Electrocardiogram monitoring method and system based on single-arm upper limb collection and twelve-lead reconstruction

Abstract

This invention discloses a method and system for electrocardiogram (ECG) monitoring based on single-arm upper limb acquisition and 12-lead reconstruction. The method includes: acquiring a unilateral upper limb ECG signal from a target user based on a preset ECG acquisition path; reconstructing the signal to obtain a single-arm lead heartbeat signal and a standard 12-lead ECG signal; performing R-wave alignment segmentation on the standard 12-lead ECG signal and alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal; and performing R-wave heartbeat aggregation and end-to-end continuous reconstruction to obtain a continuously reconstructed standard 12-lead heartbeat signal. This invention can achieve end-to-end continuous reconstruction from a single-arm ECG sequence of arbitrary length to a standard 12-lead ECG sequence, meeting the needs of long-term continuous monitoring scenarios outside hospitals. As a method and system for ECG monitoring based on single-arm upper limb acquisition and 12-lead reconstruction, this invention can be widely applied in the field of ECG monitoring technology.

Description

Technical Field

[0001] This invention relates to the field of electrocardiogram (ECG) monitoring technology, and in particular to an ECG monitoring method and system based on single-arm upper limb acquisition and twelve-lead reconstruction. Background Technology

[0002] For cardiovascular diseases, establishing a new and efficient continuous monitoring model to achieve early detection and treatment is of great significance to the modern medical system. Electrocardiography (ECG), as a fundamental tool reflecting changes in cardiac electrical activity, is a crucial basis for doctors to make early diagnoses and clinical assessments of cardiovascular diseases such as arrhythmias, myocardial ischemia, and myocardial infarction. Among them, the standard twelve-lead ECG signal can non-invasively reflect cardiac electrophysiological information, providing doctors with detailed three-dimensional spatial information of the heart in the coronal, sagittal, and cross-sectional planes, making it the "gold standard" for clinical diagnosis.

[0003] However, traditional standard 12-lead ECG signal acquisition systems are quite complex. While wearable portable devices, such as Holter monitors, can achieve long-term, continuous, and dynamic recording of ECG signals, these systems typically require ten leads connected to the patient's body via gel electrodes for signal acquisition. This multi-point, distributed, and complex cable connection method not only easily causes severe discomfort to the patient but also readily introduces lead traction interference during daily activities, often limiting it to in-hospital or out-of-hospital monitoring scenarios, making it difficult to support truly unobtrusive long-term monitoring.

[0004] To improve the portability of electrocardiogram (ECG) monitoring, numerous wearable devices have emerged, such as chest patches, clothing-integrated devices, and wristbands. However, these devices generally suffer from significant drawbacks in long-term use: chest patches are prone to skin irritation and contact instability due to sweat after prolonged wear; clothing-integrated devices are susceptible to signal instability due to differences in body shape, wearing conditions, and washing attenuation; and wristbands often require active monitoring to establish signal acquisition leads. More critically, because these devices attempt to improve patient comfort by reducing the number of leads, the effective pathological information reflected by the ECG is significantly reduced, severely limiting their clinical application value. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention aims to provide an electrocardiogram (ECG) monitoring method and system based on single-arm upper limb acquisition and twelve-lead reconstruction, which can realize end-to-end continuous reconstruction of a single-arm ECG sequence of arbitrary length into a standard twelve-lead ECG sequence, thereby meeting the needs of long-term continuous monitoring scenarios outside the hospital.

[0006] The first technical solution adopted in this invention is: an electrocardiogram monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction, comprising the following steps: The ECG signal of the target user’s unilateral upper limb is acquired based on the preset ECG acquisition path, and the signal is preprocessed and reconstructed to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal. R-wave alignment segmentation was performed on the standard 12-lead ECG signal, and alternating optimization adversarial learning was performed on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal. Based on several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal, R-wave heartbeat aggregation and end-to-end continuous reconstruction are performed to obtain a continuously reconstructed standard 12-lead heartbeat signal.

[0007] Furthermore, the step of acquiring the unilateral upper limb ECG signal of the target user based on a preset ECG acquisition path, and performing signal preprocessing and reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal specifically includes: The first ECG signal electrode and the second ECG signal electrode are respectively placed on the inner and outer sides of the target user's upper limb, and the reference driving electrode is placed between the first ECG signal electrode and the second ECG signal electrode to acquire the ECG signal of the target user's unilateral upper limb. A weak ECG signal conditioning circuit was constructed to collect and preprocess the unilateral upper limb ECG signal of the target user, and obtain the preprocessed unilateral upper limb ECG signal. The preprocessed unilateral upper limb electrocardiogram signal is transmitted to the host computer for filtering and reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead electrocardiogram signal.

[0008] Furthermore, the weak ECG signal conditioning circuit specifically includes an input protection circuit, a front-end buffer, an amplification and conditioning link, and an analog-to-digital converter, wherein: The input protection circuit is used to suppress electrostatic discharge and transient impact on the electrocardiogram signal of one side of the target user's upper limb, so as to obtain the suppressed electrocardiogram signal of one side of the upper limb. The front-end buffer is used to electrically isolate and buffer the suppressed unilateral upper limb electrocardiogram signal to obtain a buffered unilateral upper limb electrocardiogram signal. The amplification and conditioning link is used to suppress noise in the buffered unilateral upper limb electrocardiogram signal to obtain an effective unilateral upper limb electrocardiogram signal. The analog-to-digital converter is used to perform high-resolution sampling and analog-to-digital conversion processing on the valid unilateral upper limb electrocardiogram signal to obtain a preprocessed unilateral upper limb electrocardiogram signal.

[0009] Furthermore, the step of transmitting the preprocessed unilateral upper limb ECG signal to the host computer for filtering and reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal specifically includes: The preprocessed unilateral upper limb electrocardiogram signal is transmitted to the host computer for data decoding to obtain the decoded unilateral upper limb electrocardiogram signal. Based on the decoded unilateral upper limb electrocardiogram signal, FIR filtering and signal reconstruction were performed, and the data was buffered using a preset data format to obtain the unilateral lead heartbeat signal and the standard twelve-lead electrocardiogram signal.

[0010] Furthermore, the step of performing R-wave alignment segmentation processing on the standard twelve-lead ECG signal and alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal specifically includes: The standard twelve-lead electrocardiogram signal was subjected to amplitude normalization and R-wave alignment segmentation in sequence to obtain several overlapping heartbeat signals of equal length. A generative adversarial network model was constructed to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal, resulting in the reconstructed standard twelve-lead heartbeat signal.

[0011] Furthermore, the step of sequentially performing amplitude normalization and R-wave alignment segmentation on the standard twelve-lead ECG signal to obtain several overlapping heartbeat signals of equal length specifically includes: The amplitude of the standard twelve-lead electrocardiogram signal was normalized to obtain the normalized standard twelve-lead electrocardiogram signal. Based on the adaptive threshold peak detection method, the R wave is detected in the normalized standard twelve-lead ECG signal to determine the R wave location. Using the R-wave position as the time anchor point, the normalized standard twelve-lead ECG signal is synchronously truncated into heartbeat segments according to the preset length, resulting in several overlapping heartbeat signals of equal length.

[0012] Furthermore, the step of detecting the R wave and determining its location based on the adaptive threshold peak detection method for the normalized standard twelve-lead ECG signal specifically includes: The moving average of the normalized standard twelve-lead ECG signal is calculated, and the threshold is dynamically set based on the moving average at the current moment. The locations of data points exceeding the threshold in the normalized standard twelve-lead electrocardiogram signal are marked as regions of interest; Traverse all regions of interest and select the data point with the largest amplitude among all regions of interest as the R-wave position.

