Intelligent monitoring and risk early warning system for acute and critical diseases of cardiovascular diseases

Abstract

The invention relates to an intelligent monitoring and risk early warning system for cardiovascular acute and critical diseases, which is characterized in that electrocardiosignals are acquired in real time through a wearable twelve-lead ECG monitor (ADS1298 chip, 512Hz sampling), and the electrocardiosignals are encrypted and transmitted to a cloud end through an MQTT protocol in combination with edge end wavelet denoising and QRS wave detection. And the cloud end adopts a multi-modal fusion engine, integrates a random forest-XGBoost hybrid model, performs modeling based on a postoperative patient physiological index time sequence, predicts the risk through Kalman filtering and triggers secondary early warning. Clinical tests in a third-A hospital prove that the end-to-end response time is 2.8 seconds, the D2B time is shortened to 70 minutes, and the occurrence rate of main adverse cardiovascular events in 30 days after operation is reduced by 47%. According to the invention, the limitation of traditional single-mode monitoring is broken through, a whole-process closed loop of collection-analysis-intervention is constructed, and an intelligent solution is provided for prevention and control of cardiovascular acute and critical diseases.

Description

Technical Field

[0001] This invention relates to the field of medical systems, specifically to an intelligent monitoring and risk warning system for acute and critical cardiovascular diseases. Background Technology

[0002] Cardiovascular disease is the leading cause of death worldwide, with acute and critical illnesses such as acute myocardial infarction and malignant arrhythmias characterized by rapid onset, rapid progression, and high mortality. Traditional treatment models suffer from two major pain points: a lack of continuous monitoring equipment outside of hospitals, and in-hospital monitoring equipment such as Holter monitors having limited leads (typically only 1-3 leads), making it difficult to capture multidimensional ECG characteristics; existing systems rely on manual interpretation or offline analysis, resulting in time delays from data collection to risk identification, potentially missing the golden window for treatment.

[0003] In terms of data acquisition dimensions, traditional electrocardiogram (ECG) monitoring devices mostly adopt single-lead or three-lead designs, which can only acquire limited ECG vector information, thus limiting their ability to locate and assess the severity of diseases such as myocardial ischemia and arrhythmias. For example, although the myocardial infarction detection algorithm proposed by the F. Heinrich Hertz Institute in Germany achieves high accuracy, it relies on specific lead combinations; domestically developed wearable devices, such as the low-power system developed by Fuzhou University, are still mainly three-lead, which is insufficient to meet the comprehensive monitoring needs of acute and critical illnesses.

[0004] Another key issue is the weak multimodal data fusion capability. Cardiovascular disease diagnosis requires the integration of electrocardiogram signals, biomarkers (such as troponin), imaging data (CT / MRI), and genetic information. However, existing systems suffer from data silos; data from hospital information systems (HIS), picture archiving systems (PACS), and wearable devices are not interconnected, and there is a lack of real-time dynamic weighted fusion mechanisms, resulting in low feature extraction efficiency. While institutions such as Yale University in the United States have attempted multimodal prediction, they have not yet established effective fusion algorithms.

[0005] The poor generalization ability of early warning models is also a prominent issue. Existing AI diagnostic models are mostly trained on a single database, such as the PTB-ECG dataset, which carries the risk of overfitting. The accuracy decreases when validated across ethnicities and age groups, and the models do not take into account changes in indicators during the postoperative recovery period of patients, resulting in fixed early warning thresholds.

[0006] The trade-off between device portability and comfort also needs to be addressed. Traditional monitoring devices, such as the 24-hour Holter monitor, are bulky, have a high electrode detachment rate, and cause significant skin irritation, resulting in low patient compliance. While the soft, stretchable device developed by City University of Hong Kong improves comfort, the number of leads is still limited.

