Distributed intelligent electrocardiogram monitoring system based on federal learning

Abstract

The invention relates to the technical field of medical health, in particular to a federated learning-based distributed intelligent electrocardiogram monitoring system, which comprises a distributed data acquisition module, a localization processing unit, a federated learning engine and a cloud collaborative management module. The system realizes real-time signal acquisition through a high-precision acquisition terminal, carries out localized feature extraction and encrypted storage by utilizing edge calculation, and realizes model training under privacy protection in combination with federal learning. The reliability and the intelligent level of the system can be remarkably improved, the communication overhead is reduced, the system is suitable for multi-scene large-scale deployment, and the requirements of modern medical treatment for efficient, safe and intelligent electrocardiograph monitoring are met.

Description

Technical Field

[0001] The present invention belongs to the field of medical health and artificial intelligence technology, and specifically relates to a distributed intelligent electrocardiogram monitoring system based on federated learning. Background Art

[0002] The present invention relates to the technical field of electrocardiogram monitoring, and in particular to a distributed intelligent electrocardiogram monitoring system based on federated learning.

[0003] With the rapid development of ECG monitoring technology, distributed intelligent ECG monitoring systems have become a research hotspot in the healthcare field due to their ability to collect, analyze, and transmit ECG data in real time. However, existing ECG monitoring systems still have some shortcomings in data processing capabilities, privacy protection, and multi-device collaboration, which affect the system's intelligence level and practical application effectiveness.

[0004] Patent document CN107582046B discloses a real-time ECG monitoring method. This patent transmits human electrical signals to a micro-ECG device via electrode patches, and uses smart terminals and big data systems to analyze, manage, and issue alarms for the ECG data. However, in this technical solution, all ECG data must be uploaded to a centralized server for processing, which may lead to the risk of privacy leakage during data transmission. In addition, the system does not fully consider the collaborative working capabilities of multiple devices, and has certain limitations in supporting distributed data processing needs in complex scenarios, which imposes certain restrictions on its flexibility and scalability in large-scale applications.

[0005] The above issues demonstrate that existing ECG monitoring systems still have shortcomings in privacy protection, distributed computing capabilities, and intelligence. Therefore, the present invention provides a distributed intelligent ECG monitoring system based on federated learning. This system aims to achieve collaborative training and data privacy protection across multiple devices through federated learning technology, while also improving the system's intelligence and adaptability, thereby meeting the demand for efficient, secure, and intelligent ECG monitoring systems in the modern healthcare field. Summary of the Invention

[0006] The purpose of the present invention is to provide a distributed intelligent ECG monitoring system based on federated learning to solve the problems mentioned in the above background technology, such as insufficient data privacy protection, limited multi-device collaboration capabilities, and low intelligence level.

[0007] The technical solution of the present invention is: including a distributed data acquisition module, a localized processing unit, a federated learning engine and a cloud-based collaborative management module. The distributed data acquisition module realizes real-time acquisition of ECG signals through multiple independent ECG signal acquisition terminals. Each acquisition terminal is equipped with a high-precision analog-to-digital converter and a low-noise amplifier circuit. The signal sampling frequency is set to 1000Hz to ensure that the high-frequency details of the signal are retained. The localized processing unit is embedded in each acquisition terminal and adopts an edge computing architecture. It performs preliminary analysis of the collected ECG signals through a lightweight neural network, extracts key features such as R wave position, QRS complex width, etc., and stores these features in a local cache. The federated learning engine is deployed in the cloud-based collaborative management module and communicates with each localized processing unit through a secure multi-party computing protocol. It completes the update and optimization of model parameters without transmitting the original data. The cloud-based collaborative management module is responsible for coordinating the task allocation between the terminals and monitoring the operating status of the system in real time.

