A clinical data analysis method and system for hematological tumor patients

Abstract

The present invention belongs to the field of data analysis technology, and discloses a clinical data analysis method and system for hematological tumor patients. The method comprises the following steps: creating an OPC UA information instance of a clinical data acquisition device, and constructing a hematological tumor knowledge graph, a clinical data analysis model, and a data acquisition optimization model; collecting real-time clinical data of hematological tumor patients, and encrypting and uploading them to a cloud data center; writing real-time clinical data into corresponding OPC UA information instances, and extracting real-time OPC UA information data; using a clinical data analysis model to perform clinical data analysis based on real-time OPC UA information data and hematological tumor knowledge graph; and using a data acquisition optimization model to optimize data acquisition based on the real-time clinical data analysis results. The present invention solves the problems of insufficient accuracy, low data processing efficiency, and low intelligence in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to a method and system for analyzing clinical data of hematological tumor patients. Background Art

[0002] Hematological malignancies are a type of malignant tumor that originates from the blood system, including leukemia, lymphoma, multiple myeloma, etc. Clinical data analysis of hematological malignancy patients is of great significance for the treatment, recovery, prognosis assessment and personalized medicine of hematological malignancy patients.

[0003] The existing clinical data analysis technology for hematological malignancies has the following defects:

[0004] 1) Insufficient accuracy: The existing clinical data analysis of hematological malignant tumor patients relies on manual analysis by experts, which has high manpower costs and low efficiency. It is also unable to mine the deep information in the clinical data of hematological malignant tumor patients and has low accuracy.

[0005] 2) Low data processing efficiency: With the development of data processing technology, the processing of clinical data of blood cancer patients by computer has become popular. However, the data formats between different clinical data acquisition devices are not unified, resulting in a large amount of conversion work required during data integration and exchange, which reduces processing efficiency. The processing of real-time data is not fast enough and cannot meet the high requirements of data real-time in some application scenarios;

[0006] Low degree of intelligence: In the existing technology, the data collection strategy for the clinical data of blood tumor patients is preset and cannot be dynamically adjusted according to the patient's real-time condition, resulting in low efficiency and quality of data collection and low degree of intelligence. Summary of the invention

[0007] In order to solve the problems of insufficient accuracy, low data processing efficiency and low intelligence in the prior art, the present invention aims to provide a method and system for analyzing clinical data of hematological tumor patients.

[0008] The technical solution adopted by the present invention is:

[0009] A method for analyzing clinical data of hematological tumor patients, comprising the following steps:

[0010] The cloud data center creates OPC UA information instances of clinical data collection devices, and builds blood tumor knowledge graphs, clinical data analysis models, and data collection optimization models;

[0011] Clinical data collection device collects real-time clinical data of hematological tumor patients and uses OPC UA protocol to encrypt and upload the real-time clinical data to the cloud data center;

[0012] The cloud data center writes the real-time clinical data into the corresponding OPC UA information instance and extracts the real-time OPC UA information data of the OPC UA information instance;

[0013] The cloud data center uses the clinical data analysis model to perform clinical data analysis based on real-time OPC UA information data and blood tumor knowledge graph to obtain real-time clinical data analysis results;

[0014] The cloud data center uses a data collection optimization model to optimize data collection based on the real-time clinical data analysis results, obtains a real-time data collection strategy, and sends it to the clinical data collection device so that the clinical data collection device performs clinical data collection based on the received real-time data collection strategy.

[0015] Furthermore, the cloud data center creates an OPC UA information instance of the clinical data acquisition device, and constructs a blood tumor knowledge graph, a clinical data analysis model, and a data acquisition optimization model, including the following steps:

[0016] The cloud data center collects historical monitoring data from clinical data collection devices and uses OPC UA information technology to build corresponding OPC UA information models based on the historical monitoring data.

[0017] Send the OPC UA information model to all clinical data collection devices connected to the cloud data center, and deploy the OPC UA information model in each clinical data collection device;

[0018] Deploy the OPC UA server and create corresponding OPC UA information instances in the address space of the OPC UA server according to the OPC UA information model of each clinical data acquisition device;

[0019] Collecting some blood tumor knowledge and some historical clinical data of blood tumor patients, and preprocessing them to obtain some preprocessed blood tumor knowledge and some preprocessed historical clinical data;

[0020] Based on some pre-processed blood tumor knowledge, a blood tumor knowledge graph is constructed using a pre-trained named entity and entity relationship extraction model;

[0021] Based on different OPC UA information instances, according to the blood tumor knowledge graph and some pre-processed historical clinical data, a deep learning algorithm is used to build a clinical data analysis model and generate some historical clinical data analysis results;

[0022] Based on the analysis results of several historical clinical data, an enhanced reinforcement learning algorithm was used to build a data collection optimization model, and several historical data collection optimization experiences were generated.

[0023] Furthermore, the cloud data center collects historical monitoring data of the clinical data collection device, and uses OPC UA information technology to construct a corresponding OPC UA information model based on the historical monitoring data, including the following steps:

[0024] The cloud data center collects historical monitoring data from clinical data collection devices and defines the object type of the OPC UA information model based on the physical monitoring indicators in the historical monitoring data;

[0025] Define object variables of the object type of the OPC UA information model according to the monitoring indicator parameters in the historical monitoring data;

[0026] Object relationships are defined between several object types, and the corresponding OPC UA information model is constructed according to the object types, object variables and object relationships of the clinical data acquisition device.

