Non-contact heart mechanical activity event real-time monitoring method and equipment

Abstract

The invention provides a non-contact heart mechanical activity event real-time monitoring method and device. The method comprises the steps that a thoracic cavity micro-motion signal and a periodic rhythm signal are extracted from millimeter wave echo signals, collected in real time in a non-contact mode, of a target subject; and respectively inputting the chest micro-motion signals and the periodic rhythm signals into a heart event identification model, so that the model respectively extracts feature data corresponding to the chest micro-motion signals and the periodic rhythm signals and splices the feature data to obtain heart joint feature data, and decoding and multi-label classification processing are sequentially carried out on the heart joint feature data to output heart mechanical activity event identification result data. The result data comprises heart mechanical activity event type labels and time sequence positioning data. According to the invention, while the heart mechanical activity event identification efficiency is ensured, the identification accuracy of the transient event can be effectively improved and event type identification can be realized, so that the accuracy and robustness of real-time monitoring of the non-contact heart mechanical activity event can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of cardiac mechanical event monitoring technology, and in particular to a non-contact method and device for real-time monitoring of cardiac mechanical activity events. Background Technology

[0002] Dynamic monitoring of cardiac mechanical events is a core tool for assessing cardiac function and diagnosing diseases such as valvular disease and myocardial ischemia. Abnormal cardiac mechanical activity often evolves continuously from minutes to hours before its onset. If continuous monitoring at the millisecond level is possible, it may be possible to identify risks in advance and intervene accordingly.

[0003] Currently, methods such as non-contact atrial fibrillation detection using millimeter-wave radar exist, improving the convenience and efficiency of atrial fibrillation detection. However, existing millimeter-wave systems still face significant challenges in cardiac event analysis. Cardiac micromotion signals are easily masked by respiratory and body motion noise, blurring event boundaries and affecting the accuracy of real-time monitoring results. Furthermore, threshold-based peak localization methods have high resolution errors for overlapping events, further impacting the accuracy of real-time monitoring results and failing to distinguish between different types of cardiac mechanical events. These issues, to some extent, limit the practical application of millimeter-wave technology in clinical diagnosis. Summary of the Invention

[0004] In view of this, embodiments of this application provide a non-contact method and device for real-time monitoring of cardiac mechanical activity events, in order to eliminate or improve one or more defects existing in the prior art.

[0005] One aspect of this application provides a method for real-time monitoring of non-contact cardiac mechanical activity events, comprising: The thoracic micromotion signal and periodic rhythm signal are extracted from the millimeter-wave echo signal of the target subject acquired in real time using a non-contact method. The thoracic micromotion signal and the periodic rhythm signal are respectively input into the cardiac event recognition model, so that the cardiac event recognition model extracts the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively and splices them to obtain cardiac joint feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event recognition result data includes: cardiac mechanical activity event type label and temporal positioning data.

[0006] In some embodiments of this application, the cardiac event recognition model includes: a chest signal encoder, a rhythm signal encoder, a feature fusion layer, and a decoder; The input terminal of the feature fusion layer is connected to the output terminals of the chest cavity signal encoder and the rhythm signal encoder, respectively; the output terminal of the feature fusion layer is connected to the input terminal of the decoder. Correspondingly, the process of inputting the thoracic micromotion signal and the periodic rhythm signal into the cardiac event recognition model allows the model to extract the feature data corresponding to each of the thoracic micromotion signal and the periodic rhythm signal, and concatenate them to obtain cardiac joint feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output the cardiac mechanical activity event recognition result data corresponding to the millimeter-wave echo signal, including: The thoracic micromotion signal is input into the thoracic signal encoder in the cardiac event recognition model, and the periodic rhythm signal is input into the rhythm signal encoder in the cardiac event recognition model, so that the thoracic signal encoder extracts the feature data corresponding to the thoracic micromotion signal, and the rhythm signal encoder extracts the feature data corresponding to the periodic rhythm signal; the feature fusion layer concatenates the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively to obtain cardiac joint feature data; the decoder sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal.

[0007] In some embodiments of this application, the thoracic signal encoder includes three convolutional groups connected in sequence, and each convolutional group contains a contiguous one-dimensional convolutional layer and a residual block, so that the thoracic signal encoder can be used to extract local mutation feature data corresponding to the thoracic micromotion signal for distinguishing different cardiac mechanical activity events.

[0008] In some embodiments of this application, the rhythm signal encoder is provided with a gated loop unit so that the rhythm signal encoder is used to extract cross-cycle phase consistency feature data corresponding to the periodic rhythm signal to enhance the ability of the cardiac event recognition model to perceive the temporal structure of cardiac mechanical activity events.

[0009] In some embodiments of this application, the feature fusion layer is used to perform standard normal distribution modeling, sampling, and vector concatenation on the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal, respectively, to obtain corresponding feature concatenation data; and the feature concatenation data and the pre-acquired individual static feature data of the target subject are used as cardiac joint feature data to be input into the decoder; The individual static characteristic data includes data representing the gender, age, height, and weight of the target subject.

[0010] In some embodiments of this application, the decoder includes: a feature compression layer and a multi-label classifier connected in sequence; The feature compression layer is used to perform dimensional compression and semantic enhancement on the cardiac joint feature data to obtain the corresponding intermediate representation data; The multi-label classifier is used to classify the intermediate representation data into events in units of sliding time windows, so that each time window outputs in parallel multiple cardiac mechanical activity event type labels and time-series positioning data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event type labels include: mitral valve closure label, aortic valve opening label, rapid ejection label, aortic valve closure label, mitral valve opening label, rapid filling label, atrial contraction label, and isovolumetric contraction label.

[0011] In some embodiments of this application, the extraction of thoracic micromotion signals and periodic rhythm signals from millimeter-wave echo signals of the target subject acquired in real time using a non-contact method includes: The millimeter-wave radar antenna is pointed directly at the chest area of ​​the target subject to collect the millimeter-wave echo signal of the target subject in a non-contact manner in real time. The millimeter-wave echo signal is preprocessed to obtain the corresponding intermediate frequency signal, and the intermediate frequency signal is subjected to a fast Fourier transform to obtain the corresponding frequency domain signal. Range bin extraction and multi-antenna beamforming are performed on the frequency domain signal to obtain the initial thoracic micromotion signal corresponding to the millimeter wave echo signal; The initial thoracic micromotion signal is subjected to adaptive enhancement processing based on phase change characteristics and frequency components to extract periodic rhythm signals and thoracic micromotion signals representing transient displacement components corresponding to heart valve activity from the initial thoracic micromotion signal.

[0012] Another aspect of this application provides a non-contact real-time monitoring device for cardiac mechanical activity events, comprising: The real-time signal extraction module is used to extract thoracic micromotion signals and periodic rhythm signals from the millimeter-wave echo signals of the target subject acquired in real time in a non-contact manner. The cardiac event recognition module is used to input the thoracic micromotion signal and the periodic rhythm signal into the cardiac event recognition model, so that the cardiac event recognition model extracts the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively and splices them to obtain cardiac joint feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event recognition result data includes: cardiac mechanical activity event type label and temporal positioning data.

