Cerebrovascular disease health monitoring method based on intelligent agent

Abstract

The invention discloses a cerebrovascular disease health monitoring method based on an intelligent agent. The method comprises the following steps: step 1, acquiring multi-modal data in real time and preprocessing the multi-modal data; 2, converting the preprocessed multi-modal data into structured health status characterization, and identifying potential cerebrovascular risks; 3, analyzing through a cerebral apoplexy AI model, and outputting a risk level; 4, performing personalized early warning strategy planning based on the risk level; 5, converting an instruction of personalized early warning strategy planning into a cross-platform executable action; and 6, aggregating the desensitization data of the global cerebral apoplexy AI model at the cloud through federal learning, and updating the clinical knowledge graph rule base. According to the method, the health state of the patient is monitored by fusing the multi-modal data, the cerebral apoplexy AI model and the clinical mapping knowledge domain, patients with high-risk and medium-risk cerebrovascular diseases are screened out in time, and medical response is provided in time.

Description

Technical Field

[0001] The present invention relates to the technical field of medical health monitoring, and in particular to a cerebrovascular disease health monitoring method based on an intelligent agent. Background Art

[0002] Traditional medical health monitoring technology relies on single devices and intermittent manual operation (such as blood pressure monitors and electrocardiographs). This presents the following core issues: Poor real-time performance: Traditional devices cannot continuously monitor physiological parameters and struggle to capture dynamic changes. For example, in the event of a sudden cerebrovascular disease (stroke), intermittent measurements may miss the optimal early warning window. Data is isolated and fragmented, and data collected by different devices (such as blood pressure, blood oxygen, and cerebral blood flow) lacks integrated analysis, resulting in incomplete health status assessments. Traditional monitoring relies on doctors' empirical judgment of static data and lacks the support of dynamic AI-based models, resulting in a high risk of misdiagnosis or missed diagnosis. Summary of the Invention

[0003] To this end, the present invention provides an intelligent agent-based cerebrovascular disease health monitoring method to solve the problems in the prior art.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] The agent-based cerebrovascular disease health monitoring method includes the following steps:

[0006] Step 1: Acquire multimodal data in real time and preprocess the multimodal data at the edge.

[0007] Step 2: Convert the preprocessed multimodal data into a structured health status representation, model the health status, and identify potential cerebrovascular risks;

[0008] Step 3: Receive a health status representation with potential cerebrovascular risk, analyze it using the stroke AI model, and output a risk level;

[0009] Step 4: Plan a personalized early warning strategy based on the risk level output by the stroke AI model;

[0010] Step 5: Convert the instructions of the personalized early warning strategy planning into cross-platform executable actions; and feed the action execution logs into the stroke AI model;

[0011] Step 6: Aggregate the desensitized data of the global stroke AI model through federated learning in the cloud, update the clinical knowledge graph rule base, and adjust the attention weight of the stroke AI model.

[0012] Further: In step 1, the multimodal data includes user physiological parameters and environmental data; physiological parameters are collected through wearable devices, and the physiological parameters include cerebral hemodynamic indicators, HRV, SpO2 and blood pressure; environmental data is collected by calling the device API, and the environmental data includes temperature, humidity and altitude.

[0013] Further: The specific steps of modeling the health status are as follows:

[0014] (1) Fusion of feature vectors and knowledge graphs: Receive the preprocessed 128-dimensional feature vectors; extract key time series features, including the morning peak fluctuation pattern of blood pressure and the trend of changes in cerebral blood volume at night; then activate the clinical knowledge graph, query the pathology association library, and embed medical logic rules;

[0015] (2) Spatiotemporal attention weighted analysis: Calculate parameter coupling weights and use a multi-head attention mechanism to analyze cross-modal associations; generate spatiotemporal heat maps, output weight matrices to identify high-risk periods, and mark abnormal spatial patterns;

[0016] (3) Encode the health status, integrate the attention-weighted features and knowledge graph semantics, construct the state vector, and then calculate the cerebrovascular autoregulation index (CAI) to quantify the health standard;

[0017] Classify health status according to quantitative health indicators.

[0018] Furthermore: when the health status is normal, the baseline is updated using exponential smoothing; for abnormal health status data, the baseline is frozen and pathological analysis is started.

