Method for managing medical risk

Abstract

The invention discloses a method for managing medical risks, and particularly relates to the technical field of medical management, and the method comprises the steps: S1, deploying a multi-modal noninvasive biosensor, and collecting the physiological data of a patient; s2, integrating data sources, and performing cleaning and standardization processing to ensure data consistency and accuracy; s3, performing feature extraction and correlation analysis by using a deep belief network and a graph neural network, and identifying a potential risk mode; and identifying a potential risk mode through high-dimensional feature extraction and complex correlation analysis. And by utilizing self-supervised learning and a generative adversarial network, hidden risks can be efficiently detected, and potential health problems can be warned in advance. A personalized risk assessment and multi-level response mechanism provides accurate intervention measures for each patient, and the quality and efficiency of medical services are improved. Meanwhile, according to the method, continuous data collection is achieved through the noninvasive sensor network, it is guaranteed that the patient is not interfered in the monitoring process, and the comfort level and compliance of the patient are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical management, and more particularly, to a method for managing medical risks. Background Art

[0002] Existing medical management systems mainly rely on static data and traditional physiological parameters (such as heart rate, blood pressure and body temperature) for monitoring and risk assessment, which are usually obtained through regular physical examinations or clinical tests. This approach has many limitations: lack of continuous monitoring of patients' real-time physiological changes, unable to capture dynamic health data in daily life; difficult to process complex multi-source data, unable to provide a comprehensive health status assessment; risk assessment methods based on fixed rules and standards are difficult to adapt to individual differences of different patients, resulting in insufficient personalized medical intervention. Especially in acute medical events, traditional methods often fail to provide timely warnings and effective interventions, causing patients to miss the best time for treatment. For example, acute events such as heart attacks or strokes require early intervention, and traditional monitoring methods lack real-time and continuity, making it difficult to identify hidden risks in a timely manner. Therefore, there is an urgent need for a method for managing medical risks to solve the above problems. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the existing technology, the consistency and accuracy of the data are ensured by realizing the comprehensive collection and efficient analysis of the patient's physiological data, and potential risk patterns are identified through high-dimensional feature extraction and complex correlation analysis. The use of self-supervised learning and generative adversarial networks can efficiently detect hidden risks and warn of potential health problems in advance. Personalized risk assessment and multi-level response mechanisms provide each patient with precise intervention measures, improving the quality and efficiency of medical services. At the same time, this method realizes continuous data collection through a non-invasive sensor network, ensuring that the patient is not disturbed during the monitoring process, and improving the patient's comfort and compliance.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a method for managing medical risks, the method for managing medical risks comprising the following steps: S1: deploying multimodal non-invasive biosensors to collect physiological data of patients; S2: integrating data sources and performing cleaning and standardization processing to ensure data consistency and accuracy; S3: using deep belief networks and graph neural networks to perform feature extraction and association analysis to identify potential risk patterns; S4: detecting abnormal data through self-supervised learning and generative adversarial networks to identify hidden risks; S5: and identifying the degree of risk, using risk assessment algorithms to provide intervention recommendations.

[0005] In a preferred embodiment: According to a method for managing medical risks as described in claim 1, the specific steps of S1 are as follows: S101, clamp the pulse oximeter on the patient's fingertips and earlobes, and place the acoustic wave sensor above the chest to monitor the status of the heart and large blood vessels, wherein the optical sensor uses the reflection and transmission characteristics of light to detect blood oxygen saturation and heart rate by measuring the absorption changes of light in the blood, and the light source emits light, which passes through the tissue and is received by the photodetector; the acoustic wave sensor detects the blood flow velocity and blood vessel condition in the body through the principle of acoustic wave reflection, and then the sensor emits ultrasonic waves, and the waves are reflected back when encountering tissues and blood of different densities, and the blood flow velocity is calculated by receiving the reflected waves and measuring the time difference; place a non-contact electrophysiological sensor near the patient's bedside and body, and the non-contact electrophysiological sensor measures the electrophysiological signals of the human body through capacitive sensing technology, and the sensor does not need to directly Contact the skin and capture the ECG signal by measuring the change of the electric field; S102, confirm that the sensor is correctly connected and calibrated to ensure the normal operation of the equipment, start the data acquisition system, record the patient's physiological parameter data, and all sensors collect the patient's physiological data in real time, and transmit the data to the data processing unit in the form of electrical signals; S103, transmit the collected data to the central server and cloud platform through the wireless network. When the data is stored, it is classified and indexed by timestamp. The data after feature extraction is packaged into data packets and encrypted by encryption algorithm to ensure the security of the data during transmission; the encrypted data packet is transmitted to the central server and cloud platform through the wireless network. The TCP network protocol is used during the transmission process to ensure the integrity of the data packet and the transmission success rate. After receiving the data packet, the central server and cloud platform will decrypt and store the data. The data is classified and indexed by timestamp and stored in the database.