[0013] A third aspect of this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned non-contact real-time monitoring method for cardiac mechanical activity events.

[0014] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the aforementioned non-contact real-time monitoring method for cardiac mechanical activity events.

[0015] The fifth aspect of this application provides a computer program product comprising a computer program that, when executed by a processor, implements the aforementioned non-contact real-time monitoring method for cardiac mechanical activity events.

[0016] The non-contact real-time monitoring method for cardiac mechanical activity events provided in this application extracts thoracic micromotion signals and periodic rhythm signals from millimeter-wave echo signals acquired from a target subject in real time using a non-contact method. The thoracic micromotion signals and the periodic rhythm signals are then input into a cardiac event recognition model, which extracts the feature data corresponding to each signal and concatenates them to obtain combined cardiac feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification on the combined cardiac feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter-wave echo signals. The cardiac mechanical activity event identification results data include: cardiac mechanical activity event type labels and temporal positioning data; while ensuring the efficiency of cardiac mechanical activity event identification, it can effectively improve the identification accuracy of transient events and achieve event type identification by providing an event identification architecture that integrates rhythm information and micro-motion transient features. This can effectively improve the accuracy and robustness of non-contact real-time monitoring of cardiac mechanical activity events, and can be applied to various daily scenarios such as smartwatches, smartphones, and smart homes. It can provide users with a comfortable, safe, timely and accurate method for monitoring cardiac mechanical activity events, and provide a new solution for the precise monitoring of cardiovascular diseases.

[0017] Additional advantages, objectives, and features of this application will be set forth in part in the description which follows, and will in part become apparent to those skilled in the art upon review of the following description, or may be learned by practice of the application. The objectives and other advantages of this application can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0018] Those skilled in the art will understand that the purposes and advantages that can be achieved with this application are not limited to those specifically described above, and that the above and other purposes that this application can achieve will be more clearly understood from the following detailed description. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, do not constitute a limitation thereof. The components in the drawings are not drawn to scale but are merely for illustrating the principles of this application. For ease of illustration and description of certain parts of this application, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to this application. In the drawings: Figure 1 This is a schematic diagram of the first step of a non-contact real-time monitoring method for cardiac mechanical activity events according to an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of the cardiac event recognition model in a non-contact real-time monitoring method for cardiac mechanical activity events according to an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of a second process for a non-contact real-time monitoring method for cardiac mechanical activity events according to an embodiment of this application.

[0022] Figure 4 This is a schematic diagram of the execution flow of step 100 in a non-contact real-time monitoring method for cardiac mechanical activity events according to an embodiment of this application.

[0023] Figure 5 This is a flowchart illustrating the non-contact real-time monitoring process of cardiac mechanical activity events in an application example of this application.

[0024] Figure 6 This is a schematic diagram of the structure of a non-contact real-time monitoring device for cardiac mechanical activity events according to an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.

[0026] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the structures and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0027] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0028] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0029] In the following description, embodiments of the present application will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0030] Traditional monitoring techniques primarily rely on invasive cardiac catheterization or echocardiography. Cardiac catheterization can accurately capture mechanical events by directly measuring changes in intracardiac pressure; however, its invasive nature poses risks such as vascular injury and thrombosis, and it is difficult to achieve long-term continuous monitoring. While echocardiography is the gold standard for non-invasive detection, it requires specialized physicians to operate the probe for positioning, the equipment is bulky, and it is sensitive to body position, making it impossible to achieve long-term, high-frequency continuous dynamic acquisition. Therefore, it is insufficient to meet the needs for real-time monitoring of cardiac events and the capture of abnormal trends. Existing non-contact technologies, such as ballistocardiography (BCG) and optical camera monitoring using smart wearable devices, can indirectly reflect cardiac activity through surface vibration or image analysis. However, limited by sensor sensitivity and environmental interference, the temporal resolution generally cannot meet the accuracy requirements for the time-series analysis of cardiac mechanical activity. This problem is particularly prominent in certain patient groups. For example, burn patients cannot wear contact sensors due to skin damage, newborns have fragile chest walls and are not suitable for applying detection pressure, and atrial fibrillation patients often have mitral regurgitation that requires continuous monitoring of the duration of regurgitation. These clinical needs expose the adaptive limitations of existing technologies.

[0031] In the field of cardiac mechanical event monitoring, existing technologies mainly rely on two types of methods: bioelectrical signal detection and imaging examinations. However, these methods have significant limitations in practical applications.

[0032] Electrocardiography (ECG) captures the heart's electrical activity through electrodes attached to the skin. While it can identify arrhythmias, it cannot directly reflect the precise timing of mechanical movements. For example, when a patient has electromechanical asynchrony (such as left bundle branch block), the QRS width on the ECG may differ from the actual aortic valve opening time by more than 40 milliseconds, making it impossible to accurately determine the onset of ejection. Furthermore, prolonged contact between the electrodes and the skin can cause allergies or pressure marks. Especially during continuous nighttime monitoring, electrode detachment or displacement can lead to data interruption, affecting the reliability of the monitoring.

[0033] Echocardiography, which directly observes the movement of heart valves through sound wave imaging, is considered the "gold standard" for clinical diagnosis. However, the equipment is bulky and complex to operate, requiring a specialist physician to manually adjust the probe angle to capture specific sections. For example, when monitoring mitral valve closure events, the ultrasound probe must be precisely aligned with the apical four-chamber view; even slight changes in body position can lead to blurred images or signal loss. For home patients requiring long-term monitoring, this technology cannot meet the needs for convenience and continuity.

[0034] Another non-contact sensing technology study attempts to capture wrist pulse waves using optical sensors and indirectly infer cardiac rhythm through pulse wave transmission time (PTT). However, this method is essentially based on a long-range reflection of peripheral vascular waveforms. The signal becomes distorted after long-distance propagation and vascular modulation, making it impossible to reconstruct transient mechanical details occurring in the central heart, such as valve opening and closing and isovolumetric contraction. Furthermore, wrist arterial pulsation is significantly affected by factors such as arteriosclerosis, peripheral perfusion, and limb posture, making it difficult to provide high-temporal-resolution cardiac event sequences for clinical interpretation. The core problem with these technologies is that traditional methods are either limited by the comfort and reliability of contact sensors or lose crucial timing information due to insufficient signal resolution.