[0008] Furthermore, the distributed data acquisition module is designed for adaptability in multiple scenarios. Each acquisition terminal is equipped with a replaceable flexible electrode patch coated with a conductive gel layer to reduce the impact of skin impedance on signal quality. The acquisition terminals establish point-to-point connections via the Zigbee wireless communication protocol, forming a decentralized data transmission network that avoids data loss due to single points of failure. Furthermore, each acquisition terminal has a built-in power management chip that supports dynamic adjustment of operating modes to extend battery life.

[0009] Furthermore, the localized processing unit adopts a layered design. The bottom layer is the signal preprocessing module, which is responsible for filtering out power frequency interference and baseline drift, and dynamically adjusting filtering parameters using an adaptive filtering algorithm. The middle layer is the feature extraction module, which uses short-time Fourier transform and wavelet transform to perform time-frequency domain analysis on the signal. The top layer is the decision module, which generates preliminary diagnostic results based on the extracted features. The localized processing unit is also equipped with an encrypted storage module. All processed data is encrypted and stored using the AES-256 encryption algorithm to prevent unauthorized access.

[0010] Furthermore, the core of the federated learning engine is an asynchronous gradient update mechanism, which allows each terminal to flexibly participate in model training based on its own computing power. Specifically, after completing a round of local training, each terminal uploads the updated model parameters to the cloud-based collaborative management module. The cloud uses differential privacy technology to perturb the parameters, ensuring that the data of individual terminals cannot be reverse-engineered. The cloud-based collaborative management module then broadcasts the aggregated parameters to all terminals, completing a global update. This mechanism not only speeds up model convergence but also significantly reduces communication overhead.

[0011] To further validate the system's performance, both simulation experiments and real-world tests were designed. The simulation experiments were conducted in a simulated environment, using the MIT-BIH arrhythmia database to generate simulated signals and adding noise of varying intensities to simulate real-world interference. The experiments focused on evaluating the system's diagnostic accuracy, data transmission latency, and privacy protection. The real-world tests involved deploying the system in both hospital wards and home environments, testing its stability and reliability under varying network conditions.

[0012] Furthermore, the simulation environment was built on the open-source platform NS3, comprising virtual patient nodes, data acquisition endpoints, and cloud nodes. The virtual patient nodes generated ECG signals that met AAMI standards, the data acquisition endpoints simulated signal acquisition and localization processing, and the cloud nodes ran the federated learning engine. The simulation environment also incorporated dynamic network topology changes, simulating changes in signal strength as the terminal moved to assess the robustness of the system.

[0013] Furthermore, the hardware design of the local processing unit fully considers the balance between power consumption and performance. The core processor uses a low-power ARM Cortex-M7 chip with a main frequency of 216MHz and a memory capacity of 512KB, meeting the requirements of real-time signal processing. External expansion interfaces include SPI, I2C, and UART, supporting connection to a variety of sensors. In addition, the local processing unit also integrates a real-time clock module to record signal acquisition timestamps to facilitate subsequent data synchronization.

[0014] Furthermore, the cloud-based collaborative management module adopts a distributed architecture design and consists of multiple submodules, including a task scheduling module, a data synchronization module, and a security audit module. The task scheduling module dynamically assigns task priorities based on each terminal's computing power and network conditions. The data synchronization module is responsible for maintaining data consistency between the terminal and the cloud, using a hash chain-based incremental synchronization algorithm to reduce data transmission volume. The security audit module records all operation logs and regularly generates security reports to facilitate system administrators in identifying potential security risks.

[0015] Furthermore, the federated learning engine's optimization strategies include model compression and sparsification. Model compression uses quantization techniques to convert floating-point parameters into fixed-point representation, reducing storage and transmission costs. Sparsification uses a pruning algorithm to remove redundant weights, improving model inference efficiency. Furthermore, the federated learning engine supports multi-task learning, enabling the simultaneous processing of different types of ECG anomaly detection tasks within the same model.