[0006] In a preferred embodiment, the specific steps of S2 are as follows: extracting the patient's electronic health records from the hospital information system, including medical history, medication records, laboratory test results, obtaining genomic data and drug metabolism information; removing redundant items and erroneous records in the data; using the Kalman filter algorithm to remove noise in the collected data: in, is the estimated state, A is the state transfer matrix, B is the control input matrix, u k is the control input, K k is the Kalman gain, z k is the measurement value, H is the observation matrix; finally, the formats and units of different data sources are unified, and time alignment is used to ensure the consistency of timestamps of multi-source data; the time alignment algorithm is used to align data with different timestamps. The DTW algorithm aligns data of different time scales by finding the best matching path between two time series: Where DTW(X,Y) is the dynamic time warping distance between two time series X and Y, d(x i ,y i ) is the distance between two points. By minimizing the total distance, the best matching path is found. Then, a multi-view learning algorithm is used to jointly learn and fuse data from different sources. Multi-view learning handles the heterogeneity problems caused by different data sources and enhances the complementarity and consistency of data through collaborative learning: Among them, L MVL is the loss function for multi-view learning, V is the number of views, and L v (X v ,Y) is the data loss of the vth view, λ is the regularization parameter, R(W) is the regularization term, and the generalization ability of the model is enhanced by regularization; the fused data has high-dimensional features, and the feature selection algorithm is used for feature selection. LASSO regression selects important features through L1 regularization: in, is the estimated regression coefficient, y i is the response variable, x i is the explanatory variable, λ is the regularization parameter; through L1 regularization, important features are selected and redundant features are removed; recursive feature elimination recursively trains the model, removes unimportant features, and selects the best feature subset. Among them, RFE(f(X), y) is the objective function of recursive feature elimination, f(X) is the prediction function of the feature subset, y is the target variable, and the best feature subset is selected by minimizing the loss.

[0007] In a preferred embodiment: The specific steps of S3 are as follows: S301, use the deep belief network input data to perform unsupervised feature extraction through multiple RBM layers: E(v,h) = -∑ i a i v i -∑ j b j h j -∑ i,j v i W ij h j , where E(v,h) is the energy function, v and h represent the states of the visible layer and the hidden layer respectively, and a i and b j They are bias W ij S302: Then use the graph neural network to perform complex association analysis, build a graph structure to represent the nodes and relationships in the data, and apply GNN to perform relationship inference and pattern recognition: in, is the feature representation of node v at the k-th layer, is the set of neighbor nodes, c v u is the normalization coefficient, W (k) and is the weight matrix, and σ is the activation function.

[0008] In a preferred embodiment: The specific steps of S4 are as follows: S401. Perform feature learning on unlabeled data through a self-supervised learning algorithm, and train the model features by generating a target task: where, is the loss function of self-supervised learning, and x and y are the input and the generated target respectively; S402. Use a generative adversarial network to generate a normal and abnormal data comparison model, thereby training the generator and discriminator of the GAN. The generator generates forged data, and the discriminator distinguishes between real data and forged data: where, D(x) is the determination probability of the discriminator for real data, G(z) is the forged data generated by the generator, p data (x) and p z (z) are the distributions of real data and noise respectively.

[0009] In a preferred embodiment: It is characterized in that: The specific steps of S5 are as follows: Use a risk assessment algorithm, combine the information obtained from feature extraction and association analysis, and conduct a comprehensive risk assessment of the patient: Risk = f(Demographics, Vitals, History, Genomics), where Risk represents the result of the risk assessment, and the function f represents the comprehensive assessment algorithm, including demographic information, vital signs, medical history, and genomic data;

[0010] When the patient's real-time heart rate variability is lower than the set threshold, that is, HRV, the system determines it as high risk. HRV is evaluated by calculating the RR interval, and the indicators are SDNN and RMSSD; the set threshold is HRV threshold : where, SDNN is the standard deviation of heart rate variability, N is the number of heartbeats, RR i is the i-th heartbeat interval, is the average value of the heartbeat intervals. SDNN is an indicator to measure heart rate variability. The lower the value, the more damaged the cardiac autonomic nerve function; when SDNN < HRVthreshold and RMSSD < HRVthreshold, it is determined as high risk, and the medical team is immediately notified for intervention, adjusting the drug dosage and arranging emergency examinations;