[0035] How to overcome the technical bottleneck of millisecond-level cardiac event analysis without sacrificing the comfort of non-invasive monitoring, and achieve continuous, real-time, and non-invasive sensing of cardiac status, has become a core issue urgently needing to be addressed in the field of non-invasive real-time cardiovascular monitoring. Frequency Modulated Continuous Wave (FMCW) millimeter-wave radar, with its sub-millimeter wavelength and gigahertz sampling rate, can capture transient displacements on the surface of the chest cavity, with a theoretical time resolution of 0.1 ms. Its non-contact nature eliminates skin pressure artifacts caused by sensor contact and can penetrate media such as clothing and bedding, making it suitable for continuous monitoring of burn patients, newborns, and long-term bedridden individuals. More importantly, this technology has the ability to operate continuously for extended periods, supporting real-time sensing of cardiac micro-movements in various scenarios such as home, mobile, and ICU, providing new possibilities for remote clinical monitoring, personalized intervention, and early disease warning.

[0036] Current millimeter-wave systems still face significant challenges in cardiac event analysis. Cardiac micro-motion signals are easily obscured by respiratory and body motion noise. While traditional filtering algorithms can suppress baseline drift, they can cause waveform distortion in transient events, resulting in blurred event boundaries. Furthermore, threshold-based peak localization methods have high resolution errors for overlapping events and cannot distinguish between mechanical event types. These limitations restrict the practical application of millimeter-wave technology in clinical diagnosis.

[0037] Based on this, in order to solve the problems of poor accuracy of real-time monitoring results and inability to distinguish the types of cardiac mechanical events in existing non-contact cardiac activity monitoring methods, embodiments of this application provide a non-contact real-time monitoring method for cardiac mechanical activity events, a non-contact real-time monitoring device for cardiac mechanical activity events for performing the non-contact real-time monitoring method for cardiac mechanical activity events, a physical device, a computer-readable storage medium, and a computer program product, which can effectively improve the accuracy of transient event identification and achieve event type identification while ensuring the efficiency of cardiac mechanical activity event identification.

[0038] The following examples will provide a detailed description.

[0039] Based on this, embodiments of this application provide a non-contact real-time monitoring method for cardiac mechanical activity events, which can be implemented by a non-contact real-time monitoring device for cardiac mechanical activity events. See [link to relevant documentation]. Figure 1 The non-contact real-time monitoring method for cardiac mechanical activity events specifically includes the following: Step 100: Extract thoracic micromotion signals and periodic rhythm signals from the millimeter-wave echo signals of the target subject acquired in real time using a non-contact method.

[0040] It is important to note that dynamic monitoring of cardiac mechanical events is a core method for assessing cardiac function and diagnosing diseases such as valvular heart disease and myocardial ischemia. Key events in the cardiac cycle, such as mitral valve closure, aortic valve opening, rapid ejection, and isovolumetric contraction, reflect the mechanical characteristics of cardiac contraction and relaxation on a millisecond timescale. Clinical studies have confirmed that patients with acute coronary syndrome exhibit detectable mechanical disturbances several hours before the onset of symptoms. These abnormalities often precede ST segment changes on electrocardiograms and the onset of chest pain, providing a crucial real-time early warning window for cardiovascular emergencies. Especially in high-risk situations such as sudden cardiac death and acute decompensated heart failure, abnormal cardiac mechanical activity often evolves continuously for several minutes to hours before the onset. Continuous monitoring at the millisecond level may allow for early risk identification and intervention. These cardiac mechanical events can be simply referred to as cardiac events.

[0041] Before proceeding with step 100, millimeter-wave echo signals of the target subject can be acquired in real time using a non-contact method. This involves pointing the millimeter-wave radar antenna directly at the subject's chest region to collect millimeter-wave echo signals related to cardiac activity in real time; these millimeter-wave echo signals can be simply referred to as echo signals. Then, in step 100, chest cavity micromotion signals and periodic rhythm signals are extracted from the millimeter-wave echo signals.

[0042] Understandably, chest cavity micromotion signals refer to micromotion signals whose phase changes over time extracted from the spatial region corresponding to the pre-defined central chest cavity region of the target subject, in order to characterize the subtle body surface vibrations caused by cardiac mechanical activity.

[0043] The periodic rhythm signal refers to stable rhythmic features extracted from periodic thoracic micromotion signals, used for rhythm guidance and event alignment in subsequent cardiac event recognition models. Specifically, the periodic rhythm signal originates from stable patterns in continuous multi-cycle signals, used to guide coarse localization of transient events. The periodic rhythm signal can be simply referred to as the rhythm signal.

[0044] Step 200: Input the thoracic micromotion signal and the periodic rhythm signal into the cardiac event recognition model, so that the cardiac event recognition model extracts the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively and splices them to obtain cardiac joint feature data. Then, the cardiac event recognition model sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event recognition result data includes: cardiac mechanical activity event type label and temporal positioning data.

[0045] In step 200, the feature data corresponding to the thoracic micromotion signal can specifically refer to the local mutation feature data corresponding to the thoracic micromotion signal used to distinguish different cardiac mechanical activity events; the feature data corresponding to each periodic rhythm signal can specifically refer to the cross-cycle phase consistency feature data corresponding to the periodic rhythm signal used to enhance the cardiac event recognition model's ability to perceive the temporal structure of cardiac mechanical activity events.

[0046] The local mutation features include those used to display abrupt change points (jump points) in micro-motion signals. Specifically, this includes: extracting local peak points with significant instantaneous amplitude changes or phase abrupt changes in the time domain based on the short-time Fourier transform (STFT) or wavelet analysis results of millimeter-wave phase signals; and modeling parameters such as the energy amplitude, phase difference, and rate of change of the first and second derivatives of these local mutation points to form a time-frequency joint feature vector for characterizing the transient response of cardiac mechanical events (such as valve opening and closing, isovolumetric contraction initiation, etc.). The cross-cycle phase consistency feature is used to reflect multiple consecutive cardiac cycles. The stability and repeatability of the rhythm signal phase are specifically assessed by performing phase expansion and normalization on signals from multiple adjacent cycles, using the cardiac cycle as the basic unit; calculating the cross-cycle phase difference and coherence coefficient of the signal within the same phase interval to extract the consistency index of the periodic rhythm across different cycles; and splicing and fusing the transient amplitude variation features of the micro-motion signal and the cross-cycle phase consistency features of the rhythm signal in the feature dimension, and using convolutional neural networks or a two-branch Transformer structure for feature encoding to achieve multi-scale fusion of micro-motion and rhythm information.

[0047] It is understood that multi-label classification processing refers to labeling the thoracic micromotion signals with multiple cardiac mechanical activity event type labels. The temporal localization data refers to a series of data points arranged chronologically corresponding to the thoracic micromotion signals. Each data point contains a timestamp and a corresponding cardiac mechanical activity event type label, used to indicate the location and duration of each cardiac mechanical event on the time axis. This includes: establishing a time series index based on the sampling time, mapping the event probability distribution output by the model to the time series; labeling the corresponding cardiac mechanical activity event type (such as mitral valve closure, aortic valve opening, rapid ejection phase initiation, isovolumetric contraction phase end, etc.) for each moment or window; forming the localization trajectory data of cardiac events in the time domain through the mapping relationship between timestamps and event labels, used to achieve accurate identification and dynamic tracking of cardiac mechanical events; when the model identifies multiple events occurring in parallel, the temporal localization data can simultaneously record the label set of different events within the same time period, used to support multi-event concurrent judgment and complex rhythm analysis.

