0.05.

[0007] According to the technical solution provided by the present application, encoding the gene feature subset in the time dimension to generate a gene expression time series vector includes the following steps: For moments when gene expression was missing, cubic spline interpolation was used to complete the gap; The multi-scale time series features of gene expression are extracted by wavelet transform, and the multi-scale time series features include high-frequency fluctuation components and low-frequency trend components.

[0008] According to the technical solution provided in the present application, the step of aligning the gene expression timing of the gene expression timing vector with the clinical event timing of the clinical event timing feature comprises the following steps: Performing dynamic time warping on the gene expression time series and the clinical event time series to generate a time alignment path; Calculating the Pearson correlation coefficient of the gene-clinical feature pair along the time alignment path, and screening the feature pairs whose correlation coefficients are greater than a fourth preset threshold as strongly associated feature pairs; Based on the strongly associated feature pairs, construct a gene-clinical initial association matrix, wherein the matrix elements of the initial association matrix are correlation coefficient values; Performing non-negative matrix decomposition on the initial correlation matrix to obtain a gene feature subspace and a clinical feature subspace; The gene feature subspace and the clinical feature subspace are orthogonally projected and merged to generate the gene-clinical joint space.

[0009] According to the technical solution provided in this application, the extraction of effective gene loci related to the target disease and the generation of gene feature subsets include the following steps: Performing random projection dimensionality reduction on the expression time series vectors of the screened effective gene loci to generate a feature space after dimensionality reduction; Add Laplace noise to the feature space after dimensionality reduction to satisfy -Differential privacy constraints are used to obtain a subset of genetic features.

[0010] According to the technical solution provided by this application, the method also includes the following steps: When constructing the edge weights of a heterogeneous graph network, noise compensation is performed on a subset of gene features to obtain corrected weights, which are then used to generate a spatiotemporal association map.

[0011] According to the technical solution provided in this application, extracting clinical indicators associated with the target disease and generating a clinical feature subset includes the following steps: Extract clinical indicators associated with the target disease and homomorphically encrypt the patient identifier to obtain a subset of clinical features; The feature similarity between the genetic features and the encrypted clinical features is calculated in the encrypted domain, and the clinical feature subset is output.

[0012] According to the technical solution provided by this application, after homomorphically encrypting the patient identity identifier, the following steps are also included: The correlation coefficients between gene signatures and subsets of clinical signatures were calculated in the encrypted domain; In the orthogonal projection of the joint gene-clinical space, a subset of clinical features is locally decrypted, exposing only the standardized feature values ​​to participate in the spatial merging.

[0013] Compared with the prior art, the beneficial effect of this application is that this application integrates heterogeneous data such as genetic, clinical, and imaging data from different systems into a joint feature space, and establishes a multi-dimensional association mapping matrix, so that data originally scattered in different systems can establish effective connections, greatly improving the efficiency of data association analysis and reducing the time cost of data extraction and integration. The multi-dimensional association mapping matrix can be used to fully explore the potential information in multimodal medical data, provide a more accurate and valuable diagnosis and treatment decision feature set for disease diagnosis, treatment plan formulation, etc., and help to subsequently improve the accuracy and scientificity of medical decisions and achieve precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1A flowchart of the steps of the data analysis method in the process of building high-value medical data assets provided in this application. DETAILED DESCRIPTION

[0015] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0016] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0017] Example 1 As mentioned in the background technology, in response to the problems in the prior art, this application proposes to provide a data analysis method in the process of building high-value medical data assets to establish connections between multimodal medical data, such as Figure 1 As shown, the method comprises the following steps: S1. Extract effective gene loci related to the target disease and generate a gene feature subset; extract clinical indicators associated with the target disease and generate a clinical feature subset; extract quantitative features of the lesion area and generate an image feature subset; Furthermore, the extraction of effective gene loci related to the target disease comprises the following steps: Through the chi-square test, the SNP sites whose occurrence frequency in the population exceeds the second preset threshold are screened to obtain a candidate SNP list; Obtain the GWAS database, compare the candidate SNP list with the GWAS database, and retain the SNP sites with p-values ​​less than a third preset threshold as valid gene sites; wherein the third preset threshold is less than 0.05.

[0018] S2, encoding the gene feature subset in the time dimension to generate a gene expression time series vector; marking the clinical feature subset with a timestamp to extract the clinical event time series features; performing three-dimensional spatial registration on the image feature subset to extract the spatiotemporal dynamic features of the lesion area; Furthermore, encoding the gene feature subset in the time dimension to generate a gene expression time series vector comprises the following steps: For moments when gene expression was missing, cubic spline interpolation was used to complete the gap; The multi-scale time series features of gene expression are extracted by wavelet transform, and the multi-scale time series features include high-frequency fluctuation components and low-frequency trend components.

