Remote electrocardio intelligent monitoring system
By using a remote intelligent ECG monitoring system, combined with a primary abnormality detection and pathophysiological risk inference engine, the problems of high false alarm rate and delayed alarm response in ECG monitoring systems have been solved, achieving accurate alarms and rapid location, thus improving the quality and efficiency of medical services.
Patent Information
- Application Number
- CN202511418344.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-30
- Publication Date
- 2026-01-06
AI Technical Summary
Existing electrocardiogram (ECG) monitoring systems suffer from high false alarm rates, delayed alarm responses, and the inability to remotely and centrally monitor, leading to fatigue among medical staff and delays in emergency care.
A remote intelligent ECG monitoring system is adopted, which combines monitoring terminals, servers and medical staff terminals. Through a primary abnormality detection module, a clinical context perception module and a pathophysiological risk inference engine, the system dynamically assesses the risk level of ECG signals and generates interpretable structured alarm information.
It significantly reduced the false alarm rate, enabled accurate delivery and rapid location of alarm information, provided risk analysis and clinical treatment suggestions based on knowledge graphs, and improved the quality and efficiency of medical services.
Smart Images
Figure CN121265073A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of electrocardiogram (ECG) monitoring technology, and in particular relates to a remote intelligent ECG monitoring system. Background Technology
[0002] Electrocardiogram (ECG) monitoring is a core method for monitoring the vital signs of critically ill patients in intensive care units, coronary care units, and general wards. Traditional monitoring systems use ECG monitors placed at the patient's bedside to continuously monitor vital signs such as ECG, heart rate, and blood oxygenation. When the monitored data exceeds a preset fixed threshold, a local audible and visual alarm is triggered to alert medical staff.
[0003] However, existing electrocardiogram monitoring systems suffer from the following long-standing and unresolved shortcomings: First, alarm rules based on fixed thresholds cannot distinguish between physiological disturbances (such as patient movement, poor electrode contact, and lead wire detachment) and real pathological events, resulting in a massive number of false alarms. Prolonged exposure to this high-noise alarm environment can lead to "alarm fatigue" in healthcare workers, causing them to become desensitized and insensitive to alarm signals, making them highly likely to ignore genuine and critical alarms, posing a significant risk to patient safety.
[0004] Secondly, alarm information is confined to a single bedside device and cannot be remotely and centrally detected. In a large ward with dozens of beds, when multiple devices alarm simultaneously or the alarm sound comes from a distant ward, the medical staff at the nursing station cannot quickly and accurately locate which bed's device is alarming by sound alone, resulting in a delayed response and wasting precious rescue time.
[0005] Third, existing systems treat ECG signals as isolated data streams, completely detached from the patient's actual clinical condition. For example, the clinical risks of occasional premature ventricular contractions (PVCs) in a patient who has just undergone heart surgery and is using arrhythmogenic drugs are drastically different from those of the same waveform in a stable patient. However, traditional systems will issue the exact same alarm, failing to provide healthcare professionals with decision support for risk assessment. Summary of the Invention
[0006] This invention provides a remote intelligent ECG monitoring system that solves the problems of high false alarm rate and delayed alarm response.
[0007] The basic solution provided by this invention is a remote intelligent electrocardiogram monitoring system, comprising a monitoring terminal, a server, and a medical staff terminal. The server includes: The data acquisition module is used to continuously receive real-time vital sign data streams from at least one monitoring terminal; The primary anomaly detection module is used to analyze real-time vital signs data streams, determine whether there are primary anomaly events, and output the corresponding risk level. The clinical context awareness module, in response to the determination of the existence of a primary abnormal event, queries the clinical context data associated with the current patient and the type of primary abnormal event from the hospital information system in real time based on a preset context triggering and mapping strategy. The intelligent alarm decision module inputs initial abnormal events and clinical context data into a pathophysiological risk reasoning engine. The pathophysiological risk reasoning engine traverses a pre-built medical knowledge graph and, based on multi-path risk assessment rules, discovers one or more risk paths connecting initial abnormal events and potential malignant cardiac events. Based on the assessment of the risk paths, it dynamically corrects the risk level of the initial abnormal event and outputs the corrected executable alarm information. The medical staff terminal is used to receive and display executable alarm information sent by the server.