[0016] Furthermore, the distributed data acquisition module's communication protocol adopts a hybrid model, combining the advantages of Zigbee and LoRa technologies. This ensures both low latency for short-distance communication and high reliability for long-distance communication. The communication protocol stack is divided into three layers: the bottom layer is the physical layer, which uses direct sequence spread spectrum technology to improve anti-interference capabilities; the middle layer is the link layer, responsible for encapsulating and decapsulating data frames; and the top layer is the application layer, which defines the specific format and verification rules for data transmission.

[0017] Furthermore, communication between the local processing unit and the cloud-based collaborative management module utilizes the TLS 1.3 encryption protocol to ensure data security during transmission. The communication link establishment process includes three phases: identity authentication, key negotiation, and session key generation. Each phase utilizes asymmetric encryption to prevent man-in-the-middle attacks.

[0018] The present invention provides a distributed intelligent ECG monitoring system based on federated learning through improvements. Compared with the existing technology, it has the following features:

[0019] By adopting distributed data acquisition modules and localized processing units, the system can realize real-time collection and preliminary analysis of ECG signals, avoiding the traditional system's high dependence on centralized servers and significantly improving the system's reliability and flexibility.

[0020] By introducing the federated learning engine, the system can complete model training while protecting user privacy, solving the privacy leakage risks brought about by centralized storage of data in traditional systems, while improving the generalization ability of the model.

[0021] By optimizing the communication protocol and federated learning mechanism, the system reduces communication overhead while ensuring data security. It is suitable for large-scale deployment scenarios and meets the needs of the modern medical and health field for efficient, secure, and intelligent ECG monitoring systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a system architecture diagram of the present invention.

[0023] Figure 2 This is a structural diagram of the distributed data acquisition module.

[0024] Figure 3 This is the functional module diagram of the localization processing unit.

[0025] Figure 4 This is the workflow diagram of the federated learning engine.

[0026] Figure 5 Diagram of the task allocation mechanism for the cloud-based collaborative management module. DETAILED DESCRIPTION

[0027] The present invention provides a distributed intelligent ECG monitoring system based on federated learning, and its specific implementation is described in detail with reference to the accompanying drawings. Figure 1 As shown in Figure 1, the overall architecture of the system includes a distributed data acquisition module, a local processing unit, a federated learning engine, and a cloud-based collaborative management module. These modules are connected via wired or wireless communication to form a complete distributed ECG monitoring system.

[0028] The distributed data acquisition module is composed of multiple independent ECG signal acquisition terminals. The specific structure of each acquisition terminal is as follows: Figure 2 As shown. Each acquisition terminal includes a flexible electrode patch, an analog-to-digital converter, and a wireless communication interface. The flexible electrode patch directly contacts the human skin and is used to collect ECG signals. The surface of the flexible electrode patch is coated with a conductive gel layer to reduce the impact of skin impedance on signal quality. The flexible electrode patch is connected to the analog-to-digital converter through a wire. The analog-to-digital converter converts the analog signal into a digital signal. The sampling frequency is set to 1000Hz to ensure that high-frequency details are preserved. The analog-to-digital converter is connected to the wireless communication interface through metal contacts on the circuit board. The wireless communication interface uses the Zigbee protocol to establish a point-to-point connection with other acquisition terminals to form a decentralized data transmission network. Each acquisition terminal has a built-in power management chip that dynamically adjusts power consumption by adjusting the working mode to extend battery life.

[0029] The local processing unit is embedded in each acquisition terminal, and its functional modules are as follows: Figure 3 As shown. The localized processing unit includes a signal preprocessing module, a feature extraction module, and a decision module. The signal preprocessing module is connected to the analog-to-digital converter through the data bus on the circuit board, and is responsible for filtering the collected raw ECG signals. The signal preprocessing module uses an adaptive filtering algorithm to dynamically adjust the filtering parameters to filter out power frequency interference and baseline drift. The feature extraction module is connected to the signal preprocessing module through an internal data channel, and uses short-time Fourier transform and wavelet transform to perform time-frequency domain analysis on the signal to extract key features such as R wave position and QRS complex width. The decision module is connected to the feature extraction module through hardware registers to generate preliminary diagnostic results based on the extracted features. The localized processing unit is also equipped with an encrypted storage module, and all processed data are encrypted and stored using the AES-256 encryption algorithm to prevent unauthorized access.