[0007] To address the aforementioned issues, the applicant proposes an intelligent monitoring and risk warning system for acute and critical cardiovascular diseases. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent monitoring and risk warning system for acute and critical cardiovascular diseases, in order to solve the problems in the prior art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an intelligent monitoring and risk warning system for acute and critical cardiovascular diseases, comprising a clinical database for acute and critical cardiovascular diseases, a data acquisition and processing module, an AI rapid diagnostic model, a multimodal data fusion warning platform, and a signal transmission mechanism;

[0010] The cardiovascular critical care clinical database stores patients' clinical medical history data, CT imaging data, gene sequencing data, and disease outcome information;

[0011] The data acquisition and processing module consists of a wearable multi-lead intelligent ECG signal monitoring instrument and an intelligent mobile terminal. The monitoring instrument uses a twelve-lead electrode wire and an ADS1298 acquisition chip to transmit ECG signals to the mobile terminal via WiFi or Bluetooth.

[0012] The AI ​​rapid diagnostic model integrates a machine learning framework optimized with random forest, XGBoost, support vector machine, and genetic algorithm for feature selection and disease prediction.

[0013] The multimodal data fusion early warning platform integrates electrocardiogram data, biochemical data, CT data, and genetic data, and achieves real-time data interaction and risk assessment through the MQTT protocol;

[0014] The system transmits the signal to the data center server after denoising and QRS wave detection preprocessing via a mobile terminal. The AI ​​model then outputs the triage results and warning levels.

[0015] Optionally, the hardware configuration of the wearable multi-lead intelligent ECG signal monitoring instrument includes a main control chip ESP32, a disposable medical adhesive ten-electrode twelve-lead configuration, and an ADS1298 chip signal acquisition circuit with a sampling accuracy of 24 bits.

[0016] Optionally, the data transmission mechanism includes transmitting ECG signals to the mobile terminal via Bluetooth / WiFi, uploading data from the mobile terminal to the data center via the Internet or the hospital's intranet, and achieving lightweight real-time communication between the data center and the mobile terminal using the MQTT protocol.

[0017] Optionally, the training process of the AI ​​rapid diagnostic model is as follows: construct a multimodal feature set based on a cardiovascular clinical database, optimize the combined weights of random forest and XGBoost through a genetic algorithm, and combine logistic regression to classify and predict the selected features and output the risk probability of acute and critical illness.

[0018] Optionally, the functions of the multimodal data fusion early warning platform include real-time integration of electrocardiogram signals, gene markers, biochemical indicators and CT image features, calculation of the patient's comprehensive risk score using a dynamic weighted fusion algorithm, and bidirectional push of early warning information through mobile terminals and hospital systems when the risk score exceeds a threshold.

[0019] Optionally, the preprocessing operations of the smart mobile terminal include wavelet transform denoising, QRS wave detection based on the Pan-Tompkins algorithm, heart rate variability feature extraction, and signal quality assessment.

[0020] Optionally, the system also includes a personalized early warning threshold adjustment module for postoperative patients, a remote emergency data transmission interface based on a 5G network, and a clinical decision support interface for interfacing with the hospital's HIS system.

[0021] Beneficial effects: 1. Breakthrough in diagnostic efficacy

[0022] By integrating multimodal data (ECG / genes / imaging) with AI hybrid models, the accuracy of identifying acute and critical illnesses is improved, solving the problem of high false negative rates in traditional single-modal detection.

[0023] 2. Real-time early warning innovation

[0024] A real-time channel for the entire process of "collection-processing-early warning" is built, which can complete risk assessment and trigger two-way early warning (patient end / hospital end) within 3 seconds, improving the response speed compared to the manual interpretation mode.

[0025] 3. Enhanced device experience

[0026] The 12-lead wearable design achieves a balance between medical-grade precision and portability, resulting in lighter devices, longer battery life, and significantly improved patient compliance.

[0027] 4. Intelligent model adaptation

[0028] The AI ​​model, trained on multi-center data, achieves a high consistency rate in cross-institutional diagnosis, and the dynamic threshold adjustment mechanism improves the accuracy of postoperative patient early warning.

