Medical instrument adverse event risk monitoring method and system based on artificial intelligence

Abstract

The invention provides a medical instrument adverse event risk monitoring method and system based on artificial intelligence, and relates to the technical field of artificial intelligence technology monitoring, and the method comprises the steps: monitoring a medical instrument adverse event risk signal in an area under administration based on artificial intelligence; generating an investigation task of the risk signal of the adverse event of the medical device, and matching and scheduling experts to execute the task; and recommending disposal suggestions based on the investigation information of the experts. According to the medical instrument adverse event risk monitoring method and system based on artificial intelligence, active monitoring and management of the medical instrument adverse event risk are realized through the artificial intelligence technology, and the use safety and supervision efficiency of the medical instrument are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence technology monitoring, and in particular to a medical device adverse event risk monitoring method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of medical technology, medical devices are increasingly widely used in clinical diagnosis and treatment, playing an important role in improving medical quality and patient safety. However, with the increase in the use of medical devices, the frequency of medical device adverse events has also increased. These adverse events may be caused by product defects, improper use or equipment failure, etc., which pose potential risks to patient health and the medical system.

[0003] Traditional adverse event monitoring methods mainly rely on manual reporting and post-analysis, which have the disadvantages of insufficient timeliness, low efficiency and limited accuracy, and are difficult to meet the needs of modern medicine for proactive risk prevention and control.

[0004] Therefore, an innovative monitoring and management method is urgently needed to realize real-time identification and effective response to medical device adverse event risks. SUMMARY

[0005] One of the purposes of the present application is to provide a medical device adverse event risk monitoring method and system based on artificial intelligence to solve the problems pointed out in the background.

[0006] In a first aspect, the present application provides a medical device adverse event risk monitoring method based on artificial intelligence, comprising:

[0007] Based on artificial intelligence, monitoring medical device adverse event risk signals in the jurisdiction;

[0008] Generating an investigation task for a medical device adverse event risk signal and matching and scheduling experts to perform the task;

[0009] Based on the expert's investigation information, recommending a disposal suggestion.

[0010] Optionally, the monitoring of medical device adverse event risk signals in the jurisdiction based on artificial intelligence comprises:

[0011] Based on a pre-trained machine learning model, according to the device usage state information collected by the edge nodes deployed in different medical device users, the medical device adverse event risk signals in the jurisdiction are monitored.

[0012] Optionally, the generation of an investigation task for a medical device adverse event risk signal and the matching and scheduling of experts to perform the task comprise:

[0013] Generate a template based on a preset task, and generate an investigation task based on a medical device adverse event risk signal;

[0014] Match experts from a preset expert library who are suitable for performing the investigation task;

[0015] Dispatch experts to perform the investigation task.

[0016] Optionally, the investigation information of the experts is used to recommend a treatment suggestion, including:

[0017] Based on the investigation information of the experts, risk assessment is performed;

[0018] Based on the risk classification treatment engine, a treatment suggestion is determined according to the risk assessment result.

[0019] Optionally, the medical device adverse event risk monitoring method based on artificial intelligence further includes:

[0020] Based on the medical device adverse event risk signal and the investigation information of the experts, an evidence chain is generated;

[0021] The evidence chain is stored by using a block chain.

[0022] Optionally, the medical device adverse event risk monitoring method based on artificial intelligence further includes:

[0023] Based on the investigation information of the experts, a preset medical device adverse event risk signal investigation form is automatically filled in;

[0024] The preset medical device adverse event risk signal investigation form after being automatically filled in is pushed to the experts.

[0025] Optionally, the medical device adverse event risk signal at least includes product defects, improper use, and causes or possible causes of serious harm of medical device adverse events.

[0026] Optionally, the investigation information at least includes the source of the patient's ward, the original disease, the original signs, various examination data, treatment, past medical history, adverse event occurrence, occurrence time, harm / failure performance, adverse event consequences, combined drug / equipment, and risk control measures taken.

[0027] Optionally, the expert library at least contains pre-selected personnel from medical institutions, medical device manufacturers, and medical device adverse event monitoring agencies.

[0028] Optionally, the medical device adverse event risk monitoring method based on artificial intelligence further includes:

[0029] Assist experts in performing the investigation task.

[0030] In a second aspect, the embodiment of the present application provides a medical device adverse event risk monitoring system based on artificial intelligence, comprising:

[0031] An artificial intelligence monitoring module is configured to monitor medical device adverse event risk signals in a jurisdiction based on artificial intelligence.