[0048] As described above, the non-contact real-time monitoring method for cardiac mechanical activity events provided in this application can effectively improve the accuracy of transient event recognition and achieve event type recognition by providing an event recognition architecture that integrates rhythm information and micro-motion transient features, while ensuring the efficiency of cardiac mechanical activity event recognition. This can effectively improve the accuracy and robustness of non-contact real-time monitoring of cardiac mechanical activity events, and is applicable to various daily scenarios such as smartwatches, smartphones, and smart homes. It can provide users with a comfortable, safe, timely, and accurate method for monitoring cardiac mechanical activity events, and provide a new solution for the precise monitoring of cardiovascular diseases.

[0049] To further improve the effectiveness and accuracy of non-contact real-time monitoring of cardiac mechanical activity events, a non-contact real-time monitoring method for cardiac mechanical activity events is provided in this application embodiment, see [link to relevant documentation]. Figure 2 The cardiac event recognition model includes: a chest signal encoder 1, a rhythm signal encoder 2, a feature fusion layer 3, and a decoder 4.

[0050] The input terminal of the feature fusion layer 3 is connected to the output terminals of the chest cavity signal encoder 1 and the rhythm signal encoder 2, respectively; the output terminal of the feature fusion layer 3 is connected to the input terminal of the decoder 4.

[0051] It should be noted that currently, no research has proposed applying a dual-encoder-decoder model architecture to the non-contact real-time monitoring of cardiac mechanical activity events. This is because the thoracic micromotion signals and periodic rhythm signals contained in millimeter-wave signals differ significantly in time scale and signal-to-noise ratio in this scenario: thoracic micromotion signals are transient, non-stationary, and have small amplitudes, while periodic rhythm signals exhibit stable low-frequency rhythm patterns. Directly migrating the conventional dual-encoder-decoder architecture to this scenario can easily lead to the following problems: On the one hand, when dual encoders extract features of two types of signals in parallel, the large difference in the dynamic range of the two types of signals can easily lead to gradient imbalance and feature fusion distortion, which in turn causes feature drift and convergence difficulties in the training phase. On the other hand, when the decoder performs joint decoding of spliced ​​features, if it does not model the heterogeneity of micro-motion mutation features and rhythm phase features, the boundaries of the output cardiac mechanical activity events will be blurred, resulting in temporal shifts or label mismatches in the event recognition results.

[0052] To address the aforementioned technical challenges, the designers of this application innovatively propose an architecture design that integrates local mutation features and cross-cycle phase consistency features using a dual-encoder approach. A local mutation feature extraction module is introduced into the first encoder. By performing short-time time-frequency analysis and gradient-weighted modeling on the thoracic micro-motion signals, enhanced extraction of transient jump points is achieved, effectively improving the model's sensitivity to minute cardiac events (such as valve opening and closing). A cross-cycle phase consistency constraint mechanism is constructed in the second encoder to encode the cross-cycle coherence of the rhythm signal, enabling the model to automatically align the temporal structure between different cardiac cycles, thereby enhancing its stable discrimination capability for rhythmic events. In the decoding stage, an adaptive feature fusion layer is designed. An attention-weighted strategy dynamically allocates weights to the output features of different encoders, balancing the contribution ratio of transient micro-motion features and global rhythm features, avoiding information competition and fusion distortion.

[0053] Through the aforementioned technical setup, this application effectively overcomes the technical obstacles encountered by traditional models in heterogeneous signal feature fusion, such as gradient imbalance, temporal mismatch, and feature drift, achieving high-precision fusion and joint decoding of thoracic micro-motion signals and periodic rhythm signals. Experimental results show that this model significantly improves upon existing single-branch or traditional fusion models in terms of accuracy in recognizing cardiac mechanical activity events, event boundary localization error, and robustness, thus achieving the technical effect of millisecond-level cardiac event recognition under non-contact millimeter-wave signals.

[0054] Correspondingly, see Figure 3 Step 200 of the non-contact real-time monitoring method for cardiac mechanical activity events in this embodiment of the application specifically includes the following: Step 210: The thoracic cavity micromotion signal is input into the thoracic cavity signal encoder in the cardiac event recognition model, and the periodic rhythm signal is input into the rhythm signal encoder in the cardiac event recognition model, so that the thoracic cavity signal encoder extracts the feature data corresponding to the thoracic cavity micromotion signal, and the rhythm signal encoder extracts the feature data corresponding to the periodic rhythm signal; the feature fusion layer concatenates the feature data corresponding to the thoracic cavity micromotion signal and the periodic rhythm signal respectively to obtain cardiac joint feature data; the decoder sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal.

[0055] To further improve the effectiveness and accuracy of extracting local mutation feature data corresponding to the thoracic micromotion signal for distinguishing different cardiac mechanical activity events, in a non-contact real-time monitoring method for cardiac mechanical activity events provided in this application embodiment, the thoracic signal encoder 1 includes three convolutional groups connected in sequence, and each convolutional group contains a connected one-dimensional convolutional layer and a residual block, so that the thoracic signal encoder 1 can be used to extract local mutation feature data corresponding to the thoracic micromotion signal for distinguishing different cardiac mechanical activity events.

[0056] To further improve the effectiveness and accuracy of extracting cross-cycle phase consistency feature data corresponding to the periodic rhythm signal to enhance the cardiac event recognition model's ability to perceive the temporal structure of cardiac mechanical activity events, in a non-contact real-time monitoring method for cardiac mechanical activity events provided in this application embodiment, the rhythm signal encoder 2 is provided with a gated loop unit, so that the rhythm signal encoder 2 is used to extract cross-cycle phase consistency feature data corresponding to the periodic rhythm signal to enhance the cardiac event recognition model's ability to perceive the temporal structure of cardiac mechanical activity events.

[0057] To further improve the effectiveness and accuracy of extracting cardiac joint feature data, in a non-contact real-time monitoring method for cardiac mechanical activity events provided in this application embodiment, the feature fusion layer 3 is used to perform standard normal distribution modeling, sampling, and vector splicing on the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal, respectively, to obtain the corresponding feature splicing data; and the feature splicing data and the pre-acquired individual static feature data of the target subject are used as cardiac joint feature data to be input into the decoder 4; The individual static characteristic data includes data representing the gender, age, height, and weight of the target subject.

[0058] To further improve the effectiveness and accuracy of acquiring multiple cardiac mechanical activity event type labels and time-series positioning data corresponding to the millimeter-wave echo signals, a non-contact real-time monitoring method for cardiac mechanical activity events is provided in this application embodiment. (See also...) Figure 2 The decoder 4 includes: a feature compression layer and a multi-label classifier connected in sequence; The feature compression layer is used to perform dimensional compression and semantic enhancement on the cardiac joint feature data to obtain the corresponding intermediate representation data; The multi-label classifier is used to classify the intermediate representation data into events in units of sliding time windows, so that each time window outputs in parallel multiple cardiac mechanical activity event type labels and time-series positioning data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event type labels include: mitral valve closure label, aortic valve opening label, rapid ejection label, aortic valve closure label, mitral valve opening label, rapid filling label, atrial contraction label, and isovolumetric contraction label.