[0029] 5. Optimize healthcare efficiency

[0030] By establishing a closed-loop management process, the efficiency of emergency triage is improved, the misdiagnosis rate is reduced, the follow-up cost is decreased, and the transformation of the cardiovascular disease prevention and control model is promoted.

[0031] This invention achieves a significant improvement in key indicators such as diagnostic accuracy, response speed, and ease of use through collaborative innovation in hardware algorithms, providing a revolutionary solution for the prevention and control of acute and critical cardiovascular diseases. Attached Figure Description

[0032] Figure 1This is a schematic diagram of the data acquisition and processing module in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the AI ​​rapid diagnosis model process according to an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the multimodal data fusion early warning platform process according to an embodiment of the present invention;

[0035] Figure 4 This is a schematic diagram of the signal transmission mechanism in an embodiment of the present invention;

[0036] Figure 5 This is a schematic diagram of the postoperative patient personalized early warning module according to an embodiment of the present invention; Detailed Implementation

[0037] The preferred embodiments of the present invention are described below with reference to the accompanying drawings to make the technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0038] This invention originates from the Key R&D Program of Jiangxi Province, China (No. 20243BBI91022).

[0039] This invention provides an intelligent monitoring and risk warning system for acute and critical cardiovascular diseases. Through deep collaborative design of hardware, algorithms, and the system itself, it achieves intelligent management of the entire process from data acquisition to clinical decision-making. The system adopts a layered modular architecture, divided into a data acquisition layer, an edge processing layer, a cloud analysis layer, and a user interaction layer. Each layer achieves data interoperability through standardized protocols. The data acquisition layer consists of a wearable 12-lead ECG monitor, a portable biochemical detection module, and an interface to a hospital imaging system, responsible for collecting ECG signals, biochemical indicators such as troponin, and CT image data. The edge processing layer is deployed on a smart mobile terminal, incorporating signal preprocessing algorithms and a lightweight AI model to perform data denoising, QRS wave detection, and preliminary risk assessment. The cloud analysis layer integrates a multimodal fusion engine and a deep learning model to perform feature extraction, model training, and dynamic threshold calculation on uploaded data. The user interaction layer displays warning information, diagnostic reports, and rehabilitation suggestions to doctors, patients, and their families through a web interface and a mobile app. The data flow design follows a closed-loop path of "acquisition-cleaning-transmission-analysis-feedback": ECG signals are acquired by the ADS1298 chip and transmitted to the mobile device for preprocessing via WiFi / Bluetooth; biochemical data are automatically acquired through the hospital's LIS system interface; CT images are parsed using the DICOM protocol to extract key features. All data is encrypted and transmitted to the cloud server via the MQTT protocol, where it is processed in a distributed manner by a Spark cluster. The resulting structured data is stored in a MongoDB database, and the unstructured data is stored in a distributed file system.

[0040] In terms of hardware implementation, the wearable ECG monitor adopts a flexible circuit board design. The core component includes the ADS1298 chip, which integrates an 8-channel 24-bit Δ-Σ ADC, supporting right leg drive and lead dislodgement detection functions. It features an input impedance >10MΩ and a common-mode rejection ratio >110dB, ensuring accurate capture of microvolt-level ECG signals. The main control unit uses an ESP32-WROOM-32D module with a built-in dual-core 32-bit LXC processor, supporting 802.11b / g / n WiFi and Bluetooth 4.2. It communicates with the ADS1298 via an SPI interface for data packaging and wireless transmission. Power management utilizes a CR2032 coin cell battery, coupled with a TPS62740 low-dropout regulator, achieving a quiescent current <5μA and a battery life of up to 72 hours. The electrode system uses Ag / AgCl gel electrodes with a serpentine wiring layout, ensuring a skin contact impedance <5kΩ. The twelve-lead electrodes adopt a standard Colorado layout, compatible with clinical usage habits. The edge computing terminal is based on a Raspberry Pi 4B, configured with 4GB of LPDDR4 memory and 64GB of eMMC storage, running an Ubuntu Server system. The deployed signal preprocessing module employs a wavelet transform denoising algorithm combined with the Pan-Tompkins algorithm to achieve QRS wave detection, with a processing latency of <50ms. The lightweight AI model uses an optimized MobileNetV2 model, taking a 12-lead ECG signal as input and outputting characteristic parameters such as ST segment offset and QT interval. The model size is compressed to 2.8MB, and the inference speed reaches 30fps. Data encryption uses the AES-256-CBC algorithm to encrypt transmitted data, with the key dynamically generated through the hospital's PKI system.

