Intelligent re-consultation method and system based on dynamic multi-source data fusion and self-adaptive knowledge generation, terminal and medium

Abstract

The invention belongs to the technical field of medical health, and particularly discloses an intelligent re-consultation method and system based on dynamic multi-source data fusion and self-adaptive knowledge generation, a terminal and a medium, and the method comprises the steps: collecting and integrating multi-source heterogeneous data, carrying out the synchronous processing of the multi-source heterogeneous data, carrying out the feature extraction of all kinds of data through a deep neural network, and carrying out the feature extraction of the multi-source heterogeneous data; according to the method, high-dimensional feature vector representation in a unified format is generated, deep learning and big data analysis technologies are adopted, high-dimensional feature vector representation is generated, and a personalized health record of a patient is constructed. And according to the personalized health data in the personalized health record and the change trend of the personalized health data, generating and optimizing a re-visit scheme and a treatment suggestion of the patient. And more comprehensive and accurate health state evaluation can be provided for the patient. According to the method, the deep neural network is adopted to perform feature extraction and fusion on the multi-source data, high-dimensional feature vector representation in a unified format is generated, the problem of data integration is effectively solved, and the accuracy of subsequent diagnosis and treatment suggestions is ensured.

Description

Technical Field

[0001] This invention belongs to the field of medical and health technology, specifically relating to an intelligent follow-up diagnosis method, system, terminal, and medium based on dynamic multi-source data fusion and adaptive knowledge generation. Background Technology

[0002] With the increasing number of patients with chronic diseases and the growing demand for follow-up visits, traditional follow-up visit models are facing significant pressure. However, traditional offline follow-up visit models suffer from long waiting times, low efficiency, and high costs, posing a major challenge to both patients and medical institutions. The traditional offline follow-up visit process requires patients to queue at the hospital, consuming a significant amount of time, and the diagnostic process often relies on the doctor's experience, making it difficult to make precise treatment adjustments based on the patient's dynamic health condition.

[0003] With the continuous development of information technology and artificial intelligence, online follow-up consultation systems have gradually become an effective solution to this problem. However, existing online follow-up consultation systems rely on static questionnaires or scales for disease assessment and cannot dynamically track changes in the condition, thus having certain limitations. In addition, existing technologies also face challenges in data integration: although patient data can be collected through various channels such as wearable devices, test reports, and electronic medical records, the heterogeneity and complexity of this data present significant challenges in the fusion and processing of multi-source data. This makes it difficult for many online follow-up consultation systems to fully utilize patients' historical and real-time health data, resulting in lower accuracy and personalization of follow-up consultation recommendations. Summary of the Invention

[0004] This invention addresses the problems in existing technologies by providing an intelligent follow-up consultation method, system, terminal, and medium based on dynamic multi-source data fusion and adaptive knowledge generation. It solves the problem that existing technologies mainly focus on single data processing or prediction models, which cannot effectively handle the real-time fusion of heterogeneous data and health assessment. At the same time, it solves the problem that traditional follow-up consultation systems usually rely on the subjective experience of experts and lack real-time tracking and analysis of the patient's dynamic health status, resulting in a lack of flexibility and personalization in the formulation of treatment plans.

[0005] The technical solution adopted in this invention is as follows: Firstly, this application provides an intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation, which includes the following steps: Step S1: Collect and integrate multi-source heterogeneous data of patients, including static physiological indicators, medical imaging data, laboratory report data, electronic medical records and real-time physiological data; Step S2: Use a deep neural network to extract features from various types of data, generate a high-dimensional feature vector representation in a unified format, and fuse various structured high-dimensional feature vector representations through a multi-level data fusion algorithm to obtain a fused feature vector representation in a unified format. Step S3: Based on the fusion feature vector representation generated in step S2, construct a personalized health record. By integrating big data analysis technology and genomics models, record and update the personalized health data in the patient's personalized health record in real time. The personalized health data includes the fusion feature vector representation at different times. Step S4: Design a hybrid intelligent decision engine based on enhanced generation and retrieval. The hybrid intelligent decision engine generates and optimizes the patient's follow-up visit plan and treatment suggestions based on the personalized health data in the personalized health record and the changing trends of the personalized health data.

[0006] Furthermore, step S2 includes the following steps: Step S2-1: Collect and input multi-source heterogeneous data; Step S2-2: Preprocess the multi-source heterogeneous data, including noise removal, missing value filling, data format unification and data standardization. Step S2-3: Feature extraction is performed using a deep neural network, which includes multiple convolutional layers, recurrent layers, and graph convolutional layers to generate a high-dimensional feature vector representation in a unified format. Step S2-4: The preprocessed high-dimensional feature vector representations are fused using a multi-level data fusion algorithm to obtain a unified format fused feature vector representation. The multi-level data fusion algorithm performs multi-dimensional data alignment based on time series and spatial relationships.

[0007] Furthermore, in steps S2-3, dynamic features are extracted based on time series data, and recurrent neural networks or long short-term memory networks are used to process the temporal information in the data. This information is then combined with the spatial features extracted by the convolutional layer to capture the long-term dependencies of the time series. Spatial relationship modeling is performed on medical imaging data and spatial data. Graph convolutional networks are used to process the spatial structure information in graph data and to fuse spatial data with time series data, preserving the spatiotemporal dependencies in the data. By employing a multi-level feature fusion strategy, temporal features, spatial features, and fused features are integrated to generate a high-dimensional feature vector representation.