[0027] Further, deploying an OPC UA server and creating a corresponding OPC UA information instance in the address space of the OPC UA server according to the OPC UA information model of each clinical data acquisition device includes the following steps:

[0028] Deploy the OPC UA server in the cloud data center and set the corresponding address space for the OPC UA server;

[0029] Extract model metadata of each OPC UA information model, and create an OPC UA framework of a corresponding OPC UA instance in the address space of the OPC UA server based on the model metadata;

[0030] According to the object types, object variables and object relationships in the model metadata, a number of nodes, node attributes of each node and node relationships between the nodes are set in the OPC UA framework;

[0031] According to an OPC UA framework provided with a plurality of nodes, a plurality of node attributes and a plurality of node relationships, an OPC UA information instance of each clinical data acquisition device is obtained.

[0032] Further, the named entity and entity relationship extraction model is constructed based on the BERT-Double CRF algorithm, and the named entity and entity relationship extraction model includes a text feature extraction module constructed based on the BERT algorithm, a named entity annotation module constructed based on the CRF algorithm, and an entity relationship annotation module constructed based on the CRF algorithm, the text feature extraction module is respectively connected to the named entity annotation module and the entity relationship annotation module, and the named entity annotation module is connected to the entity relationship annotation module;

[0033] The clinical data analysis model is constructed based on the Double GCN-RF algorithm, and the clinical data analysis model includes a first graph feature extraction module and a second graph feature extraction module both constructed based on the GCN algorithm, and a clinical data analysis module constructed based on the RF algorithm, and the first graph feature extraction module and the second graph feature extraction module are both connected to the clinical data analysis module

[0034] The data acquisition optimization model is constructed based on the ML-DQN algorithm, and the data acquisition optimization model includes a meta-learning module constructed based on the ML algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is respectively connected to the deep Q network, the experience replay pool and the meta-learning module, and the meta-learning module is connected to the experience replay pool.

[0035] Furthermore, based on different OPC UA information instances, according to the blood tumor knowledge graph and several pre-processed historical clinical data, a deep learning algorithm is used to build a clinical data analysis model and generate several historical clinical data analysis results, including the following steps:

[0036] Writing the preprocessed historical clinical data into the corresponding OPC UA information instance, extracting the historical OPC UA information data of the OPC UA information instance, and performing graph format conversion to obtain the historical OPC UA information graph;

[0037] Traversing all pre-processed historical clinical data of different OPC UA information instances to obtain a number of historical OPC UA information graphs, and dividing the several historical OPC UA information graphs into a model training set and a model test set;

[0038] Use the Double GCN-RF algorithm to build an initial clinical data analysis model, input the model training set and blood tumor knowledge graph, train and optimize the initial clinical data analysis model, and obtain an optimized clinical data analysis model;

[0039] Input the model test set and blood tumor knowledge graph, perform model testing on the optimized clinical data analysis model, and obtain the test accuracy of the optimized clinical data analysis model;

[0040] If the test accuracy is greater than the accuracy threshold, the final clinical data analysis model is output and several historical clinical data analysis results are generated; otherwise, training optimization continues.

[0041] Furthermore, based on the analysis results of several historical clinical data, an enhanced reinforcement learning algorithm is used to construct a data collection optimization model and generate several historical data collection optimization experiences, including the following steps:

[0042] The data collection optimization problem is used as the reinforcement learning simulation environment of the DQN algorithm, the data collection optimization problem is divided into several subtasks, and the network parameter generation problem of the deep Q network of the DQN algorithm is used as the meta-learning simulation environment of the ML algorithm;

[0043] According to the preset meta-feature parameters, several corresponding historical meta-features are extracted from the historical clinical data analysis results, and based on different subtasks and several historical meta-features, the ML algorithm is used to construct the initial meta-learning module and generate several historical meta-learning experiences;

[0044] Based on the reinforcement learning simulation environment, build an intelligent agent and experience replay pool, use the initial meta-learning module to initialize the deep Q network of the DQN algorithm, and train the initial meta-learning module to obtain the final meta-learning module;

[0045] Based on the historical clinical data analysis results, define the state space of the DQN algorithm, define the action space of the DQN algorithm based on the preset data collection strategy, and define the reward function based on the impact of the action in the action space on the state in the state space;

[0046] Based on the action space, state space, intelligent agent, experience replay pool and deep Q network, the initial reinforcement learning module is constructed, and the initial reinforcement learning module is trained using several historical clinical data analysis results to obtain the final reinforcement learning module, and generate several historical data collection optimization experiences;

[0047] The final meta-learning module and reinforcement learning module are integrated to obtain the data collection optimization model, and several historical meta-learning experiences and several historical data collection optimization experiences are stored in the experience replay pool.

[0048] Furthermore, the cloud data center uses the clinical data analysis model to perform clinical data analysis based on the real-time OPC UA information data and the blood tumor knowledge graph to obtain real-time clinical data analysis results, including the following steps:

[0049] The cloud data center converts the real-time OPC UA information data into a graph format, and inputs the obtained real-time OPC UA information graph and blood tumor knowledge graph into the clinical data analysis model;

[0050] Using the clinical data analysis model, extracting a first real-time graph feature of the real-time OPC UA information graph and a second real-time graph feature of the blood tumor knowledge graph;

[0051] A plurality of real-time key features of the first real-time graph feature and the second real-time graph feature are screened, and clinical data analysis is performed according to the plurality of real-time key features to obtain corresponding real-time clinical data analysis results.