[0032] An investigation task scheduling module is configured to generate investigation tasks of medical device adverse event risk signals, and match and schedule experts to perform the tasks.

[0033] A disposal recommendation module is configured to recommend disposal suggestions based on the investigation information of the experts.

[0034] The present application has the following beneficial effects:

[0035] The artificial intelligence technology realizes the active monitoring and management of medical device adverse event risks, significantly improves the safety and efficiency of medical device use, can monitor risk signals in real time, discover potential problems in time, automatically generate investigation tasks to improve processing efficiency, intelligently match and schedule experts to ensure that professional personnel participate in the investigation, and make risk assessment and disposal recommendation based on the investigation information of the experts, providing scientific basis for decision-making. This technical scheme optimizes the risk management process by integrating machine learning and expert cooperation, and reduces the probability of adverse events and the harm they may cause.

[0036] Other features and advantages of the present application will be described in the following description, and some will become apparent from the description, or will be understood from the practice of the present application. The purpose and other advantages of the present application can be achieved and obtained by the structure specifically pointed out in the written description and the accompanying drawings.

[0037] The technical solutions of the present application will be further described in detail below with the help of the accompanying drawings and examples. DETAILED DESCRIPTION

[0038] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0039] Figure 1 It is a schematic diagram of a medical device adverse event risk monitoring method based on artificial intelligence in the embodiment of the present application;

[0040] Figure 2 It is still another schematic diagram of a medical device adverse event risk monitoring method based on artificial intelligence in the embodiment of the present application;

[0041] Figure 3Another schematic diagram of a medical device adverse event risk monitoring method based on artificial intelligence in an embodiment of the present application is provided.

[0042] Figure 4 A schematic diagram of a medical device adverse event risk monitoring system based on artificial intelligence in an embodiment of the present application is provided. DETAILED DESCRIPTION

[0043] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0044] The research and development idea of the present application is to realize the active monitoring and management of medical device adverse event risks by using artificial intelligence technology. By integrating machine learning, expert dispatching, risk assessment, and blockchain storage, etc. technologies, this method can efficiently identify risk signals, organize expert investigations and propose disposal suggestions within the jurisdiction, thereby improving the safety and efficiency of medical device use.

[0045] Figure 1 A flowchart of a medical device adverse event risk monitoring method based on artificial intelligence is provided for the embodiments of the present application, as shown in Figure 1 The method comprises:

[0046] 101. Based on artificial intelligence, monitor the medical device adverse event risk signals in the jurisdiction. The medical device adverse event risk signals at least include: product defects, improper use, causing or possibly causing serious harm of medical device adverse events.

[0047] The core of this step is to monitor the adverse event risk signals of medical devices in the jurisdiction in real time through artificial intelligence technology. These risk signals include but are not limited to product defects (such as hardware failure or design defects), improper use (such as operation error or failure to follow instructions) and events that may cause serious harm (such as patient injury or equipment failure).

[0048] Step 101 specifically includes:

[0049] Based on the pre-trained machine learning model, according to the device use state information collected by the edge nodes deployed in different medical device users, monitor the medical device adverse event risk signals in the jurisdiction.

[0050] The monitoring process relies on a pre-trained machine learning model that analyzes the medical device usage status information collected from edge nodes to identify potential risks. Edge nodes are intelligent devices deployed at medical device usage sites (e.g., hospitals, clinics) that collect real-time usage status information of the devices. This information includes device operating parameters, user operation records (e.g., key inputs), error logs, and environmental data (e.g., humidity). The machine learning model is built based on machine learning algorithms using supervised learning and anomaly detection techniques. The model is trained using historical adverse event data (e.g., known device failure cases) to learn the characteristic patterns of risk signals. Additionally, semi-supervised or unsupervised learning can be used to discover new types of risks that are not labeled. The model outputs the classification results of risk signals (e.g., "product defect" or "misuse") and their confidence levels. When the model detects an abnormal pattern (e.g., frequent device errors or parameters exceeding safety ranges), it generates a risk signal and preliminarily classifies it based on severity. For example, if the signal could cause severe harm (e.g., heart pacemaker malfunction), it is labeled as high priority.