[0008] Preferably, the primary anomaly detection module includes a signal quality assessment unit and a primary anomaly event determination unit. The signal quality assessment unit is used to analyze the signal quality of the vital signs data stream. If the signal quality is not up to standard, the detection signal is determined to be abnormal and the abnormal detection signal information is sent directly to the medical staff. If the signal meets the standard, the vital signs data stream is input to the primary abnormal event determination unit. The initial abnormal event determination unit includes an initial abnormal event determination model, which analyzes the vital sign data stream to determine the type and risk level of the initial abnormal event; wherein, the initial abnormal event determination model is a one-dimensional convolutional neural network model.
[0009] More preferably, the initial anomaly detection model performs real-time detection through the following steps: S10. Preprocess the input real-time vital signs data stream; S11. Input the preprocessed vital sign signal fragments into the trained convolutional neural network model; S12. The probability that the output of the convolutional neural network model of vital sign signal segments belongs to each predefined abnormal event type; S13. Compare the probability output by the model with the risk level mapping table. When the probability of a certain abnormal event type exceeds the threshold in the risk level mapping table, it is determined that the corresponding primary abnormal event has occurred, and the primary abnormal event type and initial risk level are output.
[0010] Preferably, the context triggering and mapping strategy is as follows: Establish and maintain an abnormal event type-clinical context mapping table; The dynamic context awareness module queries the abnormal event type-clinical context mapping table based on the identified primary abnormal event type to determine the type combination of clinical context data to be retrieved. The combination of clinical context data types includes at least one of the following: The patient's current medication list; Abnormal values in the patient's laboratory indicators; The patient's current diagnosis and past medical history.
[0011] Preferably, the intelligent alarm decision-making module includes a pathophysiological risk reasoning engine, a suggested measures generation unit, and an interpretable alarm generation unit. The pathophysiological risk reasoning engine is used to integrate physiological abnormalities and clinical context data, perform in-depth risk assessment according to multi-path risk assessment rules, and dynamically adjust alarm levels. The risk assessment includes extracting key risk factors and clinical logic based on the risk paths discovered by the risk reasoning engine. The recommended measures generation unit is used to automatically match and generate recommended measures from the medical treatment knowledge base based on the risk factors in the risk assessment. The interpretable alarm generation unit is used to convert the final risk level, risk factors, clinical logic, recommended measures, and abnormal event types output by the risk reasoning engine into structured alarm information.
[0012] More preferably, the multi-path risk assessment rule is executed through the following steps: S40. Starting from the nodes of the primary abnormal events in the graph, and taking the nodes of the retrieved clinical context data in the graph as key nodes, perform a multi-hop traversal based on the breadth-first search (BFS) algorithm to discover all paths leading to the malignant endpoint event. The malignant endpoint event is a specific physiological event that leads to cardiac arrest, including ventricular fibrillation, pulseless ventricular tachycardia, sustained ventricular tachycardia that leads to hemodynamic collapse, prolonged cardiac arrest, etc. S41. For each risk path discovered, calculate its risk value according to the following path risk scoring formula: The path risk scoring formula is as follows:
[0013] in, The risk score is given for a single path, where n is the number of relationships along that path. Let be the weight of the i-th relation on the path. The importance coefficient of the i-th context node on the path; S42. Based on the preset comprehensive risk correction function, summarize the risk scores of all discovered risk paths, and correct the initial risk level according to the risk scores; The comprehensive risk correction function is:
[0014] in, This is the revised final risk level. The initial risk level, The sum of scores for all risk paths. This is a correction factor.
[0015] More preferably, the structured alarm information includes: the corrected risk level, the core abnormal event, the associated risk analysis, and the recommended measures.
[0016] Preferably, the server further includes: The feedback optimization module is used to record the processing confirmation results of medical staff on the executable alarm information, and to iteratively update the weight parameters in the initial anomaly judgment model and / or multi-path risk assessment rules based on the processing confirmation results according to the model optimization strategy.