[0059] To further effectively distinguish between cardiac activity and respiratory movements and improve the ability to distinguish millisecond-level events, a non-contact real-time monitoring method for cardiac mechanical activity events is provided in this application embodiment, see [link to relevant documentation]. Figure 3 and Figure 4 Step 100 of the non-contact real-time monitoring method for cardiac mechanical activity events specifically includes the following: Step 110: Point the millimeter-wave radar antenna directly at the chest area of ​​the target subject to acquire the millimeter-wave echo signal of the target subject in real time in a non-contact manner.

[0060] Step 120: Preprocess the millimeter-wave echo signal to obtain the corresponding intermediate frequency signal, and perform a fast Fourier transform on the intermediate frequency signal to obtain the corresponding frequency domain signal.

[0061] Step 130: Perform range bin extraction and multi-antenna beamforming on the frequency domain signal to obtain the initial thoracic micromotion signal corresponding to the millimeter wave echo signal.

[0062] In steps 110 to 130, millimeter-wave radar can be used to achieve millimeter-wave sensing in a non-contact manner. By using millimeter-wave radar to detect the chest vibration of the target caused by the heartbeat, the raw signal containing information such as the heart activity of the target can be obtained, namely the intermediate frequency signal. Then, the intermediate frequency signal can be converted into frequency domain information by fast Fourier transform, and according to the frequency domain information and the configuration information of the millimeter-wave radar, the frequency of the frequency domain information is divided into different range compartments. According to the distance between the target and the radar antenna, the range compartment where the chest region of the target is located is extracted, and the reflected signal component corresponding to the range compartment where the chest region is located is extracted. Based on the reflected signal component, beamforming is used to map each intermediate frequency signal into three-dimensional space to extract the radar signal component corresponding to each intermediate frequency signal. The location coordinates of the heartbeat of the target are determined according to the radar signal component corresponding to each intermediate frequency signal. The phase information corresponding to the location coordinates of the heartbeat of the target is extracted, and the corresponding raw millimeter-wave chest micro-motion signal, namely the initial chest micro-motion signal, is generated based on the phase information.

[0063] Step 140: Perform adaptive enhancement processing based on phase change features and frequency components on the initial thoracic micromotion signal to extract periodic rhythm signals and thoracic micromotion signals that represent transient displacement components corresponding to heart valve activity from the initial thoracic micromotion signal.

[0064] In step 140, the adaptive enhancement processing based on phase change features and frequency components specifically includes: Phase jump detection and transient enhancement processing: The initial thoracic cavity micromotion signal is subjected to continuous phase unwrapping and differential operation to calculate the phase change rate between adjacent sampling points, and phase jump points are detected by setting an adaptive threshold; the local transient components are extracted from the detected phase jump points by high-pass filtering or short-time Fourier transform (STFT), and the signal amplitude in the phase jump region is enhanced by using a weighted window function to form a transient signal component sensitive to rapid mechanical events such as heart valve opening and closing and isovolumetric contraction.

[0065] Frequency component analysis and adaptive filtering enhancement: The power spectral density of the initial thoracic micromotion signal is estimated, and the cardiac rhythm main frequency and its harmonic components corresponding to the signal in the low frequency band (about 0.8Hz to 2Hz) are extracted. By using bandpass filtering and adaptive noise cancellation algorithm (ANC), respiratory, body movement and environmental interference components are dynamically suppressed, and the filter bandwidth is automatically adjusted according to the transient energy distribution to achieve optimal separation between rhythm components and transient components in the frequency domain.

[0066] Phase-frequency joint weighting and signal reconstruction: The transient enhanced signal obtained by phase change detection and the rhythm signal obtained by frequency adaptive filtering are normalized respectively, and then weighted and fused according to energy contribution and coherence coefficient; the enhanced thoracic micromotion signal is generated by time-frequency domain joint reconstruction algorithm, where the low frequency part corresponds to the cardiac periodic rhythm signal and the high frequency part corresponds to the transient displacement signal of the heart valve activity, realizing the separation and complementary enhancement of rhythm and transient signals.

[0067] To further illustrate the above solution, this application also provides a specific application example of a non-contact real-time monitoring method for cardiac mechanical activity events. Specifically, it is a method that uses millimeter waves to non-contactly detect micro-motion events in the human heart, thereby accurately predicting the real-time cardiac function monitoring needs of users in scenarios such as nighttime, resting, and chronic disease management. The first challenge is how to extract weak and high-speed cardiac mechanical motion signals from millimeter-wave radar echoes. To this end, this application introduces a three-dimensional spatial selection and adaptive micro-motion enhancement mechanism in the signal processing chain, effectively distinguishing cardiac activity from respiratory motion and improving the ability to resolve millisecond-level events. The core challenge then lies in how to accurately identify key cardiac events in a short time. This application example introduces an innovative technology, designing an event recognition algorithm for real-time applications. Combined with a rhythm-guided event localization module and a lightweight time window alignment mechanism, it maintains computational efficiency while ensuring the accuracy of transient event recognition, achieving accurate and robust real-time cardiac function monitoring.

[0068] See Figure 5 The specific application examples of the non-contact real-time monitoring method for cardiac mechanical activity events include the following: S1. Point the millimeter-wave radar antenna directly at the subject's chest area and collect millimeter-wave echo signals related to cardiac activity; S2. Preprocess and spatially model the millimeter-wave echo signal to extract the initial millimeter-wave micromotion signal with high temporal resolution from the central region of the chest cavity; that is, beamforming the echo signal and constructing a three-dimensional spatiotemporal model of chest micromotion to extract micromotion information related to cardiac mechanical activity.

[0069] S3. Separate the transient displacement component corresponding to the heart valve activity from the extracted signal.

[0070] S4. Input the micro-motion signal into the temporal feature encoding network in the cardiac event recognition model to extract dynamic features related to cardiac events; that is: input the extracted micro-motion signal into the temporal enhancement feature encoding network to extract key time features related to cardiac mechanical events.

[0071] S5. Utilize the decoder in the cardiac event recognition model for classification and temporal localization, outputting real-time monitoring results containing key cardiac event nodes. That is: use the event recognition and classification model to make real-time judgments on cardiac mechanical activity, generating structured output containing low-latency cardiac event information.

[0072] The extraction process of millimeter-wave micromotion signals and periodic rhythm signals from the thoracic cavity specifically includes the following: (1) Process the radar echo to obtain the intermediate frequency signal, and perform a fast Fourier transform to convert it into the frequency domain.