[0011] When the patient's blood pressure fluctuation frequency is higher than the set threshold, the system determines it as medium risk. The blood pressure fluctuation frequency is evaluated by measuring the frequency and amplitude of blood pressure changes. The set threshold is BPV. threshold : Where BPV is the frequency of blood pressure fluctuations, N is the number of measurements, and BP i is the blood pressure value measured at the i-th time. BPV is an indicator to measure the volatility of blood pressure. The higher the value, the greater the blood pressure fluctuation and the existence of cardiovascular risk. When BPV>BPVthreshold, it is judged as medium risk, and the monitoring frequency is increased, the changes in the condition are evaluated, and the intervention measures are adjusted to ensure the safety of the patient.

[0012] When the patient's physiological parameters are within the normal range but have abnormal fluctuations, the system determines that the risk is low; physiological parameters include heart rate, blood pressure, and body temperature. The set normal range and fluctuation threshold are Param normal and Param threshold : Where Deviation is the average deviation of the parameter from the normal value, N is the number of measurements, Param i is the parameter value of the ith measurement, Param normal is the normal value of the parameter. The smaller the deviation value, the smaller the fluctuation of physiological parameters and the lower the risk. <Param threshold If the patient is diagnosed with a low risk, conventional treatment and monitoring should be continued, but regular evaluation and dynamic adjustment of the intervention plan are required to prevent the risk from escalating.

[0013] Technical effects and advantages of the present invention: Through the above steps, the method can realize the comprehensive collection and efficient analysis of the patient's physiological data, ensure the consistency and accuracy of the data, and identify potential risk patterns through high-dimensional feature extraction and complex correlation analysis. By using self-supervised learning and generative adversarial networks, hidden risks can be efficiently detected and potential health problems can be warned in advance. Personalized risk assessment and multi-level response mechanisms provide each patient with precise intervention measures, improving the quality and efficiency of medical services. At the same time, the method realizes continuous data collection through a non-invasive sensor network, ensuring that the patient is not disturbed during the monitoring process, thereby improving the patient's comfort and compliance. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the steps of managing medical risks of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] An example implementation will now be described more fully with reference to the accompanying drawings. A method for managing medical risks includes the following steps: S1: deploying multimodal non-invasive biosensors to collect physiological data of patients; S2: integrating data sources and performing cleaning and standardization to ensure data consistency and accuracy; S3: using deep belief networks and graph neural networks to perform feature extraction and association analysis to identify potential risk patterns; S4: detecting abnormal data through self-supervised learning and generating adversarial networks to identify hidden risks; S5: and identifying the degree of risk, using risk assessment algorithms to provide Intervention suggestions, the specific steps of S1 are as follows: S101. Clip the pulse oximeter on the patient's fingertips and earlobes, and place the acoustic wave sensor above the chest to monitor the status of the heart and large blood vessels. Among them, the optical sensor uses the reflection and transmission characteristics of light to detect blood oxygen saturation and heart rate by measuring the absorption changes of light in the blood. The light source emits light, which passes through the tissue and is received by the photodetector; the acoustic wave sensor detects the blood flow velocity and blood vessel condition in the body through the principle of sound wave reflection. Then the sensor emits ultrasonic waves, and the waves are reflected back when they encounter tissues and blood of different densities. By receiving the reflected waves and measuring the time The blood flow velocity is calculated by the difference between the time and the time difference; non-contact electrophysiological sensors are placed near the patient's bedside and body. The non-contact electrophysiological sensors measure the electrophysiological signals of the human body through capacitive sensing technology. The sensors do not need to directly contact the skin and capture ECG signals by measuring the changes in the electric field; S102, confirm that the sensors are correctly connected and calibrated to ensure the normal operation of the equipment, start the data acquisition system, record the patient's physiological parameter data, and all sensors collect the patient's physiological data in real time and transmit the data to the data processing unit in the form of electrical signals; S103, transmit the collected data to the central server and cloud platform through the wireless network. When storing the data, it is classified and indexed by timestamp. The data after feature extraction is packaged into data packets and encrypted by encryption algorithms to ensure the security of the data during transmission; the encrypted data packets are transmitted to the central server and cloud platform through the wireless network. The TCP network protocol is used during the transmission process to ensure the integrity and transmission success rate of the data packets. After receiving the data packets, the central server and cloud platform decrypt and store the data. The data is classified and indexed by timestamp and stored in the database; deploying a multimodal non-invasive biosensor network on patients is an important initial step. These sensors include optical sensors, micro-acoustic sensors, and non-contact electrophysiological sensors. Optical sensors measure indicators such as oxygen saturation and heart rate in the blood by utilizing the reflection or transmission characteristics of light. This type of sensor is usually installed on the patient's fingertips or earlobes and fixed by clamping to ensure continuous data collection without affecting the patient's daily activities. Micro-acoustic sensors use the principle of sound wave reflection to detect blood flow velocity and blood vessel conditions in the body and obtain health information about the heart and large blood vessels.The sensor is usually placed above the patient's chest and can capture the dynamic changes of the heart and blood vessels in real time. Non-contact electrophysiological sensors use capacitive sensing technology to measure the body's electrophysiological signals, such as electrocardiograms, without direct contact with the skin. This sensor can be installed at the patient's bedside or near the body to obtain ECG signals by sensing changes in the electric field, avoiding the contact discomfort of traditional electrodes. During the sensor installation and debugging process, it is necessary to ensure that all devices are properly connected and calibrated to ensure the accuracy and reliability of data collection. After starting the data acquisition system, the sensor begins to record the patient's physiological parameter data in real time, which is transmitted to the central server or cloud platform via a wireless network for storage and processing. When the data is stored, it is classified and indexed by timestamp for subsequent analysis and retrieval. The entire data acquisition process needs to ensure continuity and stability to obtain high-quality physiological data and provide a solid foundation for subsequent analysis.