[0017] Preferably, the feedback optimization module includes a primary anomaly detection model optimization unit and a multi-path risk assessment rule optimization unit. The initial-order anomaly detection model optimization unit is used to update the weights of the initial-order anomaly detection model using the gradient descent algorithm. The multi-path risk assessment rule optimization unit is used to optimize the weights in the rules using a hybrid genetic algorithm.
[0018] More preferably, the multi-path risk assessment rule optimization unit is used to optimize the weights in the rules using a hybrid genetic algorithm; specifically, the hybrid genetic algorithm is used to optimize the weights in the multi-path risk assessment rules based on the system's continuously encrypted and stored labeled decision samples, including the complete context data of each alarm, the system decision results, and the processing confirmation results of medical staff, with the optimization objective of minimizing the overall error rate and response delay of the system. The execution process includes: The relational weights in the path risk scoring formula Importance coefficient The correction coefficient α in the comprehensive risk correction function is encoded as a real number chromosome; The quality of chromosomes is evaluated using a fitness function, which is calculated based on the false alarm rate, false negative rate, and average response delay of the system composed of the chromosome parameter set on a validation dataset. The optimal chromosome is iteratively evolved through genetic operations including selection, crossover, and mutation, combined with simulated annealing algorithm for local search. The optimal chromosome decoding parameter set is updated in the pathophysiological risk inference engine.
[0019] The principles and advantages of this invention are as follows: 1. By combining the primary anomaly detection model with the pathophysiological risk reasoning engine, the system achieves a leap from signal judgment to clinical decision-making. The system not only analyzes the characteristics of vital sign data, but also integrates the patient's real-time clinical context information (such as medication, test indicators, and diagnostic information) through a knowledge graph. This effectively distinguishes between physiological interference and real pathological events. Through a comprehensive judgment mechanism based on multi-dimensional information, the system significantly reduces the false alarm rate, thereby effectively alleviating the alarm fatigue problem of clinical medical staff.
[0020] 2. Through a multi-terminal collaborative design for medical staff, the system achieves accurate push and rapid location of alarm information. This frees alarm information from individual bedside devices, allowing for centralized display and push via the medical staff terminals. This enables medical staff to quickly and accurately obtain and locate the source of the alarm at the nurses' station, avoiding response delays caused by auditory confusion or long-distance transmission, thus saving valuable time for emergency treatment. Furthermore, the structured alarm information generated by the system not only includes risk levels and event descriptions, but more importantly, it provides association risk analysis based on knowledge graph reasoning and specific clinical treatment suggestions. This interpretable alarm output allows medical staff to clearly understand the system's decision-making logic, greatly enhancing their trust in the system's judgment results. It also provides valuable reference for clinical decision-making, improving the quality and efficiency of medical services.
[0021] 3. The system utilizes medical knowledge graphs and multi-path risk assessment rules to dynamically correct the risk level of initial abnormal events. This allows alarm information to not only reflect abnormalities in ECG data but also to output differentiated risk warnings based on the patient's specific clinical background, providing medical staff with more accurate decision support and improving the clinical relevance and practicality of alarms.
[0022] 4. Through continuous learning and feedback mechanisms, the system can dynamically adjust internal parameters and evaluation rules to adapt to changes in different patient groups and clinical environments, thereby continuously improving the accuracy of monitoring and the overall performance of the system. Attached Figure Description
[0023] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0024] The following detailed description illustrates the specific implementation method: The specific implementation process is as follows: (See details) Figure 1A remote intelligent electrocardiogram monitoring system includes a monitoring terminal, a server, and a medical staff terminal. The monitoring terminal is connected to the server network and is used to collect and output the patient's real-time vital signs data stream. The vital signs data stream includes electrocardiogram waveform data, oxygenation and circulation parameters, respiratory and blood pressure parameters, etc.; it includes a data acquisition unit and a data communication unit. The data acquisition unit is used to collect vital signs data, and the data communication unit is used to transmit the collected data to the server. The network connection between the monitoring terminal and the server can be a local area network, a cellular network, or a hybrid network. In this embodiment, a hybrid network is used for connection. Specifically, the monitoring terminal is paired with a personal smart terminal or a fixed gateway in the ward via Bluetooth. The gateway then transmits the data to the server via the hospital's Wi-Fi network.
