Hematology department clinical infection source detection system and method based on big data analysis

Abstract

The invention discloses a hematology department clinical infection source detection system and method based on big data analysis, which are used in the field of medical detection, and the hematology department clinical infection source detection system comprises an image collection module, an image analysis module, an infected part prediction module, an infection source identification module and a risk assessment module. By integrating advanced image analysis, disease identification, risk assessment and data fusion technologies, accurate positioning and risk grade division of infection sources of patients are realized, the accuracy and efficiency of clinical detection are greatly improved, resource allocation is optimized, the treatment decision process is accelerated, and the treatment efficiency is improved. More personalized and timely medical intervention is provided for the patient, and the treatment effect and safety of the patient are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of medical detection, and in particular, to a hematology clinical infection source detection system and method based on big data analysis. Background Art

[0002] Clinical practice generally refers to medical practice that is directly related to patients, including the process of diagnosing, treating and preventing diseases. Clinical work is mainly carried out in hospitals or medical facilities, involving direct interaction between doctors, nurses and other medical professionals and patients. Its goal is to apply medical theory and practice to solve specific health problems and improve the quality of life of patients.

[0003] Hematology is a branch of clinical medicine that focuses on the diagnosis and treatment of diseases of the blood and its related organ systems (such as the bone marrow, spleen, and lymphatic system). Problems treated by hematologists include various types of anemia, blood clotting disorders, leukemia, lymphoma, and other blood cancers. Hematology also involves the study of blood components such as red blood cells, white blood cells, and platelets.

[0004] The source of infection refers to the source of pathogens that can cause infection. These pathogens may include bacteria, viruses, fungi, parasites, etc. The source of infection can be not only sick individuals, but also asymptomatic carriers, or certain objects and surfaces in the environment. Identification of the source of infection is crucial to preventing and controlling the spread of infection, developing treatment plans and preventive measures.

[0005] Many patients with blood diseases, especially those undergoing chemotherapy or bone marrow transplants, may have severely suppressed immune systems, making them more susceptible to infection. For hematology patients, common infections can rapidly develop into life-threatening situations, so early diagnosis and treatment are critical. By accurately identifying the source of infection, the most effective antibiotic or antiviral treatment can be selected for a specific pathogen, thereby improving the effectiveness and efficiency of treatment. In hospitals and diagnostic and treatment environments, correct identification of the source of infection helps to take appropriate isolation and disinfection measures to prevent the spread of pathogens between patients and medical staff.

[0006] Traditional clinical diagnostic methods rely on manual interpretation of images and symptoms, which is not only time-consuming but also easily affected by the doctor's experience and subjective judgment, resulting in limited diagnostic accuracy and efficiency. In the absence of rapid and accurate diagnostic support, the formulation of effective treatment plans may be delayed, affecting the patient's treatment timing and treatment effects. Traditional methods usually adopt a "one-size-fits-all" treatment plan, which makes it difficult to formulate personalized treatment plans based on the patient's specific situation. The existing clinical decision-making process rarely uses a rich medical knowledge base, resulting in treatment plans that may not be comprehensive enough and cannot fully utilize the latest medical research results.

[0007] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0008] In order to overcome the above problems, the present invention aims to propose a hematology clinical infection source detection system and method based on big data analysis, with the aim of solving the problem that traditional clinical diagnosis methods rely on manual interpretation of images and symptoms, which is not only time-consuming but also easily affected by the doctor's experience and subjective judgment, resulting in limited diagnostic accuracy and efficiency. In the absence of rapid and accurate diagnostic support, the formulation of effective treatment plans may be delayed, affecting the patient's treatment timing and treatment effect.

[0009] To this end, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a clinical infection source detection system for hematology based on big data analysis is provided, the clinical infection source detection system for hematology comprising: an image collection module, an image analysis module, an infection site prediction module, an infection source identification module and a risk assessment module; The image collection module, the image analysis module, the infection site prediction module, the infection source identification module and the risk assessment module are connected in sequence; An image collection module, used to collect clinical medical imaging images and clinical medical infection images of patients from the hematology department; An image analysis module, used to analyze the collected clinical medical images and mark the infected image areas based on the analysis results; Infection site prediction module, which predicts the infection site based on clinical medical infection images; The infection source identification module is used to combine the image features of the infected image area with the predicted infected image location, and use the knowledge graph in the clinical medical field to identify the patient's infection source image; The risk assessment module uses big data analysis to assess the infection risk of infection source images in a clinical setting and classifies the infection risk into risk levels; Among them, the infection site prediction module includes: Collect clinical medical infection images of patients from the hematology department, extract image features and patient information features, and combine the image features and patient information features to form a comprehensive feature set; Filter the feature set, build a classification model based on the filtered features, introduce a shared weight matrix into the classification model, and design a loss function; Use the gradient descent strategy to optimize the loss function and obtain the optimal classification model parameters; Construct a multi-layer classification model based on the optimal classification model parameters; The multi-layer classification model was evaluated using the cross-validation method and the prediction accuracy of the model was verified on the test set; New clinical medical infection images are input into the multi-layer classification model to predict the patient's infection image location.

[0010] Optionally, the image analysis module analyzes the collected clinical medical image and marks the infected image area based on the analysis result, including: Acquire clinical medical imaging images of patients collected from a hematology department, and input the clinical medical imaging images into a pre-configured convolutional neural network model; The convolutional and pooling layers of the convolutional neural network model extract comprehensive image features associated with infected image regions; Based on the comprehensive image features, the convolutional neural network model outputs a segmentation mask and intuitively displays the infected image area on the clinical medical image through color coding; According to the bounding box coordinates and category labels of the infected image area, positioning and classification are performed to identify and mark the infected image area.