[0052] Further, the cloud data center uses the data collection optimization model to optimize data collection according to the real-time clinical data analysis results, obtains the real-time data collection strategy, and sends it to the clinical data collection device, so that the clinical data collection device performs clinical data collection according to the received real-time data collection strategy, including the following steps:

[0053] The cloud data center extracts corresponding real-time meta-features from the real-time clinical data analysis results according to preset meta-feature parameters;

[0054] According to a number of real-time meta-features, a number of historical meta-learning experiences in the experience playback pool of the data acquisition optimization model are matched to obtain successfully matched real-time meta-learning experiences;

[0055] According to the real-time meta-learning experience, the data collection optimization model is adjusted to obtain an adjusted data collection optimization model;

[0056] According to the real-time clinical data analysis results, several historical data collection optimization experiences in the experience playback pool of the adjusted data collection optimization model are matched to obtain successfully matched real-time data collection optimization experiences;

[0057] According to the real-time clinical data analysis results and the real-time data collection optimization experience, the adjusted data collection optimization model is used to optimize data collection and obtain the real-time data collection strategy;

[0058] The real-time data collection strategy is encrypted and sent to the corresponding clinical data collection device, so that the clinical data collection device performs clinical data collection according to the received real-time data collection strategy.

[0059] A clinical data analysis system for patients with hematological tumors is used to implement a clinical data analysis method for patients with hematological tumors. The system includes a cloud data center and several clinical data acquisition devices. The several clinical data acquisition devices are all communicatively connected to the cloud data center. The cloud data center includes an initialization unit, an information data extraction unit, a clinical data analysis unit, and a data acquisition optimization unit that are connected in sequence.

[0060] The beneficial effects of the present invention are:

[0061] The present invention discloses a clinical data analysis method and system for hematological tumor patients. By constructing an automated clinical data analysis system for hematological tumor patients, automated analysis of clinical data of hematological tumor patients is realized, which avoids reliance on experts for manual analysis, reduces cost investment, and improves analysis efficiency. By combining a hematological tumor knowledge graph and a clinical data analysis model, deep information of clinical data is mined, and predictive analysis is performed in combination with hematological tumor knowledge, thereby improving accuracy. By constructing a unified OPC UA information model, standardized conversion between data sources of different clinical data acquisition devices is realized, the data processing flow is simplified, and the processing efficiency is improved. Through the data interaction processing flow between the clinical data acquisition device and the cloud data center, the writing and query of real-time data can be quickly responded to, meeting the real-time requirements. By constructing a data acquisition optimization model, an adaptive data acquisition optimization mechanism is introduced, and the data acquisition plan is dynamically adjusted according to the real-time condition of the patient, thereby improving the data acquisition efficiency and quality as well as the intelligence of data analysis.

[0062] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flowchart of the clinical data analysis method for hematological tumor patients in the present invention.

[0064] Figure 2 It is a structural block diagram of the clinical data analysis system for blood tumor patients in the present invention. DETAILED DESCRIPTION

[0065] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0066] Embodiment 1:

[0067] like Figure 1 As shown, this embodiment provides a method for analyzing clinical data of hematological tumor patients, comprising the following steps:

[0068] S1: Cloud data center, creating an OPCUA (Object Linking and Embedding for Process Control Unified Architecture) information instance for the clinical data acquisition device, and building a blood tumor knowledge graph, clinical data analysis model, and data acquisition optimization model, including the following steps:

[0069] S1-1: The cloud data center collects historical monitoring data from the clinical data collection device and uses OPC UA information technology to build a corresponding OPC UA information model based on the historical monitoring data, including the following steps:

[0070] Since clinical data collection devices include different data source devices, such as monitors, ventilators, blood pressure monitors, etc., the collected monitoring data include basic vital signs data (body temperature, heart rate, blood pressure, respiratory rate, etc.), laboratory test data (hematological parameters, biochemical indicators, immunological indicators, etc.), drug treatment records (types of blood tumor drugs, dosages, administration time, etc.), basic patient data (name, age), etc.;

[0071] S1-1-1: The cloud data center collects historical monitoring data from clinical data collection devices and defines the object type of the OPC UA information model based on the physical monitoring indicators in the historical monitoring data;

[0072] S1-1-2: Define object variables of the object type of the OPC UA information model according to the monitoring indicator parameters in the historical monitoring data;

[0073] S1-1-3: Define object relationships between several object types, and construct corresponding OPC UA information models according to the object types, object variables and object relationships of the clinical data acquisition device;

[0074] S1-2: Send the OPC UA information model to all clinical data collection devices connected to the cloud data center, and deploy the OPC UA information model in each clinical data collection device;

[0075] S1-3: Deploy the OPC UA server and create a corresponding OPC UA information instance in the address space of the OPC UA server according to the OPC UA information model of each clinical data acquisition device, including the following steps:

[0076] S1-3-1: Deploy the OPC UA server in the cloud data center and set the corresponding address space for the OPC UA server;

[0077] S1-3-2: extracting model metadata of each OPC UA information model, and creating an OPC UA framework of a corresponding OPC UA instance in the address space of the OPC UA server according to the model metadata;

[0078] S1-3-3: According to the object types, object variables and object relationships in the model metadata, a number of nodes, node attributes of each node and node relationships between the nodes are set in the OPC UA framework;

[0079] S1-3-4: Obtaining an OPC UA information instance of each clinical data acquisition device according to an OPC UA framework provided with a plurality of nodes, a plurality of node attributes, and a plurality of node relationships;

[0080] S1-4: Collect some blood tumor knowledge and some historical clinical data of blood tumor patients, and perform preprocessing to obtain some preprocessed blood tumor knowledge and some preprocessed historical clinical data;

[0081] Preprocessing includes data cleaning, format conversion, and normalization. While eliminating duplicate and erroneous data, it also performs standardized conversion, eliminates magnitude differences and noise, and converts the data into a format that can be recognized by subsequent models.