[0013] Furthermore, the step of constructing a generative adversarial network model to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain the reconstructed standard twelve-lead heartbeat signal specifically includes: Construct a generative adversarial network (GAN) model, which includes a generator and a discriminator; A generator based on a generative adversarial network model is used to perform deep latent feature extraction and deep feature recovery processing on the single-arm lead heartbeat signal to obtain a reconstructed twelve-lead signal. A discriminator based on a generative adversarial network model is used to identify features of the reconstructed twelve-lead signal and output the probability of the true standard twelve-lead heartbeat signal. Based on the probability of real standard 12-lead heartbeat signals, the reconstructed standard 12-lead heartbeat signals are obtained by alternately optimizing adversarial learning on single-arm lead heartbeat signals through a generative adversarial network model.

[0014] Furthermore, the step of performing R-wave heartbeat aggregation and end-to-end continuous reconstruction based on several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal to obtain a continuously reconstructed standard 12-lead heartbeat signal specifically includes: Obtain the time index information of several overlapping heartbeat signals of equal length, and perform inverse mapping on the reconstructed standard twelve-lead heartbeat signal to obtain the inversely mapped standard twelve-lead heartbeat signal; By employing a continuity constraint strategy, the overlapping regions of the inversely mapped standard 12-lead heartbeat signals are aggregated to obtain continuously reconstructed standard 12-lead heartbeat signals.

[0015] The second technical solution adopted in this invention is: an electrocardiogram monitoring system based on single-arm upper limb acquisition and twelve-lead reconstruction, comprising: The first module is used to acquire the unilateral upper limb electrocardiogram signal of the target user based on the preset electrocardiogram acquisition path, and to perform signal preprocessing and signal reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead electrocardiogram signal. The second module is used to perform R-wave alignment segmentation processing on the standard twelve-lead ECG signal and to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal. The third module is used to perform R-wave heartbeat aggregation and end-to-end continuous reconstruction based on several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal, so as to obtain a continuously reconstructed standard twelve-lead heartbeat signal.

[0016] The beneficial effects of the method and system of this invention are as follows: This invention acquires the unilateral upper limb ECG signal of the target user based on a preset ECG acquisition path, and performs signal preprocessing and reconstruction to obtain a single-arm lead heartbeat signal and a standard twelve-lead ECG signal. The device is simplified to an integrated armband acquisition structure within the unilateral upper limb area, significantly reducing the number of wearing sites and exposed leads, thus reducing deployment complexity, daily activity interference, and discomfort caused by prolonged wear. Furthermore, the standard twelve-lead ECG signal undergoes R-wave alignment segmentation processing, and the single-arm lead heartbeat signal undergoes alternating optimization adversarial learning to obtain several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal. A standard twelve-lead reconstruction algorithm based on a generative adversarial network is constructed, using an adversarial mechanism between the generator and discriminator. Training establishes a deep nonlinear mapping relationship between single-arm lead signals and standard twelve-lead signals, enabling reconstruction from single-arm input to standard twelve-lead output. This not only expands the number of leads but also effectively converts limited single-arm information into richer twelve-lead information, enhancing the device's application value in anomaly identification and clinical auxiliary diagnosis. Finally, based on several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal, R-wave heartbeat aggregation and end-to-end continuous reconstruction are performed to obtain continuously reconstructed standard twelve-lead heartbeat signals. This effectively connects fixed-length network input with variable-length continuous data processing, achieving end-to-end continuous reconstruction from arbitrary-length single-arm ECG sequences to standard twelve-lead ECG sequences. This allows the system to adapt to real data input formats in long-term continuous monitoring scenarios outside the hospital. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of the electrocardiogram monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction of the present invention; Figure 2 This is a structural block diagram of the electrocardiogram monitoring system based on single-arm upper limb acquisition and twelve-lead reconstruction of the present invention; Figure 3 This is a schematic diagram of the location for acquiring electrocardiogram signals of a single upper limb according to a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the device casing provided in a specific embodiment of the present invention; Figure 5 This is a schematic diagram of an adjustable armband provided in a specific embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware circuit logic provided in a specific embodiment of the present invention; Figure 7 This is a schematic diagram of the functional flow of the host computer software provided in a specific embodiment of the present invention; Figure 8 This is a schematic diagram of the generator model structure provided in a specific embodiment of the present invention; Figure 9 This is a schematic diagram of the discriminator model structure provided in a specific embodiment of the present invention; Figure 10 This is a schematic diagram of the equipment assembly and finished product provided in a specific embodiment of the present invention; Figure 11 This is a schematic diagram of data acquisition and upper computer display test of a single-arm ECG acquisition device provided in a specific embodiment of the present invention; Figure 12 This is a schematic diagram illustrating the verification of unilateral upper limb device wearing and scene adaptability provided in a specific embodiment of the present invention; Figure 13 This is a schematic diagram of the classifier model structure provided in a specific embodiment of the present invention; Figure 14 This is a schematic diagram of the standard limb lead I, the reconstructed twelve leads, and the confusion matrix of the standard twelve leads on the classifier provided in a specific embodiment of the present invention; Figure 15 This is a schematic diagram of the F1 scores for three types of inputs provided in a specific embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0019] First, it should be noted that with the development of wearable electronics, flexible sensing, and wireless communication technologies, ECG signal acquisition is gradually extending from short-term examinations in traditional medical institutions to continuous monitoring in everyday life scenarios. Regarding the surface-based acquisition of cardiac electrical activity, existing ECG monitoring devices have developed into various technical approaches, mainly including portable monitoring devices based on multi-lead or standard lead configurations, chest patch monitoring devices, clothing-integrated monitoring devices, and wrist-integrated monitoring devices.

[0020] Portable monitoring devices based on multi-lead or standard lead configurations typically construct a standard lead system by placing electrodes at multiple locations on the body surface, such as the limbs and chest, to obtain relatively complete electrocardiographic activity information. These devices generally include multi-channel acquisition circuits, analog conditioning circuits, data recording modules, and back-end analysis terminals, enabling continuous recording of electrocardiogram signals within a certain time range. They are commonly used in clinical examinations and short-term dynamic monitoring scenarios. Chest patch-type electrocardiogram monitoring devices integrate the entire device into a patch structure attached to the chest area, acquiring electrocardiogram signals through direct contact with the chest skin. This type of solution has advantages in device miniaturization and wireless functionality, and is suitable for postoperative observation, out-of-hospital monitoring, or continuous acquisition in specific activity scenarios. Wearable clothing electrocardiogram monitoring devices typically integrate conductive fibers and electrode pads into clothing, fabric, or wearable fabric structures, acquiring electrocardiogram signals through contact with the body surface during wear. This type of solution has advantages in concealment, integration, and comfort in certain scenarios, and is suitable for combining with daily wearability for continuous monitoring. Wrist-worn integrated monitoring devices typically integrate ECG detection functionality into consumer-grade devices such as smartwatches and fitness trackers. They use wrist electrodes combined with user contact movements or specific postures to construct a data acquisition path, enabling the collection and display of ECG signals. This type of solution offers advantages such as high integration and ease of daily wear, and has become one of the important development directions for consumer-grade health monitoring devices.

[0021] Furthermore, the basic theoretical basis of lead reconstruction in related technologies is that standard 12-lead ECG signals are not completely independent; there is information redundancy among them. Lead reconstruction can be achieved by mining the common features between a few lead groups or other special lead groups and standard 12-lead ECG signals, and establishing a mapping relationship between them. Through lead reconstruction technology, the complexity of wearable acquisition systems can be greatly reduced while preserving as much information as possible from standard 12-lead ECGs, thus improving the portability of wearable systems and patient comfort.