[0030] The federated learning engine is deployed in the cloud collaborative management module, and its workflow is as follows: Figure 4As shown in Figure 2, the federated learning engine communicates with each localized processing unit via a secure multi-party computing protocol. After completing a round of training, each localized processing unit uploads the updated model parameters to the federated learning engine. The federated learning engine uses differential privacy technology to perturb the parameters, ensuring that the data of individual terminals cannot be reverse-engineered. The perturbed parameters are aggregated using an aggregation algorithm and then broadcast to all terminals, completing a global update. The federated learning engine supports an asynchronous gradient update mechanism, allowing each terminal to flexibly participate in model training based on its own computing power. This mechanism not only improves model convergence speed but also significantly reduces communication overhead.

[0031] The cloud collaborative management module adopts a distributed architecture design and consists of a task scheduling module, a data synchronization module, and a security audit module. The task scheduling module dynamically assigns task priorities based on the computing power and network conditions of each terminal. Its allocation mechanism is as follows: Figure 5 As shown in the figure, the task scheduling module determines the task allocation strategy by querying the resource status information reported by the terminal and combining it with the current network topology changes. The data synchronization module is responsible for maintaining data consistency between the terminal and the cloud, using a hash chain-based incremental synchronization algorithm to reduce data transmission volume. The security audit module records all operation logs and regularly generates security reports to facilitate system administrators to identify potential security risks.

[0032] The hardware design of the local processing unit fully considers the balance between power consumption and performance. The core processor uses a low-power ARM Cortex-M7 chip with a main frequency of 216MHz and a memory capacity of 512KB to meet the needs of real-time signal processing. External expansion interfaces include SPI, I2C and UART, which support connection with a variety of sensors. The real-time clock module is integrated inside the local processing unit to record the timestamp of signal acquisition to facilitate subsequent data synchronization. The communication between the local processing unit and the cloud collaborative management module adopts the TLS 1.3 encryption protocol. The process of establishing the communication link includes three stages: identity authentication, key negotiation and session key generation. Each stage is protected by an asymmetric encryption algorithm to prevent man-in-the-middle attacks.

[0033] The distributed data acquisition module's communication protocol uses a hybrid model, combining the advantages of Zigbee and LoRa technologies. The communication protocol stack consists of three layers. The bottom layer is the physical layer, which uses direct sequence spread spectrum technology to enhance interference resistance. The middle layer is the link layer, responsible for encapsulating and decapsulating data frames. The top layer is the application layer, which defines the specific format and validation rules for data transmission. This hybrid model ensures low latency for short-distance communication while achieving high reliability for long-distance communication.

[0034] The federated learning engine's optimization strategies include model compression and sparsification. Model compression uses quantization techniques to convert floating-point parameters into fixed-point representation, reducing storage and transmission costs. Sparsification uses a pruning algorithm to remove redundant weights, improving model inference efficiency. Furthermore, the federated learning engine supports multi-task learning, enabling the simultaneous processing of different ECG anomaly detection tasks within the same model.

[0035] To verify the system's performance, both simulation experiments and actual tests were designed. The simulation experiment established a simulation environment within the open-source platform NS3, including a virtual patient node, an acquisition terminal node, and a cloud node. The virtual patient node generated ECG signals that met AAMI standards, the acquisition terminal node simulated the signal acquisition and localization processing process, and the cloud node was responsible for running the federated learning engine. The simulation environment introduced dynamic network topology changes, simulating changes in signal strength during terminal movement to evaluate the system's robustness. The actual test deployed the system in both a hospital ward and a home environment to test its stability and reliability under different network conditions.