[0025] The monitoring terminal can be a standard bedside ECG monitor or a wearable ECG monitoring device. In this embodiment, a standard bedside ECG monitor is used. This monitor collects multi-lead ECG signals through surface electrodes and typically integrates a data communication module (such as Wi-Fi, Bluetooth, or Ethernet) for transmitting the collected real-time vital signs data. In addition to ECG, it usually integrates modules such as a blood oxygen saturation probe, a non-invasive blood pressure cuff, and respiratory impedance monitoring, enabling simultaneous collection of multiple vital signs. Wearable ECG monitoring devices are suitable for patients requiring long-term, mobile monitoring and can be dynamic ECG recorders, patch-type ECG monitors, smart bracelets / watches, or portable ECG event recorders.
[0026] The server includes: The data acquisition module is used to continuously receive real-time vital sign data streams from at least one monitoring terminal; Specifically, the data acquisition module includes a data receiving unit and a data processing unit. The data receiving unit is used to receive data transmitted by the monitoring terminal; The data processing unit is used to perform a joint denoising algorithm based on wavelet transform and bandpass filtering on the original data stream to suppress power frequency interference, baseline drift and electromyographic noise.
[0027] The primary anomaly detection module is used to analyze real-time vital signs data streams, determine whether there are primary anomaly events, and output the corresponding risk level. The initial-order anomaly detection module includes a signal quality assessment unit and an initial-order anomaly event determination unit. The signal quality assessment unit is used to analyze the signal quality of the vital signs data stream. If the signal quality is not up to standard, the detection signal is determined to be abnormal, and the abnormal detection signal information is sent directly to the medical staff. If the signal meets the standard, the vital signs data stream is input to the primary abnormal event determination unit.
[0028] The pre-positioned signal quality assessment unit can identify severe noise, signal loss, or baseline drift caused by electrode detachment, lead wire breakage, or violent patient movement. This fundamentally avoids false physiological abnormality alarms, such as arrhythmia alarms, caused by technical interference (such as body movement or electrode problems). This is a key step in solving alarm fatigue, avoiding unnecessary complex calculations on low-quality data and saving computing resources.
[0029] The initial-level abnormal event determination unit is used to determine the type and risk level of initial-level abnormal events based on the real-time vital sign data stream. The initial-level abnormal event determination unit includes an initial-level abnormal event determination model, which analyzes the vital sign data stream. The initial-level abnormal event determination model is based on a one-dimensional convolutional neural network model, and it performs real-time determination through the following steps: S10. Preprocess the input real-time vital signs data stream, including data cleaning and data standardization. S11. Input the preprocessed vital sign signal fragments into the trained convolutional neural network model; S12. The convolutional neural network model outputs the probability that a vital sign fragment belongs to each predefined abnormal event type; S13. Compare the probability output by the model with the risk level mapping table. When the probability of a certain abnormal event type exceeds the threshold in the risk level mapping table, it is determined that the corresponding primary abnormal event has occurred, and the primary abnormal event type and initial risk level are output.
[0030] The abnormal event types identified by the primary analysis module are clinical events that deviate from normal physiology and morphology, automatically detected from vital sign data. In this embodiment, the abnormal event types include at least one of the following: ventricular arrhythmia risk events, myocardial ischemia / infarction risk events, bradycardia / conduction disorder risk events, tachyarrhythmia risk events, and QT interval prolongation.