[0051] An implementation example is provided, in which a hospital deploys multiple infusion pumps, and edge nodes collect real-time flow rate, battery status, and user setting data for each pump. A pre-trained machine learning model analyzes these data and finds that a batch of infusion pumps frequently exhibit flow rate abnormalities (more than 10% of the set value) within a specific time period. The model labels this as a "product defect" risk signal with a confidence level of 0.9. If another pump causes a dose error due to a nurse's incorrect parameter setting, the model labels it as a "misuse" risk signal with a confidence level of 0.85. These signals are then pushed to subsequent steps for processing.

[0052] 102. Generating an investigation task for a medical device adverse event risk signal and matching and dispatching experts to perform the task.

[0053] As shown in Figure 2 Step 102 is specifically divided into the following three sub-steps:

[0054] 201. Generating an investigation task based on a pre-set task generation template based on a medical device adverse event risk signal.

[0055] Using a pre-set template, a structured investigation task is automatically generated based on the attributes of the risk signal (e.g., type, severity, location of occurrence). The template includes fields such as task title, description, objectives, and priority to ensure that the task is clear and actionable.

[0056] 202. Matching experts from a pre-set expert database to perform the investigation task. The expert database at least includes pre-selected personnel from medical institutions, medical device manufacturers, and medical device adverse event monitoring agencies.

[0057] The expert pool contains pre-screened personnel from medical institutions (e.g. doctors, nurses), medical device manufacturers (e.g. engineers) and monitoring agencies (e.g. supervisors). When matching, the most suitable candidate is selected considering the expert's professional field (e.g. cardiovascular device expertise), workload and geographical location.

[0058] 203. Schedule the expert to perform the investigation task.

[0059] According to the task urgency and the expert's schedule, the task execution time is determined and the expert is notified through a notification (e.g. email or app push). Continuing the above implementation example, after detecting the "product defect" risk signal, a task is generated based on the template: "investigate infusion pump flow rate anomaly". The task description includes the hospital name where the anomaly occurred, the device batch number and the preliminary detection data. When matching, a medical device engineer is found from the expert pool (professional field: infusion device, distance from the hospital: 50 km, current number of tasks: 2), and a cardiologist who is already fully loaded is excluded. The task scheduling sets a 48-hour completion deadline and notifies the engineer through a mobile app.

[0060] 103. Based on the expert's investigation information, recommend a treatment suggestion. The investigation information at least includes: the patient's ward source, the original disease, the original signs, various examination data, treatment, past medical history, adverse event occurrence, occurrence time, injury / failure performance, adverse event consequences, combination of drugs / devices, and risk control measures taken.

[0061] As shown in FIG. 10, step 103 is specifically divided into the following two sub-steps: Figure 3

[0062] 301. Based on the expert's investigation information, perform risk assessment.

[0063] A plurality of risk assessment indicators (e.g. severity, degree of derivative impact, etc.) are pre-set, and risk assessment is performed according to the investigation information provided by the expert.

[0064] 302. Based on the risk classification treatment engine, determine the treatment suggestion according to the risk assessment result.

[0065] The risk classification treatment engine contains a plurality of pre-set risk classification treatment rules, respectively corresponding to different risk assessment results. According to the assessment result, the risk is classified into different levels (e.g. low, medium, high), and the corresponding treatment measures are recommended, such as device repair, use training or product recall, etc.

[0066] ​Providing such an implementation example, the engineer investigates the infusion pump and finds that the flow rate anomaly is caused by an internal sensor failure, affecting 10 patients (with no serious consequences). During risk assessment, the analysis of the investigation information determines the severity as "medium", the probability of occurrence as "high", and the scope of impact as "local". The risk classification and disposal engine accordingly recommends: "replace the faulty sensor and conduct a comprehensive inspection of the same batch of equipment." If the investigation finds that improper use is the cause, it is recommended to "strengthen the training of nurses in operation".

[0067] The above method realizes the active monitoring and management of medical device adverse event risks through artificial intelligence technology, significantly improving the safety and efficiency of medical device use. This method can monitor risk signals in real time, discover potential problems in a timely manner; automatically generate investigation tasks to improve processing efficiency; intelligently match and dispatch experts to ensure that professional personnel participate in the investigation; and conduct risk assessment and recommend disposal suggestions based on expert investigation information, providing a scientific basis for decision-making. This technical solution optimizes the risk management process by integrating machine learning and expert collaboration, reducing the probability of adverse events and the potential harm they may cause.