[0022] In the history of ECG lead reconstruction research, some researchers initially used limb lead I, limb lead II, and chest lead V2 as a baseline lead group, employing least squares and cardiac vector projection theory to calculate the linear transformation coefficients of individual chest leads, thus reconstructing the chest lead signals. However, the human body is a complex, dynamic, nonlinear physical system, and cardiac electrical activity is not a linear process; therefore, linear transformation is not entirely suitable for ECG signal reconstruction. Other researchers used limb lead I, limb lead II, and chest lead V2 as inputs to an artificial neural network (ANN), using the lead signals to be reconstructed as outputs to train the model, completing the reconstruction of a standard twelve-lead ECG signal from three-lead ECG signals. For example, wavelet transform was used to decompose the limb lead I signal into four signals with different characteristics, which were then used as inputs to a time-domain convolutional neural network to generate chest lead signals. Furthermore, some researchers used convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to reconstruct a standard twelve-lead ECG signal from a single lead. Following the significant success of Generative Adversarial Networks (GANs) in image generation, some researchers have employed Conditional Generative Adversarial Networks (GANs) for lead reconstruction. One study used limb lead I as input, generating the remaining eleven leads of a standard twelve-lead ECG signal using eleven generator models. Another example is the use of an R-wave alignment strategy to segment the ECG signal into multiple sets of equal-length heartbeat signals, which were then superimposed as the image input network to generate chest lead signals, improving the accuracy of the generated signals. Overall, existing research has demonstrated the feasibility of reconstructing standard twelve-lead ECG signals from fewer leads, and various technical approaches, ranging from linear methods to deep learning methods, have gradually emerged. This provides an important technical foundation for simplifying wearable front-end acquisition systems while preserving as much information as possible from the standard twelve-lead ECG signal.

[0023] However, the relevant technologies still have the following shortcomings, for example: 1) It is difficult to balance the comfort and stability of electrode arrangement and equipment form during long-term continuous monitoring; Multi-lead dynamic monitoring devices like Holter monitors require electrodes to be placed on multiple parts of the body and connected to the main unit via multiple leads or connecting structures. Due to the large number of electrodes and their dispersed locations, the device typically requires a standardized installation and fixation process before use, with high requirements for electrode placement, lead routing, and fixation methods. While this type of solution can acquire relatively complete ECG activity information and has medical-grade diagnostic value, it often suffers from cumbersome deployment procedures, inconvenience in repeated wear, and significant limitations in movement when used in home environments, community settings, or by non-professionals. This makes it unsuitable for continuous use by individuals requiring long-term, repeated monitoring. In contrast, wearable devices such as clothing, patches, and wristbands offer advantages in portability, concealment, and ease of daily use by reducing the number of electrodes, shrinking the acquisition area, and simplifying the wearing method. Although these solutions align better with the development direction of wearable devices, due to limited acquisition areas, fewer leads, or unstable contact interfaces, the spatial distribution information of ECGs they can acquire is significantly less than that of a standard twelve-lead system, making it difficult to comprehensively reflect the changes in cardiac electrical activity in different directions.

[0024] Therefore, related technologies often face a dilemma in choosing device form factors: While multi-electrode, multi-site placement can obtain richer ECG information, it struggles to meet the requirements of portability, comfort, and high compliance for long-term continuous outpatient monitoring; conversely, a wearable structure with fewer leads, localization, and integration significantly improves ease of use and long-term wearability, but it is prone to reducing effective pathological information due to insufficient lead count and limited acquisition range, failing to achieve the complete clinical information of a 12-lead ECG. In other words, current device form factors struggle to simultaneously meet the two key requirements of device portability and multi-dimensional information acquisition capabilities.

[0025] 2) Limitations of existing lead reconstruction algorithms; While existing standard 12-lead ECG signal reconstruction techniques have been preliminarily proven feasible, they still have significant limitations. Firstly, early lead reconstruction studies were largely based on linear assumptions. However, the human body is a nonlinear system with complex dynamic characteristics; cardiac electrical activity is influenced by multiple physiological factors and does not conform to a simple linear relationship. Therefore, although traditional linear transformation methods can perform lead deduction under certain conditions, their ability to accurately reconstruct complex abnormal waveforms and diverse pathological features is limited, making it difficult to meet the requirements of clinical auxiliary diagnosis for information completeness and detailed accuracy.

[0026] In the field of deep learning, existing research on lead reconstruction using nonlinear methods such as artificial neural networks, convolutional neural networks, long short-term memory networks, and generative adversarial networks has enhanced the ability to model complex mapping relationships. However, existing models often focus more on the overall waveform fitting effect and are still insufficient in preserving key pathological features such as abnormal rhythms, conduction blocks, and specific changes in precordial leads, which can easily lead to weakening or distortion of local details in the reconstruction results. In addition, existing nonlinear lead reconstruction methods usually require a fixed and consistent input signal length, while the data collected in out-of-hospital continuous monitoring scenarios are often dynamic data streams of variable length that last for hours or even days. To adapt to the model input limitations, existing studies can usually only cut long sequences into multiple equal-length segments, or even isolate them into individual heartbeats, which can easily disrupt the temporal continuity of ECG signals and cause loss of information. Therefore, although existing lead reconstruction algorithms have demonstrated the feasibility of extrapolating twelve leads from a few leads, they still lack mature and systematic solutions for preserving pathological features and reconstructing continuous data of variable length.

[0027] Based on this, this invention proposes a unilateral upper limb weak ECG acquisition system for daily life, and introduces an end-to-end multi-lead reconstruction algorithm to reconstruct ECG signals of arbitrary length single arm into standard twelve-lead ECG signals containing rich pathological features. This makes up for the lack of information while ensuring that the device is extremely simple to wear and has high compliance.

[0028] Reference Figure 1 This invention provides a method for electrocardiogram monitoring based on single-arm upper limb acquisition and twelve-lead reconstruction, the method comprising the following steps: S100: Acquire the ECG signal of the target user's unilateral upper limb based on the preset ECG acquisition path, and perform signal preprocessing and signal reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal. S110. The first ECG signal electrode and the second ECG signal electrode are respectively placed on the inner and outer sides of the target user's upper limb, and the reference driving electrode is placed between the first ECG signal electrode and the second ECG signal electrode to obtain the ECG signal of the target user's unilateral upper limb. In this embodiment, the ECG acquisition area is limited to the upper arm of one side of the body, using an armband-type wearing structure. The armband is arranged in a ring around the upper limb and is integrated with the ECG acquisition electrodes, circuit module, and adjustable fixing structure to improve the stability of electrode-skin contact while ensuring ease of wear. The ECG electrodes are disposed on the inner surface of the armband, with two ECG signal electrodes respectively positioned on the inner and outer sides of the upper limb to form the potential difference required for ECG acquisition within the unilateral upper limb area. A reference driving electrode is disposed in the area between the two signal electrodes to form a stable reference driving path in the single-arm structure, thereby enhancing the system's ability to suppress common-mode interference.