[0011] Optionally, filter the feature set, build a classification model based on the filtered features, introduce a shared weight matrix into the classification model, and design a loss function including: Preset the target variable, and use information gain to evaluate the association between each feature in the feature set and the target variable, and select the feature subset that is most relevant to the target variable; Use the filtered feature subset to build a classification model and adjust the parameters of the classification model; Create a shared weight matrix and capture the interdependencies of infected image parts based on the shared weight matrix: Integrate the shared weight matrix into the classification model and design a loss function that includes a regularization term.

[0012] Optionally, a target variable is preset, and the information gain is used to evaluate the association between each feature in the feature set and the target variable. The feature subset most relevant to the target variable is selected, including: Preset the target variable of the infected image part and convert all features in the feature set into a unified format; Calculate the entropy of the target variable; For each feature, calculate the entropy of the target variable given the feature value; The information gain of each feature is calculated by subtracting the conditional entropy of the target variable after the given feature from the entropy of the target variable; Select the features whose information gain of each feature is higher than the preset threshold, and combine the screened features into a new feature subset; Among them, the formula for calculating the entropy of the target variable is: ; In the formula, Represents the target variable Y Get a specific value The probability of Represents the target variable Y Entropy of Y represents the target variable.

[0013] Optionally, the loss function is optimized using a gradient descent strategy, and the optimal classification model parameters are obtained including: Initialize the parameters of the classification model; Select an optimizer and set the learning rate; Calculate the gradient of the loss function under the current parameters through the back-propagation algorithm; Update the parameters of the classification model according to the calculated gradient and the set learning rate, and iterate in the direction of minimizing the loss; Gradient calculation and parameter update are repeated until the parameters of the classification model reach the predetermined number of iterations.

[0014] Optionally, the infection source identification module includes: combining the image features of the infected image area with the predicted infected image location and using the knowledge graph in the clinical medical field to identify the infection source image of the patient. Obtaining image features of the infected image area and predicted infected image parts; Collect knowledge in the medical field and build a clinical medical knowledge graph; Entity extraction and entity linking techniques are used to connect the image features of the infected image area with the predicted infected image parts and related entities in the knowledge graph; A hybrid prediction model is constructed based on the image features of the infected image area, the predicted infected image location and the information in the knowledge graph, and an attention mechanism is introduced to identify the patient's infection source image.

[0015] Optionally, using entity extraction and entity linking technology to connect the image features of the infected image area and the predicted infected image part with related entities in the knowledge graph includes: Using natural language processing tools, medical entities are extracted from medical documents and image descriptions, including infection type, lesion location, and symptom manifestation; Compare and match the names of infection types, lesion sites, and symptom manifestations with the medical terminology database, and based on the matching results, convert the names of infection types, lesion sites, and symptom manifestations into standardized forms in the standard terminology database; Using graph embedding technology, the standardized medical entities are accurately linked to the corresponding entities in the constructed clinical medical knowledge graph, and the association relationship between the entities is established; The method of using graph embedding technology to accurately link the standardized medical entities with the corresponding entities in the constructed clinical medical knowledge graph and establish the association relationship between the entities includes the following steps: In the clinical medical knowledge graph, according to the standardized infection type, lesion site, and symptom manifestation, corresponding nodes and their edges are updated or added; Select a graph embedding model based on the data size and sparsity of infection type, lesion site, symptom manifestation, and actual analysis needs; Using the current graph structure, train the selected graph embedding model to encode the nodes in the clinical medical knowledge graph into vector representations; Using the node vectors obtained through training, the similarity between the vector of the target medical entity and the existing node vectors in the graph is calculated; Based on the vector similarity results, the most relevant nodes are identified and linked to ensure that the standardized medical entities accurately match the corresponding entities in the knowledge graph; Based on the linking results, new relationships between entities are updated or defined to strengthen and clarify the interactions and associations between nodes in the clinical medical knowledge graph.

[0016] Optionally, a hybrid prediction model is constructed based on the image features of the infected image area, the predicted infected image location and the information in the knowledge graph, and an attention mechanism is introduced to identify the infection source image of the patient, including: Integrate and normalize the image features of the infected image area, the predicted infected image location, and the information in the knowledge graph; Through the fusion layer technology in the deep learning network, the image and semantic features are combined to create a comprehensive hybrid prediction model; Design and apply a multi-head attention mechanism to evaluate and optimize the importance of each feature in identifying the source of infection; Use labeled data sets to train the model and optimize the model parameters through cross-validation techniques; The models were tested on an independent validation set and thoroughly evaluated using multiple performance metrics.

[0017] Optionally, the risk assessment module assesses the infection risk of the infection source image by big data analysis in a clinical environment and classifies the infection risk into risk levels, including: Identify key indicators for measuring infection risk; Analyze infection sources using statistical methods and assess infection risks of infection sources under various conditions based on key indicators; Level thresholds are preset at each stage of infection risk, and risk levels of infection are divided based on the preset level thresholds.

[0018] According to another aspect of the present invention, a method for detecting clinical infection sources in hematology based on big data analysis is also provided, and the method for detecting clinical infection sources in hematology includes: S1. Collect clinical medical imaging images and clinical medical infection images of patients from the Department of Hematology; S2, analyzing the collected clinical medical imaging images, and marking the infected image areas based on the analysis results; S3, predicting the infected image location based on clinical medical infection images; S4, combining the image features of the infected image area with the predicted infected image location, and using the knowledge graph in the clinical medical field to identify the patient's infection source image; S5. Use big data analysis to assess the infection risk of infection source images in a clinical setting and classify the infection risk into risk levels.

[0019] Compared with the prior art, this application has the following beneficial effects: 1. The present invention integrates advanced image analysis, disease identification, risk assessment and data fusion technologies to achieve accurate positioning of the patient's infection source and risk level classification, greatly improving the accuracy and efficiency of clinical testing, optimizing resource allocation, accelerating the treatment decision-making process, providing patients with more personalized and timely medical intervention, and significantly improving the patient's treatment effect and safety.

