A lupus disease intelligent monitoring system based on data fusion and its application in a smartwatch

Abstract

This invention discloses a data fusion-based intelligent lupus monitoring system and its application in a smartwatch, covering the field of lupus monitoring. The system includes modules for data acquisition, preprocessing, fusion, disease assessment, early warning, storage and management, user interaction, model updating, personalized analysis, and telemedicine. Data is first collected from a smartwatch and medical devices, then denoised, normalized, and features extracted. A deep belief network is used to fuse the data to derive comprehensive health indicators. An LSTM and attention mechanism model is used to assess the disease condition. Threshold-based early warnings are set based on the assessment results and trend predictions. Distributed hash tables and blockchain are used to store and manage the data. A multi-terminal interface is developed, and the model is updated using federated learning and transfer learning. Lupus disease is assessed based on multiple datasets, and 5G communication technology enables remote doctor-patient interaction. This invention monitors the disease condition in real time, provides timely warnings of abnormalities, offers accurate disease assessment, optimizes the allocation of medical resources, promotes the development of telemedicine, and contributes to lupus research.

Description

Technical Field

[0001] This invention relates to the field of lupus disease monitoring technology, and in particular to a lupus disease intelligent monitoring system based on data fusion and its application in watches. Background Technology

[0002] Systemic lupus erythematosus (SLE) is a complex autoimmune disease characterized by frequent relapses and significant individual variability. Accurate monitoring of lupus disease progression is crucial for developing appropriate treatment plans and improving patient prognosis. However, current lupus disease monitoring faces numerous challenges.

[0003] In terms of data collection, traditional monitoring mainly relies on regular hospital checkups, resulting in limited data types and long time intervals between collections. While blood and urine tests can obtain some key indicators, they cannot reflect real-time changes in the condition. Data on the patient's daily physiological state and living environment, which significantly impact the condition, is difficult to collect effectively. Although wearable devices are becoming increasingly common, their functionality for lupus monitoring is insufficient, failing to accurately collect physiological parameters closely related to lupus. Furthermore, data compatibility between different devices is poor, making integration and utilization difficult.

[0004] The disease assessment process is also fraught with problems. Doctors often rely on experience and limited test results to make subjective judgments about the condition, lacking objective and quantifiable assessment standards. The existing disease assessment indicator system is inadequate and cannot comprehensively cover the factors affecting lupus. Moreover, due to the lack of long-term, continuous disease data, it is difficult to accurately grasp the disease's progression, which can easily lead to misdiagnosis or untimely adjustments to treatment plans, affecting the patient's treatment outcomes and quality of life.

[0005] With technological advancements, data fusion, artificial intelligence, and other technologies have offered possibilities for solving these problems. However, research on applying these technologies to lupus disease monitoring is still in the exploratory stage, lacking mature systems and solutions. There is an urgent market need for an intelligent system capable of comprehensively, in real-time, and accurately monitoring lupus symptoms, and its application in smartwatches, to meet clinical needs and improve healthcare services for lupus patients. Summary of the Invention

[0006] The present invention proposes an intelligent lupus disease monitoring system based on data fusion to solve the problems mentioned in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent lupus disease monitoring system based on data fusion, comprising:

[0008] Multi-source data acquisition module: Collects physiological data with the help of a smartwatch, obtains key indicator data by connecting to external devices via Bluetooth or Wi-Fi, and obtains symptom self-assessment data from patient input and electronic medical record system synchronously, using new microelectromechanical technology;

[0009] Data preprocessing module: Adaptive filtering algorithm is used to remove noise, improved Z-score formula is used to normalize data of different dimensions, and physiological signal features are automatically extracted using a convolutional neural network (CNN) feature extraction model.

[0010] Data fusion module: Construct a data fusion model based on deep belief network (DBN). Preprocessed data is used as input, and feature vectors FV are obtained by RBM feature learning. The comprehensive health index HI is calculated using a formula, and the weights are optimized by the backpropagation algorithm.

[0011] Disease assessment module: A learning model based on a combination of Long Short-Term Memory (LSTM) network and attention mechanism is used to learn the long-term dependence of disease conditions, focusing on key data features, outputting the probability distribution of disease conditions and classifying them into three levels, and using the cross-entropy loss function to train on labeled data;

[0012] Early warning module: Based on the probability distribution of disease assessment, data change trends and individual characteristics, dynamic early warning thresholds are set. A gray prediction model is used to predict the disease condition. When the threshold is reached, an early warning is issued through multiple methods, along with the disease condition and treatment suggestions.

[0013] Data storage and management module: Utilizes Distributed Hash Table (DHT) technology to build a distributed database, employs homomorphic encryption algorithms during storage, and provides a blockchain-based access control mechanism to access and manipulate data based on permission information recorded on the blockchain.

[0014] User interaction module: Develop a unified interface that adapts to smartwatches, mobile apps and computers, adopt adaptive layout technology to automatically adjust the displayed content according to the device screen size and resolution, and introduce emotional interaction design;

[0015] Model update module: It adopts a federated learning framework to train the disease assessment model with local data while protecting privacy. The model parameters are exchanged and updated through encryption technology, and the model is fine-tuned by combining transfer learning algorithms.