[0031] The clinical context awareness module, in response to the determination of the existence of a primary abnormal event, queries the clinical context data associated with the current patient and the type of primary abnormal event from the hospital information system in real time based on a preset context triggering and mapping strategy. The context triggering and mapping strategy is as follows: Establish and maintain an abnormal event type-clinical context mapping table; The dynamic context awareness module queries the abnormal event type-clinical context mapping table based on the identified primary abnormal event type to determine the type combination of clinical context data to be retrieved. The combination of clinical context data types includes at least one of the following: The patient's current medication list; Abnormal values in the patient's laboratory indicators; The patient's current diagnosis and past medical history.
[0032] Specifically, the mapping table includes at least the following mapping relationships: When the abnormal event type is "ventricular arrhythmia risk event", the mapped clinical context data are: [Current medication list] Screening for Class I and III antiarrhythmic drugs, digitalis, antipsychotics, and other drugs known to have ventricular arrhythmia-causing side effects; [Abnormal values of laboratory indicators] Obtaining serum potassium, serum magnesium, and serum calcium concentrations to rule out electrolyte imbalances; [Current diagnosis and past medical history] Obtaining diagnoses of organic heart diseases such as acute myocardial infarction, heart failure, cardiomyopathy, and valvular heart disease.
[0033] When the abnormal event type is "myocardial ischemia / infarction risk event", the mapped clinical context data are as follows: [Abnormal values of laboratory indicators] Obtain myocardial enzyme profile, including the values and dynamic trends of troponin and creatine kinase isoenzymes; [Current diagnosis and past medical history] Obtain history of coronary artery disease, angina pectoris, previous myocardial infarction, coronary intervention, diabetes, and hypertension; [Current medication list] View the use of antiplatelet, anticoagulant, and statin drugs to assess the current treatment strategy.
[0034] When the abnormal event type is "bradycardia / conduction disorder risk event", the mapped clinical context data are: [Current medication list] Screening for beta-blockers, non-dihydropyridine calcium channel blockers (such as verapamil, diltiazem), digoxin, amiodarone, and other drugs that may inhibit the function of the sinoatrial node or atrioventricular node; [Abnormal values of laboratory indicators] Obtaining thyroid function indicators (such as TSH) to rule out hypothyroidism; [Current diagnosis and past medical history] Obtaining history of sick sinus syndrome, atrioventricular block, and hyperkalemia.
[0035] When the abnormal event type is "risk event for rapid atrial arrhythmia", the mapped clinical context data is as follows: [Current diagnosis and past medical history] obtains information on heart failure, hypertensive heart disease, chronic obstructive pulmonary disease, hyperthyroidism, and cardiac surgery; [Abnormal values of laboratory indicators] obtains thyroid function indicators and electrolytes; [Current medication list] checks for any violations such as the use of sympathomimetic drugs.
[0036] When the abnormal event type is "QT interval prolongation risk event", the mapped clinical context data are as follows: [Current medication list] Screening against the known list of QT interval prolonging drugs (such as specific antibiotics, antihistamines, antipsychotics, Class III antiarrhythmic drugs, etc.) one by one; [Abnormal values of laboratory indicators] Obtaining the concentration values of serum potassium, serum magnesium, and serum calcium; [Current diagnosis and past medical history] Obtaining congenital long QT syndrome and family history of sudden cardiac death.