[0020] 2. The present invention integrates the infection site prediction module of clinical medical infection image features and patient information, and effectively improves the accuracy of identifying and predicting infection sites through sophisticated data processing and advanced machine learning techniques (such as XGBoost and shared weight matrix). It can not only accurately capture the specific location and nature of the infection, but also further refine the feature selection and model training process through information gain and optimized loss function, and finally achieve efficient clinical decision support. The use of cross-validation and multi-layer classification technology ensures the robustness and wide applicability of the model, significantly improves the speed and accuracy of medical detection, and provides patients with more accurate and personalized treatment plans.

[0021] 3. The present invention effectively identifies and predicts infection source images through the comprehensive application of advanced image feature extraction and knowledge graphs, and uses a hybrid prediction model and attention mechanism to accurately analyze and identify the infection source. At the same time, it combines a broad knowledge base in the clinical medical field to enhance the accuracy and reliability of detection. In addition, through entity linking technology, the image features of the infected area are closely combined with relevant clinical data, making the model not just a simple image recognition, but an intelligent system that fully considers the clinical context, which greatly enhances doctors' understanding and handling of infectious diseases, optimizes treatment strategies, and improves patients' treatment effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown: Figure 1 This is a principle block diagram of a hematology clinical infection source detection system based on big data analysis according to an embodiment of the present invention; Figure 2 The present invention is a flowchart of a method for detecting clinical infection sources in hematology based on big data analysis according to an embodiment of the present invention.

[0023] In the figure: 1. Image collection module; 2. Image analysis module; 3. Infection site prediction module; 4. Infection source identification module; 5. Risk assessment module. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] According to an embodiment of the present invention, a system and method for detecting clinical infection sources in hematology based on big data analysis are provided.

[0026] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a hematology clinical infection source detection system based on big data analysis is provided, and the hematology clinical infection source detection system includes: an image collection module 1, an image analysis module 2, an infection site prediction module 3, an infection source identification module 4 and a risk assessment module 5; The image collection module 1, the image analysis module 2, the infection site prediction module 3, the infection source identification module 4 and the risk assessment module 5 are connected in sequence; The image collection module 1 is used to collect clinical medical imaging images and clinical medical infection images of patients from the hematology department.

[0027] It should be explained that clinical medical imaging images include the following types: 1) X-rays: used to detect fractures, lung diseases, etc.

[0028] 2) CT scan (computed tomography): provides detailed cross-sectional images of the body's internal structures and is often used to detect tumors, bleeding, trauma, etc.

[0029] 3) MRI scan (magnetic resonance imaging): used to show the condition of soft tissues in the body in detail, such as the brain, spinal cord, internal organs, etc.

[0030] 4) Ultrasound examination: often used to observe pregnancy, blood vessel conditions, abdominal organs, etc.

[0031] Clinical medical infection images include the following types: 1) X-ray or CT scan of the infection site: For example, pneumonia may show abnormal shadowy areas on X-ray images.

[0032] 2) MRI or ultrasound: used to check for soft tissue infection or infection of internal organs (such as the liver and kidneys).

[0033] 3) PET scan (positron emission tomography): Sometimes used to find the source of infection, especially in complex cases, it can help locate potential areas of infection.

[0034] 4) Angiography: If the infection involves the vascular system, such as sepsis or vasculitis.

[0035] The image collection module allows these clinical medical images to be collected and organized. For infection images, doctors and researchers can conduct more in-depth analysis of the type, severity and spread of the infection, as well as its correlation with other medical conditions of the patient.

[0036] The image analysis module 2 is used to analyze the collected clinical medical images and mark the infected image areas based on the analysis results.

[0037] Preferably, the image analysis module 2 analyzes the collected clinical medical image and marks the infected image area based on the analysis result, including: Acquire clinical medical imaging images of patients collected from a hematology department, and input the clinical medical imaging images into a pre-configured convolutional neural network model; The convolutional and pooling layers of the convolutional neural network model extract comprehensive image features associated with infected image regions; Based on the comprehensive image features, the convolutional neural network model outputs a segmentation mask and intuitively displays the infected image area on the clinical medical image through color coding; According to the bounding box coordinates and category labels of the infected image area, positioning and classification are performed to identify and mark the infected image area.

[0038] It needs to be explained that clinical medical images of patients are collected from the hematology department and input into the pre-configured convolutional neural network model to ensure that all images that need to be analyzed can be correctly processed by the model; The convolutional layers and pooling layers in the convolutional neural network are used to extract comprehensive image features related to infection from the images. The convolutional layer is responsible for capturing local features in the image, while the pooling layer is used to reduce the feature dimension to extract higher-level abstract features.

[0039] Based on the comprehensive image features, the network outputs a segmentation mask to distinguish infected and non-infected areas, and intuitively displays the infected areas on the original image through color coding, enhancing the visual expression of the results.

[0040] Based on the infected image areas marked in the segmentation mask, accurate localization and classification are performed using bounding box coordinates and category labels. This ensures that the infected areas are not only identified but also correctly classified, providing important information for clinical detection.

[0041] The infection site prediction module 3 predicts the infection site of the image based on the clinical medical infection image.

[0042] The infection site prediction module 3 includes: Clinical medical infection images of patients are collected from the hematology department, and image features and patient information features are extracted and combined to form a comprehensive feature set.

[0043] It needs to be explained that clinical medical infection images of patients are collected from the hematology department. Clinical medical infection images include images generated by imaging technologies such as X-ray, CT, and MRI, which are specifically used to show the details of the infected area. Image processing techniques (such as edge detection, texture analysis, and morphological operations) are used to extract key features in the images. The key features can depict the shape, size, location, and contrast with surrounding tissues of the infected area. At the same time, the patient's basic information and relevant medical data, such as age, gender, medical history, laboratory test results, etc., are collected. This information can be obtained from the electronic medical record (EMR) and is the key to understanding the patient's overall health status. The comprehensive feature set can be used to support clinical decisions, such as infection severity assessment and treatment plan selection.