[0008] Furthermore, during the fusion process in steps S2-4 by comparing the high-dimensional feature vector representations: For time series, time interpolation, dynamic time warping, or sliding time window methods are used to align and merge time series data of different frequencies into a unified time grid. For spatial relationships, a graph neural network is used to process medical image data and spatial data. The spatial relationships between data are learned through the graph structure, and an attention mechanism is used to perform weighted fusion of different data sources.

[0009] Furthermore, step S3 includes the following steps: Step S3-1: Based on the fusion feature vector representation generated in step S2, construct the patient's personalized health record; Step S3-2: Integrate big data analytics technology, which includes data mining, pattern recognition and prediction models. Based on high-dimensional feature vector representation, it mines potential disease patterns and trends and generates personalized health reports for decision support. Step S3-3: Analyze the patient's genetic data based on a genomics model. The genetic data includes the patient's genotype, mutation information, and pharmacogenomics data to assess the patient's disease risk and personalized treatment plan.

[0010] Furthermore, step S4 includes the following steps: Step S4-1: Through continuous monitoring and feedback of real-time health data, dynamically update personalized health records. The dynamic update mechanism includes collecting and updating patients' real-time data through health monitoring devices and smart terminals. Step S4-2: Continuously learn from the updated personalized health record using a machine learning model. Based on the patient's historical health data and real-time physiological data, dynamically adjust and optimize the health assessment and prediction models in the personalized health record to generate the final follow-up consultation suggestions and treatment plans.

[0011] Furthermore, step S4-2 includes the following steps: Step S4-2-1: Collect and input the patient's historical health data and real-time physiological data; Step S4-2-2: Train historical health data and real-time physiological data using a machine learning model. The machine learning model can be a deep neural network, random forest, or support vector machine. Step S4-2-3: Based on the machine learning model obtained from training, update the patient's personalized health record in real time. The update process includes automatically adjusting the patient's health assessment model, optimizing personalized treatment plans and follow-up visit recommendations. Step S4-2-4: Continuously update and optimize the machine learning model based on feedback from changes in the patient's health, and retrain the model based on new health data; Step S4-2-5: Apply the updated machine learning model to generate a personalized follow-up plan for the patient, based on the patient's health trends, predicted risks, and the latest health data.

[0012] Secondly, this application provides an intelligent follow-up diagnosis system based on dynamic multi-source data fusion and adaptive knowledge generation, the system comprising: Data collection module: used to collect and integrate multi-source heterogeneous data of patients, including static physiological indicators, medical imaging data, laboratory report data, electronic medical records and real-time physiological data; Data processing module: Uses deep neural networks to extract features from various types of data, generates high-dimensional feature vector representations in a unified format, and performs fusion processing on the high-dimensional feature vector representations through multi-level data fusion algorithms; Personalized health record construction module: Used to construct personalized health records for patients based on fusion feature vector representation, and record and update health data in the patient's personalized health record in real time by integrating big data analysis technology and genomics model; Intelligent Decision Engine Module: Based on a hybrid intelligent decision engine that combines enhanced generation and retrieval, the engine generates and optimizes patients' follow-up visit plans and treatment suggestions based on personalized health data and its changing trends in personalized health records. Real-time health data update module: used to collect patients' real-time data through health monitoring devices and smart terminals, and to dynamically update the data to ensure that personalized health records reflect the patient's current health status; Machine learning optimization module: Used to continuously learn and optimize the updated personalized health records based on machine learning models, automatically adjust the patient health assessment model, and optimize personalized treatment plans and follow-up visit suggestions; User interaction module: Used to push updated follow-up consultation plans and treatment suggestions to patients or doctors via smart terminals, providing real-time decision support.

[0013] Thirdly, this application provides a terminal, including: The memory is used to store the intelligent follow-up diagnosis program based on dynamic multi-source data fusion and adaptive knowledge generation. The processor is configured to implement the steps of the intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation as described in the first aspect when executing the intelligent follow-up diagnosis system based on dynamic multi-source data fusion and adaptive knowledge generation.

[0014] Fourthly, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the intelligent re-examination method based on dynamic multi-source data fusion and adaptive knowledge generation as described in the first aspect.

[0015] As can be seen from the above technical solutions, the advantages of the present invention are: (1) This invention integrates multi-source heterogeneous data, including static physiological indicators, medical images, laboratory reports, and electronic medical records, to provide patients with a more comprehensive and accurate assessment of their health status. This invention employs deep neural networks to extract and fuse features from multi-source data, generating a unified high-dimensional feature vector representation, effectively solving the data integration problem and ensuring the accuracy of subsequent diagnostic and treatment recommendations. By dynamically updating patients' personalized health records and using an intelligent decision-making engine in real time, the system can quickly respond to changes in patients' health, significantly improving follow-up visit efficiency and treatment effectiveness.

[0016] (2) This invention significantly improves the integration capability of data from different sources, solving the problem of effective fusion of multi-source data in existing technologies. This method ensures data consistency and accuracy by uniformly formatting high-dimensional feature vector representations (such as static physiological indicators, medical images, and test reports). Furthermore, the use of time-series and spatial alignment techniques, such as Dynamic Time Warping (DTW) and Graph Neural Networks (GNNs), enables the system to effectively handle heterogeneous data in both time and space dimensions, overcoming the limitations of traditional methods in processing large-scale multidimensional data. This innovation greatly improves the quality of feature extraction and the predictive accuracy of subsequent models, enhancing the precision of personalized health assessments and treatment plans, thereby providing doctors and patients with more scientific and timely medical decision support.