[0037] The intelligent alarm decision module inputs initial abnormal events and clinical context data into a pathophysiological risk reasoning engine. The pathophysiological risk reasoning engine traverses a pre-built medical knowledge graph and, based on multi-path risk assessment rules, discovers one or more risk paths connecting initial abnormal events and potential malignant cardiac events. Based on the assessment of the risk paths, it dynamically corrects the risk level of the initial abnormal event and outputs the corrected executable alarm information. The intelligent alarm decision-making module includes a pathophysiological risk reasoning engine, a suggested measures generation unit, and an interpretable alarm generation unit. The pathophysiological risk reasoning engine is used to integrate physiological abnormalities and clinical context data, perform in-depth risk assessment according to multi-path risk assessment rules, and dynamically adjust alarm levels. The risk assessment includes extracting key risk factors and clinical logic based on the risk paths discovered by the risk reasoning engine. The multi-path risk assessment rules are implemented through the following steps: S40. Starting from the nodes of the primary abnormal events in the graph, and taking the nodes of the retrieved clinical context data in the graph as key nodes, perform a multi-hop traversal based on the breadth-first search (BFS) algorithm to discover all paths leading to the malignant endpoint event. The malignant endpoint event is a specific physiological event that leads to cardiac arrest, including ventricular fibrillation, pulseless ventricular tachycardia, sustained ventricular tachycardia that leads to hemodynamic collapse, prolonged cardiac arrest, etc. S41. For each risk path discovered, calculate its risk value according to the following path risk scoring formula: The path risk scoring formula is as follows:
[0038] in, The risk score is given for a single path, where n is the number of relationships along that path. Let be the weight of the i-th relation on the path. The importance coefficient of the i-th context node on the path; S42. Based on the preset comprehensive risk correction function, summarize the risk scores of all discovered risk paths, and correct the initial risk level according to the risk scores; The comprehensive risk correction function is:
[0039] in, This is the revised final risk level. The initial risk level, The sum of scores for all risk paths. This is a correction factor.
[0040] The medical knowledge graph is constructed based on clinical practice guidelines (such as cardiovascular-related guidelines published by ACC / AHA / ESC), medical textbooks, peer-reviewed high-impact academic journal articles, and a standardized medical terminology system. It also utilizes anonymized electronic medical record data and vital sign databases (such as electrocardiogram databases) to quantify the strength of relationships between entities.
[0041] The recommended measures generation unit is used to automatically match and generate specific recommended measures from the medical treatment knowledge base based on the risk factors in the risk assessment. The medical treatment knowledge base is meticulously constructed based on the rules of clinical guidelines. Specifically, it breaks down the treatment process in authoritative cardiovascular disease diagnosis and treatment guidelines into structured "condition-action" rule pairs and incorporates the experience of clinical experts to form a precise, standardized, and actionable treatment suggestion mapping library.
[0042] The interpretable alarm generation unit is used to convert the final risk level, risk factor, clinical logic, recommended measures and abnormal event types output by the risk reasoning engine into structured alarm information. The structured alarm information includes: the corrected risk level, the core abnormal event (bed-patient-abnormal event type), the correlation risk analysis (listing the key risk factors inferred from the knowledge graph and their clinical logic), and the recommended measures.
[0043] The medical staff terminal is connected to the server network and is used to receive and display executable alarm information sent by the server.
[0044] The medical staff terminal includes at least one of the following: a central display screen at the nurse station, a computer workstation client, and a mobile smart terminal application. Specifically, the central display screen at the nurse station is deployed in the core location of the ward for global monitoring. When an alarm occurs, the corresponding bed information is highlighted as a warning, and the complete structured alarm information is displayed for rapid location. The computer workstation client is integrated with the hospital's existing clinical information system. Alarm information is presented in the form of a strong reminder pop-up, interrupting the current operation of medical staff to ensure that the information is noticed and processed immediately, and supports one-click feedback. The mobile smart terminal application is installed on the smart devices worn by medical staff. In addition to pushing complete alarm information, it can also be combined with voice broadcast to achieve "information finding people," ensuring that medical staff can be informed of the alarm as soon as possible even when on the move.
[0045] Example 2 The server also includes: The feedback optimization module is used to record the processing confirmation results of medical staff on the executable alarm information, and to iteratively update the weight parameters in the initial anomaly judgment model and / or multi-path risk assessment rules based on the processing confirmation results according to the model optimization strategy.
[0046] The feedback optimization module includes an initial anomaly detection model optimization unit and a multi-path risk assessment rule optimization unit. The initial-order anomaly detection model optimization unit is used to update the weights of the initial-order anomaly detection model using the gradient descent algorithm. Specifically, alarm records confirmed by medical staff (including raw physiological signs data and final labels) are used as labeled training data; Calculate the loss between the model's current prediction and the true label; The gradient of the loss with respect to the weights of each layer of the model is calculated using the backpropagation algorithm; The optimizer updates the model weights in the opposite direction of the gradient to reduce the loss.