[0019] Furthermore: The specific reasoning and analysis process of the stroke AI model is as follows:

[0020] (1) Dual-branch feature extraction, including LSTM temporal branch processing and CNN waveform branch processing;

[0021] LSTM time series branch processing: Input health status code; analyze 72-hour physiological parameter trends, detect morning blood pressure fluctuation patterns, and identify abnormal cerebral hemodynamic cycles;

[0022] CNN waveform branch processing: Analyze high-resolution raw waveforms, specifically capturing arrhythmia segments and detecting abnormal cerebral blood flow pulsation;

[0023] (2) Multimodal feature fusion, using temporal interpolation to resolve sampling rate differences and construct a three-dimensional feature cube; the three dimensions of the three-dimensional feature cube are: time window × physiological parameter × spatial position;

[0024] (3) Identify risk hotspots by calculating attention weight distribution;

[0025] (4) Based on the identified risk hotspots, a risk decision tree is used to make hierarchical decisions on the health status data; the risks are divided into three levels: high risk, medium risk and low risk.

[0026] Further: The judgment of high risk, medium risk and low risk is as follows:

[0027] When systolic blood pressure is >180 mmHg and cerebral blood volume is <40 ml / 100 g / min, the patient is directly marked as high risk;

[0028] When the cerebral blood flow regulation index CAI is less than 0.5, it is directly marked as medium risk; when the cerebral blood flow regulation index is greater than 0.5, but there are more than two abnormal labels, it is marked as medium risk;

[0029] When the cerebral blood flow regulation index is greater than 0.5 and there are no more than two abnormal labels, it is marked as low risk;

[0030] In addition, when there is a history of thrombolytic therapy and D-dimer is >500 μg / L, the risk level will be increased by one level.

[0031] Furthermore: in step 4, when the risk is high, it is necessary to activate the thrombolysis warning, push it to the emergency center, and share the HIS case;

[0032] When the risk is medium, 24-hour intensive monitoring will be initiated, the data sampling frequency will be increased, and monitored individuals will be reminded to take medication. At the same time, remote consultations will be scheduled.

[0033] When risk is low, update personalized baseline; generate health improvement recommendations.

[0034] Further: In step 6, the parameters of the updated global stroke AI model are sent to the terminal, and the terminal adapts to the differences between the population through transfer learning.

[0035] The present invention has the following advantages: By integrating multimodal data, stroke AI models and clinical knowledge graphs, the present invention monitors the patient's health status, timely screens out patients with high-risk and medium-risk cerebrovascular diseases, and provides timely medical responses, thereby significantly advancing the detection time of cerebrovascular disease patients and improving early warning efficiency.

[0036] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] To more intuitively illustrate the prior art and the present application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be considered as limiting conditions for implementing the present application; for example, based on the technical concepts disclosed in this application and the exemplary drawings, those skilled in the art are capable of easily making routine adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components).

[0038] Figure 1 This is a flowchart of the implementation of the intelligent agent-based cerebrovascular disease health monitoring method provided in this application.

[0039] Figure 2 The present invention provides a flowchart for preprocessing multimodal data in an intelligent agent-based cerebrovascular disease health monitoring method according to an embodiment of the present invention.

[0040] Figure 3 This is a flowchart for implementing health status modeling in the agent-based cerebrovascular disease health monitoring method according to an embodiment of the present invention.

[0041] Figure 4 This is a flowchart for implementing the stroke AI model in the agent-based cerebrovascular disease health monitoring method according to an embodiment of the present invention.

[0042] Figure 5 This is an architectural diagram of the stroke AI model in the agent-based cerebrovascular disease health monitoring method according to an embodiment of the present invention.

[0043] Figure 6 Flowchart for the implementation of personalized early warning strategy planning in the agent-based cerebrovascular disease health monitoring method according to an embodiment of the present invention

[0044] Figure 7 Flowchart for the implementation of multi-protocol action execution in the agent-based cerebrovascular disease health monitoring method according to an embodiment of the present invention

[0045] Figure 8 This is a block diagram of the memory bank in the intelligent agent-based cerebrovascular disease health monitoring method of the present invention. DETAILED DESCRIPTION

[0046] The following specific embodiments illustrate the implementation of the present invention. People familiar with this technology can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. It should be understood that these embodiments are only to further illustrate the present invention and cannot be understood as limiting the scope of protection of the present invention. Technical engineers in this field can make some non-essential improvements and adjustments to the present invention based on the content of the above invention; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0047] See also Figures 1-8 , an agent-based cerebrovascular disease health monitoring method, comprising the following steps:

[0048] Step 1: Multimodal data acquisition and preprocessing: real-time acquisition of user physiological and environmental data;

[0049] Physiological parameters are collected through wearable devices (fNIRS headband, ECG patch, etc.), including cerebral hemodynamic indicators (CBF, HbO2), HRV, SpO2, and blood pressure; environmental data is collected by calling the device API to collect temperature, humidity, and altitude.