[0029] Device signal acquisition location selection, such as Figure 3 As shown, this invention preferably utilizes a relatively stable area on the posterior or upper arm to balance signal acquisition, wearing comfort, and long-term stability. This area has a relatively large arm circumference, facilitating the formation of a longer lead path within a single arm; simultaneously, local tissue deformation is minimal under daily wear conditions, which is more conducive to maintaining stable contact between the electrode and the skin. To accommodate differences in upper limb size among users, the device structure features an adjustable armband structure. Tightness can be adjusted via Velcro or other adjustable fasteners, and the electrode position can be fine-tuned using electrode fixing pads to improve fit. The electrode leads are preferably embedded within the armband, and a serpentine routing structure can be used to provide some flexibility, reducing the impact of lead traction caused by upper limb movement on the contact state and signal acquisition process. The armband material is preferably a skin-friendly, breathable, and elastic fabric to enhance comfort and stability during prolonged wear. The device casing and armband structure are described below. Figure 4 as well as Figure 5 As shown, where, Figure 4 (a) in the image is a front view of the device casing. Figure 4 (b) in the image is the front view of the device casing. Figure 4 (c) in the diagram is the front view of the device casing. Figure 4 (d) in the figure is the front view of the device casing.

[0030] S120. Construct a weak ECG signal conditioning circuit to collect and preprocess the ECG signal of the target user's unilateral upper limb to obtain the preprocessed unilateral upper limb ECG signal. In this embodiment, focusing on the characteristics of low amplitude, unconventional lead paths, and significant dynamic noise interference in single-arm upper limb ECG signals, a collaborative design was implemented for front-end acquisition, analog conditioning, analog-to-digital conversion, power management, and wireless transmission. Single-arm upper limb ECG signals typically range from tens to hundreds of microvolts, with their main effective frequency band concentrated in the 1-40 Hz range. Therefore, the front-end link needs to simultaneously possess high input impedance, low noise amplification, high common-mode rejection capability, and strong anti-interference capability. To address these issues, this embodiment designed the following... Figure 6 The diagram shows a weak signal conditioning circuit. First, the signal acquisition end employs an active electrode and a high input impedance front-end circuit design. The output signal from the ECG acquisition electrode first enters the input protection circuit, where a transient voltage suppressor inhibits electrostatic discharge and transient surges. The protected signal is then further connected to a front-end buffer, positioned close to the acquisition electrode. This buffer provides electrical isolation and buffering for the weak ECG signal acquired from the body surface, thereby reducing the load effect on the weak signal source from subsequent circuits and minimizing signal amplitude attenuation and waveform distortion caused by electrode-skin contact impedance fluctuations.

[0031] The pre - processed ECG signal enters the amplification and conditioning link. While ensuring a high input impedance, it can effectively extract weak differential signals and enhance the ability to suppress common - mode noise. To further improve the anti - interference ability of the front - end, differential equal - length routing design is preferably adopted for key signal channels. The analog conditioning link is constructed in the order of two - stage amplification and two - stage filtering to further suppress out - of - band noise and meet the anti - aliasing requirements. The conditioned signal is input into a high - precision analog - to - digital converter for digital processing. The analog - to - digital converter is preferably a 24 - bit high - precision ADC, which is used to sample the amplified and filtered analog ECG signal with high resolution, enhancing the digital expression ability of low - amplitude differential signals and subtle morphological changes. After analog - to - digital conversion, the signal is further transmitted to the main control module for local storage and communication with the host computer.

[0032] In addition, at the power supply design level, the present invention adopts a partitioned power supply design for the analog link and the digital link. The working voltage of the digital circuit is preferably 3.3 V, and the analog front - end preferably adopts a ±2.5 V symmetric power supply to meet the power supply requirements of the amplifier, conditioning link, and high - precision analog - to - digital converter. At the PCB implementation level, the present invention preferably adopts a four - layer circuit board structure and partitions the layout of the analog acquisition area, digital processing area, and wireless radio frequency area. Among them, the ECG analog front - end is arranged on one side far from the radio frequency antenna to reduce the coupling interference of the wireless communication module on the front - end weak analog signal; the analog ground and digital ground preferably adopt an isolated single - point connection method to reduce ground - loop noise; the power supply routing preferably adopts a star - shaped power supply idea to reduce the mutual interference between different functional modules. Through the above circuit and wiring design, the anti - interference ability and operation stability of the system under wearable and small - volume conditions can be further improved.

[0033] S130: Transmit the pre - processed unilateral upper - limb ECG signal to the host computer for filtering and reconstruction to obtain single - arm lead heartbeat signals and standard 12 - lead ECG signals.

[0034] Specifically, transmit the pre - processed unilateral upper - limb ECG signal to the host computer for data decoding to obtain the decoded unilateral upper - limb ECG signal; perform FIR filtering and signal reconstruction based on the decoded unilateral upper - limb ECG signal respectively, and cache them in a preset data format to obtain single - arm lead heartbeat signals and standard 12 - lead ECG signals.

[0035] In this embodiment, the present invention also includes a host computer software platform supporting the single - arm ECG acquisition system, which is used to implement device connection management, wireless data reception, real - time waveform display, pre - processing, signal reconstruction call, and local storage. The functional flow of the host computer software is as Figure 7 shown.

[0036] After the host computer software starts, it first checks whether the current computer is connected to the Wi-Fi local area network established by the device. Upon successful connection, it establishes a communication link with the device and begins receiving data frames sent by the device. At the receiving end, the host computer continuously reads the wireless data stream and decodes it according to a preset frame format. Data frames may contain fixed frame headers and trailers for frame boundary identification. After receiving a complete data frame, the host computer further parses the ECG data, status information, and time sequence information within it, and sends the ECG data to subsequent modules.

[0037] In the preprocessing section, a 50 Hz FIR power frequency notch filter module can be configured, with a stopband range preferably between 49-51 Hz, to perform preliminary interference suppression on the raw ECG data before signal reconstruction. In the signal reconstruction section, the host computer calls the pre-trained twelve-lead reconstruction model to process the input single-arm ECG signal and output the corresponding standard twelve-lead reconstruction result. In the data display section, a graphical chart control can be used to continuously refresh waveforms. The display interface includes a control area, a device status area, and an ECG display area. Simultaneously, the host computer supports real-time saving of both the raw and reconstructed signals; saved files can be in .txt or binary data file format, with each record including a time sequence number, the raw ECG value, and the reconstructed ECG value. The software can create a data directory named after the date locally by default, and a path configuration option can also be provided in the interface for users to customize the save path.

[0038] S200, R-wave alignment and segmentation processing of standard twelve-lead ECG signal and alternating optimization adversarial learning of single-arm lead heartbeat signal, to obtain several overlapping heartbeat signals of equal length and reconstructed standard twelve-lead heartbeat signal; S210. The standard twelve-lead ECG signal is subjected to amplitude normalization and R-wave alignment segmentation in sequence to obtain several overlapping heartbeat signals of equal length. Specifically, the amplitude of the standard 12-lead ECG signal is normalized to obtain a normalized standard 12-lead ECG signal; based on the adaptive threshold peak detection method, R-wave detection is performed on the normalized standard 12-lead ECG signal to determine the R-wave position; using the R-wave position as the time anchor point, the normalized standard 12-lead ECG signal is synchronously truncated into heartbeat segments according to a preset length to obtain several overlapping heartbeat signals of equal length.

[0039] In this embodiment, by analyzing the information redundancy between standard 12-lead ECG signals and considering practical application requirements, this invention selects single-lead ECG signals acquired by the device to complete the reconstruction of standard 12-lead ECG signals. Since the amplitude of ECG signals varies significantly among different individuals, in order to standardize the input and output signal paradigms of the nonlinear neural network, the standard 12-lead ECG signals in the database undergo standardization preprocessing to reduce signal amplitude differences. The standardization formula is as follows: In the above formula, Indicates electrocardiogram (ECG) signal. Indicates the maximum value. Represents absolute value. This indicates a single-arm electrocardiogram signal.