[0041] In terms of core algorithm design, the multimodal data fusion engine includes data layer fusion, feature layer fusion, and decision layer fusion. In data layer fusion, ECG signals are analyzed using time-frequency analysis to extract HRV features; CT images are analyzed using the ResNet50 model to extract indicators such as left ventricular ejection fraction and myocardial thickness; and transcriptome data are analyzed using DESeq2 to screen for differentially expressed genes. Feature layer fusion constructs a heterogeneous feature matrix, which is then reduced to 16 dimensions using the t-SNE algorithm. The fusion formula is as follows:

[0042] Ffused = αFecg + βFct + γFgene

[0043] α, β, and γ are dynamic weighting coefficients, optimized using gradient descent. The decision layer integrates the prediction results of random forest and XGBoost, employing a weighted voting mechanism to ultimately output the risk probability. A dynamic threshold adjustment algorithm models postoperative patients, collecting HRV, troponin I, and B-type natriuretic peptide data within 7 days post-surgery to construct a time-series model. A Kalman filter algorithm is used to predict physiological indicators for the next time period, triggering an early warning when the predicted value exceeds an individualized threshold.

[0044] In terms of software system implementation, the cloud-based analytics platform is deployed using Docker containers, and its architecture includes a data receiving layer, a storage layer, a computation layer, and an interface layer. The data receiving layer uses a RabbitMQ message queue to buffer and transmit data, with a peak processing capacity of 1000 messages per second. The storage layer uses MongoDB to store structured data, MinIO to store image files, and ElasticSearch to support fast retrieval. The computation layer uses Spark Streaming to process real-time data streams and TensorFlow Serving to load pre-trained models. The interface layer provides RESTful APIs and WebSocket interfaces to support calls from third-party systems. Key code implementations include the ECG preprocessing module, the multimodal fusion model, and the dynamic threshold calculation module, all written in Python and integrating libraries such as PyWavelets and PyTorch.

[0045] In terms of system testing and validation, a multi-center clinical trial was conducted in three tertiary hospitals, enrolling 500 patients with acute myocardial infarction. The control group used traditional Holter monitoring. Results showed that the ECG signal acquisition quality had a lead dropout rate of <2%, a diagnostic accuracy AUC of 0.94, a sensitivity of 92%, and a specificity of 88%. Multimodal fusion performance improved the early myocardial infarction detection rate by 27%. In timeliness testing, the average end-to-end latency was 2.8 seconds, the data upload speed reached 50Mbps under 5G network conditions, and the system's concurrency capacity increased to 1000 users / second. Device stability testing showed an average battery life of 68 hours, normal operation in temperatures ranging from -10℃ to 45℃, and an IP67 waterproof rating.

[0046] Example 1: Real-time monitoring and early warning in pre-hospital emergency care scenarios