[0073] (2) Based on the signal’s performance in the frequency domain and the characteristics of FMCW radar, the chest cavity region is locked by combining range selection and multi-antenna beamforming methods.

[0074] (3) Extract the micro-motion signal whose phase changes with time from the spatial region corresponding to the thoracic cavity to characterize the subtle body surface vibration caused by cardiac mechanical activity.

[0075] (4) Adaptive enhancement processing is performed on the extracted thoracic micromotion signals to improve the expression intensity and signal-to-noise ratio of cardiac-related components.

[0076] (5) Simultaneously, stable rhythm features are extracted from multi-period data for use in subsequent rhythm guidance and event alignment modules. The cardiac event recognition process specifically includes the following: (1) The thoracic cavity micromotion signal and the periodic rhythm signal are respectively input into the feature encoder to obtain the cardiac joint features; (2) Combine personal information with joint features; (3) Predict user heart-fluttering events through decoder and multi-event classifier.

[0077] In this application example, multi-source micro-motion signals and a rhythm guidance mechanism are combined. Through a specially designed time-series encoder architecture and a lightweight decoder network, high-precision identification of multiple key events in a complete cardiac cycle is achieved. First, this application example defines a dual-channel input path: thoracic micro-motion signals and periodic rhythm information. The thoracic micro-motion signals capture high-speed phase changes from millimeter-wave radar to characterize the minute vibrations on the body surface caused by cardiac mechanical activity; while the rhythm signals originate from stable patterns in continuous multi-cycle signals to guide coarse localization of transient events.

[0078] The two input channels are processed by a time-sequential encoder. The encoder adopts a design that combines time-sequential convolution and residual structure. The chest cavity signal encoder consists of three sets of one-dimensional convolutional layers and residual connections. Each set of convolutional layers includes: a one-dimensional convolutional layer with a kernel size of 3 and a stride of 1, a batch normalization layer, and a ReLU activation function.

[0079] The first set of convolutional layers is used to extract the local amplitude variation features of the input thoracic cavity micro-motion signal to enhance the distinguishability of the micro-vibration signal; The second set of convolutional layers further captures energy change regions in the time dimension based on the output of the first set, and preserves the original temporal information through residual connections, so that the phase structure of the signal is not destroyed; the third set of convolutional layers uses dilation conv to expand the receptive field, thereby capturing complex waveform patterns across time windows.

[0080] Residual connections are introduced between each set of convolutional layers to prevent gradient vanishing during deep network training and to enhance the representational stability of micro-movement mutation points.

[0081] After processing by three sets of convolutional and residual units, the output is a 128-dimensional time-series feature tensor, where each time step corresponds to a local mutation feature vector of the thoracic cavity micromotion signal, providing fine-grained representation for subsequent event identification. This vector can extract local mutation features in the micromotion signal to distinguish different events.

[0082] The rhythm signal encoder introduces a gated recurrent unit (GRU) to extract phase consistency features across cycles, enhancing the model's ability to perceive the temporal structure of events. This encoder consists of two stacked bidirectional GRU layers and a temporal attention layer. The first-layer bidirectional GRU receives the time-series input of the rhythm signal (input dimension is 1, time step length is T) and outputs a dimension of 64, which is used to model the short-term time dependence of the rhythm signal. The second-layer bidirectional GRU takes the hidden state sequence of the first layer as input and further learns the long-term time-dependent features across cycles in order to capture the phase consistency changes of cardiac rhythm in different cycles. Following two GRU layers, a temporal attention layer is introduced to adaptively allocate weights based on the phase stability of each time step, making the model focus more on time periods with high phase consistency across cycles. The final output is a 128-dimensional rhythm-encoded feature vector. The output features of the two encoders are normalized and modeled using a standard normal distribution in the latent space, then reparameterized sampling is used to form "multi-event sensing features." The sampled latent vector contains local abrupt changes in the thoracic micromotion signal and global temporal structure information in the rhythm signal.

[0083] In the feature fusion stage, this application example concatenates the latent vectors of the two channels and uses them together with individual static features (such as gender, age, height, and weight) as input to the decoder to form "cardiac joint features". This fused feature retains both the rapidly changing high-speed micro-motion information and the stable rhythmic structure across cycles, and makes full use of individual physiological differences to achieve more robust event discrimination.

[0084] The decoder consists of two fully connected layers and a set of multi-label classifiers: —The first fully connected layer has an input dimension of 256 and an output dimension of 128. It embeds a ReLU activation function to achieve non-linear mapping of the feature space and introduces a Dropout mechanism (dropout rate of 0.3) to prevent overfitting; —The second fully connected layer has an input dimension of 128 and an output dimension of 64. It also embeds a ReLU activation function to further compress the feature dimension and enhance the semantic representation of events; —After the two fully connected layers, a set of independent multi-label classifiers is introduced. These classify events using a sigmoid activation function to achieve parallel multi-label output. Each output node corresponds to a cardiac mechanical event type label (e.g., MC, AO, AC, MO, RF, AS, IVC, etc.). It classifies each signal segment using a sliding time window. Each time window can output multiple event labels and their corresponding confidence scores, thus achieving continuous identification of multiple key physiological events throughout the cardiac cycle. The output event labels and timestamps together construct a complete cardiac event timeline for subsequent estimation of cardiac function parameters and analysis of clinical indicators.

[0085] During training, this application uses multi-source synchronized phonocardiograms / electrocardiograms as annotation criteria, and the occurrence of each key event is used as positive and negative samples in the multi-label output. Time windows containing corresponding events are labeled as positive samples, and time windows without events are labeled as negative samples, thus constructing a multi-label supervision signal. The multi-label cross-entropy loss function is used as the optimization objective during training to improve the model's generalization ability in concurrently recognizing multiple events.

[0086] In other words, this application provides a method for low-latency monitoring of critical cardiac events by using millimeter-wave non-contact sensing of cardiac micro-motion signals. This method can capture important events such as the opening and closing of heart valves using chest cavity micro-motion signals combined with periodic rhythm information without contacting the skin or disturbing the user. This helps identify potential cardiac dysfunctions and is particularly suitable for continuous observation during daily sleep and rest. Compared to traditional patches or ultrasound devices, this method is lighter, more comfortable, and has lower latency, making it suitable for embedded deployment in smart devices. It has good effects on chronic cardiovascular disease risk management and early warning of emergencies.

[0087] In addition, to verify the effectiveness of the multi-event recognition method based on millimeter-wave non-contact signals described in this application in terms of real-time performance and recognition accuracy, systematic experimental verification and performance evaluation were conducted. Two types of data sources were used in the experiments: a pre-training dataset and a fine-tuning / validation dataset. The pre-training dataset selected a publicly available open-source millimeter-wave-physiological joint signal dataset, covering data from approximately 500 subjects, for pre-training the feature extraction network to enhance the model's generalization ability under different individual body types, respiratory rhythms, and posture differences. During the fine-tuning and validation phases, a self-built synchronous database was used, containing millimeter-wave signals and electrocardiogram (ECG) signals from 90 subjects, with each subject continuously acquiring signals for 5 to 20 minutes. The annotation process was conducted by professional technicians referring to the synchronous ECG, annotating key events such as valve opening (MO), closure (MC), aortic valve opening (AO), aortic valve closure (AC), and ejection start and stop points (FE start and stop) frame by frame. The average annotation time per person was approximately ten minutes, and the annotation work was completed collaboratively by two to three professionals and reviewed to ensure consistency and accuracy. This annotation process is a necessary manual labor component in the experimental verification of this invention.