[0036] The entire system operates as follows: the flexible electrode patch collects ECG signals and transmits them to the analog-to-digital converter, which converts the analog signals into digital signals and transmits them to the local processing unit. The signal preprocessing module of the local processing unit filters the signals, the feature extraction module extracts key features, and the decision module generates preliminary diagnostic results and stores the data in the encrypted storage module. The local processing unit uploads the updated model parameters to the federated learning engine, which completes the parameter update using differential privacy technology and aggregation algorithms and broadcasts the results to all terminals. The task scheduling module of the cloud-based collaborative management module dynamically assigns task priorities based on the terminal resource status, the data synchronization module maintains data consistency, and the security audit module records operation logs and generates security reports.

[0037] In order to better enable relevant personnel in this technical field to fully understand and implement the present invention, the implementation principle of the present invention is supplemented below with reference to a specific application scenario.

[0038] In a hospital ward environment, patients wear an ECG signal acquisition terminal consisting of a flexible electrode patch. The flexible electrode patch is connected to an analog-to-digital converter via a wire. The analog-to-digital converter converts the collected analog ECG signal into a digital signal at a sampling frequency of 1000Hz and transmits it to the local processing unit through a wireless communication interface. At this time, the signal preprocessing module in the local processing unit starts the adaptive filtering algorithm and dynamically adjusts the filtering parameters to filter out power frequency interference and baseline drift. The filtered signal is passed to the feature extraction module, which uses short-time Fourier transform and wavelet transform to perform time-frequency domain analysis on the signal and extract key features such as R wave position and QRS complex width. Subsequently, the decision module generates a preliminary diagnosis result based on the extracted features and stores the data in an encrypted storage module to ensure data security.

[0039] At the same time, another user in a home environment also wore the same ECG signal acquisition terminal. Its local processing unit also completed the signal acquisition, preprocessing, feature extraction, and preliminary diagnosis process. Notably, each terminal's local processing unit embedded a lightweight neural network model for real-time ECG signal analysis. This edge computing architecture significantly reduced reliance on centralized servers, thereby improving system responsiveness and reliability.

[0040] After each terminal completes a round of local model training, the updated model parameters are uploaded to the federated learning engine via a secure multi-party computation protocol. The federated learning engine employs differential privacy techniques to perturb the uploaded parameters, ensuring that the data from individual terminals cannot be reverse-engineered. Subsequently, the task scheduling module within the cloud-based collaborative management module dynamically prioritizes tasks based on the computing power and network conditions of each terminal. For example, the task scheduling module allocates more training tasks to terminals with stronger computing capabilities; whereas, terminals with poorer network conditions are less frequently included in global updates to reduce communication overhead.

[0041] In the cloud-based collaborative management module, the federated learning engine integrates parameters uploaded by all terminals using an aggregation algorithm and broadcasts the optimized model back to each terminal. This process supports an asynchronous gradient update mechanism, allowing different terminals to flexibly participate in model training based on their own conditions. For example, some terminals may pause training due to low battery, but the system can still continue to optimize the model through the participation of other terminals. This mechanism not only speeds up model convergence but also significantly reduces communication costs.

[0042] To validate the system's performance, a set of simulation experiments was designed. Within the simulation environment, virtual patient nodes generated ECG signals that met AAMI standards and added varying levels of noise to simulate interference in real-world scenarios. Acquisition endpoints simulated the signal acquisition and localization processes, while cloud nodes ran the federated learning engine. Experimental results demonstrated that the system maintained high robustness despite dynamic network topology changes. Actual testing further demonstrated that the system operated stably under diverse network conditions, whether in hospital wards or at home, and achieved efficient ECG anomaly detection.

[0043] Furthermore, the distributed data acquisition module's communication protocol utilizes a hybrid model, combining the advantages of Zigbee and LoRa technologies. For short-range communications, Zigbee ensures low-latency data transmission, while for long-range communications, LoRa technology provides highly reliable data connections. This hybrid model not only improves the system's adaptability but also enhances its anti-interference capabilities in complex environments.