[0019] S3, aligning the gene expression time series of the gene expression time series vector with the clinical event time series of the clinical event time series feature, and calculating the time alignment similarity; Furthermore, the aligning the gene expression timing of the gene expression timing vector with the clinical event timing of the clinical event timing feature comprises the following steps: Performing dynamic time warping on the gene expression time series and the clinical event time series to generate a time alignment path; Calculating the Pearson correlation coefficient of the gene-clinical feature pair along the time alignment path, and screening the feature pairs whose correlation coefficients are greater than a fourth preset threshold as strongly associated feature pairs; Based on the strongly associated feature pairs, construct a gene-clinical initial association matrix, wherein the matrix elements of the initial association matrix are correlation coefficient values; Performing non-negative matrix decomposition on the initial correlation matrix to obtain a gene feature subspace and a clinical feature subspace; The gene feature subspace and the clinical feature subspace are orthogonally projected and merged to generate the gene-clinical joint space.

[0020] S4, based on the spatiotemporal dynamic characteristics, obtaining the image lesion coordinates, mapping the image lesion coordinates to the gene-clinical joint space, and calculating the spatial correlation; S5, constructing a heterogeneous graph network including gene nodes, clinical nodes, and imaging nodes, and calculating edge weights between nodes based on the time alignment similarity and the spatial correlation; S6. In the heterogeneous graph network, redundant edges corresponding to the edge weights that are lower than a first preset threshold are pruned to generate a spatiotemporal correlation graph, and then obtain a correlation mapping matrix.

[0021] Furthermore, the generating of the spatiotemporal correlation map and obtaining the correlation mapping matrix comprises the following steps: Converting the spatiotemporal association graph into a weighted adjacency matrix, wherein the matrix element value of the weighted adjacency matrix is ​​the weight of the corresponding edge; Performing symmetric normalization processing on the weighted adjacency matrix to obtain a normalized matrix; Perform singular value decomposition on the normalized matrix, retain the first k principal components, and generate a low-dimensional mapping matrix as the association mapping matrix.

[0022] Furthermore, performing singular value decomposition on the normalized matrix, retaining the first k principal components, and generating a low-dimensional mapping matrix as the association mapping matrix comprises the following steps: Performing singular value decomposition on the normalized matrix, retaining the first k principal components, generating a low-dimensional mapping matrix, and generating feature clusters based on the low-dimensional mapping matrix; The cluster labels of the feature clusters are fed back to the singular value decomposition as auxiliary information, and the truncation rank k value in the singular value decomposition process is iteratively optimized to obtain the association mapping matrix.

[0023] In the existing technology, hospitals have accumulated a large amount of multimodal data corresponding to patients. Often, diagnosis and treatment data of different modalities are stored in different systems. When doctors want to study the relationship between specific gene mutations and heart disease types, it is difficult to quickly and accurately extract relevant data from different systems and establish connections, resulting in low analysis efficiency. In addition, when hospitals need to cooperate with external scientific research institutions to study the treatment effects of diseases, they do not protect patient privacy enough and directly provide the original clinical data of patients to the other party, which poses a risk of patient privacy leakage. At the same time, in traditional technologies, when doctors formulate treatment plans for new patients, they lack high-value mining of previous medical data and cannot fully support clinical diagnosis and treatment decisions. They can only refer to simple clinical indicators and past experience, making it difficult to formulate personalized treatment plans, which affects the treatment effect. Based on this, this solution is proposed.

[0024] Specifically, these different modal data are associated based on the patient's unique identifier (such as medical record number) and time information (such as time of consultation, time of symptom onset, etc.). Then, these associations are converted into an association mapping matrix, and the elements in the matrix represent the degree of association between different feature subsets, such as the degree of association between genetic features and clinical symptom features, and the degree of association between imaging features and clinical diagnosis features. For example, genetic features are used as one dimension, clinical features as another dimension, and imaging features as the third dimension. The elements in the association mapping matrix determine the relationship and connection between these dimensions. In this way, the originally scattered multimodal features are integrated into a unified space to form a structured joint feature space. In this space, the features of different modalities no longer exist in isolation, but are interrelated and influence each other.