[0017] (3) By combining big data analytics and genomics models, personalized health records are generated for patients and updated in real time, making patients' health data more dynamic and accurate. Existing online follow-up consultation systems mostly rely on static data and cannot achieve a comprehensive dynamic assessment of patients' health. This invention uses big data analytics to mine potential health patterns and trends in high-dimensional feature vector representations, providing a more scientific basis for the generation of personalized treatment plans. The introduction of genomics models can provide targeted disease risk assessments based on patients' genetic data, especially in pharmacogenomics, helping doctors to better select suitable drugs and treatment plans for patients. This method not only improves the accuracy and timeliness of personalized health records but also provides patients with tailored and precise treatment plans, which helps improve treatment effectiveness and reduce adverse reactions.

[0018] (4) This invention achieves automatic updates to health records by integrating real-time health data monitoring and intelligent terminal feedback mechanisms, ensuring that the system always reflects the patient's current health status. Continuous learning and optimization of the machine learning model enable the system to continuously adjust the personalized health assessment model based on historical health data and real-time physiological data, generating more accurate treatment suggestions and follow-up plans. This mechanism not only improves the accuracy of follow-up plans but also enhances the adaptability and flexibility of treatment, maximizing patient treatment effectiveness and compliance. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, this invention provides an intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation, including the following steps: In step S1, the system first collects and integrates patient data from various sources, including static physiological indicators, medical imaging data, laboratory report data, electronic medical records, and real-time physiological data. Static physiological indicators are collected by wearable devices and terminals, including data such as the patient's heart rate, blood glucose, blood pressure, body temperature, and weight; medical imaging data includes CT, MRI, and X-ray images, used to assist in disease diagnosis and treatment plan formulation; laboratory report data comes from laboratory tests, such as complete blood count, urinalysis, and liver and kidney function tests; electronic medical records store the patient's historical medical records, medication records, and treatment plans; real-time physiological data is collected through real-time monitoring devices (such as Holter monitoring, smart wristbands, etc.). The system integrates these different types of data and constructs a multi-source data pool to provide comprehensive data support for subsequent analysis and decision-making.

[0023] In step S2, the system classifies the collected multi-source heterogeneous data and selects appropriate deep neural network models for feature extraction based on the data type. For structured physiological parameters and test data, a multilayer perceptron model is used to extract their statistical features; for medical imaging data, such as CT or MRI images, a convolutional neural network is used to extract their spatial hierarchical features; for continuous time series data such as electrocardiograms, a long short-term memory network is used to model their temporal dependencies; and for unstructured text data such as electronic medical records, a Transformer-based semantic representation model is introduced for semantic vectorization. After processing by the corresponding models, the above types of data are transformed into structured high-dimensional feature vector representations. Subsequently, the system performs unified multi-level data fusion processing on the aforementioned feature vectors. Through fusion strategies such as feature dimension alignment, vector concatenation, attention mechanism weighting, and residual connections, it achieves dynamic integration of various features within a unified semantic space. The fused output is a unified formatted fused feature vector, which serves as the input basis for subsequent personalized health record construction and intelligent follow-up consultation decision engine.

[0024] Step S3: Based on the high-dimensional feature vector representation generated in step S2, the system constructs a personalized health record. This record includes the patient's medical history, real-time physiological data, medical imaging data, and genetic data. By integrating big data analytics, the system can process and update the patient's health record in real time, storing historical health data and dynamically updating the data based on the patient's latest health status. Furthermore, the system combines genomics models to analyze the patient's genetic data, including variant detection on whole-exome sequencing (WES) or whole-genome sequencing data to identify key single nucleotide polymorphisms, copy number variations, and structural variations; functional annotation and pathogenicity classification of identified variants using a knowledge graph-based disease-related gene database; and further, using a multi-factor risk scoring model combined with other patient health characteristics to analyze disease susceptibility and classify subtypes, assessing the patient's disease risk and providing personalized treatment plans. The personalized health data includes high-dimensional feature vector representations generated at different times, ensuring that the patient's health record dynamically reflects changes in their health status.

[0025] Step S4 involves designing a hybrid intelligent decision-making engine based on augmented generation and retrieval. This engine generates and optimizes follow-up appointment plans and treatment recommendations for patients based on health data and trends in their personalized health records. Specifically, augmented generation technology helps the engine generate preliminary follow-up appointment plans based on the patient's current health data, historical medical records, and treatment feedback; while the retrieval method utilizes a medical knowledge base, treatment guidelines, and the patient's historical data to further refine the treatment recommendations, ensuring the accuracy and personalization of the plans. Through this hybrid intelligent decision-making engine, the system can not only optimize and adjust the initial recommendations but also dynamically modify follow-up appointment plans and treatment recommendations in real time based on changes in the patient's health data, ensuring that the treatment plan generated at each follow-up appointment is the optimal plan based on the latest health data and trends.