[0068] In some embodiments, the artificial intelligence-based medical device adverse event risk monitoring method further comprises:

[0069] Based on the medical device adverse event risk signals and the expert investigation information, an evidence chain is generated.

[0070] The evidence chain is an orderly record of the entire risk monitoring process, including risk signals, investigation tasks, expert reports, risk assessments, and disposal suggestions. The timestamps and data signatures of each link ensure integrity.

[0071] The evidence chain is stored using blockchain technology.

[0072] The evidence chain is uploaded to the blockchain (usually a permissioned chain), taking advantage of its distributed ledger characteristics to achieve non-tamperable and traceable storage. Each record is encrypted and linked to the previous block, forming a secure chain.

[0073] Providing such an implementation example, in the above infusion pump case, the evidence chain includes: risk signal (flow rate anomaly, 09:00), investigation task (generation time: 09:05), expert report (sensor failure, 10-02 15:00), risk assessment (medium, 10-02 16:00), and disposal suggestion (replace the sensor, 10-02 16:30). These data generate hash values (such as SHA-256) and are stored in the blockchain nodes, which can be accessed by regulatory agencies at any time.

[0074] The above method ensures the integrity and traceability of the entire risk monitoring process by generating a chain of evidence and utilizing blockchain storage. The tamper-proof nature of the blockchain significantly enhances data security and credibility, allowing reliable recording and storage of data at various stages such as risk signals, investigation tasks, expert reports, and disposal recommendations. This method facilitates the review of relevant information by regulatory agencies at any time, thereby strengthening regulatory efforts and providing technical support for subsequent audits and responsibility tracing.

[0075] In some embodiments, the artificial intelligence-based medical device adverse event risk monitoring method further comprises:

[0076] Based on the expert investigation information, the pre-set medical device adverse event risk signal investigation form is automatically filled in.

[0077] The system extracts key fields (such as event time and injury performance) from the investigation information and uses natural language processing (NLP) technology to fill in the pre-set template.

[0078] The pre-set medical device adverse event risk signal investigation form after automatic filling is pushed to the expert.

[0079] The filled-in form is pushed to the expert electronically, and the expert can edit or confirm and submit it.

[0080] An implementation example is provided, in which after the infusion pump investigation is completed, the system automatically fills in the form: event time (10-0108:50), injury performance (flow rate exceeds standard), and consequence (no serious injury). The form is pushed to the engineer's mobile phone, and the engineer adds a note "suggest regular calibration" and submits it.

[0081] The above method of automatically filling in the investigation form and pushing it to the expert reduces the workload of the expert and improves work efficiency. By using natural language processing technology, the system can extract key content from the investigation information and accurately fill in the pre-set form, ensuring consistency and accuracy of filling. This automated process facilitates the expert to quickly confirm and submit the investigation results, shortens the processing time, and thus accelerates the entire risk monitoring and response cycle.

[0082] In some embodiments, the artificial intelligence-based medical device adverse event risk monitoring method further comprises:

[0083] 104、When the expert performs the investigation task, physical behavior data, digital operation trajectory data, and environmental context data are collected in real time through a sensor group carried by the expert and wearable devices; the sensor group includes an accelerometer, a gyroscope, a microphone, and a GPS positioning unit.

[0084] When experts perform investigation tasks, they collect three types of data in real time through sensors on their mobile terminals (such as smartphones, tablets) and wearable devices (such as smart bracelets, smart glasses):

[0085] Physical behavior data: Records the expert's actions such as walking, standing, and operating device frequency and amplitude through accelerometers and gyroscopes. These data reflect the expert's activity intensity and behavior patterns, for example, rapid movement may indicate time pressure.

[0086] Digital operation trajectory data: Records the expert's interaction behavior on the mobile terminal, such as screen tapping, interface sliding, and text input frequency and mode. These data reflect the expert's interaction habits and work efficiency with the survey tool.

[0087] Environmental context data: Captures the noise level of the surrounding environment (such as the degree of noise in a hospital ward) through a microphone and determines the expert's geographic location (such as in a ward or conference room) through a GPS positioning unit. These data provide potential environmental influences on the investigation work.

[0088] The sensor group includes:

[0089] Accelerometer: Measures the acceleration of the expert's actions in m / s 2 .

[0090] Gyroscope: Detects the direction and angular velocity of the expert's actions in rad / s.

[0091] Microphone: Records environmental sounds and analyzes noise decibels (dB).

[0092] GPS positioning unit: Provides latitude and longitude coordinates.