[0025] For example, a heart hospital has accumulated a large amount of patient genomic data, clinical text data (such as medical records, diagnostic reports), and medical imaging data such as cardiac ultrasound and electrocardiogram. The target disease is determined to be coronary heart disease. In the coronary heart disease patient population, the gene loci related to atherosclerosis (such as APOE, PCSK9) are screened through whole genome sequencing data, and the SNP loci with a frequency of >5% (the second preset threshold) in the population are retained. For example, the rs7412 locus (APOE gene) is screened as a candidate gene feature. Clinical indicators related to coronary heart disease are extracted from electronic medical records. For structured data: left ventricular ejection fraction (LVEF) and BNP (brain natriuretic peptide) levels are directly recorded. For unstructured text: key descriptions such as "chest pain duration >30 minutes" and "ST segment elevation" are identified through NLP, and standardized into ICD-10 codes (such as I21.0 acute anterior wall myocardial infarction). The cardiac CT images are reconstructed in three dimensions to extract quantitative features such as coronary artery calcification score (CAC Score) and myocardial ischemic area volume. For example, the degree of stenosis of the left anterior descending artery (such as stenosis rate of 70%) is calculated by deep learning segmentation algorithm. Then, time dimension encoding and time series alignment are performed: gene time series vector: record the APOE gene expression of patients 1 week, 1 month, and 3 months after surgery, and use cubic spline interpolation to fill in missing data; clinical event time series: mark key time points (such as "arrhythmia occurred on the 3rd day after surgery"); dynamic time warping alignment: align the gene expression peak time (such as APOE expression increased sharply 1 month after surgery) with the clinical index deterioration time (such as LVEF decreased to 40% 1 month after surgery), calculate the time offset similarity (such as DTW distance = 0.15), and construct a weighted adjacency matrix. Three types of nodes are defined in the heterogeneous graph network: gene nodes: APOE, PCSK9 and other gene sites; clinical nodes: LVEF, BNP and other indicators; imaging nodes: coronary artery calcification score, myocardial ischemic volume. Edge weight calculation: Time alignment similarity: DTW correlation coefficient between APOE gene expression and LVEF change = 0.82; Spatial correlation: Map the coordinates of the myocardial ischemic area to the gene-clinical joint space, and calculate the spatial correlation between the APOE gene and the ischemic volume (such as Pearson r = 0.75). Then, perform graph pruning and matrix decomposition: prune redundant edges with weight < 0.6 (the first preset threshold) (such as the association weight between PCSK9 and BNP = 0.45, which was removed); symmetric normalization of the remaining adjacency matrix (formula: D - ¹ / ²AD -¹ / ²), singular value decomposition (SVD) retains the first k=3 principal components (cumulative contribution rate 92%) to generate a 3D mapping matrix. For example, the first principal component represents the "core pathway of atherosclerosis progression", which includes APOE gene, LVEF and calcification score. After the initial decomposition, it was found that the characteristic cluster silhouette coefficient of APOE gene and calcification score = 0.68 (<0.7 threshold), triggering k value optimization, specifically increasing k=5, and the silhouette coefficient increased to 0.73 after re-decomposition. In the final association mapping matrix, the association weight of APOE gene and calcification score increased from 0.71 to 0.89. For example, the APOE gene expression time series (1 week to 3 months after surgery) and the LVEF change time series were aligned; the gene expression peak (1 month after surgery) corresponds to the lowest LVEF value (40%); the optimal alignment path was generated, the correlation coefficient r=0.85 (>0.7 threshold) was calculated, and the gene-clinical initial correlation matrix was constructed (APOE and LVEF correlation coefficient=0.85, PCSK9 and BNP=0.72); NMF decomposition obtained: gene subspace: weight vectors of APOE and PCSK9 genes (0.9, 0.1); clinical subspace: weight vectors of LVEF and BNP (0.8, 0.2); after orthogonal projection merging, the gene-clinical joint space coordinates of APOE-LVEF are (0.72, 0.15) reflects a strong association. Through the association mapping matrix, it can be obtained that among patients with high APOE gene expression, the risk of cardiac mortality increased by 3.2 times in the group with LVEF < 50% and calcification score > 400. Doctors can intervene in high-risk patients in advance (such as intensive lipid-lowering therapy) to support diagnosis and treatment decisions. The gene-imaging joint space can reveal that coronary artery calcification progresses faster in APOE ε4 allele carriers (p = 0.003), providing a basis for targeted treatment and having certain scientific research value.

[0026] Embodiment 2: Based on Example 1, considering the high sensitivity of medical data (traditional medical data analysis often regards data processing and privacy protection as separate independent stages, resulting in either sacrificing data utility to achieve strong privacy protection or exposing sensitive information in exchange for analysis accuracy), this example proposes the following solution to enable this solution to simultaneously implement a privacy enhancement mechanism in the process of constructing a spatiotemporal correlation graph. Specifically: In a preferred embodiment, extracting effective gene loci associated with the target disease and generating a gene feature subset comprises the following steps: Performing random projection dimensionality reduction on the expression time series vectors of the screened effective gene loci to generate a feature space after dimensionality reduction; Add Laplace noise to the feature space after dimensionality reduction to satisfy -Differential privacy constraints are used to obtain a subset of genetic features.

[0027] In a preferred embodiment, extracting clinical indicators associated with the target disease and generating a clinical feature subset comprises the following steps: Clinical indicators associated with the target disease are extracted, and patient identifiers are homomorphically encrypted to obtain a subset of clinical features.

[0028] The feature similarity between the genetic features and the encrypted clinical features is calculated in the encrypted domain, and the clinical feature subset is output.

[0029] Specifically, for the perturbed gene feature subset, the random projection transformation algorithm is used to transform the features of the effective gene loci, and then noise is injected according to the requirements of (ε,δ)-differential privacy. For example, random noise within a certain range is added to the gene expression data. The ciphertext features use the Paillier homomorphic encryption algorithm to encrypt the highly sensitive patient identity features in the clinical indicator feature subset, and complete the feature similarity comparison in the encrypted domain, so that even if the data is obtained, the true identity of the patient cannot be known without the decryption key. The original form of the image feature subset is directly retained, because the privacy sensitivity of the image feature itself is relatively low, and retaining the original form is conducive to subsequent analysis.