[0026] In some embodiments, step S2 includes the following steps: In step S2-1, the system first collects and inputs multi-source heterogeneous data from the patient, including static physiological indicators, medical imaging data, laboratory report data, electronic medical records, and real-time physiological data. Static physiological indicators come from wearable devices, smart monitoring terminals, and other devices, recording the patient's health indicators such as heart rate, blood glucose, and blood oxygen saturation. Medical imaging data includes images collected by CT, MRI, X-ray, and other equipment to aid in disease diagnosis. Laboratory report data comes from hospital laboratory tests, including various indicators such as complete blood count, urinalysis, and liver and kidney function tests. Electronic medical records record the patient's medical history, medication records, and treatment process. Real-time physiological data comes from dynamic monitoring devices, collecting real-time changes in the patient's physiological indicators. All collected heterogeneous data is uniformly input into the system, providing a foundation for subsequent data processing and decision-making.

[0027] Step S2-2 involves preprocessing the collected multi-source heterogeneous data. The preprocessing includes noise removal, missing value imputation, data format unification, and standardization. Due to significant differences in data format and scale from different devices and sources, the system first denoises the data to remove inaccurate data caused by device errors or external environmental interference. Next, it imputes missing values ​​using strategies such as interpolation or mean imputation. Then, it unifies the data format, converting heterogeneous data from different sources into a unified format to ensure that the data can be processed within the same data structure. Finally, it standardizes the data to allow comparison and analysis of various data types on the same scale. This step ensures data quality and consistency, laying the foundation for subsequent feature extraction and data fusion.

[0028] In steps S2-3, the system employs a deep neural network (DNN) for feature extraction. The DNN includes multiple convolutional layers, recurrent layers, and graph convolutional layers to extract meaningful features from the fused data. Convolutional layers process medical image data, extracting spatial features from the images; recurrent layers (such as LSTM or GRU) process time-series data, capturing temporal dependencies and long-term trends; graph convolutional layers process spatial relationship data, especially medical images and other spatially structured data, uncovering spatial correlations between data. Through this hierarchical deep learning network, the system can effectively extract key features from multidimensional data, generating a unified high-dimensional feature vector representation, providing high-quality input data for subsequent health record construction and intelligent decision engines.

[0029] In steps S2-4, the system uses a multi-level data fusion algorithm to fuse the preprocessed high-dimensional feature vector representations to obtain a fused feature vector representation. During this process, the system synchronously processes and fuses data with different time series and spatial characteristics using a multi-dimensional data alignment method based on time series and spatial relationships. For time series data, methods such as Dynamic Time Warping (DTW) are used for time alignment to ensure that data with different frequencies and time scales are aligned on a unified time grid. For spatial data, Graph Neural Networks (GNNs) are used to learn the spatial relationships between data using graph structures, ensuring consistency of data in the spatial dimension. The fusion algorithm uses methods such as weighted averaging or attention mechanisms to weight and fuse data from different sources, optimizing the contribution of various data types, and finally generating a fused dataset, providing high-quality input for subsequent feature extraction.

[0030] In some embodiments, steps S2-3 involve feature extraction and processing. First, based on feature extraction from time-series data, the system uses a recurrent neural network (RNN) or a long short-term memory network (LSTM) to process the temporal information in the data. Time-series data typically exhibits long-term dependencies, and RNNs and LSTMs can effectively capture dynamic features in time series, addressing the problem of traditional neural networks struggling to maintain long-term memory when processing long sequences. The system trains on the time-series data using these networks to extract dynamic trends and important temporal information from the sequence, which is valuable for subsequent health assessment and prediction. The extracted temporal features are then combined with spatial features extracted by convolutional layers. Convolutional layers (CNNs) are responsible for extracting spatial features from image data, especially when processing medical image data, where they effectively extract local features from images. By combining the dynamic features of time-series data with the local features of spatial data, the system can capture the spatiotemporal dependencies in the data, thereby making the model more expressive and accurate.

[0031] Next, the system models the spatial relationships between medical imaging data and other spatial data, utilizing Graph Convolutional Networks (GCNs) to process the spatial structure information in the graph data. Medical imaging data and other spatial data typically exhibit significant spatial dependencies; the relationships between pixels or voxels determine the overall structure of the image. GCNs can learn the spatial relationships between data points through graph structures, identify the correlations between different data points, and effectively extract spatial features. The system models the spatial information in the images using graph convolutional networks, fusing these spatial features with the dynamic features of time-series data, preserving the spatiotemporal dependencies in the data. This process enables the system to fully utilize information in both spatial and temporal dimensions, thereby providing a more comprehensive analysis and prediction of the patient's health status.

[0032] A multi-level feature fusion strategy enables the effective combination of features from different levels within a unified space, ensuring the complementarity of various features. The system fuses features from different levels—including time-series features, spatial features, and cross-modal fused features—to generate a final high-dimensional feature vector representation. This representation comprehensively and accurately reflects the patient's health status, providing accurate input for subsequent personalized health record construction, decision engines, and follow-up consultation plan generation.