[0017] Next, the specific steps of S2 are as follows: extract the patient's electronic health records from the hospital information system, including medical history, medication records, laboratory test results, obtain genomic data and drug metabolism information; remove redundant items and erroneous records in the data; use the Kalman filter algorithm to remove noise in the collected data: in, is the estimated state, A is the state transfer matrix, B is the control input matrix, u k is the control input, K k is the Kalman gain, z k is the measurement value, H is the observation matrix; finally, the formats and units of different data sources are unified, and time alignment is used to ensure the consistency of timestamps of multi-source data; the time alignment algorithm is used to align data with different timestamps. The DTW algorithm aligns data of different time scales by finding the best matching path between two time series: Where DTW(X,Y) is the dynamic time warping distance between two time series X and Y, d(x i ,y i ) is the distance between two points. By minimizing the total distance, the best matching path is found. Then, a multi-view learning algorithm is used to jointly learn and fuse data from different sources. Multi-view learning handles the heterogeneity problems caused by different data sources and enhances the complementarity and consistency of data through collaborative learning: Among them, L MVL is the loss function for multi-view learning, V is the number of views, and L v (X v,Y) is the data loss of the vth view, λ is the regularization parameter, R(W) is the regularization term, and the generalization ability of the model is enhanced by regularization; the fused data has high-dimensional features, and the feature selection algorithm is used for feature selection. LASSO regression selects important features through L1 regularization: in, is the estimated regression coefficient, y i is the response variable, x i is the explanatory variable, λ is the regularization parameter; through L1 regularization, important features are selected and redundant features are removed; recursive feature elimination recursively trains the model, removes unimportant features, and selects the best feature subset. Among them, RFE(f(X), y) is the objective function of recursive feature elimination, f(X) is the prediction function of the feature subset, y is the target variable, and the best feature subset is selected by minimizing the loss; cross-domain data integration and preprocessing are important links in data analysis. By integrating data from different data sources, cleaning and standardizing them, the consistency and accuracy of the data are ensured. First, the electronic health records of patients, including medical history, drug records, laboratory test results, etc., are extracted from the hospital information system. In addition, genomic data and drug metabolism information are also needed to obtain a more comprehensive understanding of the patient's health status. These data come from different sources, with different formats and units, and need to be processed uniformly. Data cleaning is a key step to ensure data quality, including removing redundant data, correcting erroneous records, and processing missing values. Through cleaning, noise data can be removed and the accuracy and reliability of data can be improved. In the process of data standardization, the formats and units of different data sources need to be processed uniformly, such as converting blood glucose values ​​in different units into standard units. Time alignment is an important part of standardization, ensuring that the timestamps of multi-source data are consistent for synchronous analysis. After data cleaning and standardization, the integrity and consistency of the data have been significantly improved, laying the foundation for subsequent high-dimensional feature extraction and complex correlation analysis. In this process, it is also necessary to ensure data privacy and security, use encryption technology to protect patient sensitive information, follow relevant laws and regulations, and ensure compliance with data processing.