[0030] Furthermore, the noise injection amount of different feature dimensions is dynamically adjusted through the privacy budget monitor to ensure that the overall privacy budget consumption does not exceed the first preset budget. The following steps are included: Based on the privacy budget monitor, the information entropy of each feature dimension is calculated in real time; Specifically, the information entropy calculation formula is used to analyze the distribution of gene loci. If the distribution of gene loci in the patient population is relatively uniform, that is, the frequencies of occurrence of different gene types are similar, then the information entropy is high, indicating that the uncertainty contained in this feature dimension is large; conversely, if the gene loci are concentrated, the information entropy is low. For the clinical indicator feature dimension, such as the age distribution of patients, if the age span is large and evenly distributed, the information entropy is high; if most patients are concentrated in a certain age group, the information entropy is low. The same is true for the image feature dimension. The information entropy is calculated by analyzing the distribution and diversity of the lesion features in the image.

[0031] Establish an inverse rule for allocating privacy budgets. Based on this inverse rule, allocate privacy budgets according to information entropy. Specifically, because features with high information entropy have high uncertainty, the need for privacy protection is relatively low; while features with low information entropy have high certainty and may cause greater privacy risks once leaked, so more privacy protection resources are needed. Therefore, feature dimensions with high information entropy are allocated with less privacy budget, and feature dimensions with low information entropy are allocated with more privacy budget. For example, in the gene sequence feature dimension, some highly polymorphic gene regions have high information entropy and are allocated with less privacy budget; while some gene loci related to key diseases and relatively stable have low information entropy and are allocated with more privacy budget. In actual operation, according to the pre-set algorithm, the overall privacy budget is allocated according to the information entropy ratio of each feature dimension.

[0032] When it is detected that the privacy budget allocation fluctuation exceeds the second preset budget within N consecutive cycles, the feature dimension isolation mechanism is automatically triggered to suspend the data of this dimension from participating in feature evolution.

[0033] Specifically, the system will continuously track the fluctuations in the privacy budget allocation. When it is detected that the fluctuation of the privacy budget allocation of a certain feature dimension exceeds the second preset budget (such as 0.1) for N consecutive cycles (assuming N = 5), the feature dimension isolation mechanism is automatically triggered.

[0034] For example, in the data processing process of analyzing the relationship between the treatment effect of heart disease patients and genes and clinical indicators, it is found that the privacy budget allocation of a clinical indicator feature dimension fluctuates by more than 0.1 for 5 consecutive cycles, and the data of this dimension is suspended from participating in feature evolution. During the isolation period, the noise injection amount of this feature dimension will no longer be adjusted, and other operations related to feature evolution will not be performed until the fluctuation returns to normal or after further safety assessment, and then decide whether to re-include the data of this dimension in the feature evolution.

[0035] In a preferred embodiment, the method further comprises the following steps: When constructing the edge weights of a heterogeneous graph network, noise compensation is performed on a subset of gene features to obtain corrected weights, which are then used to generate a spatiotemporal association map.

[0036] In a preferred embodiment, after homomorphically encrypting the patient identifier, the following steps are further included: The correlation coefficients between gene signatures and subsets of clinical signatures were calculated in the encrypted domain; In the orthogonal projection of the joint gene-clinical space, a subset of clinical features is locally decrypted, exposing only the standardized feature values ​​to participate in the spatial merging.

[0037] Further, a trainable weight matrix is ​​obtained, and the trainable weight matrix is ​​combined with the association mapping matrix to obtain a feature association strength matrix; and a feature dimension risk score is comprehensively generated; including the following steps: Applying the trainable weight matrix to the association mapping matrix, assigning attention weights to different feature nodes, and generating feature importance scores, wherein the feature importance scores are used to measure the closeness between features; Integrating the feature importance scores to obtain a feature association strength matrix; Based on the feature dimensions of the feature association strength matrix, potential privacy attack behaviors are simulated and the possibility of privacy leakage of each feature dimension is evaluated to output the risk probability distribution corresponding to each feature dimension, thereby obtaining the feature dimension risk score.

[0038] Specifically, the trainable weight matrix can be obtained through machine learning algorithms, such as the training process of a neural network. Apply it to the association mapping matrix, and assign attention weights to different feature nodes. For example, in a disease diagnosis model, gene feature A and clinical indicator B have a greater impact on the diagnosis result, so they are assigned higher attention weights to generate feature importance scores. These scores are integrated to obtain a feature association strength matrix, and then based on this matrix, potential privacy attack behaviors are simulated to evaluate the possibility of privacy leakage in each feature dimension. For example, by simulating hackers invading a data system, analyze which feature dimensions are more likely to be cracked and leak privacy, output risk probability distribution, and obtain feature dimension risk scores.