[0093] The real-time nature of data collection depends on the device's sampling frequency (such as 100 samples per second for the accelerometer), ensuring the capture of subtle changes in the expert's behavior.

[0094] This step realizes the comprehensive capture of the expert's physical behavior, operation habits, and environmental conditions through multi-dimensional data collection, providing a rich data foundation for subsequent state analysis.

[0095] Provide an implementation example, a medical device engineer investigates the abnormal flow rate of an infusion pump in a hospital. He carries a smartphone and a smart bracelet. The accelerometer on the bracelet records that the engineer has moved rapidly 10 times in 5 minutes (average acceleration 2 m / s 2The data showed he was frequently moving around the ward; the gyroscope detected a wrist rotation frequency of 20 times per minute, indicating he was frequently operating equipment or recording data. His smartphone recorded 5 clicks per minute and 3 notes entered in the survey app, indicating a high frequency of digital operations. The microphone detected an ambient noise level of 70 decibels (a noisy ward environment), and GPS location showed he was on the 3rd floor of the hospital ward.

[0096] 105. Transform physical behavior data, digital operation trajectory data, and environmental context data into multidimensional feature vectors; wherein, the multidimensional feature vectors shall include at least operation intensity features, cognitive attention features, and environmental risk level features.

[0097] The collected raw data (physical behavior, digital operation trajectories, environmental context) is transformed into multi-dimensional feature vectors for processing by machine learning models. The feature vectors must contain at least the following three types of features:

[0098] Operational intensity characteristics: Quantifying the frequency of an expert's actions and digital operations per unit of time. For example, the sum of clicks and wrist rotations per minute reflects workload.

[0099] Cognitive focus characteristics: This assesses an expert's level of concentration by analyzing the consistency and error rate of actions. For example, repeatedly clicking the same button may indicate focus, while frequently undoing actions may indicate distraction.

[0100] Environmental risk level characteristics: The degree of interference from the environment on work is determined based on noise levels and location information. For example, high noise (>60 dB) may increase cognitive burden.

[0101] Feature extraction involves data preprocessing (such as filtering and denoising) and statistical calculations. For example, a sliding window (window size of 1 second) is applied to acceleration data to calculate the mean and variance; the decibel mean is calculated for noise data; and the click interval time distribution is calculated for operation trajectories.

[0102] This step transforms complex raw data into structured features, making it easier for machine learning models to understand and process. Furthermore, the multi-dimensional design of feature vectors (intensity, focus, risk) comprehensively reflects the expert's state, avoiding the limitations of a single indicator.

[0103] Continuing with the above implementation example, the engineer's raw data over 5 minutes is processed into feature vectors:

[0104] Operational intensity characteristics: Average acceleration 2 m / s² 2 The wrist rotation frequency was 20 times / minute, the clicking frequency was 5 times / minute, and the overall operation intensity was calculated as "high" (standardized score 0.8, range 0-1).

[0105] Cognitive attention characteristics: The variance of the click interval time is 0.5 seconds, there are no obvious errors in the input notes, and the attention score is "moderate" (score 0.6). There may be slight distraction due to frequent movement.

[0106] Environmental risk level characteristics: The average noise level is 70 decibels, the location is a ward, and the environmental risk level is "high" (score 0.75), indicating that the noisy environment may interfere with the judgment.

[0107] The final multidimensional feature vector is generated as [0.8, 0.6, 0.75], for subsequent model analysis.

[0108] 106. An ensemble machine learning model is used to process multidimensional feature vectors and output the probability distribution of expert working states; among which, working states include high-pressure decision-making state, evidence-deficient state, cognitive bias state, and deep analysis state.

[0109] An ensemble machine learning model (such as a random forest or gradient boosting tree) is used to process multidimensional feature vectors, outputting a probability distribution of the expert's current working state. Working states include:

[0110] High-pressure decision-making state: Experts need to make multiple key decisions in a short period of time, characterized by high operational intensity and moderate concentration.

[0111] The state of lack of evidence: Insufficient key information leads to difficulty in judgment, which is manifested as low concentration and repetitive operation.

[0112] Cognitive bias state: Fatigue or environmental interference leads to judgment errors, which are characterized by high operational intensity but low consistency.

[0113] Deep analysis state: Experts focus on detailed analysis, characterized by high concentration and low operational intensity.