[0018] Then, the specific steps of S3 are as follows: S301, use the deep belief network input data to perform unsupervised feature extraction through multiple RBM layers: E(v,h)=-∑ i a i v i -∑ j b j h j -∑ i,j v i W ij h j, where E(v,h) is the energy function, v and h represent the states of the visible layer and the hidden layer respectively, and a i and b j They are bias W ij S302: Then use the graph neural network to perform complex association analysis, build a graph structure to represent the nodes and relationships in the data, and apply GNN to perform relationship inference and pattern recognition: in, is the feature representation of node v at layer k, is the set of neighbor nodes, c v u is the standardized coefficient, W (k) and is the weight matrix, σ is the activation function; high-dimensional feature extraction and complex association analysis are the core steps of data processing. High-dimensional feature extraction and complex association analysis of data are performed through deep belief networks and graph neural networks, so as to identify potential risk patterns. First of all, deep belief networks are an unsupervised learning model that can extract high-dimensional features from a large amount of data and capture the intrinsic structure of the data. The input data is extracted through multiple restricted Boltzmann machine layers, abstracted layer by layer, and finally obtains a high-dimensional feature representation. This process does not require manual annotation and can efficiently process a large amount of data. Graph neural networks are used for complex association analysis. By constructing a graph structure to represent the nodes and relationships in the data, the complex associations between the data are captured. In the graph structure, the nodes represent the different characteristics of the patient, and the edges represent the relationship between the features. Through the training of graph neural networks, implicit patterns and potential relationships in the data can be discovered, so as to identify potential high-risk patterns. The results of high-dimensional feature extraction and complex association analysis provide an important basis for risk assessment and the formulation of intervention measures. In this process, it is also necessary to continuously optimize the model to improve its accuracy and generalization ability to cope with the individual differences of different patients. Through high-dimensional feature extraction and complex correlation analysis, we can deeply explore the value of data, provide scientific decision-making support, and improve the level of intelligent medical management.

[0019] At this time, the specific steps of S4 are as follows: S401, feature learning of unlabeled data is performed through a self-supervised learning algorithm, and model features are trained by generating target tasks: in, is the loss function of self-supervised learning, x and y are the input and generated targets respectively; S402, using the generative adversarial network, generate normal and abnormal data comparison models, so as to train the generator and discriminator of GAN, the generator generates forged data, and the discriminator distinguishes between real data and forged data: Among them, D(x) is the probability of the discriminator to judge the real data, G(z) is the forged data generated by the generator, and p data (x) and pz (z) are the distributions of real data and noise respectively; self-supervised learning and implicit risk detection are key steps to further improve the accuracy of risk identification. Self-supervised learning is a feature learning method for unlabeled data, which trains model features by generating target tasks (such as predicting the next value of data). Different from traditional supervised learning, self-supervised learning does not require a large amount of labeled data, can learn useful features from unlabeled data, and improve the generalization ability of the model. Generative adversarial networks are used to generate a comparison model of normal and abnormal data. Through the adversarial training of the generator and discriminator, the generator generates forged data, and the discriminator distinguishes real data from forged data. In this way, abnormal patterns in data can be effectively detected and potential implicit risks can be identified. During the risk monitoring process, by using the comparative analysis of normal and abnormal data generated by generative adversarial networks, abnormal patterns in patient data can be accurately identified, and potential health risks can be warned in advance. This process requires continuous optimization of the performance of the generator and discriminator to improve the accuracy and sensitivity of abnormal detection. By combining self-supervised learning and generative adversarial networks, efficient detection of implicit risks can be achieved, providing more accurate health management services for patients.