[0044] Filter the feature set, build a classification model based on the filtered features, introduce a shared weight matrix into the classification model, and design a loss function.

[0045] Preferably, the feature set is screened, a classification model is constructed based on the screened features, and a shared weight matrix is ​​introduced into the classification model. At the same time, the loss function is designed to include: Preset the target variable, and use information gain to evaluate the association between each feature in the feature set and the target variable, and select the feature subset that is most relevant to the target variable; Use the filtered feature subset to build a classification model and adjust the parameters of the classification model; Create a shared weight matrix and capture the interdependencies of infected image parts based on the shared weight matrix: Integrate the shared weight matrix into the classification model and design a loss function that includes a regularization term.

[0046] Preferably, the target variable is preset, and the association between each feature in the feature set and the target variable is evaluated using information gain, and the feature subset most relevant to the target variable is screened, including: Preset the target variable of the infected image part and convert all features in the feature set into a unified format; Calculate the entropy of the target variable; For each feature, calculate the entropy of the target variable given the feature value; The information gain of each feature is calculated by subtracting the conditional entropy of the target variable after the given feature from the entropy of the target variable; Select the features whose information gain of each feature is higher than the preset threshold, and combine the screened features into a new feature subset; Among them, the formula for calculating the entropy of the target variable is: ; In the formula, Represents the target variable Y Get a specific value The probability of Represents the target variable Y Entropy of Y represents the target variable.

[0047] The formula for calculating the entropy of the target variable under the condition of a given eigenvalue is: ; In the formula, Indicates that the known variable X After the value of Y The conditional entropy of Representation variables X Get a specific value x j The probability of Indicates in the variable X equal x j Under the condition that Y Entropy of x j Representation variables X A specific value of i ,y Both represent indexes.

[0048] It should be explained that the infected image site is first set as the target variable, which will guide the training process of the entire model and ensure that all features have a unified format before analysis. This is an important step in data preprocessing and helps avoid deviations in subsequent analysis. The entropy of the target variable is calculated to obtain its initial information state. For each feature, the conditional entropy of the target variable under the condition of the feature value is calculated. The information gain is calculated by subtracting the conditional entropy from the entropy of the target variable. The amount of information provided by each feature for the target variable is measured, and features with information gain higher than the preset threshold are selected to form a new feature subset. The features screened out in this way are more critical for predicting the target variable.

[0049] A classification model is constructed using the filtered feature subset, and the classification model is an XGBoost classification model. XGBoost is an efficient gradient boosting algorithm that is widely used in classification tasks. Its advantages lie in its speed and effect when processing large-scale data sets. The parameters of the XGBoost model are adjusted through methods such as cross-validation to find the optimal parameter configuration to improve model performance; a shared weight matrix is ​​designed to capture the interdependence between different infected image parts, which can help the model to be more accurate when dealing with multi-label classification problems. The shared weight matrix is ​​integrated into the XGBoost model so that the model can take into account the correlation between different labels during training. A loss function containing a regularization term is designed to optimize the XGBoost model. The regularization term helps control the complexity of the model and avoid overfitting, so that the model performs better on unseen data.

[0050] The method of combining information gain, XGBoost, shared weight matrix and regularized loss function provides a powerful framework to enhance the performance and robustness of the model when processing complex clinical imaging data.

[0051] The gradient descent strategy is used to optimize the loss function and obtain the optimal classification model parameters.

[0052] Preferably, the loss function is optimized using a gradient descent strategy, and the optimal classification model parameters are obtained including: Initialize the parameters of the classification model; Select an optimizer and set the learning rate; Calculate the gradient of the loss function under the current parameters through the back-propagation algorithm; Update the parameters of the classification model according to the calculated gradient and the set learning rate, and iterate in the direction of minimizing the loss; Gradient calculation and parameter update are repeated until the parameters of the classification model reach the predetermined number of iterations.

[0053] It should be explained that model parameters (such as weights and biases) are usually initialized to small random numbers or use specific initialization methods (such as He or Xavier initialization) to help the algorithm converge more effectively; commonly used optimizers include SGD (stochastic gradient descent), Adam, RMSprop, etc. Different optimizers are suitable for different application scenarios. Among them, Adam is popular for its adaptive learning rate adjustment. The learning rate is a hyperparameter that controls the step size of parameter update. It is crucial to set an appropriate learning rate. Too high a learning rate may lead to instability during training, while too low a learning rate may lead to too slow training speed. The backpropagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters. This process involves a forward pass on the model to calculate the loss, and then backward pass the gradient to obtain the gradient of the parameter. According to the calculated gradient and the set learning rate, the model parameters are updated. The update rule is usually that the new parameter value is equal to the original value minus the learning rate multiplied by the gradient. This step is iterative, and the goal is to gradually reduce the loss of the model. The gradient calculation and parameter update process are repeated until the predetermined number of iterations is reached, or the iteration is stopped when the performance of the model is no longer significantly improved. Usually, the performance on the validation set is also monitored to avoid overfitting of the model.

[0054] A multi-layer classification model is constructed based on the optimal classification model parameters.

[0055] The multi-layer classification model was evaluated using the cross-validation method, and the prediction accuracy of the model was verified on the test set.

[0056] New clinical medical infection images are input into the multi-layer classification model to predict the patient's infection image location.