[0114] The model is trained on historical data, taking a feature vector as input and outputting the probability of each state (e.g., [0.4, 0.2, 0.3, 0.1]). The ensemble model improves prediction robustness through voting among multiple decision trees.

[0115] In this step, the ensemble model provides a probability distribution rather than a single classification, preserving the uncertainty of the states to facilitate subsequent processing. The four state designs cover common scenarios in expert surveys, ensuring the comprehensiveness of the analysis.

[0116] Continuing with the above implementation example, for the engineer's feature vector [0.8, 0.6, 0.75], the random forest model (containing 50 trees) predicts the state probability distribution:

[0117] High-pressure decision state: 0.45 (This judgment is supported by high operational intensity and high environmental risk);

[0118] State of missing evidence: 0.15;

[0119] Cognitive bias state: 0.25;

[0120] Deep analysis state: 0.15;

[0121] Maximum probability 0.45 (high pressure decision state), but below the subsequent threshold, indicating that the state is uncertain.

[0122] 107、When the maximum value of the probability distribution is below the preset threshold, activate the data gap identification signal and generate a list of missing key variables.

[0123] When the maximum value of the probability distribution is below the preset threshold (such as 0.7), it indicates that the model cannot confidently determine the state of the expert, which may be due to insufficient or ambiguous data. At this time, the data gap identification signal is activated, and a list of missing key variables is generated, listing the information that needs to be supplemented (such as "environmental interference degree" or "operation purpose").

[0124] In this step, the threshold mechanism avoids low confidence prediction, ensuring the reliability of state judgment. The missing variable list accurately locates the data deficiency point, providing a clear direction for subsequent supplementation.

[0125] Continuing the above implementation example, the maximum value of the state probability of the engineer is 0.4 < 0.7, triggering the data gap identification signal. The system analyzes the feature vector and finds that the "cognitive concentration" feature (0.6) is not clear due to the lack of operation purpose data, generating a list of missing variables: ["current task pressure perception", "environmental interference subjective evaluation"].

[0126] 108、In response to the data gap identification signal, perform:

[0127] 401、Generate an interaction request with no more than two questions, which are dynamically generated based on the list of missing key variables.

[0128] 402、Based on voice shortcut options or single gesture swipe input methods, assist the expert in inputting supplementary data based on the mobile terminal and wearable devices carried.

[0129] 403、Fuse the supplementary data with the multi-dimensional feature vector, and use the integrated machine learning model to recalculate the expert's working state based on the fusion result.

[0130] In steps 401-402, no more than two questions are dynamically generated based on the list of missing variables, such as "Do you feel stressed by the task?" or "Is the environmental noise disturbing you?" The expert quickly answers through voice shortcut options (such as "Yes / No") or gesture sliding (such as left swipe No / right swipe Yes), reducing the operational burden. The supplementary data is fused with the original feature vector, and the integrated model is re-run to calculate the state. The interactive request is concise and efficient, with a maximum of two questions to reduce the burden on the expert. Voice / gesture input adapts to the on-site environment, improving response speed. In addition, data fusion ensures that the state judgment considers both new and old information, improving accuracy.

[0131] Continuing the above implementation example, the system generates the questions: "Do you feel stressed by the task?" and "Is the environmental noise disturbing you?" The engineer answers "Yes" and "Yes" by voice. After quantifying the supplementary data (stress perception: 0.8, disturbance evaluation: 0.7), it is fused into a new vector: [0.8, 0.68, 0.75]. The model recalculates the probability distribution: [0.65, 0.1, 0.2, 0.05], confirming the "high-pressure decision state" (0.65).

[0132] 404. According to the final determination of the expert's working state, the optimal execution strategy is matched from the pre-defined strategy library; wherein the strategy library contains cognitive load reduction protocol, counter perspective injection protocol, evidence chain completion protocol and deep analysis support protocol.

[0133] According to the final state, the optimal execution strategy is matched from the strategy library:

[0134] Cognitive load reduction protocol: simplify the interface, provide automatic summary, reduce the load.

[0135] Counter perspective injection protocol: provide counterexamples or different viewpoints to correct biases.

[0136] Evidence chain completion protocol: search for supplementary information to fill gaps in the evidence.

[0137] Deep analysis support protocol: provide advanced tool support for detailed analysis.

[0138] The strategy is converted into instructions (such as adjusting the interface layout) and pushed to the device.

[0139] In this step, personalized strategies solve expert problems and improve survey efficiency. Instruction pushing realizes seamless support and reduces manual adjustment time.