[0016] Furthermore, it also includes:

[0017] Personalized analysis module: Based on patient genetic data (GD), it searches for gene markers related to lupus disease through a gene-disease association database and constructs a gene risk score (GRS) formula. w gi It is the weight of gene marker i, g i The genotype score of gene marker i; combined with lifestyle data such as diet frequency (DF) and exercise intensity (EI), the weights of personalized assessment indicators are determined using the analytic hierarchy process (AHP) to correct the disease assessment results.

[0018] Furthermore, it also includes:

[0019] The telemedicine module uses 5G communication technology combined with video stream encoding and compression technology to support real-time remote consultation. The system allows users to view various monitoring data and high-definition videos of patients in real time, and uses virtual annotation technology to annotate and write on patient body parts or test images. It also introduces remote operation assistance technology to guide medical operations remotely.

[0020] Furthermore, the multi-source data acquisition module adds environmental data acquisition capabilities, collecting data on ultraviolet (UVI) intensity, allergen concentration (AC), and noise level (NL) through miniature environmental sensors; it also integrates environmental data with other physiological and medical data using an improved Bayesian fusion formula. H indicates the disease status, E i Represents various types of data;

[0021] The formula for calculating the integrated health index is as follows: w i FV represents the weights of each dimension of the fused feature vector.

[0022] Furthermore, the improved Z-score normalization formula in the data preprocessing module. μ s and σ s These are the mean and standard deviation of the data within a specific time period, respectively, and k and b are coefficients adjusted according to the characteristics of the data;

[0023] Generative Adversarial Networks (GANs) are used to repair missing data. The generator learns the distribution of real data to generate missing data values, the discriminator distinguishes between generated data and real data, and wavelet transform is used to perform multi-scale analysis on the data to extract features of different frequency components.

[0024] Furthermore, the cross-entropy loss function is used in the disease assessment module. y j These are actual disease severity labels, p j It is the model's predicted probability;

[0025] A multimodal fusion Transformer model was used to assess the condition. Image data was processed by extracting features through a convolutional neural network, text data was converted into vector representation using a word vector model, and numerical data was normalized before being input into the Transformer model. The multi-head attention mechanism was used to fuse and analyze the data from different modalities.

[0026] Furthermore, the early warning module uses the grey prediction model GM(1,1) to predict the trend of changes in disease indicators, with the formula as follows: x (0) Given the original data sequence, x (1) The sequence is accumulated and generated, where a and b are model parameters;

[0027] An algorithm combining fuzzy logic and reinforcement learning is used to determine the warning threshold. Fuzzy logic is used to process the disease assessment results, data change trends and individual patient characteristics to construct a fuzzy rule base. Reinforcement learning algorithm is used to optimize the fuzzy rule base and warning threshold by using the actual changes in the patient's condition as feedback.

[0028] Furthermore, the data storage and management module uses a combination of blockchain and the InterPlanetary File System (IPFS) to store data. The blockchain stores data metadata and hash values, while IPFS stores the actual data files. Data visualization tools are developed to display data statistical analysis results through interactive charts.

[0029] Furthermore, the user interaction module uses brain-computer interface (BCI) technology to achieve interaction, allowing users to operate the smartwatch by thinking and viewing data or issuing commands, and using voice emotion recognition technology to analyze voice emotions.

[0030] Furthermore, the watch integrates physiological and some environmental data collection functions, uses Bluetooth technology to transmit data with external medical testing equipment and mobile phones, has a built-in microprocessor to preprocess some data in real time, and alerts patients through vibration, sound and light signals. It provides a simple interactive interface, allowing users to view medical data and receive warning information through touch operation.

[0031] Compared with existing technologies, the beneficial effects of this invention are:

[0032] From the patient's perspective, this greatly improves the convenience of disease monitoring. The smartwatch integrates multiple functions, allowing patients to collect data anytime, anywhere, eliminating the need for frequent hospital visits. The system monitors the condition in real time, providing timely alerts upon detecting abnormalities, enabling patients to take immediate action, enhancing their self-management capabilities and reducing psychological burden. For example, patients can be aware of their health status in daily life and respond promptly to sudden discomfort.

[0033] In terms of disease assessment and treatment, the system provides more accurate data. Multi-source data fusion and advanced algorithms comprehensively and objectively assess the condition, avoiding the limitations of doctors relying solely on experience. Doctors can more accurately grasp the changing trends of the disease, adjust treatment plans in a timely manner, and improve treatment outcomes. For patients with complex conditions, the comprehensive data integrated by the system helps doctors make quick and accurate diagnoses, avoiding misdiagnosis and delays in treatment.