[0033] In some embodiments, steps S2-4 specifically address the alignment of time series and spatial relationship data. First, the system processes the time series data. Since data from different devices or sensors may have different sampling frequencies and inconsistent timestamps, the system uses multiple methods for alignment. For time series data, the system employs time interpolation techniques to supplement data at different time scales, ensuring that data at each time point is presented on a unified time grid. Furthermore, the system uses the Dynamic Time Warping (DTW) algorithm, which measures the similarity between two time series and dynamically adjusts them to ensure effective alignment of the time series data on the time axis. For time series data with significantly different sampling frequencies, the system uses a sliding time window method, cutting the data into fixed time windows and processing and fusing the data features of each time window to solve the problem of data frequency mismatch. The combination of these methods enables time series data of different frequencies to be processed synchronously on a unified time grid, providing accurate data input for subsequent feature extraction and analysis. In spatial data processing, the system uses graph neural network (GNN) technology to process medical image data and other spatial data. Medical image data (such as CT, MRI, X-ray films, etc.) typically exhibit spatial relationships, with spatial dependencies between pixels or voxels. Traditional neural network methods struggle to fully capture this spatial information. Therefore, the system employs graph neural networks to construct graph structures to represent spatial relationships in image data and learn the spatial dependencies between individual pixels. Graph neural networks effectively capture complex relationships between data points, showing significant advantages, particularly for spatial feature extraction in medical images. To further improve the fusion effect of different data sources, the system incorporates an attention mechanism. By assigning different weights to data from different sources, the system optimizes the data fusion process. The attention mechanism allows the system to dynamically evaluate the importance of each data source, adjusting its weights based on its contribution to the prediction results during fusion. This ensures that the final generated high-dimensional feature vector representation comprehensively considers the characteristics of various data sources.

[0034] To enhance the ability to structure and dynamically model multi-source heterogeneous health data of patients, the system introduces a feature generation, alignment, and knowledge generation framework with a clear mathematical modeling foundation in the process of data fusion and intelligent decision-making.

[0035] First, the system will process the patient's multi-source heterogeneous data. The data are input into the corresponding deep learning sub-models for feature extraction. Different data sources (such as static physiological indicators, medical images, test reports, medical records, and real-time physiological data) are mapped to feature vectors in a unified vector space. The features are weighted and fused together using an attention mechanism to obtain a unified high-dimensional feature representation.

[0036] in Attention weights dynamically reflect the importance of each data source in modeling the current health status.

[0037] Considering the typical spatiotemporal coupling characteristics of multi-source health data, the system further introduces a spatiotemporal alignment mechanism to map the aforementioned fused feature F into a representation with time dependence and spatial structure: (Based on time series alignment) (Based on spatial graph structure modeling) Among them, the function Implement time alignment strategies such as Dynamic Time Warping (DTW) and sliding time windows, functions By jointly mapping the above-mentioned temporal and spatial features, an embedded representation of patient health status is constructed:

[0038] This high-dimensional vector representation integrates the dynamic trends, spatial relationships of organs, and historical evolutionary characteristics of multimodal health data, and is used to support the subsequent construction of personalized health records and knowledge generation.

[0039] To achieve real-time dynamic updates of patient health records, the system introduces an incremental update mechanism based on a Bayesian feedback framework to update the health status embedding vector over time steps.

[0040] in This represents the mapping result of the health data collected at the current moment. The learning rate serves as feedback, adjusting the weighting of responses between historical and new states. This mechanism continuously incorporates new health data, enabling automatic evolution and personalized modeling of health records.

[0041] In the process of generating follow-up consultation suggestions, the system further introduces a dual-channel intelligent engine based on variational inference for generation and recognition, deeply optimizing personalized treatment strategies. Its modeling objective is to maximize the lower bound of evidence (ELBO), i.e.:

[0042] Where y represents the candidate follow-up strategy; z is a latent variable representing the patient's potential health status; For model recognition; This is a joint generative model.

[0043] In some embodiments, step S3 includes the following steps: Step S3-1 constructs a personalized health record for the patient based on the fused feature vector representation generated in step S2. This record contains comprehensive health information about the patient, including but not limited to historical medical records, real-time physiological data, medical imaging data, laboratory reports, and genomic data. The system integrates this data from different sources to generate a unified health record, where each health data item is timestamped to ensure the system can dynamically update according to changes in the patient's health status. The personalized health record not only records the patient's basic health information but also continuously tracks the patient's health trends, providing a basis for subsequent health management and treatment decisions.

[0044] In step S3-2, the system integrates big data analytics, primarily including data mining, pattern recognition, and predictive models. By analyzing high-dimensional feature vector representations, it uncovers potential disease patterns and health trends. Through big data analytics, the system can extract regular information from patients' health records, discover potential correlations between different health indicators, and thus identify possible disease risks. Pattern recognition algorithms can identify patterns of change in patients' health status based on historical and real-time data, predict future health trends, and generate personalized health reports as a reference for doctors' decision-making. These personalized health reports not only help doctors develop treatment plans but also help patients understand their own health status and take effective preventative and intervention measures.

[0045] In step S3-3, the system performs in-depth analysis of the patient's genetic data using a genomics model. This genetic data includes the patient's genotype, mutation information, and pharmacogenomics data, which reveal the patient's responsiveness and risks under different treatment regimens. Supported by the genomics model, the system can assess the patient's potential disease risk based on their genetic characteristics and provide personalized treatment plans based on genotype. For example, gene mutation information can help determine a patient's response to a specific drug, avoiding side effects or poor treatment outcomes caused by unsuitable medications. Pharmacogenomic analysis ensures that patients receive personalized drug dosages and drug selections during medication, thereby maximizing treatment effectiveness and minimizing adverse reactions. By combining genomic data, the system can not only assess the patient's health risks but also provide precise, personalized treatment plans, thus achieving true precision medicine.