[0082] S1-5: Based on some pre-processed blood tumor knowledge, use the pre-trained named entity and entity relationship extraction model to construct a blood tumor knowledge graph;

[0083] The named entity and entity relationship extraction model is constructed based on the Bidirectional Encoder Representations from Transformers (BERT)-Double Conditional Random Fields (CRF) algorithm, and the named entity and entity relationship extraction model includes a text feature extraction module constructed based on the BERT algorithm, a named entity annotation module constructed based on the CRF algorithm, and an entity relationship annotation module constructed based on the CRF algorithm. The text feature extraction module is connected to the named entity annotation module and the entity relationship annotation module respectively, and the named entity annotation module is connected to the entity relationship annotation module;

[0084] The BERT module can capture deep semantic information in the text, which is very useful for identifying different types of entities (such as names, places, organizations, etc.) and extracting relationships between entities. The CRF module can consider the dependencies between adjacent labels, which can help the model learn the sequence dependencies of entity labels, thereby improving the accuracy of named entity labeling. In the entity relationship labeling task, the CRF module handles the relationship classification between entity pairs, especially when dealing with multiple entities and relationships, to achieve entity relationship labeling.

[0085] Based on some pre-processed blood tumor knowledge, a blood tumor knowledge graph is constructed using a pre-trained named entity and entity relationship extraction model, including the following steps:

[0086] S1-5-1: Use the text feature extraction module to extract the text features of the preprocessed blood tumor knowledge, and send the text features to the named entity annotation module and the entity relationship annotation module;

[0087] S1-5-2: Use the named entity annotation module to annotate the preprocessed blood tumor knowledge according to the text features to obtain several named entities;

[0088] S1-5-3: Use the entity relationship annotation module to annotate several named entities based on text features, and obtain entity relationships between several named entities;

[0089] S1-5-4: traverse all pre-processed blood tumor knowledge to obtain a number of named entities of each pre-processed blood tumor knowledge and entity relationships between the named entities;

[0090] S1-5-5: Take named entities as nodes and entity relationships as edges of nodes to construct a knowledge graph to obtain a blood tumor knowledge graph;

[0091] S1-6: Based on different OPC UA information instances, according to the blood tumor knowledge graph and several pre-processed historical clinical data, a deep learning algorithm is used to build a clinical data analysis model and generate several historical clinical data analysis results, including the following steps:

[0092] S1-6-1: Write the pre-processed historical clinical data into the corresponding OPC UA information instance, extract the historical OPC UA information data of the OPC UA information instance, and perform graph format conversion to obtain the historical OPC UA information graph;

[0093] S1-6-2: traverse all pre-processed historical clinical data of different OPC UA information instances to obtain a number of historical OPC UA information graphs, and divide the number of historical OPC UA information graphs into a model training set and a model test set;

[0094] S1-6-3: Use the Double Graph Convolutional Network (GCN)-Random Forest (RF) algorithm to build an initial clinical data analysis model, input the model training set and blood tumor knowledge graph, train and optimize the initial clinical data analysis model, and obtain an optimized clinical data analysis model;

[0095] The clinical data analysis model is constructed based on the Double GCN-RF algorithm, and the clinical data analysis model includes a first graph feature extraction module and a second graph feature extraction module both constructed based on the GCN algorithm, and a clinical data analysis module constructed based on the RF algorithm, and the first graph feature extraction module and the second graph feature extraction module are both connected to the clinical data analysis module;

[0096] The GCN network propagates features on graph data through operations similar to convolution, extracts node features of graph data and edge features between nodes, and forms graph features of graph data; the first graph feature extraction module is used to extract the first graph feature of the OPC UA information graph, and the second graph feature extraction module is used to extract the second graph feature of the blood tumor knowledge graph; the RF structure screens graph features through the internal Classification And Regression Tree (CART), which can process a large number of graph features and select the most stable and discriminative key features. After the RF structure is trained, new graph features are input, which can be screened and predicted based on the pre-selected key features;

[0097] S1-6-4: Input the model test set and blood tumor knowledge graph, perform model testing on the optimized clinical data analysis model, and obtain the test accuracy of the optimized clinical data analysis model;

[0098] S1-6-5: If the test accuracy is greater than the accuracy threshold, the final clinical data analysis model is output and several historical clinical data analysis results are generated; otherwise, training optimization is continued;

[0099] S1-7: Based on the analysis results of several historical clinical data, an enhanced reinforcement learning algorithm was used to build a data collection optimization model and generate several historical data collection optimization experiences;

[0100] The data collection optimization model is constructed based on the Meta-Learning (ML)-Deep Q Network (DQN) algorithm, and the data collection optimization model includes a meta-learning module constructed based on the ML algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is connected to the deep Q network, the experience replay pool and the meta-learning module respectively, and the meta-learning module is connected to the experience replay pool.