[0034] From the perspective of medical resource utilization, this system optimizes resource allocation. The telemedicine module enables patients and doctors to communicate remotely, reducing unnecessary hospital visits and alleviating hospital pressure. Data storage and management functions facilitate medical data sharing, promote collaboration among medical institutions, and improve the overall level of medical services. Simultaneously, the large amount of patient data continuously collected by the system provides rich material for lupus research, contributing to a deeper understanding of the disease mechanism and driving the development of new diagnostic methods and treatments. Attached Figure Description

[0035] Figure 1 This is a schematic block diagram of a lupus disease intelligent monitoring system based on data fusion proposed in this invention;

[0036] Figure 2 A comparative diagram illustrating the timeliness of patient condition monitoring;

[0037] Figure 3 A diagram comparing the accuracy of doctors' diagnoses of patients' conditions;

[0038] Figure 4 A diagram illustrating the distribution of patient access to medical care scores. Detailed Implementation

[0039] 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.

[0040] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0041] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0042] Reference Figures 1 to 4 A lupus disease intelligent monitoring system based on data fusion, comprising:

[0043] Multi-source data acquisition module: For physiological data acquisition, the core device is a smartwatch integrating multiple high-precision sensors. The watch's built-in heart rate sensor uses photoelectric transmission measurement, illuminating the skin with specific wavelengths of light and accurately capturing changes in blood volume caused by heartbeats based on the periodic changes in blood absorption of light, thus obtaining real-time and accurate heart rate (HR) data. Blood pressure measurement utilizes a novel oscilloscope sensor that can sensitively detect vibration waves in arterial walls, combined with complex algorithms to dynamically monitor blood pressure (BP) values. The skin temperature (ST) sensor uses infrared sensing technology, quickly and accurately measuring body surface temperature without skin contact, capturing subtle temperature changes. Step count (SS) relies on a high-precision accelerometer, accurately counting each step by recognizing information such as wrist movement acceleration and direction. When connecting to external medical monitoring devices, stable and fast data transmission is achieved using Bluetooth Low Energy 5.0 or high-speed Wi-Fi 6 technology. Connecting to devices that detect the concentration of anti-double-stranded DNA antibodies (dsDNA) in the blood, this key autoimmune indicator can be acquired in real time; communicating with instruments that detect erythrocyte sedimentation rate (ESR) allows for the monitoring of information related to inflammatory responses. Furthermore, the module supports patients actively inputting self-assessment data on symptoms, such as fatigue level (FS) and joint pain level (JPS). Patients can record their feelings at any time through a simple and user-friendly interface. Simultaneously, through standardized interfaces and deep integration with electronic medical record systems, it automatically synchronizes information such as medical history (DHS) and medication records (MR). The sensors on the smartwatch utilize novel microelectromechanical systems (MEMS) technology. By optimizing the sensor's microstructure design, such as reducing the size of sensitive elements and increasing integration, the accuracy and stability of data acquisition are greatly improved, providing a solid data foundation for subsequent medical analysis and diagnosis.

[0044] Data preprocessing module: In the denoising stage, an adaptive filtering algorithm is used. This algorithm is highly intelligent and can dynamically adjust the filtering parameters based on the real-time characteristics of the acquired physiological signals (such as heart rate and blood pressure). When motion artifacts or electromagnetic interference generate noise, the adaptive filter will sensitively capture abnormal fluctuations in the signal. By analyzing the noise frequency, amplitude, and other characteristics, a matched filtering strategy is adopted to accurately remove noise from the original physiological signal, ensuring the purity of the signal and providing reliable basic data for subsequent analysis. For data normalization processing of different dimensions, an improved Z-score normalization formula is used. In this formula, μ s and σ s These represent the mean and standard deviation of data within a specific time period, respectively. They are key statistics for measuring the central tendency and dispersion of data. The coefficients k and b are not fixed values ​​but are finely adjusted based on the distribution characteristics and range of variation of different types of physiological data. For example, the values ​​of k and b will differ for data on movement steps with large fluctuations and relatively stable skin temperature data, thus ensuring comparability of various data after normalization and facilitating subsequent unified analysis. In terms of feature extraction, a feature extraction model based on convolutional neural networks (CNNs) is used. The convolutional layers in the CNN model slide convolution on the physiological signal data through different convolutional kernels, automatically mining the characteristics of heart rate variability in the time domain (intervals between adjacent heartbeats) and the distribution characteristics of different frequency components in the frequency domain. For blood pressure fluctuation data, the model can accurately capture its periodic changes, identifying the rising and falling phases of blood pressure within a cycle, as well as the amplitude of fluctuations. This deep learning-based feature extraction method is more efficient and accurate than traditional methods, capable of uncovering subtle features that are difficult for humans to perceive, providing rich and valuable information for medical applications such as disease diagnosis and health assessment.