[0040] Through the above standardization process, the amplitude differences between different samples can be reduced while preserving the relative morphological relationships of each lead, thereby improving the convergence stability and reconstruction consistency during model training.

[0041] Since generative nonlinear reconstruction models typically require input data of uniform length, and single-arm ECG signals acquired in out-of-hospital continuous monitoring scenarios often have variable durations, ranging from seconds to hours or even longer, the original continuous data cannot be directly input into the reconstruction network. Therefore, this invention employs a heartbeat segmentation strategy based on R-wave alignment to divide the variable-length continuous ECG signal into fixed segments of uniform length, thus balancing network input requirements with the alignment needs of key ECG events. Specifically, R-wave detection is first performed on the original ECG signal, using each R-wave position as a time anchor point. Heartbeat segments are synchronously extracted before and after these anchor points according to a preset length, forming multiple overlapping heartbeat signals of equal length. In a preferred embodiment, the length of each heartbeat segment is set to 1024 sampling points; for heartbeat segments located at the beginning or end of the original signal that result in insufficient extraction length, the length can be supplemented using zero-padding. It should be noted that 1024 sampling points is only one preferred segmentation length of this invention; in actual implementation, other fixed length parameters can be selected based on the sampling rate, network input dimension, target application scenario, and reconstruction accuracy requirements.

[0042] For R-wave detection, firstly, a moving average is calculated on the normalized standard 12-lead ECG signal, and a threshold is dynamically set based on the moving average at the current moment; then, the data points in the normalized standard 12-lead ECG signal that exceed the threshold are marked as regions of interest; finally, all regions of interest are traversed, and the data point with the largest amplitude in all regions of interest is selected as the R-wave position.

[0043] More specifically, an adaptive threshold peak detection method is used to identify R-waves. This method calculates a moving average for the input sequence in real time and dynamically calculates a threshold based on the current moving average. Data points exceeding the threshold are marked as Regions of Interest (ROIs), and then each ROI is traversed to select the point with the largest amplitude as the R-wave location. This detection method can stably identify R-waves even under conditions of baseline drift and noise interference, and has advantages such as simple implementation, low computational cost, and suitability for continuous data processing. By employing an R-wave alignment segmentation strategy, this invention can establish an effective connection between variable-length continuous inputs and fixed-length network inputs, while maintaining the temporal consistency of key ECG events within each heartbeat segment as much as possible, providing a more stable data foundation for subsequent generative models to learn the correspondence between single-arm leads and twelve leads.

[0044] S220. Construct a generative adversarial network model and perform alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain the reconstructed standard twelve-lead heartbeat signal.

[0045] Specifically, a generative adversarial network (GAN) model is constructed, comprising a generator and a discriminator. Based on the generator of the GAN model, deep latent feature extraction and deep feature recovery processing are performed on the single-arm lead heartbeat signal to obtain a reconstructed twelve-lead signal. Based on the discriminator of the GAN model, feature recognition is performed on the reconstructed twelve-lead signal to output the probability of the real standard twelve-lead heartbeat signal. Based on the probability of the real standard twelve-lead heartbeat signal, the single-arm lead heartbeat signal is subjected to alternating optimization adversarial learning through the GAN model to obtain the reconstructed standard twelve-lead heartbeat signal.

[0046] In this embodiment, after preprocessing and R-wave alignment segmentation, the present invention further employs a deep model based on a generative adversarial network (GAN) to reconstruct the standard twelve-lead heartbeat signal from the single-arm heartbeat signal. The model comprises a generator and a discriminator, which establishes a nonlinear mapping relationship between the single-arm lead heartbeat signal and the standard twelve-lead heartbeat signal through adversarial training. Unlike conventional networks that rely solely on mean square error for point-by-point fitting, this embodiment introduces a discriminator to impose distribution constraints on the generated results. This ensures that the generator not only approximates the true twelve-lead signal numerically but also closely approximates the true signal in terms of overall morphology, local details, and multi-lead spatial relationships, thereby improving the effectiveness of the reconstructed signal in expressing pathological features.

[0047] The input signal size of the generator is That is, a single-arm heartbeat signal with a length of 1024 sampling points; the output signal magnitude is This refers to a standard twelve-lead heartbeat signal with 1024 sampling points. Based on the actual generation results, the generator model was improved by introducing ResNet residual connections into the one-dimensional U-Net framework. The overall structure includes an eight-layer G-block coding structure and an eight-layer Upblock decoding structure, as follows... Figure 8 As shown. The encoding part is used to progressively extract deep latent features from the input single-arm leads, while the decoding part is used to restore the deep features to the standard twelve-lead output. Skip connections between corresponding layers preserve shallow local details, thus balancing global semantic features and the ability to restore local waveform details. The introduction of residual structures helps improve the training stability of deep networks and enhances the model's ability to learn complex nonlinear mapping relationships. Regarding the training objective, the generator loss function consists of two parts: generative adversarial loss and reconstruction loss, which can be expressed as: in, This indicates the input single-arm heartbeat signal. This indicates a true standard 12-lead heartbeat signal. This represents the reconstructed twelve-lead signal output by the generator. This indicates the probability that the discriminator determines the reconstructed result to be a true signal. Represents the mathematical expectation. To generate weighting coefficients between adversarial loss and reconstruction loss. In a specific embodiment, Set to 200 to enhance the realism of the generated results in terms of data distribution while ensuring the overall waveform fitting ability.

[0048] like Figure 9 As shown, the discriminator is used to distinguish whether the input signal is a genuine standard 12-lead heartbeat signal or a reconstructed heartbeat signal output by the generator. The input size of the discriminator is... The output value is 1, representing the probability that the input signal is a true standard 12-lead heartbeat signal. To enhance the ability to discriminate multi-scale features of ECG signals, this embodiment employs a multi-scale fusion discriminator structure. By fusing local and global features extracted from different levels by a convolutional network, the ability to identify differences between the generated signal and the real signal is improved. The discriminator loss function can be expressed as: in, This indicates the discriminator's judgment result on the true standard twelve-lead heartbeat signal. This indicates the judgment result of the generator output.

[0049] During training, the generator and discriminator engage in adversarial learning through alternating optimization. The discriminator is trained to distinguish between the real and reconstructed signals as accurately as possible, while the generator is trained to "deceive" the discriminator as much as possible while maintaining overall similarity to the real signal, making the reconstructed result approximate the real 12-lead signal in both distribution and morphology. Through this adversarial training process, the generator can gradually capture deep latent features shared with the standard 12-lead heartbeat signal from the single-arm lead heartbeat signal, establishing a stable and effective nonlinear mapping relationship. Compared to methods that rely solely on a single reconstruction error constraint, this method is more effective in recovering complex pathological morphologies in the 12-lead signal, and is particularly suitable for improving the auxiliary identification capabilities of single-arm wearable systems for arrhythmias and other abnormal events.

[0050] S300: Based on several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal, R-wave heartbeat aggregation and end-to-end continuous reconstruction are performed to obtain a continuously reconstructed standard twelve-lead heartbeat signal.

[0051] S310. Obtain the time index information of several overlapping heartbeat signals of equal length, and perform reverse mapping on the reconstructed standard twelve-lead heartbeat signal to obtain the reverse-mapped standard twelve-lead heartbeat signal. S320. By using a continuity constraint strategy, the overlapping regions of the standard twelve-lead heartbeat signals after inverse mapping are aggregated to obtain continuously reconstructed standard twelve-lead heartbeat signals.