[0088] In model performance evaluation, precision, recall, and F1 score are used as the main recognition metrics to measure the accuracy and completeness of event recognition. Meanwhile, to assess the system's real-time response capability, two latency metrics are defined: average recognition latency and end-to-end system latency. Average recognition latency represents the time difference between the occurrence of a real event and the system reporting the result, while end-to-end system latency represents the complete time from millimeter-wave signal acquisition, enhancement, encoding, decoding, post-processing, to the output result.

[0089] Under the same experimental conditions, comparing the single-channel convolutional baseline model (CNN baseline) and the dual-encoder fusion model (combining convolution and gated recurrent units (GRUs), the results show that the single-channel baseline model has an accuracy of 0.70, a recall of 0.74, and an F1 score of 0.72. After adopting the dual-encoder structure (convolution and GRU), the accuracy improves to 0.82, the recall improves to 0.88, and the F1 score reaches 0.85. Compared to the baseline model, the dual-encoder structure improves the F1 score by approximately 13 percentage points (absolute value) and approximately 18.1% relatively. Therefore, the dual encoder and enhancement structure described in this application can significantly improve the accuracy and robustness of the system in recognizing heart valve motion and rhythmic events.

[0090] To verify the system's real-time performance, latency tests were conducted on various hardware platforms. The test platforms included three types: a Raspberry Pi 4B edge platform equipped with a BCM2711 Cortex-A72 1.5GHz processor and 4GB of memory; an embedded single-board computer equipped with an ARM Cortex-A72-like 2.0GHz processor; and a laptop computer equipped with an Intel i7-12700H processor. Test results showed that on the Intel i7 platform, the average latency for a single forward inference iteration was approximately 42 milliseconds, and the end-to-end system latency was approximately 60 milliseconds; on the ARM A72 platform, the forward latency was approximately 60 milliseconds, and the end-to-end latency was approximately 110 milliseconds; and on the Raspberry Pi 4B platform, the forward latency was approximately 85 milliseconds, and the end-to-end system latency was approximately 140 milliseconds.

[0091] The results above demonstrate that even on the resource-constrained Raspberry Pi 4B platform, the average end-to-end latency remains below 200 milliseconds, fully meeting the requirements for real-time recognition. This result indicates that the method described in this application not only enables real-time processing on embedded edge devices but also maintains high recognition accuracy on low-power platforms, demonstrating excellent application scalability and deployment value.

[0092] In summary, this application presents a non-contact, low-latency technique for monitoring cardiac events using millimeter-wave radar. This technique is applicable to various everyday scenarios such as smartwatches, smartphones, and smart homes, providing users with a comfortable, safe, timely, and accurate method for monitoring cardiac events, offering a novel solution for precise monitoring of cardiovascular diseases. Specifically, it provides a non-contact and highly timely method for identifying cardiac micro-motion events based on millimeter-wave radar. This method is characterized by designing a three-dimensional spatial modeling and adaptive enhancement mechanism in the millimeter-wave radar echo signal processing, combining chamber selection and beamforming techniques to achieve spatial focusing on the thoracic region and extract cardiac micro-motion signals with high-speed phase changes. In the signal enhancement stage, an adaptive module based on phase abrupt change features and frequency components is introduced to effectively suppress respiratory interference and highlight cardiac details. Furthermore, it provides an event recognition architecture that integrates rhythm information and transient micro-motion features. This architecture features dual-channel input, including micro-motion signals and stable patterns from continuous multi-cycle signals. These inputs are processed by an encoder composed of multi-layer convolutional neural networks and residual structures. The high-dimensional features extracted during the encoding stage are modeled together with the rhythm guidance signal through a feature fusion unit and then input into an event recognition module built on a lightweight attention mechanism network to achieve real-time multi-classification output of typical cardiac mechanical events during the cardiac cycle.

[0093] From a software perspective, this application also provides a non-contact cardiac mechanical activity event real-time monitoring device for performing all or part of the aforementioned non-contact cardiac mechanical activity event real-time monitoring method, see [link to relevant documentation]. Figure 6 The non-contact real-time monitoring device for cardiac mechanical activity events specifically includes the following components: The real-time signal extraction module 10 is used to extract thoracic micromotion signals and periodic rhythm signals from the millimeter-wave echo signals of the target subject acquired in real time in a non-contact manner.

[0094] The cardiac event recognition module 20 is used to input the thoracic micromotion signal and the periodic rhythm signal into the cardiac event recognition model, so that the cardiac event recognition model extracts the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively and splices them to obtain cardiac joint feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event recognition result data includes: cardiac mechanical activity event type label and temporal positioning data.

[0095] The embodiments of the non-contact real-time monitoring device for cardiac mechanical activity events provided in this application can be used to execute the processing flow of the embodiments of the non-contact real-time monitoring method for cardiac mechanical activity events described above. Its functions will not be repeated here, but can be referred to the detailed description of the embodiments of the non-contact real-time monitoring method for cardiac mechanical activity events described above.

[0096] The non-contact real-time monitoring of cardiac mechanical activity events in the aforementioned device can be performed on a server or client device. The device communicates with a millimeter-wave radar. The specific choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are performed on the client device, the client device may further include a processor for specific processing of the non-contact real-time monitoring of cardiac mechanical activity events.

[0097] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0098] The server and the client device can communicate using any suitable network protocol, including those not yet developed as of the date of this application. Such network protocols may include, for example, TCP / IP, UDP / IP, HTTP, HTTPS, etc. Furthermore, such network protocols may also include RPC (Remote Procedure Call Protocol) and REST (Representational State Transfer Protocol) protocols used on top of the aforementioned protocols.

[0099] As described above, the non-contact real-time monitoring device for cardiac mechanical activity events provided in this application embodiment can effectively improve the accuracy of transient event recognition and achieve event type recognition by providing an event recognition architecture that integrates rhythm information and micro-motion transient features, while ensuring the efficiency of cardiac mechanical activity event recognition. This can effectively improve the accuracy and robustness of non-contact real-time monitoring of cardiac mechanical activity events, and can be applied to various daily scenarios such as smartwatches, smartphones, and smart homes. It can provide users with a comfortable, safe, timely, and accurate method for monitoring cardiac mechanical activity events, and provide a new solution for the precise monitoring of cardiovascular diseases.