[0045] Data Fusion Module: Utilizing advanced Deep Belief Network (DBN) technology, various physiological and medical data are organically integrated. First, a data fusion model based on DBN is constructed. The DBN consists of multiple layers of Restricted Boltzmann Machines (RBMs). A Restricted Boltzmann Machine is an energy-based model with a unique structure of no connections within layers and fully connected connections between layers. During data fusion, pre-processed data such as heart rate, blood pressure, blood indicators, and symptom self-assessment are input into the DBN. Each RBM layer undertakes an important feature learning task. During the learning process, the RBM captures latent patterns and features in the data by adjusting the connection weights between neurons. For example, for heart rate variability data, the RBM mines features such as periodicity and trends in its time series; for blood pressure data, it learns features such as fluctuation amplitude and frequency of change. Features learned by one layer of RBM serve as input for the next layer. As the number of layers increases, the model can abstract and refine the data from different levels and perspectives, achieving progressive learning from low-level features to high-level comprehensive features. After multiple layers of RBM processing, the final fused feature vector FV is obtained. To quantify an individual's overall health status, a formula for calculating the fused comprehensive health index is introduced. Among them, FV i It integrates the various dimensions of the feature vector (FV), representing key health features extracted from different data; w i The weights for each dimension are not arbitrarily set, but optimized using the backpropagation algorithm. Based on the gradient descent principle, the backpropagation algorithm continuously adjusts the weight parameters, guided by the error between the model's predicted and actual values, so that the comprehensive health indicators can more accurately reflect an individual's true health status. Through this series of complex and sophisticated operations, the data fusion module achieves deep integration and value enhancement of diverse data, providing a more comprehensive and accurate basis for medical decision-making.

[0046] The disease assessment module employs a deep learning model combining Long Short-Term Memory (LSTM) networks with an attention mechanism. LSTM is a special type of Recurrent Neural Network (RNN) unique in that it possesses memory units and gating structures. The memory units store information, acting like an "information warehouse" that can store key information from the disease data for extended periods. The input gate, forget gate, and output gate act like "smart valves," controlling the inflow, outflow, and retention of information. When processing disease data, such as changes in patient symptoms over time and fluctuations in vital signs—data with time-series characteristics—LSTM networks can effectively learn long-term dependencies in the disease data through these gating mechanisms, avoiding the gradient vanishing or exploding problems common in traditional RNNs. This allows for accurate capture of the disease's evolution over a longer time span. The integration of the attention mechanism makes the model more "intelligent." In massive amounts of disease data, not all data is equally important. The attention mechanism enables the model to automatically focus on key data features. For example, when assessing a heart patient's condition, the model will pay more attention to recent abnormal fluctuations in electrocardiograms and rapid changes in blood pressure, while relatively downplaying some routine and stable physiological indicators. In this way, the model can extract the most valuable parts from complex data, improving the accuracy of the assessment. The model's output is a probability distribution of disease severity, classifying the disease into three levels: remission (RP), stable (SP), and active (AP). To ensure accurate predictions, the model needs to be trained on a large amount of labeled historical data. During training, the cross-entropy loss function is used. Among them, y j These are labels indicating the actual severity of the illness, representing the actual state of the condition; p j This is the probability predicted by the model. By continuously adjusting its parameters and minimizing the loss function value, the model aims to make the predicted probability as close as possible to the true label, thereby improving its performance on disease assessment tasks and providing reliable reference for doctors to formulate treatment plans and predict patient prognosis.

[0047] Early Warning Module: The early warning threshold setting abandons the traditional fixed threshold mode and adopts a dynamic threshold. First, based on the probability distribution of disease severity output by the disease assessment module, the likelihood and trend of the disease being at different levels are analyzed. For example, if the probability of the disease being in the active phase (AP) gradually increases, the early warning threshold needs to be adjusted accordingly. Simultaneously, the trends of various physiological data, such as the fluctuation amplitude and rate of heart rate and blood pressure over a period of time, are considered. Furthermore, individual patient characteristics cannot be ignored; factors such as age, underlying medical history, and genetic factors all affect the setting of the early warning threshold. The early warning threshold will inevitably differ between young patients in good health and elderly patients with multiple chronic diseases. For predicting the trend of disease indicator changes, a grey prediction model GM(1,1) is used, with the formula: This model has a unique advantage in making effective predictions based on a small amount of data. Original data sequence x (0) It records historical data related to the disease, and obtains x through an accumulation operation. (1) The sequence operation weakens the randomness of the original data, revealing potential regularities. Parameters 'a' and 'b' in the model are determined by a specific algorithm, which dictates the shape and trend of the prediction curve. This model can predict changes in disease indicators over a future period, such as whether blood pressure will rise in the next few days or whether heart rate will fluctuate abnormally. When the grey prediction model determines that the condition is worsening and relevant indicators reach a pre-set dynamic warning threshold, the warning module acts swiftly. Warnings are issued through multiple methods: the smartwatch vibrates to alert the patient, accompanied by voice prompts, ensuring immediate awareness. Furthermore, detailed warning information is pushed through a mobile app, including not only potential changes in the condition, such as blood pressure possibly exceeding the normal range or symptoms possibly worsening, but also emergency treatment suggestions, such as reminding the patient to rest and informing them of simple first aid measures, ensuring timely intervention in the early stages of deterioration and buying valuable time for medical intervention.