[0039] Further, generating a dynamic feature weight matrix according to the feature association strength matrix and the feature dimension risk score comprises the following steps: Through the dual-objective reward function R = α·U(F)-β·P(F), the dynamic feature weight matrix is ​​obtained; Among them, U(F) is the clinical utility evaluation value obtained based on the feature subset F involved in the feature association strength matrix, P(F) is the privacy leakage risk evaluation value quantified based on the privacy leakage risk reflected by the feature dimension risk score, and α and β are dynamic adjustment coefficients. The dynamic adjustment coefficient in the dual-objective reward function is determined by the following method: Establish the medical data value density function V(x)=log(1+IC(x)·SC(x)), where IC(x) is the information content index and SC(x) is the clinical significance index; Specifically, for each medical data x, its information content index IC (x) and clinical significance index SC (x) must be determined first. IC (x) can be measured by calculating the information entropy of the data. The higher the information entropy, the greater the uncertainty and information contained in the data. For example, for genetic testing data, the polymorphism of certain gene loci is complex, and its IC (x) value is higher. SC (x) needs to be evaluated in combination with the opinions of clinical experts and medical research results. For example, for ejection fraction data, which plays a key role in diagnosing heart failure, its SC (x) value is higher. By substituting IC (x) and SC (x) into the formula, the value density of each data can be obtained.

[0040] According to the security level requirements of real-time data access scenarios, the α / β ratio is dynamically adjusted through the fuzzy logic controller; Specifically, the fuzzy logic controller monitors the security level requirements of data access scenarios in real time. The security level can be divided according to factors such as the identity of the visitor (such as internal doctors, external researchers), the purpose of the visit (daily diagnosis and treatment, scientific research analysis), etc. For example, when external researchers conduct large-scale data access for scientific research projects, the security level may be set to a medium level. The fuzzy logic controller dynamically adjusts the α / β ratio based on the security level according to pre-set fuzzy rules. If the security level is high, more attention is paid to privacy protection, and the ratio of β to α will be appropriately increased; if the security level is low, more emphasis is placed on the clinical utility of the data, and the ratio of α to β will be increased. According to different data usage scenarios, the relationship between clinical utility and privacy protection can be flexibly balanced to improve the security and effectiveness of data use.

[0041] When abnormal access behavior is detected, the β value multiplication mechanism is automatically triggered to enhance the privacy protection strength.

[0042] Specifically, when the system detects abnormal access behavior, such as a large number of unauthorized access requests in a short period of time, or access from suspicious IP addresses, the β value multiplication mechanism will be automatically triggered. Assuming that the original β = 0.3, after detecting the abnormality, β is increased to 0.6. This mechanism can quickly enhance the privacy protection strength when facing potential data leakage risks, reduce the losses caused by data leakage, and ensure the security of patient data.

[0043] For example, suppose a heart hospital collects a batch of patients' diagnosis and treatment data, and extracts a feature subset F from it, which includes genetic test results, electrocardiogram data, and clinical symptom descriptions. After evaluation by professional doctors and calculation by data analysis models, the accuracy assessment value of feature subset F for judging the type of heart disease (such as coronary heart disease, cardiomyopathy, etc.) is U (F) = 0.8. By analyzing the patient's sensitive information contained in feature subset F, such as family genetic history, personal identity information, etc., its privacy leakage risk assessment value P (F) = 0.4 is evaluated. Assuming that the current scenario is a scientific research project with high requirements for privacy protection, for this scenario, α is taken as 0.4 and β is taken as 0.6. According to the calculation of the dual-objective reward function, R is 0.08.

[0044] Optionally, determining the contribution ratio of each feature dimension in the feature subset F (such as gene test results, electrocardiogram data, clinical symptom description) to the determination of the type of heart disease, and obtaining a dynamic feature weight matrix; For example, through multiple experiments and data analysis, it is determined that the contribution ratio of genetic test results to the determination of heart disease type is 0.5, the contribution ratio of electrocardiogram data is 0.3, and the contribution ratio of clinical symptom description is 0.2. The calculated R value (0.08) is distributed to each feature dimension according to these contribution ratios to obtain the initial dynamic feature weight matrix ; Optionally, the original data of each feature dimension in the feature subset may be normalized so that its value range is between [0, 1], and the R value is distributed according to the normalized data ratio to obtain a dynamic feature weight matrix; For example, the normalized value of the gene test result is 0.6, the normalized value of the electrocardiogram data is 0.3, and the normalized value of the clinical symptom description is 0.1. Then the weight of the gene test result is 0.048, the weight of the electrocardiogram data is 0.024, and the weight of the clinical symptom description is 0.008, and the dynamic feature weight matrix is ​​obtained. ; Optionally, the analytic hierarchy process can be used to determine the relative importance weights of each feature dimension in the feature subset, construct a judgment matrix, and determine the relative importance comparison between each feature dimension through expert scoring, thereby obtaining a dynamic feature weight matrix.

[0045] For example, compared with electrocardiogram data, the importance of genetic test results is 3 (indicating that genetic test results are slightly more important than electrocardiogram data), compared with clinical symptom descriptions, the importance of genetic test results is 5 (indicating that genetic test results are significantly more important than clinical symptom descriptions), and compared with electrocardiogram data and clinical symptom descriptions, the importance is 2 (indicating that electrocardiogram data is slightly more important than clinical symptom descriptions). By calculation, the weight of genetic test results is 0.5396, the weight of electrocardiogram data is 0.2970, and the weight of clinical symptom description is 0.1634. Then the R value is allocated according to these weights. The weight of genetic test results is 0.043168, the weight of electrocardiogram data is 0.02376, and the weight of clinical symptom description is 0.013072, and the dynamic feature weight matrix is ​​obtained. .