[0046] In some embodiments, step S4 includes the following steps: Step S4-1 dynamically updates personalized health records through continuous monitoring and feedback of real-time health data. Specifically, the system collects real-time physiological data from patients through smart health monitoring devices (such as smartwatches, blood glucose meters, and dynamic electrocardiogram monitoring devices) and smart terminal devices (such as mobile phones and smart health management apps). This data includes key health indicators such as heart rate, blood pressure, blood glucose, body temperature, and weight. The system tracks changes in the patient's health status in real time and updates the personalized health record promptly based on the new health data. Each update incorporates new health data and trends into the patient's record, ensuring that the record reflects the patient's latest health status. This dynamic update mechanism keeps the patient's health record up-to-date, enabling doctors and patients to make more timely and accurate medical decisions.

[0047] In step S4-2, the system uses a machine learning model to continuously learn from the updated personalized health record. Based on the patient's historical health data and real-time physiological data, the machine learning model is trained and optimized through algorithms, dynamically adjusting and refining the health assessment and prediction model. After each health data update, the model reassesses the patient's health based on the latest data, identifying potential health risks and trends, and automatically generating corresponding follow-up recommendations and treatment plans based on this information. By continuously learning from changes in the patient's health data, the machine learning model can improve the accuracy of assessments and the personalization of treatment plans. This system not only provides real-time feedback on the patient's health status but also adjusts treatment plans based on a combination of historical and real-time data, ensuring optimal treatment outcomes. Each updated recommendation and treatment plan reflects the patient's current health status and future health trends, making follow-up plans more accurate and personalized, greatly improving patient treatment outcomes and the quality of medical services.

[0048] In some embodiments, step S4-2 includes the following steps: Step S4-2-1 first collects and inputs the patient's historical health data and real-time physiological data. The system obtains the patient's historical health data, including diagnostic records, treatment history, and medication usage, through interfaces with the hospital information system, electronic medical record system, and intelligent health monitoring devices. Simultaneously, it collects the patient's real-time physiological data, such as heart rate, blood pressure, blood sugar, and body temperature, through wearable devices and smart terminals. The system transmits this data to the backend in real time and inputs it into the personalized health record, providing a comprehensive information foundation for subsequent data processing and analysis.

[0049] In step S4-2-2, the system trains a machine learning model on historical health data and real-time physiological data. To improve the accuracy and predictive ability of data processing, the system uses machine learning algorithms such as deep neural networks (DNN), random forests (RF), or support vector machines (SVM). Through these algorithms, the system can extract valuable features from patient health data and establish relationship models between the data. Deep neural networks can capture complex nonlinear relationships, random forests improve predictive stability and accuracy through ensemble learning methods, and support vector machines can handle small-sample learning tasks, helping to provide effective predictions even when health data is incomplete or noisy.

[0050] In step S4-2-3, based on the trained machine learning model, the system updates the patient's personalized health record in real time. Whenever new health data is collected and input into the system, the machine learning model automatically adjusts the patient's health assessment model according to the latest input data, optimizing personalized treatment plans and follow-up recommendations. Specifically, the model analyzes the patient's current health status and compares it with historical health data to identify trends in health changes and potential risk points. The system then automatically generates follow-up plans based on these analysis results and adjusts treatment recommendations to ensure the plans are personalized and targeted.

[0051] In step S4-2-4, as the patient's health status changes, the system continuously receives health feedback information. This feedback is transmitted to the system in real time via smart devices or health reports submitted by the patient, triggering the optimization and updating of the machine learning model. Using new health data, the system retrains the existing model to adapt to changes in the patient's health condition. After each model update, the system reassesses the patient's health risks and dynamically adjusts treatment plans and follow-up recommendations to ensure the patient receives the most appropriate medical support during treatment.

[0052] Step S4-2-5 involves applying the updated machine learning model to generate a personalized follow-up appointment plan for the patient. After each change in health status, the machine learning model comprehensively analyzes the patient's health trends, predicted risks, and the latest health data to provide a personalized follow-up appointment plan. This plan considers not only the current health status but also the patient's historical health data and disease progression trends, thereby generating targeted treatment plans and follow-up appointment recommendations. In this way, the follow-up appointment plan accurately reflects the patient's individual needs, providing doctors with scientific decision support and improving patient treatment effectiveness and satisfaction.

[0053] In some embodiments, this application provides an intelligent follow-up diagnosis system based on dynamic multi-source data fusion and adaptive knowledge generation. The functional modules of the system are designed as follows: First, the data collection module is used to collect and integrate multi-source heterogeneous patient data. The system automatically collects patients' static physiological indicators, medical imaging data, laboratory report data, electronic medical records, and real-time physiological data by connecting with various health monitoring devices, hospital information systems, and smart terminals. Static physiological indicators, such as heart rate, blood glucose, and blood oxygen, are monitored through wearable devices and smart terminals; medical imaging data comes from hospital imaging examinations, such as CT scans, MRI scans, and X-rays; laboratory report data is provided by the hospital laboratory and includes routine blood, urine, and liver and kidney function indicators; electronic medical records store patients' historical medical records, medical history, and treatment information; and real-time physiological data comes from the health monitoring devices worn by the patient and is collected in real-time through smart terminals. This multi-source heterogeneous data will be integrated and transmitted to the backend for further processing.