[0140] Continuing the above implementation example, the engineer's state is "high-pressure decision state", and the "cognitive load reduction protocol" is matched. Instruction pushing: hide non-critical information on the smartphone interface and display an automatic summary of flow rate anomalies (such as "Abnormal times: 5, Average deviation: 12%").

[0141] 405、Convert the optimal execution strategy into executable instructions and push it to the carried mobile terminal and wearable device.

[0142] The above method uses artificial intelligence technology to monitor the behavior and environmental data of medical device adverse event investigation experts in real time, dynamically analyzes their working state, and provides personalized support strategies. This innovative application has unique value in the medical device adverse event investigation scenario of the present application: through multi-dimensional data collection (such as physical behavior, digital operation trajectory, and environmental context), the system can accurately capture the state changes of experts in high-pressure, insufficient evidence, or noisy environments, and immediately push customized strategies, such as simplifying the interface or supplementing the evidence chain. This not only significantly improves the investigation efficiency and accuracy, but also effectively reduces the cognitive burden of experts in complex scenarios, embodying the deep integration of technology and professional needs, and injecting intelligent innovation into traditional investigation methods.

[0143] Figure 4 A schematic diagram of a medical device adverse event risk monitoring system based on artificial intelligence is provided for the embodiments of the present application, as shown in Figure 4 The system comprises:

[0144] An artificial intelligence monitoring module 100 for monitoring medical device adverse event risk signals in the jurisdiction based on artificial intelligence;

[0145] An investigation task scheduling module 200 for generating investigation tasks for medical device adverse event risk signals, and matching and scheduling experts to perform tasks;

[0146] A disposal recommendation module 300 for recommending disposal recommendations based on the investigation information of the experts.

[0147] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. An artificial intelligence-based medical device adverse event risk monitoring method, characterized by, The method comprises the following steps: Based on artificial intelligence, monitor the medical device adverse event risk signal in the jurisdiction; Generate an investigation task for the medical device adverse event risk signal, and match and dispatch experts to perform the task; Based on the investigation information of the experts, recommend a disposal suggestion. 2.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein, The method based on artificial intelligence, monitoring the medical device adverse event risk signal in the jurisdiction, comprises the following steps: Based on the pre-trained machine learning model, according to the device use state information collected by the edge nodes deployed in different medical device users, monitor the medical device adverse event risk signal in the jurisdiction. 3.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein, The method for generating an investigation task for a medical device adverse event risk signal and matching and dispatching experts to perform the task comprises the following steps: Based on the preset task generation template, generate an investigation task according to the medical device adverse event risk signal; Match experts suitable for performing the investigation task from a preset expert library; Dispatch the experts to perform the investigation task. 4.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein, The method for recommending a disposal suggestion based on the investigation information of the experts comprises the following steps: Based on the investigation information of the experts, perform risk assessment; Based on the risk classification disposal engine, determine a disposal suggestion according to the risk assessment result. 5.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein, The method further comprises the following steps: Based on the medical device adverse event risk signal and the investigation information of the experts, generate an evidence chain; Use a blockchain to store the evidence chain. 6.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein The method further comprises the following steps: Based on the investigation information of the experts, automatically fill in a preset medical device adverse event risk signal investigation form; Push the automatically filled preset medical device adverse event risk signal investigation form to the experts. 7.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein The medical device adverse event risk signal at least includes product defects, improper use, and causes or may cause serious harm of medical device adverse events. 8.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein, The investigation information at least includes the source of the reported patient ward, the original disease, the original signs, various examination data, treatment conditions, past medical history, adverse event occurrence, occurrence time, damage / failure performance, adverse event consequences, combined drug / device, and risk control measures taken. 9.The artificial intelligence-based medical device adverse event risk monitoring method of claim 1, wherein, The expert library at least contains pre-selected personnel from medical institutions, medical device manufacturers, and medical device adverse event monitoring agencies.

10. An artificial intelligence-based medical device adverse event risk monitoring system, characterized by, The method comprises the following steps: An artificial intelligence monitoring module is used to monitor the medical device adverse event risk signal in the jurisdiction based on artificial intelligence; An investigation task scheduling module is used to generate an investigation task for the medical device adverse event risk signal, and match and dispatch experts to perform the task; A disposal suggestion recommendation module is used to recommend a disposal suggestion based on the investigation information of the experts.

Images