[0046] Further, based on the feature dimension risk score and the dynamic feature weight matrix, the sensitivity of each feature dimension is calculated, the privacy budget is dynamically allocated in combination with the privacy protection strength of each feature in the polymorphic feature space, and the disturbance feature is noise-recalibrated to generate an enhanced privacy feature set; the following steps are included: By formula , calculate the sensitivity weight of each feature dimension; where, w i For the i The sensitivity weight of feature dimensions, , is the i-th feature dimension in the dynamic feature weight matrix, Score the characteristic dimension risk for the i-th characteristic dimension; k is the sensitivity adjustment coefficient, which is a predefined hyperparameter. Its function is to control the response intensity of the sensitivity weight to the sensitivity score of the feature dimension by adjusting the slope of the function. For example, when k is large, the sensitivity weight changes more steeply with the sensitivity score of the feature dimension, and the discrimination is high; when k is small, the sensitivity weight changes slowly, and the discrimination is low. k is usually set based on experimental experience (such as k=1) or determined through cross-validation optimization.

[0047] By formula Dynamically allocate the privacy budget allocated to each feature dimension, and use the formula , a momentum factor is introduced to adjust the noise injection intensity so that the privacy budget of the current cycle refers to the historical allocation record; wherein, is the privacy budget allocated to the i-th feature dimension, where the privacy budget represents the amount of resources allocated for adding privacy protection operations, For the overall privacy budget, is the sum of the sensitivity weights of all feature dimensions, is the adjusted noise injection intensity at time period t, is the noise injection intensity of the previous time period, is the momentum coefficient.

[0048] For example, assuming that the risk score of the feature dimension has been obtained, for example, the risk score of the feature dimension of the genetic test result is 0.7, the value of this dimension in the dynamic feature weight matrix is ​​0.9, and the sensitivity weight of this feature dimension is 0.9 / 0.7=1.29. Assuming that the overall privacy budget is 1.0, the sum of the sensitivity weights of all feature dimensions is 5.0 (such as the three feature dimensions of genetic test results, electrocardiogram data, and clinical symptom descriptions mentioned above), the privacy budget allocated to the feature dimension of genetic test results is calculated as , it is known that the noise injection intensity in the previous time period is 0.2, the momentum coefficient is 0.3, the currently allocated privacy budget is 0.258, and the last allocated privacy budget is 0.2. Substitute it into the formula It can be obtained that the noise injection intensity of the feature dimension of the gene test results in the current time period after adjustment is 0.2174. According to the privacy budget of each feature dimension calculated above and the adjusted noise injection intensity, the noise of the disturbed gene feature is recalibrated. For example, the genetic test result data is adjusted on the basis of the original noise according to the new noise injection intensity of 0.2174. The electrocardiogram data and other features describing clinical symptoms are also processed for privacy protection accordingly, and these processed features are combined to generate an enhanced privacy feature set. This enhanced privacy feature set not only retains valuable information for the diagnosis and treatment of heart disease, but also protects patient privacy to a large extent. This implementation method can realize the dynamic adjustment of privacy protection intensity and optimize the privacy protection strategy in combination with historical data.

[0049] Furthermore, under the constraint of the dynamic feature weight matrix, a target feature subset is searched in the enhanced privacy feature set, and a composite feature expression is generated from the target feature subset through genetic programming to obtain a diagnosis and treatment decision feature set; the following steps are included: A multi-objective optimization algorithm is used to maximize the clinical utility of the feature and minimize the privacy risk, and a non-dominated solution is searched in the feature space corresponding to the privacy-enhanced feature set to generate a Pareto frontier solution set; Calculate the contribution value of each feature to clinical utility and privacy risk under different feature subset combinations of the Pareto frontier solution set, and use the contribution value as the Shapley value; eliminate redundant features whose Shapley value is lower than a third preset threshold to obtain retained features; The retained features are nonlinearly combined through genetic programming to generate a composite feature expression that satisfies the operator combination generated by symbolic regression, and the feature set corresponding to the composite feature expression is used as a diagnosis and treatment decision feature set.

[0050] Specifically, in the diagnosis and treatment feature set for heart failure patients in a heart disease hospital, multi-objective optimization algorithms such as NSGA-II are used to search for non-dominated solutions in the feature space corresponding to the privacy-enhanced feature set, with the goal of maximizing the clinical utility of the feature (such as accurately judging the severity of heart failure) and minimizing the privacy risk. Some feature subsets may have a significant effect on accurately judging the severity of heart failure, but the privacy risk is high; some are the opposite. The algorithm finds non-dominated solutions that achieve a good balance between clinical utility and privacy risk, and generates a Pareto frontier solution set. The contribution value (Shapley value) of each feature to clinical utility and privacy risk under different feature subset combinations of the Pareto frontier solution set is calculated. Assuming that in a feature subset, the Shapley value of feature A (such as a less relevant blood indicator) is 0.005, which is lower than the third preset threshold (such as 0.01), it is removed as a redundant feature to obtain the retained features, simplify the feature set, and improve the analysis efficiency. The retained features are nonlinearly combined through genetic programming to generate a composite feature expression that satisfies the operator combination generated by symbolic regression.