[0054] Next, the data processing module is responsible for synchronously processing the aforementioned high-dimensional feature vector representations using a multi-level data fusion algorithm. The system first preprocesses the data to ensure consistency and quality. Then, a deep neural network (DNN) is used to extract features from the processed data, mining complex patterns within the data through different levels of network structure to generate high-dimensional feature vector representations in a unified format. These high-dimensional features can cover multiple dimensions of patient health, providing a foundation for subsequent data analysis and the construction of personalized health records.

[0055] Based on the fused feature vector representation generated in step S2, the personalized health record construction module integrates big data analytics and genomics models to build and update patients' personalized health records in real time. The system first integrates data from different sources into a comprehensive personalized record, recording the patient's health data, medical history, treatment records, and genomic data. Big data analytics is used to mine potential disease patterns and health trends in the patient's health data in real time, while the genomics model analyzes the patient's genetic data to help assess disease risk and provide personalized treatment recommendations. Whenever the patient's health condition changes, the health record is automatically updated to ensure accuracy and timeliness.

[0056] The intelligent decision engine module operates based on a hybrid intelligent decision engine that combines augmented generation and retrieval. This engine first generates preliminary follow-up consultation plans and treatment suggestions by retrieving relevant medical knowledge bases, treatment guidelines, and patient medical records. Then, using augmented generation technology, it further optimizes the follow-up consultation plan based on the patient's health data and trends. Each time health data is updated, the system automatically adjusts the treatment plan based on the new health status and feedback information, ensuring the personalization, accuracy, and timeliness of the follow-up consultation plan and treatment suggestions.

[0057] To ensure the timeliness of personalized health records, a real-time health data update module continuously collects patients' real-time data through health monitoring devices and smart terminals. This module dynamically inputs this data into the system, automatically updating patients' health records to ensure that the records always reflect the patient's current health status. Whether through hospital monitoring equipment or patients' smart terminals, all real-time health data is synchronized to the personalized health record, supporting the decision engine in generating more accurate treatment recommendations.

[0058] Building upon this foundation, the machine learning optimization module continuously learns and optimizes the updated personalized health records. Based on historical health data and real-time physiological data, the system continuously optimizes the health assessment model through machine learning and automatically adjusts follow-up consultation recommendations and treatment plans. Each time a patient's health data changes, the machine learning model is retrained based on the new data, and the health assessment and treatment plans are adjusted to ensure the system can adapt to real-time changes in the patient's health condition and provide the most accurate personalized medical services.

[0059] Finally, the user interaction module pushes updated follow-up consultation plans and treatment suggestions to patients or doctors via smart terminals, providing real-time decision support. Whether through a mobile app, smartwatch, or the medical platform used by doctors, patients and doctors can view updated treatment plans in real time, facilitating more timely medical decisions. Through this module, the system ensures the efficient delivery and execution of treatment plans and follow-up consultation suggestions, enhances interaction between patients and doctors, and improves the overall efficiency of medical services.

[0060] In some embodiments, this application provides a terminal, including: The memory is used to store the intelligent follow-up diagnosis program based on dynamic multi-source data fusion and adaptive knowledge generation. A processor is used to implement the steps of the intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation when executing the intelligent follow-up diagnosis system based on dynamic multi-source data fusion and adaptive knowledge generation.

[0061] In some embodiments, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the intelligent re-examination method based on dynamic multi-source data fusion and adaptive knowledge generation.

[0062] It is understood that the systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can be a personal computer, a laptop computer, a personal digital assistant, a tablet computer, a wearable device, or any combination of these devices.

[0063] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0064] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0065] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0066] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0067] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0068] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0069] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this specification. The singular forms “a,” “described,” and “the” used in one or more embodiments of this specification and in the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0070] It should be understood that although the terms first, second, third, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of one or more embodiments of this specification, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "in response to a determination," or "when," or "in the event of a determination."

[0071] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. An intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation, characterized in that, Includes the following steps: Step S1: Collect and integrate multi-source heterogeneous data of patients, including static physiological indicators, medical imaging data, laboratory report data, electronic medical records and real-time physiological data; Step S2: Use a deep neural network to extract features from various types of data, generate a high-dimensional feature vector representation in a unified format, and fuse various structured high-dimensional feature vector representations through a multi-level data fusion algorithm to obtain a fused feature vector representation in a unified format. Step S3: Based on the fusion feature vector representation generated in step S2, construct a personalized health record. By integrating big data analysis technology and genomics models, record and update the personalized health data in the patient's personalized health record in real time. The personalized health data includes the fusion feature vector representation at different times. Step S4: Design a hybrid intelligent decision engine based on enhanced generation and retrieval. The hybrid intelligent decision engine generates and optimizes the patient's follow-up visit plan and treatment suggestions based on the personalized health data in the personalized health record and the changing trends of the personalized health data.