[0057] It should be explained that a multi-layer classification model is constructed based on the best classification model parameters obtained by previous optimization; the multi-layer classification model includes multiple hidden layers, and each layer uses an activation function, such as ReLU or Sigmoid, to increase the nonlinear processing capability of the model; a cross-validation method is used to evaluate the performance of the multi-layer classification model; this usually includes dividing the data set into several subsets (such as five-fold or ten-fold cross-validation), where each subset is used as a test set in turn, and the rest are used as training sets to ensure that the model can be effectively verified on different data subsets, and appropriate evaluation indicators are selected, such as accuracy, recall rate, F1 score, etc., to comprehensively evaluate the performance of the model in all aspects, verify the predictive accuracy of the multi-layer classification model on an independent test set, and input the newly collected clinical medical infection images into the trained multi-layer classification model to predict the patient's infection image site, which can help doctors quickly identify the infected area and start treatment early.

[0058] The infection source identification module 4 is used to combine the image features of the infected image area with the predicted infected image location, and use the knowledge graph in the clinical medical field to identify the patient's infection source image.

[0059] Preferably, the infection source identification module 4 includes: combining the image features of the infected image area with the predicted infected image location and using the knowledge graph in the clinical medical field to identify the infection source image of the patient. Obtain the image features of the infected image area and the predicted infected image location.

[0060] Collect knowledge in the medical field and build a clinical medical knowledge graph.

[0061] It needs to be explained that, first, the infected image area is preprocessed, including denoising, contrast enhancement, etc., to improve the image quality. Image analysis techniques, such as edge detection, texture analysis, color analysis, etc., are used to extract key image features of the infected area. The key image features include area size, shape, boundary clarity, texture uniformity, etc. The extracted image features are input into the previously trained classification model to predict the specific location of the infected image; information about diseases, symptoms, treatment methods, etc. is collected from medical literature, clinical trial records, online medical databases and other sources, and the collected information is organized to define entities (such as diseases, symptoms, drugs, treatment methods) and relationships between entities (such as the association between symptoms and diseases). Knowledge graph technology is used to convert structured data into a graph database, in which nodes represent entities and edges represent relationships between entities. This can form a complex network to support clinical decision-making and research. Knowledge graphs are used to support clinical decisions, such as personalized medical recommendations. At the same time, knowledge graphs can also be used for education and research to help medical professionals and scholars better understand complex medical information and knowledge structures.

[0062] Entity extraction and entity linking techniques are used to connect the image features of the infected image area with the predicted infected image parts and related entities in the knowledge graph.

[0063] Preferably, using entity extraction and entity linking technology to connect the image features of the infected image area and the predicted infected image part with the relevant entities in the knowledge graph includes: Natural language processing tools are used to extract medical entities from medical documents and image descriptions, where medical entities include infection type, lesion location, and symptom manifestation.

[0064] It should be explained that before entity extraction, medical documents and image descriptions need to be preprocessed, including text cleaning (removing irrelevant characters, punctuation marks, etc.), format unification, encoding conversion, etc., and regular expressions are used to remove or replace special characters and noise in the text. For example, the re library in Python can be used for this purpose. Natural language processing tools are used to identify medical entities in the text, such as infection type, lesion site, and symptom manifestation. Entities are usually proper nouns or technical terms. The identified medical entities may have multiple expressions. These entities need to be converted into standardized medical terms. Usually, the extracted entities are mapped to standard terms by matching to an authoritative medical terminology library (such as ICD-10, SNOMED CT, etc.).

[0065] The names of infection types, lesion sites, and symptom manifestations are compared and matched with the medical terminology library. Based on the matching results, the names of infection types, lesion sites, and symptom manifestations are converted into standardized forms in the standard terminology library.

[0066] It should be explained that a suitable medical terminology library should be selected, such as ICD-10, SNOMED CT or UMLS. These medical terminology libraries contain a wide range of medical terms and their standardized forms. The most appropriate terminology library should be selected according to the data requirements and the goals of the project. The extracted infection types, lesion sites, and symptom names should be matched with the terms in the selected medical terminology library. String matching technology or more advanced semantic matching tools, such as natural language processing algorithms, should be used to find the closest terms. Once a matching term is found, the extracted non-standard medical entity names should be converted into the standardized form in the terminology library.

[0067] Using graph embedding technology, the standardized medical entities are accurately linked to the corresponding entities in the constructed clinical medical knowledge graph, and the association relationship between the entities is established; The method of using graph embedding technology to accurately link the standardized medical entities with the corresponding entities in the constructed clinical medical knowledge graph and establish the association relationship between the entities includes the following steps: In the clinical medical knowledge graph, according to the standardized infection type, lesion site, and symptom manifestation, corresponding nodes and their edges are updated or added; Select a graph embedding model based on the data size and sparsity of infection type, lesion site, symptom manifestation, and actual analysis needs; Using the current graph structure, train the selected graph embedding model to encode the nodes in the clinical medical knowledge graph into vector representations; Using the node vectors obtained through training, the similarity between the vector of the target medical entity and the existing node vectors in the graph is calculated; Based on the vector similarity results, the most relevant nodes are identified and linked to ensure that the standardized medical entities accurately match the corresponding entities in the knowledge graph; Based on the linking results, new relationships between entities are updated or defined to strengthen and clarify the interactions and associations between nodes in the clinical medical knowledge graph.

[0068] It should be explained that according to the infection type, lesion site and symptom manifestation extracted and standardized from the documents, the corresponding nodes and edges are updated or added to the clinical medical knowledge graph, and a graph database management system, such as Neo4j or ArangoDB, is used to manage and update the graph; nodes represent medical entities, and edges represent the relationships between entities, such as correlation or causality. In the graph, the accurate representation of nodes and edges is crucial for subsequent analysis and links.