[0051] For example, the operator combination includes a combination of a gene sequence feature subset and a clinical indicator feature subset. The expression of genes related to heart disease is combined with cardiac function indicators (such as ejection fraction) to generate a composite feature expression such as "Gene A expression*Ejection fraction+Gene C expression / average heart rate". Correspondingly, the diagnosis and treatment decision feature set for heart diseases such as heart failure includes gene A expression, gene C expression, ejection fraction, and average heart rate. When processing the diagnosis and treatment data of heart diseases such as heart failure, matching features are extracted from a large amount of patient data, such as the expression of genes A and C, as well as ejection fraction, heart rate average, etc. These features are combined and calculated according to the above composite feature expression to obtain a comprehensive feature value. This feature set integrates the key information of multi-source data and contains richer and more relevant information than a single feature or a simple feature set. Doctors can use this diagnosis and treatment decision feature set in combination with machine learning models or other data analysis methods to more accurately judge the patient's condition. For example, when judging the severity of heart failure in patients, comprehensive feature values ​​are input into the model. The model outputs a judgment result on the severity of the patient's condition based on the patterns and rules learned during training, and assists doctors in formulating personalized treatment plans, such as whether the drug dosage needs to be adjusted or whether surgical treatment is suitable.

[0052] In addition, the diagnosis and treatment decision feature set has other uses: for example, heart failure is a chronic disease, and patients need long-term tracking and management. The diagnosis and treatment decision feature set can help doctors continuously monitor changes in patients' conditions, compare the latest examination data with the historical diagnosis and treatment decision feature set, evaluate the treatment effect, and determine whether the condition is stable, worsening, or improving. By comparing the ejection fraction and the comprehensive changes of gene expression and cardiac function indicators in different periods, the treatment plan can be adjusted in time. The diagnosis and treatment decision feature set of historical medical data is a valuable research resource. Medical researchers can conduct large-scale data analysis and research based on it to explore the pathogenesis of heart failure, the relationship between genetic factors and disease development, and the long-term effects of different treatment methods. By analyzing a large amount of patient data, it is helpful to discover new treatment targets, optimize existing treatment plans, and promote the progress of medicine. The rules and patterns of the diagnosis and treatment decision feature set obtained based on the medical data of historical patients are also of great reference value for the diagnosis and treatment of new patients. When facing new heart failure patients, doctors can compare the characteristics of new patients with similar cases in the historical diagnosis and treatment decision feature set, quickly obtain possible diagnostic directions and treatment ideas, and improve the efficiency and accuracy of diagnosis and treatment. In summary, the potential value of multimodal medical data has been explored while taking into account patient privacy protection.

[0053] This solution achieves an organic integration of value mining and privacy protection from the following perspectives: Endogenous privacy protection: In core links such as feature extraction (step S1), time alignment (step S3), and spatial mapping (step S4), privacy protection operations such as differential privacy perturbation and homomorphic encryption are directly embedded in the data processing process (such as adding Laplace noise to genetic features and encrypting clinical identifiers), rather than adding a protection layer afterwards; Dynamic balance mechanism: Through the privacy budget monitor, the information entropy and sensitivity of the feature dimension are evaluated in real time, and a dynamic inverse relationship between privacy budget allocation and data value density is established, so that high-value features can obtain more privacy resource investment; Risk-aware modeling: When constructing a heterogeneous graph network (step S5), the clinical utility and privacy risk are simultaneously evaluated through the feature association strength matrix, so that the final generated diagnosis and treatment decision feature set is naturally privacy-safe.

[0054] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and its core ideas of this application. The above is only the preferred implementation method of this application. It should be pointed out that due to the limitations of textual expression and the objective existence of infinite specific structures, ordinary technicians in this technical field can make several improvements, modifications or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes or combinations, or the direct application of the inventive concept and technical solution to other occasions without improvement, should be regarded as the scope of protection of this application.

Claims

1. A data analysis method in the process of building high-value medical data assets, characterized in that: The following steps are involved: Extract effective gene loci related to the target disease and generate gene feature subsets; Extract clinical indicators associated with the target disease and generate a clinical feature subset; extract quantitative features of the lesion area and generate an image feature subset; The gene feature subset is encoded in the time dimension to generate a gene expression time series vector; the clinical feature subset is marked with a timestamp to extract the clinical event time series features; the image feature subset is registered in three-dimensional space to extract the spatiotemporal dynamic features of the lesion area; Aligning the gene expression time series of the gene expression time series vector with the clinical event time series of the clinical event time series feature, and calculating the time alignment similarity; Based on the spatiotemporal dynamic characteristics, image lesion coordinates are obtained, the image lesion coordinates are mapped to the gene-clinical joint space, and the spatial correlation is calculated; Constructing a heterogeneous graph network including gene nodes, clinical nodes, and imaging nodes, and calculating edge weights between nodes based on the time alignment similarity and the spatial correlation; In the heterogeneous graph network, redundant edges corresponding to the edge weights below a first preset threshold are pruned to generate a spatiotemporal association graph, thereby obtaining an association mapping matrix.