[0101] The meta-learning module enables the model to learn quickly on a small number of samples. This allows the model to quickly adapt to new working environments and improve the generalization ability of the model for clinical data analysis results of different blood cancer patients. Even under unseen data, the deep Q network can be updated based on previous learning experience. Through meta-learning, the DQN algorithm can converge faster, reducing the amount of data and time required for training, which can significantly improve learning efficiency for data acquisition optimization strategy generation tasks.

[0102] Based on the analysis results of several historical clinical data, an enhanced reinforcement learning algorithm is used to build a data collection optimization model and generate several historical data collection optimization experiences, including the following steps:

[0103] S1-7-1: Use the data collection optimization problem as a reinforcement learning simulation environment for the DQN algorithm, divide the data collection optimization problem into several subtasks, and use the network parameter generation problem of the deep Q network of the DQN algorithm as a meta-learning simulation environment for the ML algorithm;

[0104] S1-7-2: According to the preset meta-feature parameters, extract several corresponding historical meta-features from the historical clinical data analysis results, and based on different subtasks and several historical meta-features, use ML algorithms to build an initial meta-learning module and generate several historical meta-learning experiences;

[0105] S1-7-3: Based on the reinforcement learning simulation environment, build the intelligent agent and experience replay pool, use the initial meta-learning module to initialize the deep Q network of the DQN algorithm, and train the initial meta-learning module to obtain the final meta-learning module;

[0106] S1-7-4: Based on the historical clinical data analysis results, define the state space of the DQN algorithm, define the action space of the DQN algorithm based on the preset data collection strategy, and define the reward function based on the impact of the action in the action space on the state in the state space;

[0107] S1-7-5: Based on the action space, state space, intelligent agent, experience replay pool and deep Q network, construct the initial reinforcement learning module, and use some historical clinical data analysis results to train the initial reinforcement learning module to obtain the final reinforcement learning module, and generate some historical data collection optimization experience;

[0108] S1-7-6: Integrate the final meta-learning module and reinforcement learning module to obtain a data collection optimization model, and store several historical meta-learning experiences and several historical data collection optimization experiences into the experience replay pool;

[0109] S2: Clinical data collection device collects real-time clinical data of blood cancer patients and uses the OPC UA protocol to encrypt and upload the real-time clinical data to the cloud data center, including the following steps:

[0110] S2-1: A clinical data collection device that collects real-time monitoring data of blood tumor patients according to a preset data collection strategy;

[0111] S2-2: Convert the real-time monitoring data into corresponding real-time clinical data according to the OPC UA information model deployed by the clinical data acquisition device;

[0112] S2-3: Use an asymmetric encryption algorithm to encrypt the real-time clinical data to obtain the encrypted real-time clinical data, and use the OPC UA protocol to upload the encrypted real-time clinical data to the cloud data center;

[0113] S3: The cloud data center writes the real-time clinical data into the corresponding OPC UA information instance and extracts the real-time OPC UA information data of the OPC UA information instance, including the following steps:

[0114] S3-1: The cloud data center uses an asymmetric decryption algorithm to decrypt the encrypted real-time clinical data to obtain the decrypted real-time clinical data;

[0115] S3-2: Parse the decrypted real-time clinical data to locate the real-time clinical data value of the dynamic node, where the real-time clinical data value includes the real-time node value and the real-time node attribute value of the dynamic node;

[0116] S3-3: Encapsulate the real-time clinical data value to obtain the real-time node encapsulation data of the dynamic node, and locate the target node corresponding to the OPC UA information instance and the dynamic node;

[0117] S3-4: Use the write service of the OPC UA protocol to write the real-time node encapsulation data of the dynamic node to the target node of the OPC UA information instance;

[0118] S3-5: traverse the real-time node encapsulation data of all dynamic nodes of the decrypted real-time clinical data until all target nodes of the OPCUA information instance are written;

[0119] S3-6: After the data is written, the real-time OPC UA information data of the OPC UA information instance is extracted using the extraction service of the OPC UA protocol;

[0120] S4: The cloud data center uses the clinical data analysis model to perform clinical data analysis based on the real-time OPC UA information data and the blood tumor knowledge graph to obtain real-time clinical data analysis results, including the following steps:

[0121] S4-1: The cloud data center converts the real-time OPC UA information data into a graph format, and inputs the obtained real-time OPC UA information graph and blood tumor knowledge graph into the clinical data analysis model;

[0122] S4-2: using the first graph feature extraction module of the clinical data analysis model to extract the first real-time graph feature of the real-time OPC UA information graph, and using the second graph feature extraction module to extract the second real-time graph feature of the blood tumor knowledge graph;

[0123] S4-3: using the clinical data analysis module, screening a number of real-time key features of the first real-time graph feature and the second real-time graph feature, and performing clinical data analysis based on the number of real-time key features to obtain corresponding real-time clinical data analysis results;

[0124] The blood tumor knowledge graph can provide the normal range of all monitoring indicators in real-time clinical data and the potential connection between each monitoring indicator. By combining the blood tumor knowledge graph, it can accurately detect abnormalities in real-time clinical data. The real-time clinical data analysis results obtained include real-time abnormal monitoring parameter values, patient condition assessment (patient's overall clinical condition score), disease progression prediction (staging and typing information of blood tumors, change trends of tumor markers, evaluation of cancer cell proliferation and spread, etc.), treatment effect analysis (real-time evaluation of drug treatment effects, prediction of reactions and side effects after radiotherapy or chemotherapy, comparison of changes in clinical indicators before and after treatment, etc.);