[0048] Data storage and management module: A distributed database is built using Distributed Hash Table (DHT) technology. DHT technology distributes data evenly across multiple nodes, avoiding performance bottlenecks and single points of failure associated with centralized data storage. Each node is responsible for storing and managing only a portion of the data, and nodes communicate and collaborate with each other through specific network protocols. When data needs to be stored or retrieved, the system calculates the data's storage location based on a hash algorithm, quickly locating the corresponding node for efficient data access. This distributed storage method greatly improves data availability; even if some nodes fail, the system can still operate normally, ensuring continuous data accessibility. Simultaneously, data redundancy and replica management mechanisms ensure data fault tolerance and prevent data loss. Homomorphic encryption is used during data storage. Homomorphic encryption is a special encryption method that allows specific computational operations to be performed directly on encrypted data without first decrypting it. For example, in medical data statistical analysis scenarios, analysts can directly perform statistical calculations such as summation and averaging on encrypted patient physiological data, obtaining results consistent with those obtained from calculating plaintext data. This feature ensures data privacy without affecting normal data analysis and processing, meeting the dual needs of the medical field in data security and utilization. Regarding data access control, a blockchain-based access control mechanism is provided. Blockchain possesses characteristics such as decentralization and immutability. Patients and authorized doctors each possess different private keys, which act as personal digital identity credentials. The system manages data access and operations based on the permission information recorded on the blockchain. When a doctor needs to access a patient's medical data, their request is compared and verified against the permission records on the blockchain. Only when the permissions match can the doctor decrypt and access the corresponding data using their private key. This mechanism fundamentally eliminates unauthorized data access, ensuring the security and privacy of medical data and providing comprehensive protection for patients' health information.

[0049] User Interaction Module: In terms of interface development, a unified yet distinctive interface was created for different devices. For smartwatches, as convenient wearable devices, the interface design prioritizes simplicity and intuitiveness. Large fonts and high-contrast colors are used to display key health data, such as real-time heart rate and blood pressure values, allowing patients to quickly access important health information. Warning messages are presented with eye-catching icons and concise text to ensure immediate attention. Simultaneously, the watch's vibration and sound alerts are used to promptly push warnings, ensuring no important health signals are missed. The mobile app further expands its functionality, providing more detailed data charts and analysis reports. Various physiological data, such as steps taken and sleep quality, are presented in intuitive line graphs and bar charts, allowing patients to easily understand trends in their health data. Analysis reports, based on professional algorithms, provide in-depth analysis of the data, such as assessing autonomic nervous system function based on heart rate variability, offering patients a comprehensive health interpretation. Furthermore, the app supports detailed data queries and historical record tracking, meeting patients' needs for comprehensive control over their health information. The desktop client is primarily designed for doctors, enabling them to comprehensively view patient data and perform diagnostic operations. The interface layout is designed according to medical workflows, integrating all patient medical information, including physiological data, disease assessment results, and historical medical records. Doctors can use the desktop client for precise data filtering and comparative analysis, and utilize professional diagnostic tools to assess conditions and formulate treatment plans. It also supports the editing and storage of electronic medical records, achieving efficient management of medical information. In terms of interface display, adaptive layout technology is employed. By detecting parameters such as the device's screen size and resolution, the size, position, and display method of interface elements are automatically adjusted. For example, on a large-screen desktop client, multiple data windows and detailed diagnostic tools can be displayed simultaneously; while on a small-screen smartwatch, only key information is highlighted, ensuring optimal visual effects and user experience on any device. Furthermore, emotional interaction design is incorporated. The system obtains patient emotional data through voice tone analysis technology. When it detects negative emotions such as anxiety or frustration caused by the patient's condition, the interface provides personalized encouragement and comfort. For example, when patients are reviewing their medical reports, a friendly reminder can pop up, such as "Don't worry, your condition is under control. Continue treatment and you will gradually improve." This allows patients to feel humanistic care during their medical treatment, improving their experience and treatment adherence.

[0050] Model Update Module: Employing a federated learning framework, each medical institution trains its disease assessment model using local data while protecting patient data privacy. This framework breaks away from the traditional centralized data training model. Medical institutions no longer need to upload patient data to a central server; instead, they train their disease assessment models locally using their own data. During local training, medical institutions use encryption techniques to process model parameters, ensuring data privacy. For example, homomorphic encryption or differential privacy techniques are used to securely encrypt model parameters without disclosing the original data. Subsequently, each medical institution exchanges the encrypted model parameters through a secure communication channel. These parameters are then fused at an aggregation center and updated globally using specific aggregation algorithms, such as weighted averaging. This approach avoids centralized exposure of patient data while fully utilizing the diverse data from each medical institution, improving the model's generalization ability. To further enhance the model's training efficiency and adaptability, a transfer learning algorithm is introduced. First, pre-training is performed on large-scale general data covering a wide range of patient groups and disease types. After training with a deep neural network, an initial model with a certain generalization ability is obtained. When a specific medical institution needs to optimize its model for its own patient population, this initial model is used as a foundation. Since different medical institutions have different patient populations, fine-tuning is performed on the institution's local data to allow the model to quickly adapt to the characteristics of that specific patient group. During fine-tuning, the main focus is on adjusting certain model parameters, especially those of layers closely related to the characteristics of the local data. For example, for medical institutions with a high proportion of elderly patients, the focus is on adjusting model parameters related to common geriatric diseases. This accelerates model convergence, reduces training time and computational resources, and improves the accuracy and adaptability of the model in assessing the condition of different patient groups, enabling it to more accurately serve various types of patients.