[0020] Finally, the specific steps of S5 are as follows: Use a risk assessment algorithm, combined with the information obtained from feature extraction and correlation analysis, to conduct a comprehensive risk assessment of the patient: Risk = f(Demographics, Vitals, History, Genomics), where Risk represents the result of the risk assessment, the function f represents the comprehensive assessment algorithm, including demographic information, vital signs, medical history, and genomic data; when the patient's real-time heart rate variability is lower than the set threshold, that is, HRV, the system determines it as high risk. HRV is evaluated by calculating the RR interval, and the indicators are SDNN and RMSSD; the set threshold is HRV threshold : Among them, SDNN is the standard deviation of heart rate variability, N is the number of heartbeats, RR i is the i-th inter-beat interval, is the average value of the inter-beat intervals. SDNN is an indicator to measure heart rate variability. The lower the value, the more damaged the cardiac autonomic nerve function; when SDNN < HRVthreshold and RMSSD < HRVthreshold, it is determined as high risk, and the medical team is immediately notified for intervention, adjusting the drug dosage and arranging emergency examinations; when the frequency of the patient's blood pressure fluctuations is higher than the set threshold, the system determines it as medium risk. The frequency of blood pressure fluctuations is evaluated by measuring the frequency and amplitude of blood pressure changes, and the set threshold is BPV threshold : Among them, BPV is the frequency of blood pressure fluctuations, N is the number of measurements, BP iis the blood pressure value measured for the i-th time. BPV is an indicator to measure the volatility of blood pressure. The higher the value, the greater the blood pressure fluctuation and the existence of cardiovascular risk. When BPV>BPVthreshold, it is judged as medium risk, and the monitoring frequency is increased, the condition changes are evaluated, and the intervention measures are adjusted to ensure the safety of the patient. When the patient's physiological parameters are within the normal range but there are abnormal fluctuations, the system determines it as low risk. Physiological parameters include heart rate, blood pressure, and body temperature. The set normal range and fluctuation threshold are Param normal and Param threshold : Where Deviation is the average deviation of the parameter from the normal value, N is the number of measurements, Param i is the parameter value of the ith measurement, Param normal is the normal value of the parameter. The smaller the deviation value, the smaller the fluctuation of physiological parameters and the lower the risk. <Param thresholdWhen the patient is diagnosed with low risk, conventional treatment and monitoring will continue, but regular evaluation and dynamic adjustment of intervention plans are required to prevent risk escalation; personalized risk assessment and multi-level response mechanisms are important steps to ensure that each patient receives precision medicine. In this process, it is necessary to first establish and continuously update the patient's personalized health record. The health record includes comprehensive data such as real-time physiological data, medical history records, and genetic information, which can reflect the patient's current health status and historical change trends. By dynamically updating the health record, the system can obtain the latest data in real time, providing a reliable basis for subsequent risk assessment. Once the health record is updated, the system will use the risk assessment algorithm to conduct a comprehensive assessment of the patient's health status. The risk assessment algorithm takes into account multiple factors, including demographic information (such as age, gender, race, etc.), vital signs (such as heart rate, blood pressure, body temperature, etc.), medical history records (such as previous diseases, medication, surgical history, etc.), and genomic data (such as genetic markers, gene mutations, etc.). By integrating this information, the algorithm can conduct a comprehensive analysis of the patient's health risks and derive a risk score for each patient. Specifically, the risk assessment first assesses the heart health status by analyzing the patient's real-time heart rate variability (HRV). HRV is assessed by measuring the time intervals between heartbeats. Higher heart rate variability generally means better heart health, while lower heart rate variability may indicate impaired cardiac autonomic function. The system determines the patient's heart health risk by monitoring HRV and comparing it to the set health threshold. If HRV is lower than the set threshold, the system will determine the patient as high risk and immediately notify the medical team for emergency intervention, such as adjusting medication dosage or arranging an emergency examination. In addition to HRV, the system also monitors the patient's blood pressure fluctuation frequency. Blood pressure fluctuation frequency is assessed by measuring the frequency and amplitude of blood pressure changes. Large blood pressure fluctuations may indicate increased cardiovascular risk, so the system compares the blood pressure fluctuation frequency with the set threshold. If the blood pressure fluctuation frequency is higher than the set threshold but still within the safe range, the system will determine the patient as medium risk. At this time, it is recommended to increase the monitoring frequency, regularly evaluate changes in the condition, and adjust intervention measures as needed to ensure the safety of the patient. In addition, the system also evaluates the fluctuations of other physiological parameters, including heart rate, blood pressure, and body temperature. If these physiological parameters are all within the normal range but have slight abnormal fluctuations, the system will determine the patient as low risk. This means that although there are no obvious health problems at present, routine treatment and monitoring still need to be continued, and the patient's condition needs to be reassessed regularly to prevent the escalation of potential risks. After the risk assessment is completed, the system provides personalized intervention measures based on the patient's risk level. For high-risk patients, the system will immediately take emergency intervention measures, such as notifying the medical team, adjusting the treatment plan, etc., to prevent the condition from worsening. For medium-risk patients, the system recommends increasing the frequency of monitoring and evaluation to ensure that the condition is tracked and managed in a timely manner.For low-risk patients, the system recommends continuing routine treatment and monitoring, but regular assessments are still required to ensure that any potential risks can be discovered and dealt with in a timely manner. The core of this process is to ensure that every patient can get the most suitable medical service through comprehensive data integration and efficient algorithm analysis. The efficiency and reliability of the system come from the application of a variety of advanced algorithms, including heart rate variability analysis, blood pressure fluctuation assessment, and physiological parameter fluctuation detection. Each step is scientifically designed and rigorously verified to ensure the accuracy and timeliness of medical decisions. Through personalized risk assessment and multi-level response mechanisms, medical teams can better manage patients' health, improve treatment outcomes and patient satisfaction, and ultimately achieve the goal of precision medicine.