[0050] See Figure 2 ,The preprocessing of multimodal data is performed at the edge end, and the ,preprocessing process of multimodal data is as follows: Kalman filter ,noise reduction→time stamp alignment→time domain / frequency domain feature extraction→generate ,standardized feature vector (128-dimensional feature vector).

[0051] Step 2: Dynamic health state modeling: Convert multimodal physiological data into structured health state representations to identify potential cerebrovascular risks.

[0052] See Figure 3 ,The specific implementation of dynamic health status modeling is as follows:

[0053] (1) Fusion of feature vectors and knowledge graphs: Receive the preprocessed 128-dimensional feature vector (including indicators such as cerebral hemodynamics, HRV, and blood pressure); extract key time series features, including the morning peak fluctuation pattern of blood pressure and the trend of changes in cerebral blood volume at night; then activate the clinical knowledge graph, query the pathological association library (such as "sudden drop in systolic blood pressure → cerebral hypoperfusion → ischemic stroke"), and embed medical logic rules (IF blood oxygen saturation <90% AND cerebral blood flow decrease rate >15% / min THEN risk label = "acute hypoxia").

[0054] (2) Spatiotemporal attention weighted analysis: Calculate parameter coupling weights and use a multi-head attention mechanism to analyze cross-modal associations.

[0055] Among them, key relationship 1: phase synchronization between heart rate variability (HRV) and cerebral blood flow velocity; key relationship 2: time lag effect between blood pressure fluctuation and oxygenated hemoglobin (HbO) concentration.

[0056] Generate spatiotemporal heat maps and output weight matrices to identify high-risk periods (such as the blood pressure trough period between 3 and 5 a.m.) and mark abnormal spatial patterns (such as left hemisphere blood flow asymmetry >25%).

[0057] (3) Encode the health status, fuse the attention-weighted features and knowledge graph semantics, construct the state vector, and then calculate the cerebrovascular autoregulation index (CAI) to quantify the health standard.

[0058] (4) Classify health status according to quantitative health indicators. The health status classification is shown in Table 1 below:

[0059] Table 1

[0060] Status Category Judgment conditions Response Action normal CAI∈[0.8,1.2] and no abnormal labels Update the personalized baseline model Low-risk abnormalities Single tag trigger and CAI>0.6 Mark risks and push health advice to the app High-risk abnormalities Multi-label superposition or CAI≤0.5 Activating AI emergency reasoning for stroke

[0061] When the health status is normal, exponential smoothing is used to update the baseline; when the health status data is abnormal, the baseline is frozen and pathological analysis is started.

[0062] Step 3: Read Figure 4-Figure 6 ,Stroke AI model reasoning: Receive data encoded with risk labels (abnormal states) and analyze it through the stroke AI model.

[0063] The specific reasoning and analysis process of the stroke AI model is as follows:

[0064] 1) Dual-branch feature extraction, including LSTM temporal branch processing and CNN waveform branch processing.

[0065] LSTM time-series branch processing: Input health status code (including spatiotemporal attention weighted features); analyze 72-hour physiological parameter trends, detect morning blood pressure surge patterns (morning systolic blood pressure rise rate >10 mmHg / h), and identify abnormal cerebral hemodynamic cycles (such as nocturnal CBF sustained <45 ml / 100 g / min). CNN waveform branch processing: parse high-resolution raw waveforms, specifically including ECG and fNIRS;

[0066] ECG: capture arrhythmia fragments (QT interval prolongation >450ms);

[0067] fNIRS: Detects abnormal cerebral blood flow pulsatility (waveform peak-to-trough ratio > 2.5).

[0068] 2) Multimodal feature fusion, including cross-domain feature alignment and 3D feature cube construction.

[0069] Cross-domain feature alignment: using temporal interpolation to address sampling rate differences (ECG 250Hz → fNIRS 10Hz);

[0070] The key feature correlation matrix is ​​shown in Table 2 below:

[0071] Table 2

[0072]

[0073]

[0074] Three-dimensional feature cube construction: The three dimensions are: time window × physiological parameter × spatial position (such as left / right cerebral hemisphere). In this embodiment, the output fusion feature tensor is 32×16×8.

[0075] See Figure 5 、 Figure 8 ,The memory bank is an intelligent medical memory system that supports real-time decision-making and long-term learning,

[0076] Short-term memory function: It can realize real-time status caching and rapid retrieval; create a rolling buffer area with a time window of 30 minutes to store the original physiological data stream (sampling rate 10Hz).

[0077] Long-term memory function: It can build medical knowledge graphs and user health records; in addition, when doctors correct system warnings, they mark the error type (false positive / false negative) and extract key features to generate new rules.