[0100] This application also provides an electronic device that may include a processor, a memory, a receiver, and a transmitter. The processor is used to execute the non-contact real-time monitoring method for cardiac mechanical activity events mentioned in the above embodiments. The processor and memory can be connected via a bus or other means, taking a bus connection as an example. The receiver can be connected to the processor and memory via wired or wireless means.

[0101] The processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0102] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the non-contact real-time monitoring method for cardiac mechanical activity events in the embodiments of this application. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby realizing the non-contact real-time monitoring method for cardiac mechanical activity events in the above method embodiments.

[0103] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0104] The one or more modules are stored in the memory, and when executed by the processor, they perform the non-contact real-time monitoring method for cardiac mechanical activity events in the embodiment.

[0105] In some embodiments of this application, the user equipment may include a processor, a memory, and a transceiver unit. The transceiver unit may include a receiver and a transmitter. The processor, memory, receiver, and transmitter may be connected via a bus system. The memory is used to store computer instructions, and the processor is used to execute the computer instructions stored in the memory to control the transceiver unit to send and receive signals.

[0106] As one implementation method, the functions of the receiver and transmitter in this application can be implemented by transceiver circuits or dedicated transceiver chips, and the processor can be implemented by dedicated processing chips, processing circuits or general-purpose chips.

[0107] As another implementation approach, the server provided in this application embodiment can be implemented using a general-purpose computer. That is, the program code implementing the processor, receiver, and transmitter functions is stored in memory, and the general-purpose processor implements the processor, receiver, and transmitter functions by executing the code in memory.

[0108] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned non-contact real-time monitoring method for cardiac mechanical activity events. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0109] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned non-contact real-time monitoring method for cardiac mechanical activity events.

[0110] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave.

[0111] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0112] In this application, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0113] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to the embodiments of this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A non-contact method for real-time monitoring of cardiac mechanical activity events, characterized in that, include: The thoracic cavity micromotion signal and periodic rhythm signal are extracted from the millimeter wave echo signal of the target subject acquired in real time in a non-contact manner. The thoracic micromotion signal and the periodic rhythm signal are respectively input into the cardiac event recognition model, so that the cardiac event recognition model extracts the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively and splices them to obtain cardiac joint feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event recognition result data includes: cardiac mechanical activity event type label and temporal positioning data.

2. The non-contact real-time monitoring method for cardiac mechanical activity events according to claim 1, characterized in that, The cardiac event recognition model includes: a chest signal encoder, a rhythm signal encoder, a feature fusion layer, and a decoder; The input terminal of the feature fusion layer is connected to the output terminals of the chest cavity signal encoder and the rhythm signal encoder, respectively; the output terminal of the feature fusion layer is connected to the input terminal of the decoder. Correspondingly, the process of inputting the thoracic micromotion signal and the periodic rhythm signal into the cardiac event recognition model allows the model to extract the feature data corresponding to each of the thoracic micromotion signal and the periodic rhythm signal, and concatenate them to obtain cardiac joint feature data. The cardiac event recognition model then sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output the cardiac mechanical activity event recognition result data corresponding to the millimeter-wave echo signal, including: The thoracic micromotion signal is input into the thoracic signal encoder in the cardiac event recognition model, and the periodic rhythm signal is input into the rhythm signal encoder in the cardiac event recognition model, so that the thoracic signal encoder extracts the feature data corresponding to the thoracic micromotion signal, and the rhythm signal encoder extracts the feature data corresponding to the periodic rhythm signal; the feature fusion layer concatenates the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal respectively to obtain cardiac joint feature data; the decoder sequentially decodes and performs multi-label classification processing on the cardiac joint feature data to output cardiac mechanical activity event recognition result data corresponding to the millimeter wave echo signal.

3. The non-contact real-time monitoring method for cardiac mechanical activity events according to claim 2, characterized in that, The thoracic signal encoder includes three convolutional groups connected in sequence, and each convolutional group contains a contiguous one-dimensional convolutional layer and a residual block, so that the thoracic signal encoder can be used to extract local mutation feature data corresponding to the thoracic micromotion signal to distinguish different cardiac mechanical activity events.

4. The non-contact real-time monitoring method for cardiac mechanical activity events according to claim 2, characterized in that, The rhythm signal encoder is equipped with a gated loop unit, which enables the rhythm signal encoder to extract cross-cycle phase consistency feature data corresponding to the periodic rhythm signal, which is used to enhance the cardiac event recognition model's ability to perceive the temporal structure of cardiac mechanical activity events.

5. The non-contact real-time monitoring method for cardiac mechanical activity events according to claim 2, characterized in that, The feature fusion layer is used to perform standard normal distribution modeling, sampling, and vector concatenation on the feature data corresponding to the thoracic micromotion signal and the periodic rhythm signal, respectively, to obtain the corresponding feature concatenation data; and the feature concatenation data and the pre-acquired individual static feature data of the target subject are used as cardiac joint feature data to be input into the decoder; The individual static characteristic data includes data representing the gender, age, height, and weight of the target subject.

6. The non-contact real-time monitoring method for cardiac mechanical activity events according to claim 2, characterized in that, The decoder includes: a feature compression layer and a multi-label classifier connected in sequence; The feature compression layer is used to perform dimensional compression and semantic enhancement on the cardiac joint feature data to obtain the corresponding intermediate representation data; The multi-label classifier is used to classify the intermediate representation data into events in units of sliding time windows, so that each time window outputs in parallel multiple cardiac mechanical activity event type labels and time-series positioning data corresponding to the millimeter wave echo signal. The cardiac mechanical activity event type labels include: mitral valve closure label, aortic valve opening label, rapid ejection label, aortic valve closure label, mitral valve opening label, rapid filling label, atrial contraction label, and isovolumetric contraction label.

7. The method for real-time monitoring of non-contact cardiac mechanical activity events according to any one of claims 1 to 6, characterized in that, The extraction of thoracic micromotion signals and periodic rhythm signals from millimeter-wave echo signals of the target subject acquired in real time using a non-contact method includes: The millimeter-wave radar antenna is pointed directly at the chest area of ​​the target subject to collect the millimeter-wave echo signal of the target subject in a non-contact manner in real time. The millimeter-wave echo signal is preprocessed to obtain the corresponding intermediate frequency signal, and the intermediate frequency signal is subjected to a fast Fourier transform to obtain the corresponding frequency domain signal. Range bin extraction and multi-antenna beamforming are performed on the frequency domain signal to obtain the initial thoracic micromotion signal corresponding to the millimeter wave echo signal; The initial thoracic micromotion signal is subjected to adaptive enhancement processing based on phase change characteristics and frequency components to extract periodic rhythm signals and thoracic micromotion signals representing transient displacement components corresponding to heart valve activity from the initial thoracic micromotion signal.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the non-contact real-time monitoring method for cardiac mechanical activity events as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the non-contact real-time monitoring method for cardiac mechanical activity events as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the non-contact real-time monitoring method for cardiac mechanical activity events as described in any one of claims 1 to 7.

Images