[0069] Select a suitable graph embedding model based on the scale, sparsity and analysis requirements of the entity data, and evaluate different graph embedding models such as Node2Vec, GraphSAGE or DeepWalk. The graph embedding model can convert the nodes in the graph structure into vectors in a low-dimensional space. The vectors can capture the topological relationship and node attributes between nodes. Use the current structure of the clinical medical knowledge graph to train the selected graph embedding model, input the graph data into the model for training, and use the node and edge information in the graph structure to generate the vector representation of each node. During the training process, the model learns the optimal vector representation of the node by optimizing an objective function, which usually involves neighbor nodes. Maximize the similarity between points, use the node vectors obtained through training to calculate the similarity between the vector of the target medical entity and the existing node vectors in the graph, identify and link the most relevant nodes based on the vector similarity results to ensure that the medical entity accurately matches the corresponding entity in the knowledge graph, select the node with the highest similarity score as the matching node, and create or update the links between the nodes in the graph. Based on the link results, update or define new relationships between entities, strengthen and clarify the interactions and associations between nodes in the clinical medical knowledge graph, and update the attributes or types of the edges in the graph based on the newly discovered links between entities, such as adding new causal or correlation labels.

[0070] Suppose data about the infection type "pneumonia" is extracted from a batch of medical documents, including the common lesion site "lower respiratory tract" and the symptoms "cough" and "fever".

[0071] The specific example steps are as follows: 1) Update or add nodes and edges: In the clinical medical knowledge graph, add nodes such as "pneumonia", "lower respiratory tract", "cough", and "fever".

[0072] Establish edges between nodes. For example, "pneumonia" is connected with "lower respiratory tract" to indicate the location of the lesion; "pneumonia" is connected with "cough" and "fever" to indicate the symptoms.

[0073] 2) Choose a graph embedding model: Given that the entities involved in the dataset are relatively clear and closely related, the Node2Vec model is selected for graph embedding to optimize the structural and functional expressions between nodes.

[0074] 3) Training graph embedding model: Input the graph data constructed above into the Node2Vec model and train the model to obtain the vector representation of each node.

[0075] 4) Calculate vector similarity: The cosine similarity calculation method is used to compare the similarity between the "pneumonia" node vector and the node vectors of other diseases or symptoms in the graph.

[0076] 5) Identify and link the most relevant nodes: Identify other respiratory infection disease nodes that are highly similar to "pneumonia", such as "bronchitis", and link these nodes, indicating that they have similar symptoms or lesion sites.

[0077] 6) Update or define new relationships between entities: In the knowledge graph, the attributes of the edges are updated based on the new link relationship, such as adding labels such as "possible complications" or "common treatment methods" to strengthen the practical relevance between nodes. A hybrid prediction model is constructed based on the image features of the infected image area, the predicted infected image location, and the information in the knowledge graph, and an attention mechanism is introduced to identify the patient's infection source image.

[0078] Preferably, a hybrid prediction model is constructed based on the image features of the infected image area, the predicted infected image location and the information in the knowledge graph, and an attention mechanism is introduced to identify the infection source image of the patient, including: Integrate and normalize the image features of the infected image area, the predicted infected image location, and the information in the knowledge graph; Through the fusion layer technology in the deep learning network, the image and semantic features are combined to create a comprehensive hybrid prediction model; Design and apply a multi-head attention mechanism to evaluate and optimize the importance of each feature in identifying the source of infection; Use labeled data sets to train the model and optimize the model parameters through cross-validation techniques; The models were tested on an independent validation set and thoroughly evaluated using multiple performance metrics.

[0079] It needs to be explained that, first, the image features of the infected image area, the predicted infected image parts and the information in the knowledge graph are integrated together. The data sources have different scales and formats and need to be processed uniformly so that the model can learn effectively. The integrated data is normalized to ensure the stability and efficiency of the model training. Normalization usually includes scaling the feature values ​​to the same numerical range, such as between 0 and 1 or -1 to 1.

[0080] Use the fusion layer technology in the deep learning network to combine image and semantic features. This can be achieved in many ways, such as serial fusion, weighted fusion, etc. The purpose is to create a comprehensive model that can understand both image features and semantic information at the same time, and design the model architecture according to the needs, including choosing the appropriate deep learning framework and network structure (such as CNN, RNN, Transformer, etc.).

[0081] A multi-head attention mechanism is designed and applied, which enables the model to focus on features that are particularly important for identifying infection source images. The multi-head attention mechanism allows the model to analyze input data from multiple angles, enhances the model's ability to capture key information, and evaluates the importance of each feature in identifying infection sources through attention scores, thereby optimizing the model's interpretability and performance.

[0082] Risk Assessment Module 5: evaluates the infection risk of infection source images in a clinical setting through big data analysis and classifies the infection risk into risk levels; Preferably, the risk assessment module assesses the infection risk of the infection source image by big data analysis in a clinical environment and classifies the infection risk into risk levels, including: Identify key indicators for measuring infection risk; Analyze infection sources using statistical methods and assess infection risks of infection sources under various conditions based on key indicators; Level thresholds are preset at each stage of infection risk, and risk levels of infection are divided based on the preset level thresholds.

[0083] It should be explained that we should first identify key indicators that can reflect the severity of infection and the potential for transmission. Key indicators may include the size and depth of the infected area, the type of infection (such as bacteria, viruses), known drug resistance characteristics, etc. We should collect data on historical infection cases, including imaging features, treatment outcomes, patient responses, etc. We should apply statistical analysis methods, such as regression analysis and survival analysis, to evaluate the infection risks of different sources of infection under various conditions, help identify which factors are most likely to lead to serious infection outcomes, and preset different risk level thresholds according to the severity of the infection risk. For example, we can set three levels of risk assessment standards: low, medium, and high. High risks may require immediate intervention, while low risks require routine monitoring. Based on the preset level thresholds and statistical analysis results, we should classify the risks of infection sources. This needs to be implemented on a big data platform to process and analyze large amounts of clinical data. The risk assessment results can directly support medical decision-making and help doctors choose the most appropriate treatment plan. High-risk infections may require more medical resources and emergency responses, and through risk level classification, hospitals can allocate resources more effectively.