[0078] Through the "short-term-long-term" memory collaboration and federated evolution mechanism, the problems of "data islands" and "static knowledge bases" in traditional medical systems have been solved.

[0079] 3) Multi-headed attention pathological focus: identifying risk hotspots by calculating attention weight distribution; the high-weight period is 3-5 am (high-risk period for blood pressure fluctuations); the high-weight parameter is cerebral vascular resistance (CVR>1.2 mmHg·s / ml).

[0080] 4) Risk decision tree classification: Based on the identified risk hotspots, the risk decision tree is used to make hierarchical decisions on health status data; risks are divided into three levels: high risk, medium risk and low risk.

[0081] When systolic blood pressure is >180 mmHg and cerebral blood volume is <40 ml / 100 g / min, the patient is directly marked as high risk (confidence >99%); when the cerebral blood flow regulation index CAI is less than 0.5, the patient is directly marked as medium risk; when the cerebral blood flow regulation index is greater than 0.5 but there are two or more abnormal labels, the patient is marked as medium risk;

[0082] When the cerebral blood flow regulation index is greater than 0.5 and there are no more than two abnormal labels, it is marked as low risk.

[0083] In addition, if there is a history of thrombolytic therapy and the D-dimer is >500 μg / L, the risk level is increased by one level (i.e., low risk is increased to medium risk, and medium risk is increased to high risk).

[0084] Step 4: Personalized early warning strategy planning, generating the optimal response path based on the risk level output by the stroke AI model.

[0085] When the risk is high, it is necessary to activate the thrombolysis warning (pushed to the emergency center) and share the HIS case;

[0086] When the risk is medium, 24-hour intensive monitoring will be initiated, the data sampling frequency will be increased, and monitored individuals will be reminded to take medication. At the same time, remote consultations will be scheduled.

[0087] When the risk is low, the personalized baseline is updated (exponential smoothing algorithm); health improvement recommendations (such as sleep optimization plans) are generated.

[0088] Step 5: Multi-protocol action execution: Convert planning instructions into cross-platform executable actions to achieve seamless collaboration between the medical system and terminal devices; distribute response instructions across platforms based on personalized early warning strategy planning, and feed back action execution logs to the stroke AI model.

[0089] During the execution process, it is necessary to monitor the instruction status and track the "delivery-execution-confirmation" chain in real time.

[0090] If the command fails to be issued, a retry is required, and the backup communication protocol (such as MQTT→CoAP) is automatically switched.

[0091] In addition, the command semantics need to be analyzed to parse the JSON commands (including action type, target device, and urgency) output by the stroke AI model.

[0092] The key fields identify the following: action_type: warning push / device control / data synchronization;

[0093] target_protocol: HL7 / BLE / FCM; priority_level: 1-5 (1 is the highest).

[0094] The protocol adaptation layer conversion execution includes HL7 / FHIR engine processing and IoT control interface execution.

[0095] The HL7 / FHIR engine processes the risk warning, converting it into an FHIR resource type. When the message is encapsulated, the patient ID, timestamp, and institution code are automatically filled in.

[0096] IoT control interface execution includes compiling device instructions and regulating biofeedback.

[0097] In this embodiment, for example, the wearable device control instruction (in binary format) can be expressed as: 0x010xA20xFF; which means increasing the fNIRS sampling rate to 10 Hz.

[0098] In this embodiment, the transcranial direct current stimulation (tDCS) parameters may be dynamically adjusted; for example, the current intensity parameter may be adjusted as follows: current intensity = min (2 mA, 0.5*CBF_deficit).

[0099] The key distribution paths for cross-platform instructions include hospital HIS systems, wearable devices, and doctor terminals.

[0100] Among them, the hospital HIS system: real-time push thrombolysis warning package (including 3D cerebral blood flow heat map); automatic reservation of emergency resources (including CT room, stroke green channel);

[0101] Wearable devices: Emergency mode activated, vibration alarm intensity increased by 300%, LED red light flashing frequency 5Hz;

[0102] Doctor terminal: Multimodal warning information push, including APP pop-up window, voice call and smartwatch vibration.

[0103] Step 6: Federated learning model evolution: The cloud aggregates desensitized data through federated learning, updates the clinical knowledge graph rule base, and adjusts the attention weight of the stroke AI model; updates the parameters of the global stroke AI model and sends them to the terminal. The terminal adapts to population differences (elderly / hypertensive patients) through transfer learning.