[0125] S5: The cloud data center uses the data collection optimization model to optimize data collection according to the real-time clinical data analysis results, obtains the real-time data collection strategy, and sends it to the clinical data collection device, so that the clinical data collection device performs clinical data collection according to the received real-time data collection strategy, including the following steps:

[0126] S5-1: The cloud data center extracts corresponding real-time meta-features from the real-time clinical data analysis results according to the preset meta-feature parameters;

[0127] S5-2: According to a number of real-time meta-features, a number of historical meta-learning experiences in the experience playback pool of the data acquisition optimization model are matched to obtain successfully matched real-time meta-learning experiences;

[0128] S5-3: According to the real-time meta-learning experience, the data collection optimization model is adjusted to obtain an adjusted data collection optimization model, including the following steps:

[0129] S5-3-1: Based on the real-time meta-learning experience, use the meta-learning module to generate network parameters and obtain the adjusted network parameters of the deep Q network;

[0130] S5-3-2: adjusting the deep Q network of the reinforcement learning module according to the adjusted network parameters of the deep Q network to obtain an adjusted reinforcement learning module;

[0131] S5-3-3: According to the adjusted reinforcement learning module, an adjusted data acquisition optimization model is obtained;

[0132] S5-4: According to the real-time clinical data analysis results, several historical data collection optimization experiences in the experience playback pool of the adjusted data collection optimization model are matched to obtain successfully matched real-time data collection optimization experiences;

[0133] S5-5: Based on the real-time clinical data analysis results and the real-time data collection optimization experience, use the adjusted data collection optimization model to optimize data collection and obtain the real-time data collection strategy, including the following steps:

[0134] S5-5-1: Based on the real-time data collection optimization experience, the action space of the updated reinforcement learning module is adjusted to obtain the updated action space ,in, For the updated action, is the action indicator, is the total number of action space dimensions;

[0135] S5-5-2: According to the real-time clinical data analysis results, adjust the state space of the adjusted reinforcement learning module to obtain an updated state space ,in, For the updated state, is the state indicator, is the total number of state space dimensions;

[0136] S5-5-3: Input the updated state space into the adjusted reinforcement learning module, and use the adjusted deep Q network to generate the Q value of each possible action in the updated action space based on the updated state space;

[0137] The formula is:

[0138]

[0139] In the formula, For updated status and update actions The corresponding updated Q value; Status and actions The corresponding predicted Q value; is the learning rate; is the highest predicted Q value; It is a comprehensive indicator; To update the parameters; Action-based The status Transition to the updated state The reward value of

[0140] S5-5-4: Repeat the above Q value update steps until the number of iterations reaches the iteration number threshold, use the greedy strategy to select the possible action with the highest updated Q value as the execution action, and obtain the real-time data collection optimization strategy based on the execution action;

[0141] Real-time data collection optimization strategies include sensor and device selection, clinical data monitoring goals, data integration and interface development decisions, and data collection decisions (including collection time, cycle, frequency, etc.);

[0142] S5-6: Encrypting the real-time data collection strategy and sending it to the corresponding clinical data collection device, so that the clinical data collection device performs clinical data collection according to the received real-time data collection strategy, including the following steps:

[0143] S5-6-1: Use an asymmetric encryption algorithm to encrypt the real-time data collection strategy, obtain the encrypted real-time data collection strategy, and send it to the corresponding clinical data collection device;

[0144] S5-6-1: The clinical data collection device uses an asymmetric decryption algorithm to decrypt the encrypted real-time data collection strategy to obtain the decrypted real-time data collection strategy;

[0145] S5-6-2: Clinical data collection device, collects clinical data according to the decrypted real-time data collection strategy.

[0146] Embodiment 2:

[0147] like Figure 2 As shown, this embodiment provides a clinical data analysis system for blood tumor patients, which is used to implement a clinical data analysis method for blood tumor patients, and is characterized in that: the system includes a cloud data center and a plurality of clinical data acquisition devices, the plurality of clinical data acquisition devices are all communicatively connected to the cloud data center, and the cloud data center includes an initialization unit, an information data extraction unit, a clinical data analysis unit, and a data acquisition optimization unit connected in sequence;

[0148] Clinical data collection device, used to collect real-time clinical data of hematological tumor patients, and use OPC UA protocol to encrypt and upload the real-time clinical data to the cloud data center;

[0149] Initialization unit, used to create OPC UA information instances of clinical data acquisition devices, and build blood tumor knowledge graphs, clinical data analysis models, and data acquisition optimization models;

[0150] An information data extraction unit, used to write real-time clinical data into a corresponding OPC UA information instance and extract real-time OPC UA information data of the OPC UA information instance;

[0151] A clinical data analysis unit is used to perform clinical data analysis based on real-time OPC UA information data and blood tumor knowledge graph using a clinical data analysis model to obtain real-time clinical data analysis results;

[0152] The data collection optimization unit is used to optimize data collection based on the real-time clinical data analysis results using the data collection optimization model to obtain a real-time data collection strategy and send it to the clinical data collection device.