[0052] In this embodiment, an R-wave alignment strategy is used to segment continuous single-arm ECG signals of different lengths into multiple overlapping heartbeat segments with 1024 sampling points each. A generator is then used to reconstruct a standard 12-lead heartbeat signal for each segment. To further achieve end-to-end output from raw single-arm continuous data to a standard 12-lead continuous signal, this invention further constructs a reverse aggregation mechanism based on heartbeat-level reconstruction, restoring multiple reconstructed heartbeat segments back into a standard 12-lead continuous signal with the same length as the original input.

[0053] Specifically, after generating the twelve leads for each cardiac beat segment, the system reverse-maps each reconstructed segment based on the time index information recorded during the aforementioned R-wave alignment and segmentation phase, restoring it to the continuous time axis according to its time position in the original ECG signal. For regions where adjacent cardiac beat segments overlap, weighted averaging, smooth splicing, or other continuity constraint strategies can be used for aggregation to reduce potential waveform abrupt changes and splicing traces at segment boundaries, thereby improving the smoothness and overall consistency of the reconstructed signal on the time axis.

[0054] Through the aforementioned reverse aggregation process, this embodiment can restore fragmented beat-level reconstruction results into a standard twelve-lead continuous ECG signal with the same length as the original single-arm input, completing end-to-end reconstruction of sequences of arbitrary length. This mechanism overcomes the limitation of fixed-length input in nonlinear neural networks, making this invention suitable not only for standardized data processing under short-term static recording conditions but also for dynamic data streams of varying lengths lasting several hours or even days in out-of-hospital continuous monitoring scenarios. Therefore, while achieving extreme simplification of the front-end acquisition system, this invention can still output twelve-lead ECG signals with continuous temporal structure and high pathological information value, providing a more complete data foundation for subsequent anomaly screening, trend tracking, and clinical auxiliary interpretation.

[0055] In summary, this invention provides a single-arm ECG acquisition and standard 12-lead reconstruction system for long-term continuous monitoring in out-of-hospital settings. This technical solution uses one upper limb as the wearing and acquisition area. Through coordinated design of the device's hardware and software systems and the ECG signal reconstruction method, it achieves stable acquisition of weak single-arm ECG signals in dynamic scenarios and continuous reconstruction of standard 12-lead ECG signals. This technical solution can complete single-arm ECG signal acquisition while minimizing wearing burden and deployment complexity, and can output standard 12-lead ECG signals with richer pathological information through backend reconstruction methods.

[0056] Therefore, the embodiments of the present invention have the following distinguishing technical features compared to the prior art: 1) While ensuring comfort during long-term single-arm wear, it achieves high-quality and stable acquisition of weak ECG signals. It integrates electrode acquisition and arm-cuff structure within a single upper limb, greatly simplifying system deployment and reducing the burden on patients during long-term wear, while maintaining continuous stability of the electrode-skin contact interface during dynamic activities. Furthermore, considering the characteristics of unconventional lead pathways, it enhances the system's ability to sense and amplify weak ECG signals without loss through collaborative optimization of the underlying hardware circuitry.

[0057] 2) To achieve continuous reconstruction of single-arm signals into standard 12-lead signals while ensuring information integrity and preserving key features, a segmentation and aggregation mechanism suitable for neural network processing and maintaining temporal continuity is constructed for variable-length single-arm ECG data acquired continuously outside the hospital. This enables continuous reconstruction of sequences of arbitrary length. Simultaneously, a stable and effective nonlinear mapping relationship is established between single-arm signals and standard 12-lead signals. This ensures that the reconstructed results not only closely approximate the overall waveform of the real 12-lead signal but also retain key pathological features relevant to disease identification as much as possible, thereby enhancing the application value of single-arm wearable devices in clinical auxiliary diagnosis.

[0058] The embodiments of the present invention have the following advantages over the prior art: 1) The single-arm wearable integrated data acquisition structure significantly improves the convenience and compliance of long-term continuous monitoring; This invention simplifies the device into an integrated armband acquisition structure within a single upper limb, significantly reducing the number of wearing sites and exposed leads, thus lowering deployment complexity, interference with daily activities, and discomfort caused by prolonged wear. This single-arm armband acquisition solution, while ensuring ECG signal availability, centrally fixes the device to one upper limb, balancing concealment, comfort, and continuity, better meeting the practical needs of long-term continuous use in out-of-hospital environments. Therefore, this invention significantly surpasses traditional multi-lead portable devices in terms of product form and user experience, providing a reliable front-end carrier foundation for achieving high-compliance long-term monitoring.

[0059] 2) The high-precision acquisition link design improves the quality of raw data and system stability; This invention focuses on the characteristics of weak single-arm ECG signals and optimizes input protection, front-end buffering, instrumentation amplification, multi-stage filtering, analog-to-digital conversion, power management, and PCB layout in a coordinated manner. By setting a front-end buffer circuit with high input impedance and low bias current, the impact of load effects and contact impedance variations on the raw signal is effectively reduced. The combination of instrumentation amplification and multi-stage analog conditioning links suppresses low-frequency drift and high-frequency noise while increasing the effective signal amplitude. High-resolution ADC, analog / digital partitioned power supply, and isolation between the front-end and RF areas enhance the acquisition capability and anti-interference ability of weak ECG signals. This ensures that the device can stably acquire high-quality raw ECG signals under simplified single-arm acquisition conditions, providing a reliable data foundation for subsequent twelve-lead reconstruction.

[0060] 3) The twelve-lead reconstruction algorithm for pathological feature preservation breaks through the application bottleneck of insufficient signal information in a single arm; This invention constructs a standard 12-lead reconstruction algorithm based on a generative adversarial network. Through adversarial training between the generator and discriminator, a deep nonlinear mapping relationship is established between single-arm lead signals and standard 12-lead signals, achieving reconstruction from single-arm input to standard 12-lead output. The algorithm takes single-arm heartbeat signals as input and standard 12-lead heartbeat signals as the target output, further restoring and preserving key pathological features based on overall waveform reconstruction. In this way, this invention not only expands the number of leads but also achieves an effective conversion from limited single-arm information to richer 12-lead information, enhancing the application value of the device in anomaly identification and clinical auxiliary diagnosis.

[0061] 4) The end-to-end continuous reconstruction mechanism supporting variable-length sequences improves the practicality in long-range monitoring scenarios outside the hospital; This invention proposes a segmentation and reverse aggregation mechanism based on an R-wave alignment strategy to process variable-length dynamic electrocardiogram (ECG) data commonly encountered in out-of-hospital continuous monitoring scenarios. Specifically, the original continuous single-arm ECG signal is first segmented into heartbeat segments of uniform length using the R-wave as the time anchor point, and each segment is input into a reconstruction model to generate a 12-lead heartbeat. Subsequently, the segments are reverse-aggregated according to the original time index to restore a continuous 12-lead ECG signal with the same input length. Through this mechanism, this invention effectively connects fixed-length network input with variable-length continuous data processing, achieving end-to-end continuous reconstruction of arbitrary-length single-arm ECG sequences into standard 12-lead ECG sequences, enabling the system to adapt to real-world data input formats in long-term out-of-hospital continuous monitoring scenarios.

[0062] Finally, referring to the accompanying drawings of specific embodiments, to verify the feasibility and effectiveness of the unilateral upper limb ECG monitoring system and its signal reconstruction method proposed in this invention, three parts of experiments were conducted: device hardware assembly and software verification, wearing and scenario adaptability verification, and signal reconstruction verification. Experimental results show that the unilateral upper limb wearable ECG acquisition device proposed in this invention can achieve stable data acquisition, transmission, and display under actual wearing conditions, and can maintain good wearing stability under different body positions and usage states. Simultaneously, the single-arm ECG residual reconstruction method oriented towards preserving pathological features can significantly improve the quality of the original single-arm ECG signal, better preserving key ECG features while reducing noise pollution, providing a more reliable data foundation for subsequent abnormality identification and clinical interpretation.