[0104] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cerebrovascular disease health monitoring method based on an intelligent agent, characterized in that: The following steps are involved: Step 1: Acquire multimodal data in real time; And pre-process the multimodal data at the edge; Step 2: Convert the preprocessed multimodal data into a structured health status representation, model the health status, and identify potential cerebrovascular risks; Step 3: Receive a health status representation with potential cerebrovascular risk, analyze it using the stroke AI model, and output a risk level; Step 4: Plan a personalized early warning strategy based on the risk level output by the stroke AI model; Step 5: Convert the instructions of the personalized early warning strategy planning into cross-platform executable actions; and feed the action execution logs into the stroke AI model; Step 6: Aggregate the desensitized data of the global stroke AI model through federated learning in the cloud, update the clinical knowledge graph rule base, and adjust the attention weight of the stroke AI model.

2. The agent-based cerebrovascular disease health monitoring method according to claim 1, characterized in that: In step 1, the multimodal data includes user physiological parameters and environmental data; physiological parameters are collected through wearable devices, and the physiological parameters include cerebral hemodynamic indicators, HRV, SpO2, and blood pressure; environmental data is collected by calling the device API, and the environmental data includes temperature, humidity, and altitude.

3. The agent-based cerebrovascular disease health monitoring method according to claim 1, characterized in that: The specific steps for modeling the health status are as follows: (1) Fusion of feature vectors and knowledge graphs: Receive the preprocessed 128-dimensional feature vectors; extract key time series features, including the morning peak fluctuation pattern of blood pressure and the trend of changes in cerebral blood volume at night; then activate the clinical knowledge graph, query the pathology association library, and embed medical logic rules; (2) Spatiotemporal attention weighted analysis: Calculate parameter coupling weights and use a multi-head attention mechanism to analyze cross-modal associations; Generate spatiotemporal heat maps, output weight matrices to identify high-risk periods, and mark abnormal spatial patterns; (3) Encode the health status, integrate the attention-weighted features and knowledge graph semantics, construct the state vector, and then calculate the cerebrovascular autoregulation index (CAI) to quantify the health standard; Classify health status according to quantitative health indicators.

4. The agent-based cerebrovascular disease health monitoring method according to claim 3, characterized in that: When the health status is normal, the baseline is updated using exponential smoothing; for abnormal health status data, the baseline is frozen and case analysis is started.

5. The agent-based cerebrovascular disease health monitoring method according to claim 1, characterized in that: The specific reasoning and analysis process of the stroke AI model is as follows: (1) Dual-branch feature extraction, including LSTM temporal branch processing and CNN waveform branch processing; LSTM time series branch processing: Input health status code; analyze 72-hour physiological parameter trends, detect morning blood pressure fluctuation patterns, and identify abnormal cerebral hemodynamic cycles; CNN waveform branch processing: Analyze high-resolution raw waveforms, specifically capturing arrhythmia segments and detecting abnormal cerebral blood flow pulsation; (2) Multimodal feature fusion, using temporal interpolation to address sampling rate differences and construct a three-dimensional feature cube; The three dimensions of the 3D feature cube are: time window × physiological parameter × spatial position; (3) Identify risk hotspots by calculating attention weight distribution; (4) Based on the identified risk hotspots, a risk decision tree is used to make hierarchical decisions on the health status data; the risks are divided into three levels: high risk, medium risk and low risk.

6. The agent-based cerebrovascular disease health monitoring method according to claim 5, characterized in that: The judgment of high risk, medium risk and low risk is as follows: When systolic blood pressure is >180 mmHg and cerebral blood volume is <40 ml / 100 g / min, the patient is directly marked as high risk; When the cerebral blood flow regulation index CAI is less than 0.5, it is directly marked as medium risk; when the cerebral blood flow regulation index is greater than 0.5, but there are more than two abnormal labels, it is marked as medium risk; When the cerebral blood flow regulation index is greater than 0.5 and there are no more than two abnormal labels, it is marked as low risk; In addition, when there is a history of thrombolytic therapy and D-dimer is >500 μg / L, the risk level will be increased by one level.

7. The agent-based cerebrovascular disease health monitoring method according to claim 1, characterized in that: In step 4, when the risk is high, a thrombolysis warning needs to be activated, sent to the emergency center, and the HIS case is shared; When the risk is medium, 24-hour intensive monitoring will be initiated, the data sampling frequency will be increased, and monitored individuals will be reminded to take medication. At the same time, remote consultations will be scheduled. When the risk is low, update the personalized baseline; Generate health improvement recommendations.

8. The agent-based cerebrovascular disease health monitoring method according to claim 1, characterized in that: In step 6, the parameters of the updated global stroke AI model are sent to the terminal, and the terminal adapts to the differences among the population through transfer learning.

Images