[0153] The present invention discloses a clinical data analysis method and system for hematological tumor patients. By constructing an automated clinical data analysis system for hematological tumor patients, automated analysis of clinical data of hematological tumor patients is realized, which avoids reliance on experts for manual analysis, reduces cost investment, and improves analysis efficiency. By combining a hematological tumor knowledge graph and a clinical data analysis model, deep information of clinical data is mined, and predictive analysis is performed in combination with hematological tumor knowledge, thereby improving accuracy. By constructing a unified OPC UA information model, standardized conversion between data sources of different clinical data acquisition devices is realized, the data processing flow is simplified, and the processing efficiency is improved. Through the data interaction processing flow between the clinical data acquisition device and the cloud data center, the writing and query of real-time data can be quickly responded to, meeting the real-time requirements. By constructing a data acquisition optimization model, an adaptive data acquisition optimization mechanism is introduced, and the data acquisition plan is dynamically adjusted according to the real-time condition of the patient, thereby improving the data acquisition efficiency and quality as well as the intelligence of data analysis.

[0154] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing clinical data of patients with hematological malignancies, characterized by: The steps include: The cloud data center creates OPC UA information instances of clinical data collection devices, and builds blood tumor knowledge graphs, clinical data analysis models, and data collection optimization models; Clinical data collection device collects real-time clinical data of hematological tumor patients and uses OPC UA protocol to encrypt and upload the real-time clinical data to the cloud data center; The cloud data center writes the real-time clinical data into the corresponding OPC UA information instance and extracts the real-time OPC UA information data of the OPC UA information instance; The cloud data center uses the clinical data analysis model to perform clinical data analysis based on real-time OPC UA information data and blood tumor knowledge graph to obtain real-time clinical data analysis results, including the following steps: The cloud data center converts the real-time OPC UA information data into a graph format, and inputs the obtained real-time OPC UA information graph and blood tumor knowledge graph into the clinical data analysis model; Using the clinical data analysis model, extracting a first real-time graph feature of the real-time OPC UA information graph and a second real-time graph feature of the blood tumor knowledge graph; screening a number of real-time key features of the first real-time graph feature and the second real-time graph feature, and performing clinical data analysis based on the number of real-time key features to obtain corresponding real-time clinical data analysis results; Real-time clinical data analysis results include real-time abnormal monitoring parameter values, patient condition assessment, disease progression prediction, and treatment effect analysis; The cloud data center uses a data collection optimization model to optimize data collection based on the real-time clinical data analysis results, obtains a real-time data collection strategy, and sends it to the clinical data collection device so that the clinical data collection device performs clinical data collection based on the received real-time data collection strategy.

2. The method for analyzing clinical data of hematological malignancy patients according to claim 1, characterized in that: The cloud data center creates an OPC UA information instance of the clinical data collection device, and builds a blood tumor knowledge graph, a clinical data analysis model, and a data collection optimization model, including the following steps: The cloud data center collects historical monitoring data from clinical data collection devices and uses OPC UA information technology to build corresponding OPC UA information models based on the historical monitoring data. Send the OPC UA information model to all clinical data collection devices connected to the cloud data center, and deploy the OPC UA information model in each clinical data collection device; Deploy the OPC UA server and create corresponding OPC UA information instances in the address space of the OPC UA server according to the OPC UA information model of each clinical data acquisition device; Collecting some blood tumor knowledge and some historical clinical data of blood tumor patients, and preprocessing them to obtain some preprocessed blood tumor knowledge and some preprocessed historical clinical data; Based on some pre-processed blood tumor knowledge, a blood tumor knowledge graph is constructed using a pre-trained named entity and entity relationship extraction model; Based on different OPC UA information instances, according to the blood tumor knowledge graph and some pre-processed historical clinical data, a deep learning algorithm is used to build a clinical data analysis model and generate some historical clinical data analysis results; Based on the analysis results of several historical clinical data, an enhanced reinforcement learning algorithm was used to build a data collection optimization model, and several historical data collection optimization experiences were generated.

3. A method for analyzing clinical data of hematological tumor patients according to claim 2, characterized in that: The cloud data center collects historical monitoring data from the clinical data collection device and uses OPC UA information technology to build a corresponding OPC UA information model based on the historical monitoring data, including the following steps: The cloud data center collects historical monitoring data from clinical data collection devices and defines the object type of the OPC UA information model based on the physical monitoring indicators in the historical monitoring data; Define object variables of the object type of the OPC UA information model according to the monitoring indicator parameters in the historical monitoring data; Object relationships are defined between several object types, and the corresponding OPC UA information model is constructed according to the object types, object variables and object relationships of the clinical data acquisition device.

4. A method for analyzing clinical data of hematological malignancy patients according to claim 3, characterized in that: Deploy the OPC UA server and create corresponding OPC UA information instances in the address space of the OPC UA server according to the OPC UA information model of each clinical data acquisition device, including the following steps: Deploy the OPC UA server in the cloud data center and set the corresponding address space for the OPC UA server; Extract model metadata of each OPC UA information model, and create an OPC UA framework of a corresponding OPC UA instance in the address space of the OPC UA server based on the model metadata; According to the object types, object variables and object relationships in the model metadata, a number of nodes, node attributes of each node and node relationships between the nodes are set in the OPC UA framework; According to an OPC UA framework provided with a plurality of nodes, a plurality of node attributes and a plurality of node relationships, an OPC UA information instance of each clinical data acquisition device is obtained.