2. The data analysis method in the process of constructing high-value medical data assets according to claim 1 is characterized by: The generating of the spatiotemporal correlation map and obtaining the correlation mapping matrix comprises the following steps: Converting the spatiotemporal association graph into a weighted adjacency matrix, wherein the matrix element value of the weighted adjacency matrix is ​​the weight of the corresponding edge; Performing symmetric normalization processing on the weighted adjacency matrix to obtain a normalized matrix; Perform singular value decomposition on the normalized matrix, retain the first k principal components, and generate a low-dimensional mapping matrix as the association mapping matrix.

3. The data analysis method in the process of constructing high-value medical data assets according to claim 2 is characterized in that: The step of performing singular value decomposition on the normalized matrix, retaining the first k principal components, and generating a low-dimensional mapping matrix as the association mapping matrix comprises the following steps: Performing singular value decomposition on the normalized matrix, retaining the first k principal components, generating a low-dimensional mapping matrix, and generating feature clusters based on the low-dimensional mapping matrix; The cluster labels of the feature clusters are fed back to the singular value decomposition as auxiliary information, and the truncation rank k value in the singular value decomposition process is iteratively optimized to obtain the association mapping matrix.

4. The data analysis method in the process of constructing high-value medical data assets according to claim 1 is characterized in that: The method of extracting effective gene loci related to the target disease comprises the following steps: Through the chi-square test, the SNP sites whose occurrence frequency in the population exceeds the second preset threshold are screened to obtain a candidate SNP list; Obtain the GWAS database, compare the candidate SNP list with the GWAS database, and retain the SNP sites with p-values ​​less than a third preset threshold as valid gene sites; wherein the third preset threshold is less than 0.

05.

5. The data analysis method in the process of constructing high-value medical data assets according to claim 1 is characterized in that: The step of encoding the gene feature subset in the time dimension to generate a gene expression time series vector comprises the following steps: For moments when gene expression was missing, cubic spline interpolation was used to complete the gap; The multi-scale time series features of gene expression are extracted by wavelet transform, and the multi-scale time series features include high-frequency fluctuation components and low-frequency trend components.

6. The data analysis method in the process of constructing high-value medical data assets according to claim 1 is characterized in that: The step of aligning the gene expression timing of the gene expression timing vector with the clinical event timing of the clinical event timing feature comprises the following steps: Performing dynamic time warping on the gene expression time series and the clinical event time series to generate a time alignment path; Calculating the Pearson correlation coefficient of the gene-clinical feature pair along the time alignment path, and screening the feature pairs whose correlation coefficients are greater than a fourth preset threshold as strongly associated feature pairs; Based on the strongly associated feature pairs, construct a gene-clinical initial association matrix, wherein the matrix elements of the initial association matrix are correlation coefficient values; Performing non-negative matrix decomposition on the initial correlation matrix to obtain a gene feature subspace and a clinical feature subspace; The gene feature subspace and the clinical feature subspace are orthogonally projected and merged to generate the gene-clinical joint space.

7. The data analysis method in the process of constructing high-value medical data assets according to claim 4 is characterized in that: The method of extracting effective gene loci related to the target disease and generating a gene feature subset comprises the following steps: Performing random projection dimensionality reduction on the expression time series vectors of the screened effective gene loci to generate a feature space after dimensionality reduction; Add Laplace noise to the feature space after dimensionality reduction to satisfy -Differential privacy constraints are used to obtain a subset of genetic features.

8. The data analysis method in the process of constructing high-value medical data assets according to claim 7, characterized in that: The method further comprises the following steps: When constructing the edge weights of a heterogeneous graph network, noise compensation is performed on a subset of gene features to obtain corrected weights, which are then used to generate a spatiotemporal association map.

9. The data analysis method in the process of constructing high-value medical data assets according to claim 1, characterized in that: The step of extracting clinical indicators associated with the target disease and generating a clinical feature subset comprises the following steps: Extract clinical indicators associated with the target disease and homomorphically encrypt the patient identifier to obtain a subset of clinical features; The feature similarity between the genetic features and the encrypted clinical features is calculated in the encrypted domain, and the clinical feature subset is output.

10. The data analysis method in the process of building high-value medical data assets according to claim 9 is characterized in that: After homomorphically encrypting the patient identifier, the following steps are also included: The correlation coefficients between gene signatures and subsets of clinical signatures were calculated in the encrypted domain; In the orthogonal projection of the joint gene-clinical space, a subset of clinical features is locally decrypted, exposing only the standardized feature values ​​to participate in the spatial merging.

Images