[0047] Patient Zhang, 58 years old, with a history of coronary heart disease, suddenly experienced chest pain accompanied by profuse sweating while at home. His family immediately assisted him in wearing the wearable 12-lead ECG monitor provided by this invention. This device adopts a flexible circuit board design, and its core components include the ADS1298 chip with an integrated 8-channel 24-bit Δ-Σ type ADC, supporting right leg drive and lead dislodgement detection functions. It has an input impedance >10MΩ and a common-mode rejection ratio >110dB, ensuring accurate capture of microvolt-level ECG signals. The main control unit uses an ESP32-WROOM-32D module with a built-in dual-core 32-bit LXC processor, supporting 802.11b / g / n. WiFi and Bluetooth 4.2 are supported. The device communicates with the ADS1298 via the SPI interface for data packetization and wireless transmission. Power management is achieved using a CR2032 coin cell battery and a TPS62740 low-dropout regulator, resulting in a quiescent current of <5μA and a battery life of up to 72 hours. The electrode system uses Ag / AgCl gel electrodes with a serpentine wiring layout to ensure a skin contact impedance of <5kΩ. The twelve-lead electrode uses a standard Colorado layout, compatible with clinical usage habits. After family members activate the real-time monitoring mode via a mobile app, the ECG monitor acquires 12-lead ECG signals at a sampling rate of 512Hz and transmits them to the mobile phone via WiFi. The mobile preprocessing module performs wavelet transform denoising on the signals and uses the Pan-Tompkins algorithm to detect QRS waves with a processing latency of <50ms. The preprocessed data is encrypted and uploaded to the cloud server via the MQTT protocol, simultaneously triggering an optimized MobileNetV2 lightweight model for preliminary risk assessment. This model takes 12-lead ECG signals as input and outputs characteristic parameters such as ST segment offset and QT interval. The model size is compressed to 2.8MB, and the inference speed reaches 30fps. The cloud-based multimodal fusion engine retrieves historical patient data, including previous CT images and gene testing results. Combined with real-time ECG features, it integrates prediction results from random forests and XGBoost, and outputs risk probabilities using a weighted voting mechanism. When the risk probability is greater than 85%, the system sends an alert to the emergency department of the nearest tertiary hospital via both APP push and SMS. The alert includes the patient's location, real-time ECG waveform, and preliminary diagnosis. After receiving the alert, the emergency department doctor remotely guides the family to administer initial treatments such as nitroglycerin sublingual administration and dispatches a 120 ambulance for priority response. The time from signal acquisition to alert triggering is 2.3 seconds, which is shorter than the traditional 120 dispatch process. Patients can directly enter the catheterization lab after admission, reducing the D2B time to 70 minutes.

[0048] Example 2: Personalized early warning and rehabilitation management for postoperative patients

[0049] Patient Li, 65 years old, underwent PCI after acute myocardial infarction and was transferred to the cardiovascular rehabilitation ward. The hospital configured the system of this invention for postoperative monitoring, with a planned continuous monitoring period of 30 days. The system retrieves the patient's HRV, cTnI, and BNP data within 24 hours postoperatively, constructs an individualized time series model, and extracts features such as SDNN, RMSSD, and pNN50 from the HRV data through time-frequency analysis. After normalization, the cTnI and BNP data are used to predict the baseline values ​​of physiological indicators for the next day using a Kalman filter algorithm, with a threshold range set at ±20% of the baseline value. The wearable device continuously collects ECG signals from 8:00 to 20:00 daily, synchronously connecting to the hospital's LIS system to obtain daily morning biochemical indicators, and a CT image review is performed every Monday. The system automatically analyzes the left ventricular ejection fraction and the extent of myocardial edema. On the 10th day post-surgery, the patient's ECG showed new Q waves, and LVEF decreased by 15% compared to the previous reading, triggering a level-two alert. The doctor's interface simultaneously displayed gene testing results, showing a positive TNNI3 gene mutation, and recommended adjusting the antiplatelet therapy regimen and increasing the ACE inhibitor dosage. Within 30 days post-surgery, two mild myocardial ischemia events were successfully detected, neither of which developed into acute myocardial infarction. Based on the multimodal data provided by the system, the doctor included the patient in the high-risk recurrence group and developed a plan to extend the dual antiplatelet therapy cycle. No major adverse cardiovascular events occurred during the 6-month follow-up.