[0084] Practical application example: Pneumonia infection detection and risk assessment, the steps are as follows: The following is a hypothetical application scenario: A patient is admitted to the hospital with acute respiratory discomfort. The doctor suspects pneumonia and uses X-rays and CT scans to collect chest images of the patient; Operation 1: The image collection module automatically obtains the latest patient images from the hospital's imaging system; analyzes the collected chest X-rays and CT scans, and analyzes the patient images to determine whether there are signs of infection; Operation 2: Analyze the image through a pre-configured deep learning model (such as a convolutional neural network), identify and mark areas showing abnormal shadows and typical signs of infection; predict the specific infection site based on the marked image features; Operation 3: The module analyzes details in the image, such as shadow size, location, and morphology, and combines the patient's clinical information (such as age and medical history) to predict the specific location and severity of the infection and determine the specific type of infection, such as bacterial pneumonia or viral pneumonia; Operation 4: Use the features of the infected image area in combination with the knowledge graph to identify possible pathogens; for example, by comparing image features with historical data, the module may predict that the infection is Streptococcus pneumoniae; assess the risk level of the pneumonia infection and provide a basis for treatment decisions; Action 5: Assess the risk level of infection based on information such as the location, size, patient age and comorbidities of the infection; use big data analysis to determine that the infection is medium risk and recommend hospitalization and the use of antibiotics.

[0085] According to another embodiment of the present invention, Figure 2As shown, a method for detecting clinical infection sources in hematology based on big data analysis is also provided, and the method for detecting clinical infection sources in hematology includes: S1. Collect clinical medical imaging images and clinical medical infection images of patients from the Department of Hematology; S2, analyzing the collected clinical medical imaging images, and marking the infected image areas based on the analysis results; S3, predicting the infected image location based on clinical medical infection images; S4, combining the image features of the infected image area with the predicted infected image location, and using the knowledge graph in the clinical medical field to identify the patient's infection source image; S5. Use big data analysis to assess the infection risk of infection source images in a clinical setting and classify the infection risk into risk levels.

[0086] In summary, with the help of the above technical solutions of the present invention, the present invention realizes the precise positioning of the infection source of patients and the risk level classification by integrating advanced image analysis, disease identification, risk assessment and data fusion technology, greatly improves the accuracy and efficiency of clinical detection, optimizes resource allocation, accelerates the treatment decision-making process, provides patients with more personalized and timely medical intervention, and significantly improves the treatment effect and safety of patients; the present invention integrates the infection site prediction module of clinical medical infection image features and patient information, and effectively improves the accuracy of identifying and predicting infection sites through sophisticated data processing and advanced machine learning technologies (such as XGBoost and shared weight matrix); it can not only accurately capture the specific location and nature of the infection, but also further refine the feature selection and model training process through information gain and optimized loss function, and finally achieve Efficient clinical decision support uses cross-validation and multi-layer classification technology to ensure the robustness and wide applicability of the model, significantly improves the speed and accuracy of medical testing, and provides patients with more accurate and personalized treatment plans; the present invention effectively identifies and predicts infection source images through advanced image feature extraction and comprehensive application of knowledge graphs, and uses hybrid prediction models and attention mechanisms to accurately analyze and identify infection sources. At the same time, combined with a broad knowledge base in the clinical medical field, it enhances the accuracy and reliability of detection. In addition, through entity linking technology, the image features of the infected area are closely combined with relevant clinical data, making the model not just a simple image recognition, but an intelligent system that fully considers the clinical context, which greatly enhances doctors' understanding and handling of infectious diseases, optimizes treatment strategies, and improves patient treatment outcomes.

[0087] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A hematology clinical infection source detection system based on big data analysis, characterized in that: The hematology clinical infection source detection system includes: image collection module, image analysis module, infection site prediction module, infection source identification module and risk assessment module; The image collection module, the image analysis module, the infection site prediction module, the infection source identification module and the risk assessment module are connected in sequence; An image collection module, used to collect clinical medical imaging images and clinical medical infection images of patients from the hematology department; An image analysis module, used to analyze the collected clinical medical images and mark the infected image areas based on the analysis results; Infection site prediction module, which predicts the infection site based on clinical medical infection images; The infection source identification module is used to combine the image features of the infected image area with the predicted infected image location, and use the knowledge graph in the clinical medical field to identify the patient's infection source image; The risk assessment module uses big data analysis to assess the infection risk of infection source images in a clinical setting and classifies the infection risk into risk levels; Wherein, the infection site prediction module includes: Collect clinical medical infection images of patients from the hematology department, extract image features and patient information features, and combine the image features and patient information features to form a comprehensive feature set; Filter the feature set, build a classification model based on the filtered features, introduce a shared weight matrix into the classification model, and design a loss function; Use the gradient descent strategy to optimize the loss function and obtain the optimal classification model parameters; Construct a multi-layer classification model based on the optimal classification model parameters; The multi-layer classification model was evaluated using the cross-validation method and the prediction accuracy of the model was verified on the test set; New clinical medical infection images are input into the multi-layer classification model to predict the patient's infection image location.

2. According to claim 1, a hematology clinical infection source detection system based on big data analysis is characterized in that: The image analysis module analyzes the collected clinical medical images and marks the infected image area based on the analysis results, including: Acquire clinical medical imaging images of patients collected from a hematology department, and input the clinical medical imaging images into a pre-configured convolutional neural network model; The convolutional and pooling layers of the convolutional neural network model extract comprehensive image features associated with infected image regions; Based on the comprehensive image features, the convolutional neural network model outputs a segmentation mask and intuitively displays the infected image area on the clinical medical image through color coding; According to the bounding box coordinates and category labels of the infected image area, positioning and classification are performed to identify and mark the infected image area.