[0047] The multi-path risk assessment rule optimization unit is used to optimize the weights in the rules using a hybrid genetic algorithm; Specifically, a hybrid genetic algorithm is used to continuously encrypt and store labeled decision samples, including complete context data for each alarm, system decision results, and confirmation results from medical staff (such as "confirmed as true alarm" or "marked as false alarm"). The optimization objective is to minimize the overall error rate and response latency of the system. The weights in the multi-path risk assessment rules are optimized, and the execution process includes: The relational weights in the path risk scoring formula Importance coefficient The correction coefficient α in the comprehensive risk correction function is encoded as a real number chromosome; The quality of chromosomes is evaluated using a fitness function, which is calculated based on the false alarm rate, false negative rate, and average response delay of the system composed of the chromosome parameter set on a validation dataset. The optimal chromosome is iteratively evolved through genetic operations including selection, crossover, and mutation, combined with simulated annealing algorithm for local search. The optimal chromosome decoding parameter set is updated in the pathophysiological risk inference engine.
[0048] The comprehensive risk correction function is:
[0049] in, For the fitness of the parameter set, , , To balance the weights of the different error types, including false positive rate, false negative rate, and average response delay, Scores for false alarm rate, false negative rate, and average response delay. for In this embodiment, =0.5, =2, =0.5.
[0050] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. A remote electrocardio intelligent monitoring system, characterized in that, The monitoring terminal, the server, and the medical terminal are included, The server comprises: a data acquisition module, configured to continuously receive a real-time vital sign data stream of at least one monitoring terminal; a primary abnormality detection module, configured to analyze the real-time vital sign data stream, determine whether a primary abnormality event exists, and output a corresponding risk level; a clinical context perception module, configured to, in response to a determination that the primary abnormality event exists, query, based on a preset context triggering and mapping strategy, clinical context data associated with a current patient and a type of the primary abnormality event from a hospital information system in real time; an intelligent alarm decision module, configured to input the primary abnormality event and the clinical context data into a pathophysiological risk reasoning engine; wherein the pathophysiological risk reasoning engine discovers one or more risk paths connecting the primary abnormality event and a potential malignant cardiac event by traversing a pre-constructed medical knowledge graph based on multi-path risk assessment rules, dynamically corrects the risk level of the primary abnormality event based on an assessment of the risk paths, and outputs corrected executable alarm information; the medical terminal, configured to receive and display the executable alarm information sent by the server. 2.The remote intelligent ECG monitoring system according to claim 1, characterized in that, The primary abnormality detection module comprises a signal quality assessment unit and a primary abnormality event determination unit, The signal quality assessment unit is configured to perform signal quality analysis on the vital sign data stream, determine a signal abnormality if the signal quality is not up to standard, and directly send signal abnormality detection information to the medical terminal, or input the vital sign data stream to the primary abnormality event determination unit if the signal quality is up to standard. The primary abnormality event determination unit comprises a primary abnormality determination model, which is configured to analyze the vital sign data stream by the primary abnormality determination model to determine a type of the primary abnormality event and a risk level; wherein the primary abnormality determination model is a one-dimensional convolutional neural network model. 3.The remote intelligent ECG monitoring system of claim 2, wherein, The primary abnormality determination model performs real-time determination by the following steps: S10, pre-processing the input real-time vital sign data stream; S11, inputting the pre-processed vital sign signal segment into the trained convolutional neural network model; S12, the convolutional neural network model outputs probabilities of the vital sign signal segment belonging to each pre-defined abnormality event type; S13, comparing the probabilities output by the model with a risk level mapping table, and determining that a corresponding primary abnormality event occurs and outputting the type of the primary abnormality event and the initial risk level when the probability of a certain abnormality event type exceeds a threshold value in the risk level mapping table. 4.The remote intelligent ECG monitoring system of claim 1, wherein, The context triggering and mapping strategy is establishing and maintaining an abnormality event type-clinical context mapping table; The dynamic context perception module queries the abnormality event type-clinical context mapping table according to the identified type of the primary abnormality event to determine the type combination of the clinical context data that needs to be retrieved; The type combination of the clinical context data includes at least one of the following: a current medication list of the patient; abnormal values of test indicators of the patient; current diagnosis and past medical history of the patient.