[0050] Example 3: Multimodal Diagnostic Support for Primary Healthcare Institutions

[0051] A county hospital admitted Mr. Wang, a 48-year-old patient with chest pain and no prior medical history. Due to limitations in facilities, coronary angiography could not be performed. The nurse fitted Mr. Wang with a wearable ECG monitor, simultaneously collecting 12-lead ECG signals and finger-prick blood biochemical data. The data was uploaded to a regional medical cloud platform via a 4G network, triggering a multimodal fusion diagnostic process. The cloud engine retrieved the patient's ECG characteristics, biochemical indicators, and CT images from the county hospital. By integrating the prediction results of random forest and XGBoost, the system output a diagnostic probability. The system recommended activating the "Chest Pain Center Green Channel" and recommended referral to a higher-level hospital for coronary angiography. Based on the risk map provided by the system, the county hospital doctor coordinated ambulance transport. The total time from patient symptom onset to coronary angiography was 3.2 hours, below the chest pain center certification standard. Subsequent pathology confirmed non-ST-segment elevation myocardial infarction, and the system's diagnosis achieved an 89% concordance rate with the gold standard.

[0052] This invention utilizes technologies such as multi-lead wearable design, dynamic threshold algorithm, edge-cloud collaborative computing, and medical-grade data fusion to construct a full-process management solution for cardiovascular acute and critical illnesses, providing strong technical support for precise clinical diagnosis and treatment.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0054] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A cardiovascular disease critical illness intelligent monitoring and risk early warning system, characterized in that: The system comprises a cardiovascular critical illness clinical database, a data acquisition and processing module, an AI rapid diagnosis model, a multi-modal data fusion early warning platform, and a signal transmission mechanism. ​ The cardiovascular critical illness clinical database stores the clinical history data, CT image data, gene sequencing data, and disease outcome information of patients. The data acquisition and processing module is composed of a wearable multi-lead intelligent ECG signal monitoring instrument and a smart mobile terminal. The AI rapid diagnosis model integrates a machine learning framework optimized by random forest, XGBoost, support vector machine, and genetic algorithm, and is used for feature selection and disease prediction. The multi-modal data fusion early warning platform integrates ECG data, biochemical data, CT data, and genetic data, and realizes real-time data interaction and risk assessment through the MQTT protocol. The system transmits the preprocessed signals to the data center server through the mobile terminal, and outputs the triage results and warning levels by the AI model.

2. The system of claim 1, wherein: The hardware configuration of the wearable multi-lead intelligent ECG signal monitoring instrument includes a main control chip ESP32, a disposable medical adhesive ten-electrode twelve-lead configuration, and an ADS1298 chip signal acquisition circuit with a sampling accuracy of 24 bits.

3. The system of claim 1 or 2, wherein: The data transmission mechanism includes ECG signal transmission to the mobile terminal through Bluetooth / WiFi, data upload to the data center through the Internet or hospital intranet by the mobile terminal, and lightweight real-time communication between the data center and the mobile terminal using the MQTT protocol.

4. The system of claim 1, wherein: The training process of the AI rapid diagnosis model includes constructing a multi-modal feature set based on the cardiovascular clinical database, optimizing the combination weight of random forest and XGBoost through genetic algorithm, and classifying and predicting the selected features using logistic regression to output the risk probability of critical illness.

5. The system of claim 1, wherein: The functions of the multi-modal data fusion early warning platform include real-time integration of ECG signals, genetic markers, biochemical indicators, and CT image features, dynamic weighted fusion algorithm for calculating the comprehensive risk score of patients, and two-way push of early warning information through the mobile terminal and hospital system when the risk score exceeds the threshold.

6. The system of claim 1, wherein: The preprocessing operations of the smart mobile terminal include wavelet transform denoising, QRS wave detection based on Pan-Tompkins algorithm, heart rate variability feature extraction, and signal quality evaluation.

7. The system of claim 1, wherein: The system also includes a personalized early warning threshold adjustment module for postoperative patients, a remote first aid data transmission interface based on 5G network, and a clinical decision support interface connected with the hospital HIS system.

Images