3. According to the big data analysis-based hematology clinical infection source detection system of claim 1, it is characterized in that: The screening feature set will build a classification model based on the screened features, introduce a shared weight matrix into the classification model, and design a loss function including: Preset the target variable, and use information gain to evaluate the association between each feature in the feature set and the target variable, and select the feature subset that is most relevant to the target variable; Use the filtered feature subset to build a classification model and adjust the parameters of the classification model; Create a shared weight matrix and capture the interdependencies of infected image parts based on the shared weight matrix: Integrate the shared weight matrix into the classification model and design a loss function that includes a regularization term.

4. A hematology clinical infection source detection system based on big data analysis according to claim 3, characterized in that: The preset target variable and the information gain are used to evaluate the association between each feature in the feature set and the target variable, and the feature subset most relevant to the target variable is selected, including: Preset the target variable of the infected image part and convert all features in the feature set into a unified format; Calculate the entropy of the target variable; For each feature, calculate the entropy of the target variable given the feature value; The information gain of each feature is calculated by subtracting the conditional entropy of the target variable after the given feature from the entropy of the target variable; Select the features whose information gain of each feature is higher than the preset threshold, and combine the screened features into a new feature subset; Wherein, the formula for calculating the entropy of the target variable is: ； In the formula, Represents the target variable Y Get a specific value The probability of Represents the target variable Y Entropy of Y represents the target variable.

5. A hematology clinical infection source detection system based on big data analysis according to claim 4, characterized in that: The method of optimizing the loss function using the gradient descent strategy and obtaining the best classification model parameters includes: Initialize the parameters of the classification model; Select an optimizer and set the learning rate; Calculate the gradient of the loss function under the current parameters through the back-propagation algorithm; Update the parameters of the classification model according to the calculated gradient and the set learning rate, and iterate in the direction of minimizing the loss; Gradient calculation and parameter update are repeated until the parameters of the classification model reach the predetermined number of iterations.

6. A hematology clinical infection source detection system based on big data analysis according to claim 1, characterized in that: The infection source identification module combines the image features of the infected image area with the predicted infected image location and uses the knowledge graph in the clinical medical field to identify the infection source image of the patient, including: Obtaining image features of the infected image area and predicted infected image parts; Collect knowledge in the medical field and build a clinical medical knowledge graph; Entity extraction and entity linking techniques are used to connect the image features of the infected image area with the predicted infected image parts and related entities in the knowledge graph; A hybrid prediction model is constructed based on the image features of the infected image area, the predicted infected image location and the information in the knowledge graph, and an attention mechanism is introduced to identify the patient's infection source image.

7. A hematology clinical infection source detection system based on big data analysis according to claim 6, characterized in that: The method of using entity extraction and entity linking technology to connect the image features of the infected image area and the predicted infected image part with the relevant entities in the knowledge graph includes: Using natural language processing tools, medical entities are extracted from medical documents and image descriptions, including infection type, lesion location, and symptom manifestation; Compare and match the names of infection types, lesion sites, and symptom manifestations with the medical terminology database, and based on the matching results, convert the names of infection types, lesion sites, and symptom manifestations into standardized forms in the standard terminology database; Using graph embedding technology, the standardized medical entities are accurately linked to the corresponding entities in the constructed clinical medical knowledge graph, and the association relationship between the entities is established; The method of using graph embedding technology to accurately link the standardized medical entities with the corresponding entities in the constructed clinical medical knowledge graph and establish the association relationship between the entities includes the following steps: In the clinical medical knowledge graph, according to the standardized infection type, lesion site, and symptom manifestation, corresponding nodes and their edges are updated or added; Select a graph embedding model based on the data size and sparsity of infection type, lesion site, symptom manifestation, and actual analysis needs; Using the current graph structure, train the selected graph embedding model to encode the nodes in the clinical medical knowledge graph into vector representations; Using the node vectors obtained through training, the similarity between the vector of the target medical entity and the existing node vectors in the graph is calculated; Based on the vector similarity results, the most relevant nodes are identified and linked to ensure that the standardized medical entities accurately match the corresponding entities in the knowledge graph; Based on the linking results, new relationships between entities are updated or defined to strengthen and clarify the interactions and associations between nodes in the clinical medical knowledge graph.

8. A hematology clinical infection source detection system based on big data analysis according to claim 7, characterized in that: The hybrid prediction model is constructed based on the image features of the infected image area, the predicted infected image location and the information in the knowledge graph, and the attention mechanism is introduced to identify the infection source image of the patient, including: Integrate and normalize the image features of the infected image area, the predicted infected image location, and the information in the knowledge graph; Through the fusion layer technology in the deep learning network, the image and semantic features are combined to create a comprehensive hybrid prediction model; Design and apply a multi-head attention mechanism to evaluate and optimize the importance of each feature in identifying the source of infection; Use labeled data sets to train the model and optimize the model parameters through cross-validation techniques; The models were tested on an independent validation set and thoroughly evaluated using multiple performance metrics.

9. The hematology clinical infection source detection system based on big data analysis according to claim 1 is characterized in that: The risk assessment module assesses the infection risk of the infection source image by big data analysis in a clinical environment and classifies the infection risk into risk levels, including: Identify key indicators for measuring infection risk; Analyze infection sources using statistical methods and assess infection risks of infection sources under various conditions based on key indicators; Level thresholds are preset at each stage of infection risk, and risk levels of infection are divided based on the preset level thresholds.

10. A method for detecting clinical infection sources in hematology based on big data analysis, used to implement the system for detecting clinical infection sources in hematology based on big data analysis as claimed in any one of claims 1 to 9, characterized in that: The hematology clinical infection source detection method includes: S1. Collect clinical medical imaging images and clinical medical infection images of patients from the Department of Hematology; S2, analyzing the collected clinical medical imaging images, and marking the infected image areas based on the analysis results; S3, predicting the infected image location based on clinical medical infection images; S4, combining the image features of the infected image area with the predicted infected image location, and using the knowledge graph in the clinical medical field to identify the patient's infection source image; S5. Use big data analysis to assess the infection risk of infection source images in a clinical setting and classify the infection risk into risk levels.

Images