[0021] The above embodiments can be implemented in whole or in part by software, hardware, firmware and any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions and computer programs. When the computer instructions and computer programs are loaded and executed on a computer, the processes and functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, and other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server and data center to another website site, computer, server and data center by wired (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, data center, etc. that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0022] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0023] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0024] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0025] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0026] If the functions are implemented in the form of software functional units and sold and used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0027] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes and substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for managing medical risks, characterized in that: The method for managing medical risk comprises the following steps: S1: Deploy multimodal non-invasive biosensors to collect patients’ physiological data; S2: Integrate data sources and perform cleaning and standardization to ensure data consistency and accuracy; S3: Use deep belief networks and graph neural networks for feature extraction and association analysis to identify potential risk patterns; S4: Detect abnormal data and identify hidden risks through self-supervised learning and generative adversarial networks; S5: And identify the degree of risk and provide intervention recommendations using risk assessment algorithms.

2. A method for managing medical risks according to claim 1, characterized in that: The specific steps of S1 are as follows: S101. Clip the pulse oximeter on the patient's fingertips and earlobes, and place the acoustic wave sensor above the chest to monitor the status of the heart and large blood vessels. Among them, the optical sensor uses the reflection and transmission characteristics of light to detect blood oxygen saturation and heart rate by measuring the absorption changes of light in the blood. The light source emits light, which passes through the tissue and is received by the photodetector; the acoustic wave sensor detects the blood flow velocity and blood vessel condition in the body through the principle of sound wave reflection. Then the sensor emits ultrasonic waves, and the waves are reflected back when encountering tissues and blood of different densities. The blood flow velocity is calculated by receiving the reflected waves and measuring the time difference; non-contact electrophysiological sensors are placed near the patient's bedside and body. The non-contact electrophysiological sensors measure the electrophysiological signals of the human body through capacitive sensing technology. The sensor does not need to directly contact the skin and captures the electrocardiogram by measuring changes in the electric field. Signal; S102, confirm that the sensor is correctly connected and calibrated to ensure the normal operation of the equipment, start the data acquisition system, record the patient's physiological parameter data, all sensors collect the patient's physiological data in real time, and transmit the data to the data processing unit in the form of electrical signals; S103, transmit the collected data to the central server and cloud platform through the wireless network. When the data is stored, it is classified and indexed by timestamp. The data after feature extraction is packaged into data packets and encrypted by encryption algorithm to ensure the security of the data during transmission; the encrypted data packet is transmitted to the central server and cloud platform through the wireless network. The TCP network protocol is used during the transmission process to ensure the integrity of the data packet and the transmission success rate. After receiving the data packet, the central server and cloud platform will decrypt and store the data. The data is classified and indexed by timestamp and stored in the database.

3. A method for managing medical risks according to claim 1, characterized in that: The specific steps of S2 are as follows: extract the patient's electronic health records from the hospital information system, including medical history, medication records, laboratory test results, obtain genomic data and drug metabolism information; remove redundant items and erroneous records in the data; use the Kalman filter algorithm to remove noise in the collected data: in, is the estimated state, A is the state transfer matrix, B is the control input matrix, u k is the control input, K k is the Kalman gain, z k is the measurement value, H is the observation matrix; finally, the formats and units of different data sources are unified, and time alignment is used to ensure the consistency of timestamps of multi-source data; the time alignment algorithm is used to align data with different timestamps. The DTW algorithm aligns data of different time scales by finding the best matching path between two time series: DTW Where DTW(X,Y) is the dynamic time warping distance between two time series X and Y, d(x i ,y i ) is the distance between two points. By minimizing the total distance, the best matching path is found. Then, a multi-view learning algorithm is used to jointly learn and fuse data from different sources. Multi-view learning handles the heterogeneity problems caused by different data sources and enhances the complementarity and consistency of data through collaborative learning: Among them, L MVL is the loss function for multi-view learning, V is the number of views, and L v (X v ,Y) is the data loss of the vth view, λ is the regularization parameter, R(W) is the regularization term, and the generalization ability of the model is enhanced by regularization; the fused data has high-dimensional features, and the feature selection algorithm is used for feature selection. LASSO regression selects important features through L1 regularization: in, is the estimated regression coefficient, y i is the response variable, x i is the explanatory variable, λ is the regularization parameter; through L1 regularization, important features are selected and redundant features are removed; recursive feature elimination recursively trains the model, removes unimportant features, and selects the best feature subset. Among them, RFE(f(X), y) is the objective function of recursive feature elimination, f(X) is the prediction function of the feature subset, y is the target variable, and the best feature subset is selected by minimizing the loss.