[0044] Finally, the communication link between the local processing unit and the cloud-based collaborative management module utilizes the TLS 1.3 encryption protocol. The communication link establishment process includes three phases: identity authentication, key negotiation, and session key generation. Each phase utilizes asymmetric encryption to prevent man-in-the-middle attacks. This multi-layered security mechanism ensures data integrity and confidentiality during transmission.

[0045] In summary, the present invention realizes an efficient, secure and intelligent ECG monitoring system by combining distributed data collection, local processing, federated learning and cloud collaborative management, which can meet the needs of large-scale deployment in the modern medical and health field.

[0046] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0047] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A distributed intelligent ECG monitoring system based on federated learning, characterized by: It includes a distributed data acquisition module, a local processing unit, a federated learning engine and a cloud-based collaborative management module. The distributed data acquisition module is composed of multiple ECG signal acquisition terminals. Each acquisition terminal contains a flexible electrode patch, an analog-to-digital converter and a wireless communication interface. The signal sampling frequency of the analog-to-digital converter is set to 1000Hz. The local processing unit is embedded in each acquisition terminal and uses an edge computing architecture to perform preliminary analysis of the ECG signal. The federated learning engine is deployed in the cloud-based collaborative management module and updates and optimizes model parameters with each local processing unit through a secure multi-party computing protocol.

2. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The surface of the flexible electrode patch of the distributed data acquisition module is coated with a conductive gel layer, and point-to-point connections are established between the acquisition terminals through the Zigbee wireless communication protocol to form a decentralized data transmission network.

3. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The localization processing unit includes a signal preprocessing module, a feature extraction module and a decision module. The signal preprocessing module uses an adaptive filtering algorithm to dynamically adjust the filtering parameters to filter out power frequency interference and baseline drift. The feature extraction module uses short-time Fourier transform and wavelet transform to perform time-frequency domain analysis on the signal. The decision module generates preliminary diagnostic results and stores the data in an encrypted storage module.

4. The distributed intelligent ECG monitoring system based on federated learning according to claim 3 is characterized in that: The encryption storage module of the local processing unit uses the AES-256 encryption algorithm to encrypt and store the processed data.

5. The distributed intelligent ECG monitoring system based on federated learning according to claim 1 is characterized in that: The federated learning engine adopts an asynchronous gradient update mechanism, allowing each terminal to flexibly participate in model training based on its own computing power, and perturbs the uploaded model parameters through differential privacy technology.

6. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The cloud collaborative management module includes a task scheduling module, a data synchronization module and a security audit module. The task scheduling module dynamically assigns task priorities according to the computing power and network status of each terminal. The data synchronization module uses an incremental synchronization algorithm based on a hash chain to maintain data consistency between the terminal and the cloud. The security audit module records all operation logs and generates security reports regularly.

7. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The core processor of the local processing unit is a low-power ARM Cortex-M7 chip with a main frequency of 216MHz and a memory capacity of 512KB. The external expansion interfaces include SPI, I2C and UART, which support connection with a variety of sensors.

8. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The communication protocol of the distributed data acquisition module adopts a hybrid mode, combining the advantages of Zigbee and LoRa technologies. The communication protocol stack is divided into three layers. The bottom layer is the physical layer, which uses direct sequence spread spectrum technology to improve anti-interference ability. The middle layer is the link layer, which is responsible for the encapsulation and decapsulation of data frames. The top layer is the application layer, which defines the specific format and verification rules of data transmission.

9. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The communication between the local processing unit and the cloud collaborative management module adopts the TLS 1.3 encryption protocol. The process of establishing the communication link includes three stages: identity authentication, key negotiation and session key generation. Each stage is protected by an asymmetric encryption algorithm.

10. The distributed intelligent ECG monitoring system based on federated learning according to claim 1, characterized in that: The optimization strategy of the federated learning engine includes model compression and sparsification processing. Model compression converts floating-point parameters into fixed-point representation through quantization technology, and sparsification removes redundant weights through pruning algorithm.

Images