[0063] Furthermore, to verify the engineering feasibility of the hardware system proposed in this invention, the structural assembly and finished product manufacturing of the device were first completed. The assembly process mainly includes steps such as electrode assembly, hardware circuit module assembly, shell encapsulation, and armband connection and fixation. Specifically, electrodes are first installed at preset positions on the inside of the armband to ensure stable contact between the ECG acquisition electrodes and the skin while worn. Figure 10 As shown in (a) above. Subsequently, the battery module and the acquisition circuit board are assembled inside the device housing, completing the integration of the core hardware unit, as shown below. Figure 10 As shown in (b) above. Based on this, further completion of the outer casing and arm strap connection yields the finished product. (See diagram below.) Figure 10 (c) in the image is a top view of the equipment, which reflects its overall appearance after assembly; such as Figure 10 (d) in the diagram is a side view, which shows the connection between the equipment and the adjustable armband; for example... Figure 10(e) in the figure is a bottom view, which shows the bottom structure of the device and the shape of the wearing contact surface. From the assembly result, the single-arm arm ring structure proposed in this invention can complete the integrated arrangement of electrodes, acquisition circuit, power supply module and wearing components, forming a finished device with complete structure, compact size and suitable for wearing on one upper limb.

[0064] To further demonstrate the device's software performance, data acquisition and display functions were tested. In the experiment, the single-arm ECG acquisition device was worn on the subject's upper arm, and real-time data acquisition was performed on different subjects under different wearing conditions. The original acquired waveforms and processed waveforms were synchronously displayed via host computer software. The test results are as follows: Figure 11 As shown in the figure, the single-arm ECG acquisition device proposed in this invention can continuously output stable electrocardiogram signals when worn, and display them in real time through host computer software. The left side of the figure shows the wearing effect of two different subjects, and the right side shows the real-time acquisition and display of the single-arm ECG signal. Clear periodic ECG waveform changes can be observed in the acquired signal, indicating that the hardware acquisition link, wireless transmission link, and host computer receiving and display module of this invention can achieve stable collaborative operation.

[0065] The above experiments show that the present invention has not only completed the integrated design of the device at both the hardware and software levels, but also can realize the stable acquisition, transmission and display of actual electrocardiogram data. This verifies that the device has the ability to acquire ECG signals from a single-arm position, and also shows that the proposed system has good engineering feasibility and practical application value.

[0066] Based on the verification that the device can complete ECG data acquisition and display, a device wearing and acquisition scenario verification experiment was conducted to further verify the feasibility of the proposed unilateral upper limb ECG acquisition system in practical applications. In the experiment, the unilateral ECG acquisition device of this invention was worn on one arm of the subject, and a medical Holter monitor was used as a reference acquisition system. Data was simultaneously collected from the subject in different positions, including standing, natural standing posture, sitting posture, and lying position. The device verification diagram is shown below. Figure 12 As shown, the unilateral upper limb data acquisition device of the present invention can be stably fixed in the upper limb area, maintaining a good wearing condition under different body positions and postures. No obvious detachment, displacement, or failure caused by wire traction has been observed, indicating that the proposed arm-ring structure and its matching adjustment and fixation method have good wearing stability and scene adaptability. Compared with the traditional medical Holter which relies on multiple electrodes attached to the chest and multiple wire connections, the present invention concentrates the acquisition unit in the unilateral upper limb area, significantly simplifying the wearing process, reducing the burden of use caused by multiple placements, and at the same time having a usage form that is closer to that of daily wearable devices.

[0067] The above verification results show that the unilateral upper limb wearing structure of the present invention can meet the basic requirements of long-term continuous monitoring in non-clinical environments, proving the feasibility of the single-arm acquisition scheme in practical applications, and also verifying the effectiveness of the arm-ring integrated structure in terms of wearing convenience, stability and continuous acquisition capability.

[0068] To evaluate the effectiveness of the proposed 12-lead reconstruction method in waveform morphology recovery, the correlation coefficient was used to assess the similarity between the reconstructed 12-lead ECG signal and the actual standard 12-lead ECG signal. In this experiment, the standard limb lead I ECG signal of length x from the CPSC2018 test set was segmented into N segments of heartbeat signals with 1024 sampling points each. These N segments of limb lead I heartbeat signals were then input into the generator model to reconstruct the corresponding N segments of 12-lead heartbeat signals. Subsequently, the reconstructed multiple 12-lead heartbeat segments were re-aggregated into a continuous 12-lead ECG signal of length x using the R-wave aggregation algorithm, thus completing the reconstruction process from single-lead input to continuous 12-lead output. For the continuous 12-lead ECG signal after R-wave aggregation, the correlation coefficient between it and the actual standard 12-lead ECG signal was calculated to evaluate the reconstruction effect of the algorithm described in this invention at the continuous signal level. The correlation coefficient results of the reconstructed 12-lead ECG signal on nine types of data are shown in Table 1.

[0069] Table 1. Correlation coefficient data of reconstructed standard 12-lead ECG signals Furthermore, to assess whether the generated standard 12-lead ECG signal contains effective and accurate pathological feature information, a classifier model was trained to classify the standard limb lead I, the real standard 12-lead ECG, and the generated standard 12-lead ECG signal, respectively. The effectiveness of the pathological features contained in the generated standard 12-lead ECG signal was judged by comparing the accuracy of the three signal classifications.

[0070] This invention employs a classifier model trained on the CPSC2018 standard 12-lead dataset. The model performs nine classifications on the input standard limb lead I, standard 12-lead, and reconstructed standard 12-lead ECG signals. The classifier structure is as follows: Figure 13 As shown, the loss function formula is as follows: in, This is the expected output in one-hot encoding, where only the element at the actual position is 1 and all other elements are 0. Indicates input electrocardiogram signal The predicted category probability.

[0071] A confusion matrix is ​​used to illustrate the correspondence between the predicted and true classes of a classification model on the test dataset. The rows of the matrix represent the actual classes, and the columns represent the predicted classes. The confusion matrix of the classification results is shown below. Figure 14 As shown, Figure 14 In the diagram, (a) represents the confusion matrix of standard limb lead I on the classifier. Figure 14 In the diagram, (b) represents the confusion matrix of the reconstructed twelve leads on the classifier. Figure 14 In the diagram, (c) represents the confusion matrix of the standard twelve leads on the classifier.

[0072] Among them, Normal indicates healthy ECG data, AF indicates atrial fibrillation, I-AVB indicates first-degree atrioventricular block, LBBB indicates left bundle branch block, RBBB indicates right bundle branch block, PAC indicates premature atrial contractions, PVC indicates premature ventricular contractions, STD indicates ST segment depression, and STE indicates ST segment elevation.

[0073] Based on the confusion matrix, the number of true positive (TP), true negative (TN), false positive (FP), and false negative (FN) samples is counted. The F1 score is calculated using the following formula, and the F1 scores of the standard limb lead I, the true standard 12 leads, and the reconstructed standard 12 leads ECG signals on the classifier are listed as follows: Figure 15 As shown, the expression for the F1 score is: According to the above formula, the F1 score of the reconstructed standard 12-lead ECG signal on the classifier is greater than that of the standard limb lead I ECG signal but less than that of the true standard 12-lead ECG signal. In other words, the performance of the reconstructed standard 12-lead ECG signal in the classifier falls between that of the standard limb lead I ECG signal and the true standard 12-lead ECG signal. The experimental results indicate that, compared with the standard limb lead I ECG signal, the reconstructed standard 12-lead ECG signal provides more effective pathological feature information and has potential application value in the initial screening and classification of cardiovascular diseases.