4. A method for managing medical risks according to claim 1, characterized in that: The specific steps of S3 are as follows: S301, use the deep belief network input data to perform unsupervised feature extraction through multiple RBM layers: E(v,h) = -∑ i a i v i -∑ j b j h j -∑ i,j v i W ij h j , where E(v,h) is the energy function, v and h represent the states of the visible layer and the hidden layer respectively, and a i and b j They are bias W ij S302: Then use the graph neural network to perform complex association analysis, build a graph structure to represent the nodes and relationships in the data, and apply GNN to perform relationship inference and pattern recognition: in, is the feature representation of node v at layer k, is the set of neighbor nodes, c v u is the standardized coefficient, W (k) and is the weight matrix and σ is the activation function.

5. A method for managing medical risks according to claim 1, characterized in that: The specific steps of S4 are as follows: S401, perform feature learning of unlabeled data through self-supervised learning algorithm, and train model features by generating target tasks: in, is the loss function of self-supervised learning, x and y are the input and generated target respectively; S402, using the generative adversarial network to generate normal and abnormal data comparison models, so as to train the generator and discriminator of GAN, the generator generates fake data, and the discriminator distinguishes between real data and fake data: min G max D V(D,G) Among them, D(x) is the probability of the discriminator to judge the real data, G(z) is the forged data generated by the generator, and p data (x) and p z (z) are the distributions of real data and noise, respectively.

6. A method for managing medical risks according to claim 1, characterized in that: The specific steps of S5 are as follows: Use the risk assessment algorithm to combine the information obtained from feature extraction and association analysis to conduct a comprehensive risk assessment of the patient: Risk = f (Demographics, Vitals, History, Genomics), where Risk represents the result of the risk assessment, and function f represents the comprehensive assessment algorithm, which includes demographic information, vital signs, medical history, and genomic data; When the patient's real-time heart rate variability is lower than the set threshold, and the real-time heart rate variability is HRV, the system determines it as a high risk. HRV is evaluated by calculating the RR interval, and the indicators are SDNN and RMSSD; the set threshold is HRV threshold : Among them, SDNN is the standard deviation of heart rate variability, N is the number of heartbeats, RR i is the i-th heartbeat interval, is the average value of the heartbeat intervals. SDNN is an indicator to measure heart rate variability. The lower the value, the more damaged the cardiac autonomic nerve function is; when SDNN < HRVthreshold and RMSSD < HRVthreshold, it is determined as a high risk, and the medical team is immediately notified for intervention, adjusting the drug dosage and arranging emergency examinations; When the patient's blood pressure fluctuation frequency is higher than the set threshold, the system determines it as medium risk. The blood pressure fluctuation frequency is evaluated by measuring the frequency and amplitude of blood pressure changes. The set threshold is BPV. threshold : Where BPV is the frequency of blood pressure fluctuations, N is the number of measurements, and BP i is the blood pressure value measured at the i-th time. BPV is an indicator to measure the volatility of blood pressure. The higher the value, the greater the blood pressure fluctuation and the existence of cardiovascular risk. When BPV>BPVthreshold, it is judged as medium risk, and the monitoring frequency is increased, the changes in the condition are evaluated, and the intervention measures are adjusted to ensure the safety of the patient. When the patient's physiological parameters are within the normal range but have abnormal fluctuations, the system determines that the risk is low; physiological parameters include heart rate, blood pressure, and body temperature. The set normal range and fluctuation threshold are Param normal and Param threshold : Where Deviation is the average deviation of the parameter from the normal value, N is the number of measurements, Param i is the parameter value of the ith measurement, Param normal is the normal value of the parameter. The smaller the deviation value, the smaller the fluctuation of physiological parameters and the lower the risk. <Param threshold If the patient is diagnosed with a low risk, conventional treatment and monitoring should be continued, but regular evaluation and dynamic adjustment of the intervention plan are required to prevent the risk from escalating.

Images