[0051] This invention also includes the following modules:

[0052] Personalized Analysis Module: Focusing on patient genetic data (GD), this module performs a comprehensive search through a gene-disease association database, which stores a vast amount of scientifically validated information on the relationship between genes and diseases. Specialized algorithms carefully screen for gene markers closely related to lupus; these markers act like "code words" hidden in the human genetic code, indicating an individual's susceptibility to lupus and the tendency for disease progression. Once relevant gene markers are identified, a Genetic Risk Score (GRS) is constructed. Formula It is the core of the calculation, where w gi This refers to the weight of gene marker i. It is not arbitrarily set, but determined based on a large amount of medical research and clinical data, taking into account factors such as the importance and scope of influence of gene markers in the pathogenesis of lupus. iThis involves the genotype score of gene marker i, quantifying and scoring based on different gene variant types and frequencies. Ultimately, a personalized gene risk score is calculated for each patient, quantifying the potential impact of genetic factors on lupus. Besides genetic data, lifestyle data is also a crucial component of personalized analysis. For example, regarding dietary frequency (DF), the system records detailed information such as the time intervals between meals, the types of food consumed at each meal, and the amount ingested. Frequent consumption of high-sugar and high-fat foods may exacerbate inflammatory responses and affect lupus progression. Regarding exercise intensity (EI), sensors from devices such as smartwatches accurately monitor data such as heart rate and acceleration during exercise, thereby calculating exercise intensity. For patients who are sedentary and have very low exercise intensity, their immune and metabolic functions may be affected, increasing the risk of lupus exacerbation. Using the analytic hierarchy process (AHP), the system scientifically determines the weights of personalized assessment indicators. The AHP categorizes multiple factors, such as gene risk score, dietary frequency, and exercise intensity, according to their importance, and determines the relative weights of each factor through pairwise comparisons. For example, for patients with a family history of lupus, the weight of the genetic risk score will be relatively higher; while for patients with recent sudden changes in lifestyle and significant unhealthy habits, the weight of lifestyle data will be increased. The assessment results are then adjusted by comprehensively considering the weights of these factors. For instance, for patients with specific high-risk genes and low activity levels, the system will calculate and appropriately increase the severity assessment level based on the weights, providing doctors with a more personalized diagnosis and helping to develop more targeted treatment and management plans.

[0053] This invention also includes the following modules:

[0054] Remote Medical Module: Utilizing 5G communication technology, 5G's high speed, low latency, and large capacity provide a solid guarantee for high-speed data transmission between patients and doctors. Various physiological data collected by the patient's smart device, such as heart rate, blood pressure, and blood indicators, can be transmitted stably to the doctor at extremely high speeds. Problems such as untimely data transmission and lag caused by network latency in the past are effectively solved, ensuring that doctors can obtain the latest health information of patients in real time. Employing video stream encoding and compression technology (such as H.265), the H.265 encoding standard significantly reduces data transmission volume while maintaining video quality through more advanced algorithms. It can efficiently encode images in video, intelligently identify and compress redundant information. Even under limited network conditions, it can provide doctors with clear and smooth patient video footage. Doctors can observe the patient's mental state, facial features, and other conditions in detail through the video; this intuitive information is of great reference value for disease diagnosis. Real-time remote consultation is the core function of this module. Doctors can view various monitoring data and high-definition videos of patients in real time through the system. The system provides an integrated interface that integrates data and video display, facilitating comprehensive judgment by doctors. Virtual annotation technology plays a crucial role in this process. Doctors can annotate and write descriptions on patient body parts or examination images (such as X-rays and ultrasound images). For example, abnormal areas can be marked on examination images of patient joints, with added text descriptions, facilitating communication with patients and other medical personnel, accurately conveying diagnostic thinking and key points of concern, and improving consultation efficiency and accuracy. Furthermore, remote operation assistance technology, exemplified by force feedback devices, is introduced. These devices can provide real-time feedback to nurses on the force and direction of the doctor's manipulations. In simple medical procedures, such as intravenous puncture positioning and wound dressing changes, doctors can remotely operate force feedback devices to simulate the actual operation feel and remotely guide nurses. Nurses can perceive the doctor's intentions based on the force feedback, improving the standardization and accuracy of their operations, overcoming spatial limitations, and achieving precise remote medical operation guidance, allowing patients to enjoy guidance services from high-level medical experts locally.

[0055] In this invention, the multi-source data acquisition module adds environmental data acquisition functionality. Through miniature environmental sensors integrated into smartwatches or peripheral devices, data such as ultraviolet radiation intensity (UVI), allergen concentration (AC), and noise level (NL) in the environment are collected. In the data fusion module, environmental data is fused with other physiological and medical data using an improved Bayesian fusion formula. (where H represents the disease status, E) i It represents various types of data (including environmental data) to more accurately calculate comprehensive health indicators.