5. The remote intelligent ECG monitoring system of claim 1, wherein, The intelligent alarm decision module comprises a pathophysiological risk reasoning engine, a recommended measure generation unit, and an interpretable alarm generation unit, The pathophysiological risk inference engine is configured to fuse the physiological abnormalities and the clinical context data, perform deep risk assessment according to a multi-path risk assessment rule, and dynamically correct an alarm level, wherein the risk assessment includes key risk factors and clinical logic extracted based on a risk path discovered by the risk inference engine; The recommended measure generation unit is configured to automatically match and generate recommended measures from a medical treatment knowledge base according to risk factors in the risk assessment; The interpretable alarm generation unit is configured to convert a final risk level, risk factors, clinical logic, recommended measures and abnormal event types output by the risk inference engine into structured alarm information. 6.The remote ECG intelligent monitoring system according to claim 1 or 5, characterized in that, The multi-path risk assessment rule is executed by the following steps: S40. Starting from a node of the primary abnormal event in the graph and taking a node of the retrieved clinical context data in the graph as a key node, a multi-hop traversal based on a breadth-first search algorithm is performed to discover all paths that can lead to a malignant terminal event, wherein the malignant terminal event is a specific physiological event that leads to cardiac arrest, including ventricular fibrillation, pulseless ventricular tachycardia, persistent ventricular tachycardia leading to hemodynamic collapse, long-term cardiac arrest, etc. S41. For each risk path discovered, a risk value thereof is calculated according to the following path risk scoring formula: The path risk scoring formula is: wherein, is the risk score for a single path, n is the number of relationships on the path, is the weight of the i-th relationship on the path, is the importance coefficient of the i-th context node on the path; S42. Based on a preset comprehensive risk correction function, risk scores of all discovered risk paths are summarized, and an initial risk level is corrected according to the risk score. The comprehensive risk correction function is: wherein, is the final risk rating after correction, is the initial risk rating, is the sum of all risk path scores, is the correction factor.
7. The remote intelligent ECG monitoring system of claim 6, wherein, The structured alarm information includes a corrected risk level, a core abnormal event, associated risk analysis, and recommended measures. 8.The remote intelligent ECG monitoring system of claim 1, wherein, The server further includes: A feedback optimization module configured to record a processing confirmation result of the executable alarm information by medical personnel and iteratively update weight parameters in the primary abnormality determination model and / or the multi-path risk assessment rule according to a model optimization strategy using the processing confirmation result.
9. The remote intelligent ECG monitoring system of claim 8, wherein, The feedback optimization module includes a primary abnormality determination model optimization unit and a multi-path risk assessment rule optimization unit, The primary abnormality determination model optimization unit is configured to update the weight of the primary abnormality determination model using a gradient descent algorithm; The multi-path risk assessment rule optimization unit is configured to optimize the weight in the rule using a hybrid genetic algorithm.
10. The remote intelligent ECG monitoring system of claim 9, wherein, In the multi-path risk assessment rule optimization unit, a hybrid genetic algorithm is used to optimize the weight in the multi-path risk assessment rule based on system continuous encryption storage of labeled decision samples, including complete context data of each alarm, system decision results, and processing confirmation results of medical personnel, with the optimization objective being to minimize the comprehensive error rate and response delay of the system, and the execution process including: The relationship weight in the path risk score formula , the importance coefficient and the correction coefficient α in the comprehensive risk correction function are coded as real number chromosomes; Evaluating the pros and cons of chromosomes using a fitness function, wherein the fitness function is based on a comprehensive calculation of the false positive rate, false negative rate and average response delay of the system composed of the chromosome parameter set on the verification data set; Through genetic operations including selection, crossover and mutation, and combined with a simulated annealing algorithm for local search, an optimal chromosome is iteratively evolved. updating the parameter set decoded from the optimal chromosome into the physiopathological risk reasoning engine.