5. The method for analyzing clinical data of hematological tumor patients according to claim 2, characterized in that: The named entity and entity relationship extraction model is constructed based on the BERT-Double CRF algorithm, and the named entity and entity relationship extraction model includes a text feature extraction module constructed based on the BERT algorithm, a named entity annotation module constructed based on the CRF algorithm, and an entity relationship annotation module constructed based on the CRF algorithm, the text feature extraction module is connected to the named entity annotation module and the entity relationship annotation module respectively, and the named entity annotation module is connected to the entity relationship annotation module; The clinical data analysis model is constructed based on the Double GCN-RF algorithm, and the clinical data analysis model includes a first graph feature extraction module and a second graph feature extraction module both constructed based on the GCN algorithm, and a clinical data analysis module constructed based on the RF algorithm, and the first graph feature extraction module and the second graph feature extraction module are both connected to the clinical data analysis module; The data acquisition optimization model is constructed based on the ML-DQN algorithm, and the data acquisition optimization model includes a meta-learning module constructed based on the ML algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is respectively connected to the deep Q network, the experience replay pool and the meta-learning module, and the meta-learning module is connected to the experience replay pool.

6. A method for analyzing clinical data of hematological tumor patients according to claim 5, characterized in that: Based on different OPC UA information instances, according to the blood tumor knowledge graph and several pre-processed historical clinical data, a deep learning algorithm is used to build a clinical data analysis model and generate several historical clinical data analysis results, including the following steps: Writing the preprocessed historical clinical data into the corresponding OPC UA information instance, extracting the historical OPC UA information data of the OPC UA information instance, and performing graph format conversion to obtain the historical OPC UA information graph; Traversing all pre-processed historical clinical data of different OPC UA information instances to obtain a number of historical OPC UA information graphs, and dividing the several historical OPC UA information graphs into a model training set and a model test set; Use the Double GCN-RF algorithm to build an initial clinical data analysis model, input the model training set and blood tumor knowledge graph, train and optimize the initial clinical data analysis model, and obtain an optimized clinical data analysis model; Input the model test set and blood tumor knowledge graph, perform model testing on the optimized clinical data analysis model, and obtain the test accuracy of the optimized clinical data analysis model; If the test accuracy is greater than the accuracy threshold, the final clinical data analysis model is output and several historical clinical data analysis results are generated; otherwise, training optimization continues.

7. A method for analyzing clinical data of hematological tumor patients according to claim 5, characterized in that: Based on the analysis results of several historical clinical data, an enhanced reinforcement learning algorithm is used to build a data collection optimization model and generate several historical data collection optimization experiences, including the following steps: The data collection optimization problem is used as the reinforcement learning simulation environment of the DQN algorithm, the data collection optimization problem is divided into several subtasks, and the network parameter generation problem of the deep Q network of the DQN algorithm is used as the meta-learning simulation environment of the ML algorithm; According to the preset meta-feature parameters, several corresponding historical meta-features are extracted from the historical clinical data analysis results, and based on different subtasks and several historical meta-features, the ML algorithm is used to construct the initial meta-learning module and generate several historical meta-learning experiences; Based on the reinforcement learning simulation environment, build an intelligent agent and experience replay pool, use the initial meta-learning module to initialize the deep Q network of the DQN algorithm, and train the initial meta-learning module to obtain the final meta-learning module; Based on the historical clinical data analysis results, define the state space of the DQN algorithm, define the action space of the DQN algorithm based on the preset data collection strategy, and define the reward function based on the impact of the action in the action space on the state in the state space; Based on the action space, state space, intelligent agent, experience replay pool and deep Q network, the initial reinforcement learning module is constructed, and the initial reinforcement learning module is trained using several historical clinical data analysis results to obtain the final reinforcement learning module, and generate several historical data collection optimization experiences; The final meta-learning module and reinforcement learning module are integrated to obtain the data collection optimization model, and several historical meta-learning experiences and several historical data collection optimization experiences are stored in the experience replay pool.

8. A method for analyzing clinical data of hematological tumor patients according to claim 7, characterized in that: The cloud data center uses a data collection optimization model to optimize data collection based on the real-time clinical data analysis results, obtains a real-time data collection strategy, and sends it to the clinical data collection device, so that the clinical data collection device performs clinical data collection according to the received real-time data collection strategy, including the following steps: The cloud data center extracts corresponding real-time meta-features from the real-time clinical data analysis results according to preset meta-feature parameters; According to a number of real-time meta-features, a number of historical meta-learning experiences in the experience playback pool of the data acquisition optimization model are matched to obtain successfully matched real-time meta-learning experiences; According to the real-time meta-learning experience, the data collection optimization model is adjusted to obtain an adjusted data collection optimization model; According to the real-time clinical data analysis results, several historical data collection optimization experiences in the experience playback pool of the adjusted data collection optimization model are matched to obtain successfully matched real-time data collection optimization experiences; According to the real-time clinical data analysis results and the real-time data collection optimization experience, the adjusted data collection optimization model is used to optimize data collection and obtain the real-time data collection strategy; The real-time data collection strategy is encrypted and sent to the corresponding clinical data collection device, so that the clinical data collection device performs clinical data collection according to the received real-time data collection strategy.

9. A clinical data analysis system for patients with blood tumors, used to implement the clinical data analysis method for patients with blood tumors as claimed in any one of claims 1 to 8, characterized in that: The system includes a cloud data center and several clinical data acquisition devices, and the several clinical data acquisition devices are all communicatively connected to the cloud data center. The cloud data center includes an initialization unit, an information data extraction unit, a clinical data analysis unit, and a data acquisition optimization unit connected in sequence.

Images