[0056] In this invention, the data preprocessing module employs a Generative Adversarial Network (GAN) to repair a small amount of missing data. The generator learns the distribution of real data to generate reasonable missing data values, while the discriminator distinguishes between generated and real data. Adversarial training improves the accuracy of missing data repair. Simultaneously, wavelet transform is used to perform multi-scale analysis on the data, extracting features of different frequency components to provide richer information for subsequent disease assessment.

[0057] In this invention, the disease assessment module employs a multimodal fusion Transformer model for disease assessment. This model extracts features from image data (such as high-resolution images of skin erythema) using a convolutional neural network, transforms textual data (such as descriptions of the patient's symptoms) into vector representations using a word vector model, and normalizes numerical data (such as various physiological indicators) before inputting them into the Transformer model. The Transformer model uses a multi-head attention mechanism to fuse and analyze data from different modalities, improving the accuracy and reliability of disease assessment.

[0058] In this invention, the early warning module uses an algorithm combining fuzzy logic and reinforcement learning to determine the early warning threshold. Fuzzy logic is used to fuzzify factors such as disease assessment results, data trends, and individual patient characteristics, constructing a fuzzy rule base. Reinforcement learning is then used to continuously optimize the fuzzy rule base and the early warning threshold, based on feedback from actual changes in the patient's condition. For example, when a patient's condition worsens, the early warning threshold is adjusted to make the warning more timely and improve its effectiveness.

[0059] In this invention, the data storage and management module uses a combination of blockchain and IPFS (InterPlanetary File System) to store data. The blockchain stores the data's metadata and hash values, ensuring data immutability and traceability; IPFS stores the actual data files, utilizing content addressing technology to improve data storage and retrieval efficiency. Simultaneously, a data visualization tool is developed to display statistical analysis results through interactive charts, allowing doctors and patients to intuitively understand the trends in their patients' conditions.

[0060] In this invention, the user interaction module employs brain-computer interface (BCI) technology to achieve more natural interaction. Patients can operate the smartwatch to view data or issue commands through simple brain activity (such as imagining hand movements). Voice emotion recognition technology is used to analyze the emotional state of the patient's voice during interaction with the system, providing more personalized services. For example, when patient anxiety is detected, relaxation training audio or video is automatically pushed to the system.

[0061] In this invention, the watch, as the core of the wearable device, integrates physiological data acquisition and some environmental data acquisition functions from the multi-source data acquisition module. The watch uses low-power Bluetooth technology to transmit data with external medical testing equipment and mobile phones, sending the collected data to the data preprocessing module for processing. The watch has a built-in high-performance microprocessor that can perform real-time preprocessing of some data, reducing the burden on the system server. Simultaneously, the watch has an alert module that alerts patients through vibration, sound, and light signals. Furthermore, the watch provides a simple and easy-to-use user interface, allowing patients to view their condition data and receive alert information via touch operation, facilitating convenient monitoring and management of their condition anytime, anywhere.

[0062] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A lupus disease intelligent monitoring system based on data fusion, characterized in that, Includes the following modules: Multi-source data acquisition module: This module collects patient physiological data using sensors on a smartwatch, and obtains key indicator data by connecting to external devices via Bluetooth or Wi-Fi. It also automatically synchronizes past medical history and medication records from the electronic medical record system through patient-initiated symptom self-assessment data. The sensors on the smartwatch utilize novel microelectromechanical systems (MEMS) technology. The patient physiological data includes heart rate, blood pressure, body temperature, and steps taken. The key indicator data includes the concentration of anti-double-stranded DNA antibodies in the blood and the erythrocyte sedimentation rate. The symptom self-assessment data includes fatigue level and joint pain level. Data preprocessing module: Adaptive filtering algorithm is used to denoise the collected patient physiological data, an improved Z-score normalization formula is used to normalize data of different dimensions, and a feature extraction model based on convolutional neural network (CNN) is used to automatically extract physiological signal features. Data Fusion Module: Constructs a data fusion model based on Deep Belief Network (DBN). Preprocessed data of various types is used as input. The DBN is composed of multiple layers of Restricted Boltzmann Machines (RBMs). After feature learning through multiple RBMs, a fused feature vector (FV) is finally obtained. The fused comprehensive health index calculation formula is then introduced. To quantify the patient's overall health status, including FV i It is the fusion of various dimensions of the feature vector FV, w i These are the weights for the corresponding dimensions, and these weights are optimized using the backpropagation algorithm; Disease assessment module: A deep learning model based on a combination of Long Short-Term Memory (LSTM) network and attention mechanism is used to learn the long-term dependencies of disease data. The attention mechanism enables the deep learning model to automatically focus on key data features. The disease is divided into three levels: remission, stable and active. The output of the deep learning model is the probability distribution of the severity of the disease. The deep learning model is trained based on a large amount of labeled historical data, and the loss function is the cross-entropy loss function. Early warning module: Based on the probability distribution of the severity of the disease output by the disease assessment module, the changing trends of various physiological data and individual patient characteristics, a dynamic early warning threshold is set. A gray prediction model is used to predict the changing trends of disease indicators. When the gray prediction model judges that the disease has a worsening trend and the relevant indicators reach the pre-set dynamic early warning threshold, an early warning is issued through multiple methods, along with possible changes in the disease and emergency treatment suggestions. Data storage and management module: Utilizes Distributed Hash Table (DHT) technology to build a distributed database to distribute data across multiple nodes. Homomorphic encryption algorithm is used during data storage, and a blockchain-based access control mechanism is provided to control data access. Specifically, the system manages data access and operations based on the permission information recorded on the blockchain. User interaction module: Develop user interaction interfaces that are compatible with smartwatches, mobile apps and computers. Adaptive layout technology is used for interface display. The displayed content is automatically adjusted according to the device screen size and resolution. At the same time, emotional interaction design is introduced. The system obtains the patient's emotional data through voice tone analysis technology. When the system detects that the patient has negative emotions due to the illness, the interaction interface provides personalized encouragement and comfort. Model update module: Adopting a federated learning framework, under the premise of protecting patient data privacy, each medical institution uses local data to train the disease assessment model. Each medical institution processes and exchanges model parameters through encryption technology. The aggregation center aggregates the model parameters and updates the global model. At the same time, the transfer learning algorithm is introduced to fine-tune the model.

2. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, Also includes: Personalized analysis module: Acquires patient genetic data, searches for gene markers related to lupus through a gene-disease association database, and constructs the Gene Risk Score (GRS) formula. w gi It is the weight of gene marker i, g i The genotype score of gene marker i is used; combined with the patient's lifestyle data such as diet frequency and exercise intensity, the weights of gene risk score GRS, diet frequency, and exercise intensity are determined using the analytic hierarchy process (AHP) to correct the disease assessment results.

3. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, Also includes: The telemedicine module uses 5G communication technology to transmit patient physiological data, and video stream encoding and compression technology to encode patient video data. It supports real-time remote consultation, allowing users to view various monitoring data and high-definition videos of patients in real time. It also uses virtual annotation technology to annotate and write on patient body parts or test images, and introduces remote operation assistance technology to remotely guide medical operations.

4. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, The multi-source data acquisition module adds environmental data acquisition functionality, using miniature environmental sensors to collect data such as ultraviolet radiation intensity, allergen concentration, and noise level; and utilizes an improved Bayesian fusion formula. Integrating environmental data with other physiological and medical data, where H represents disease status, E i Represents various types of data.

5. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, The improved Z-score normalization formula in the data preprocessing module is as follows: μ s and σ s These are the mean and standard deviation of the data within a specific time period, respectively, and k and b are coefficients adjusted according to the distribution characteristics and range of variation of different types of data; Generative Adversarial Networks (GANs) are used to repair missing data. Specifically, the generator learns the distribution of real data to generate missing data values, and the discriminator distinguishes between generated data and real data. The accuracy of missing data repair is improved through adversarial training. At the same time, wavelet transform is used to perform multi-scale analysis on the data to extract features of different frequency components.

6. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, The cross-entropy loss function described in the disease assessment module is: , where y j These are actual disease severity labels, p j It is the model's predicted probability; A multimodal fusion Transformer model is used for disease assessment. In this model, image data is processed by extracting features through a convolutional neural network, text data is converted into vector representation using a word vector model, and numerical data is normalized before being input into the Transformer model. The multi-head attention mechanism is used to fuse and analyze the data from different modalities.

7. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, The formula for the grey prediction model GM(1,1) in the early warning module is as follows: , where x (0) This is the original data sequence, used to record historical data of disease-related indicators. The sequence is generated through an accumulation operation, and a and b are model parameters; The early warning model uses an algorithm combining fuzzy logic and reinforcement learning to determine the early warning threshold. Fuzzy logic is used to process the disease assessment results, the changing trends of various physiological data and individual patient characteristics to construct a fuzzy rule base. Reinforcement learning algorithm is used to continuously optimize the fuzzy rule base and the early warning threshold based on the actual changes in the patient's condition.

8. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, The data storage and management module uses a combination of blockchain and the InterPlanetary File System (IPFS) to store data. The blockchain stores the metadata and hash values ​​of the data, while IPFS stores the actual data files. A data visualization tool has been developed to display the statistical analysis results of the data through interactive charts.

9. The intelligent lupus disease monitoring system based on data fusion according to claim 1, characterized in that, The user interaction module uses brain-computer interface (BCI) technology to enable interaction, allowing users to operate the smartwatch by thinking and issuing commands, and using voice emotion recognition technology to analyze voice emotions.

10. A smartwatch, used in the data fusion-based intelligent lupus monitoring system according to any one of claims 1-9, characterized in that, The watch integrates physiological data collection and some environmental data collection functions. It uses Bluetooth technology to transmit data with external medical testing equipment and mobile phones. The watch's built-in microprocessor performs real-time preprocessing of some data and alerts patients through vibration, sound, and light signals. It provides a simple interactive interface, allowing patients to view their condition data and receive warning information through touch operation.

Images