2. The intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation according to claim 1, characterized in that, Step S2 includes the following steps: Step S2-1: Collect and input multi-source heterogeneous data; Step S2-2: Preprocess the multi-source heterogeneous data, including noise removal, missing value filling, data format unification and data standardization. Step S2-3: Feature extraction is performed using a deep neural network, which includes multiple convolutional layers, recurrent layers, and graph convolutional layers to generate a high-dimensional feature vector representation in a unified format. Step S2-4: The preprocessed high-dimensional feature vector representations are fused using a multi-level data fusion algorithm to obtain a unified format fused feature vector representation. The multi-level data fusion algorithm performs multi-dimensional data alignment based on time series and spatial relationships.

3. The intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation according to claim 2, characterized in that, In steps S2-3, dynamic features are extracted based on time series data. Recurrent neural networks or long short-term memory networks are used to process the temporal information in the data and combine it with the spatial features extracted by the convolutional layer to capture the long-term dependencies of the time series. Spatial relationship modeling is performed on medical imaging data and spatial data. Graph convolutional networks are used to process the spatial structure information in graph data and to fuse spatial data with time series data, preserving the spatiotemporal dependencies in the data. By employing a multi-level feature fusion strategy, temporal features, spatial features, and fused features are integrated to generate a high-dimensional feature vector representation.

4. The intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation according to claim 3, characterized in that, When fusing the high-dimensional feature vector representations in steps S2-4: For time series, time interpolation, dynamic time warping, or sliding time window methods are used to align and merge time series data of different frequencies into a unified time grid. For spatial relationships, a graph neural network is used to process medical image data and spatial data. The spatial relationships between data are learned through the graph structure, and an attention mechanism is used to perform weighted fusion of different data sources.

5. The intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation according to claim 1, characterized in that, Step S3 includes the following steps: Step S3-1: Based on the fusion feature vector representation generated in step S2, construct the patient's personalized health record; Step S3-2: Integrate big data analytics technology, which includes data mining, pattern recognition and prediction models. Based on high-dimensional feature vector representation, it mines potential disease patterns and trends and generates personalized health reports for decision support. Step S3-3: Analyze the patient's genetic data based on a genomics model. The genetic data includes the patient's genotype, mutation information, and pharmacogenomics data to assess the patient's disease risk and personalized treatment plan.

6. The intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation according to claim 5, characterized in that, Step S4 includes the following steps: Step S4-1: Through continuous monitoring and feedback of real-time health data, dynamically update personalized health records. The dynamic update mechanism includes collecting and updating patients' real-time data through health monitoring devices and smart terminals. Step S4-2: Continuously learn from the updated personalized health record using a machine learning model. Based on the patient's historical health data and real-time physiological data, dynamically adjust and optimize the health assessment and prediction models in the personalized health record to generate the final follow-up consultation suggestions and treatment plans.

7. The intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation according to claim 6, characterized in that, Step S4-2 includes the following steps: Step S4-2-1, Collect and input the patient's historical health data and real-time physiological data; Step S4-2-2: Train historical health data and real-time physiological data using a machine learning model. The machine learning model can be a deep neural network, random forest, or support vector machine. Step S4-2-3: Based on the machine learning model obtained from training, update the patient's personalized health record in real time. The update process includes automatically adjusting the patient's health assessment model, optimizing personalized treatment plans and follow-up visit recommendations. Step S4-2-4: Continuously update and optimize the machine learning model based on feedback from changes in the patient's health, and retrain the model based on new health data; Step S4-2-5: Apply the updated machine learning model to generate a personalized follow-up plan for the patient, based on the patient's health trends, predicted risks, and the latest health data.

8. An intelligent follow-up diagnosis system based on dynamic multi-source data fusion and adaptive knowledge generation, characterized in that, The system includes: Data collection module: used to collect and integrate multi-source heterogeneous data of patients, including static physiological indicators, medical imaging data, laboratory report data, electronic medical records and real-time physiological data; Data processing module: Uses deep neural networks to extract features from various types of data, generates high-dimensional feature vector representations in a unified format, and performs fusion processing on the high-dimensional feature vector representations through multi-level data fusion algorithms; Personalized health record construction module: Used to construct personalized health records for patients based on fusion feature vector representation, and record and update health data in the patient's personalized health record in real time by integrating big data analysis technology and genomics model; Intelligent Decision Engine Module: Based on a hybrid intelligent decision engine that combines enhanced generation and retrieval, the engine generates and optimizes follow-up visit plans and treatment suggestions for patients based on personalized health data and its changing trends in their personalized health records. Real-time health data update module: used to collect patients' real-time data through health monitoring devices and smart terminals, and to update it dynamically to ensure that personalized health records reflect the patient's current health status; Machine learning optimization module: Used to continuously learn and optimize the updated personalized health records based on machine learning models, automatically adjust the patient health assessment model, and optimize personalized treatment plans and follow-up visit suggestions; User interaction module: Used to push updated follow-up consultation plans and treatment suggestions to patients or doctors via smart terminals, providing real-time decision support.

9. A terminal, characterized in that, include: The memory is used to store the intelligent follow-up diagnosis program based on dynamic multi-source data fusion and adaptive knowledge generation. The processor is configured to implement the steps of the intelligent follow-up diagnosis method based on dynamic multi-source data fusion and adaptive knowledge generation as described in claim 1 when executing the intelligent follow-up diagnosis system based on dynamic multi-source data fusion and adaptive knowledge generation.

10. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions. When the computer reads the computer instructions from the storage medium, the computer executes the intelligent re-examination method based on dynamic multi-source data fusion and adaptive knowledge generation as described in claim 1.

Images