[0074] Reference Figure 2 An electrocardiogram monitoring system based on single-arm upper limb acquisition and twelve-lead reconstruction includes: The first module 201 is used to acquire the unilateral upper limb electrocardiogram signal of the target user based on the preset electrocardiogram acquisition path, and to perform signal preprocessing and signal reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead electrocardiogram signal. The second module 202 is used to perform R-wave alignment segmentation processing on the standard twelve-lead ECG signal and to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal. The third module 203 is used to perform R-wave beat aggregation and end-to-end continuous reconstruction based on several overlapping beat signals of equal length and the reconstructed standard twelve-lead beat signal to obtain a continuously reconstructed standard twelve-lead beat signal.

[0075] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0076] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for electrocardiogram monitoring based on single-arm upper limb acquisition and twelve-lead reconstruction, characterized in that, Includes the following steps: The ECG signal of the target user’s unilateral upper limb is acquired based on the preset ECG acquisition path, and the signal is preprocessed and reconstructed to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal. R-wave alignment segmentation was performed on the standard 12-lead ECG signal, and alternating optimization adversarial learning was performed on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal. Based on several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal, R-wave heartbeat aggregation and end-to-end continuous reconstruction are performed to obtain a continuously reconstructed standard 12-lead heartbeat signal.

2. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 1, characterized in that, The step of acquiring the unilateral upper limb ECG signal of the target user based on a preset ECG acquisition path, and performing signal preprocessing and reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal specifically includes: The first ECG signal electrode and the second ECG signal electrode are respectively placed on the inner and outer sides of the target user's upper limb, and the reference driving electrode is placed between the first ECG signal electrode and the second ECG signal electrode to acquire the ECG signal of the target user's unilateral upper limb. A weak ECG signal conditioning circuit was constructed to collect and preprocess the unilateral upper limb ECG signal of the target user, and obtain the preprocessed unilateral upper limb ECG signal. The preprocessed unilateral upper limb electrocardiogram signal is transmitted to the host computer for filtering and reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead electrocardiogram signal.

3. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 2, characterized in that, The weak ECG signal conditioning circuit specifically includes an input protection circuit, a front-end buffer, an amplification and conditioning link, and an analog-to-digital converter, wherein: The input protection circuit is used to suppress electrostatic discharge and transient impact on the electrocardiogram signal of one side of the target user's upper limb, so as to obtain the suppressed electrocardiogram signal of one side of the upper limb. The front-end buffer is used to electrically isolate and buffer the suppressed unilateral upper limb electrocardiogram signal to obtain a buffered unilateral upper limb electrocardiogram signal. The amplification and conditioning link is used to suppress noise in the buffered unilateral upper limb electrocardiogram signal to obtain an effective unilateral upper limb electrocardiogram signal. The analog-to-digital converter is used to perform high-resolution sampling and analog-to-digital conversion processing on the valid unilateral upper limb electrocardiogram signal to obtain a preprocessed unilateral upper limb electrocardiogram signal.

4. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 3, characterized in that, The step of transmitting the preprocessed unilateral upper limb ECG signal to the host computer for filtering and reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead ECG signal specifically includes: The preprocessed unilateral upper limb electrocardiogram signal is transmitted to the host computer for data decoding to obtain the decoded unilateral upper limb electrocardiogram signal. Based on the decoded unilateral upper limb electrocardiogram signal, FIR filtering and signal reconstruction were performed, and the data was buffered using a preset data format to obtain the unilateral lead heartbeat signal and the standard twelve-lead electrocardiogram signal.

5. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 4, characterized in that, The step of performing R-wave alignment segmentation on the standard 12-lead ECG signal and alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal specifically includes: The standard twelve-lead electrocardiogram signal was subjected to amplitude normalization and R-wave alignment segmentation in sequence to obtain several overlapping heartbeat signals of equal length. A generative adversarial network model was constructed to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal, resulting in the reconstructed standard twelve-lead heartbeat signal.

6. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 5, characterized in that, The step of sequentially performing amplitude normalization and R-wave alignment segmentation on the standard twelve-lead electrocardiogram signal to obtain several overlapping heartbeat signals of equal length specifically includes: The amplitude of the standard twelve-lead electrocardiogram signal was normalized to obtain the normalized standard twelve-lead electrocardiogram signal. Based on the adaptive threshold peak detection method, the R wave is detected in the normalized standard twelve-lead ECG signal to determine the R wave location. Using the R-wave position as the time anchor point, the normalized standard twelve-lead ECG signal is synchronously truncated into heartbeat segments according to the preset length, resulting in several overlapping heartbeat signals of equal length.

7. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 6, characterized in that, The step of detecting the R wave and determining the R wave position in the normalized standard twelve-lead ECG signal based on the adaptive threshold peak detection method specifically includes: The moving average of the normalized standard twelve-lead ECG signal is calculated, and the threshold is dynamically set based on the moving average at the current moment. The locations of data points exceeding the threshold in the normalized standard twelve-lead electrocardiogram signal are marked as regions of interest; Traverse all regions of interest and select the data point with the largest amplitude among all regions of interest as the R-wave position.

8. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 7, characterized in that, The step of constructing a generative adversarial network model to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain the reconstructed standard twelve-lead heartbeat signal specifically includes: Construct a generative adversarial network (GAN) model, which includes a generator and a discriminator; A generator based on a generative adversarial network model is used to perform deep latent feature extraction and deep feature recovery processing on the single-arm lead heartbeat signal to obtain a reconstructed twelve-lead signal. A discriminator based on a generative adversarial network model is used to identify features of the reconstructed twelve-lead signal and output the probability of the true standard twelve-lead heartbeat signal. Based on the probability of real standard 12-lead heartbeat signals, the reconstructed standard 12-lead heartbeat signals are obtained by alternately optimizing adversarial learning on single-arm lead heartbeat signals through a generative adversarial network model.

9. The ECG monitoring method based on single-arm upper limb acquisition and twelve-lead reconstruction according to claim 8, characterized in that, The step of performing R-wave heartbeat aggregation and end-to-end continuous reconstruction based on several overlapping heartbeat signals of equal length and the reconstructed standard 12-lead heartbeat signal to obtain a continuously reconstructed standard 12-lead heartbeat signal specifically includes: Obtain the time index information of several overlapping heartbeat signals of equal length, and perform inverse mapping on the reconstructed standard twelve-lead heartbeat signal to obtain the inversely mapped standard twelve-lead heartbeat signal; By employing a continuity constraint strategy, the overlapping regions of the inversely mapped standard 12-lead heartbeat signals are aggregated to obtain continuously reconstructed standard 12-lead heartbeat signals.

10. An electrocardiogram monitoring system based on single-arm upper limb acquisition and twelve-lead reconstruction, characterized in that, Includes the following modules: The first module is used to acquire the unilateral upper limb electrocardiogram signal of the target user based on the preset electrocardiogram acquisition path, and to perform signal preprocessing and signal reconstruction to obtain the single-arm lead heartbeat signal and the standard twelve-lead electrocardiogram signal. The second module is used to perform R-wave alignment segmentation processing on the standard twelve-lead ECG signal and to perform alternating optimization adversarial learning on the single-arm lead heartbeat signal to obtain several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal. The third module is used to perform R-wave heartbeat aggregation and end-to-end continuous reconstruction based on several overlapping heartbeat signals of equal length and the reconstructed standard twelve-lead heartbeat signal, so as to obtain a continuously reconstructed standard twelve-lead heartbeat signal.

Images