A non-contact method, system, and storage medium for arrhythmia detection

Abstract

The application provides a non-contact arrhythmia detection method, system and storage medium, and the method comprises the following steps: acquiring a patient heart micro-motion signal detected by a radar system, and based on the time domain phase feature, symbolic dynamics feature and motion vector feature of the heart micro-motion signal; obtaining the input of a pre-trained disease diagnosis model based on the time domain phase feature, symbolic dynamics feature and motion vector feature, inputting the input into the disease diagnosis model, and outputting the arrhythmia detection result of the patient. The application can not only capture high-frequency heart motion features, but also has high detection accuracy and user comfort by using a non-contact detection method.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a non-contact method and system for detecting cardiac arrhythmias. Background Technology

[0002] Cardiac arrhythmias such as atrial fibrillation and premature ventricular contractions seriously affect patients' health, and timely detection helps maintain heart health. Current methods for detecting arrhythmias are mainly limited to hospital examinations, such as electrocardiograms (ECG) and echocardiograms, as well as wearable devices such as smartwatches.

[0003] However, the existing detection methods have the following drawbacks: (1) The detection frequency of the signal wave is low. The traditional detection wave frequency (0.5-40Hz) cannot capture high-frequency mechanical motion characteristics. For example, the current detection method is difficult to detect paroxysmal atrial fibrillation, which easily leads to low detection accuracy; (2) Currently, contact detection methods are mostly used. For example, when patients undergo ECG detection, electrodes need to be attached, which results in poor comfort and compliance, and is easily affected by skin impedance.

[0004] Therefore, there is an urgent need for a new method for detecting arrhythmias that can not only capture high-frequency cardiac motion characteristics to improve detection accuracy, but also improve user comfort. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a non-contact arrhythmia detection method and system, which can not only capture high-frequency cardiac motion characteristics, but also adopt a non-contact detection method, thus having high detection accuracy and user comfort.

[0006] One aspect of the present invention provides a non-contact method for detecting cardiac arrhythmias, the method comprising the following steps:

[0007] The patient's cardiac micromotion signals detected by the radar system were acquired, and the temporal phase characteristics, symbolic dynamics characteristics, and motion vector characteristics were obtained based on the cardiac micromotion signals.

[0008] The input to the pre-trained disease diagnosis model is obtained based on time-domain phase features, symbolic dynamics features, and motion vector features. This input is then fed into the disease diagnosis model, and the output is the patient's arrhythmia detection result.

[0009] The temporal phase features and motion vector features are obtained through the following methods:

[0010] The cardiac micro-motion signal is divided into two orthogonal signals. The phase signal is obtained based on the phase information of the two orthogonal signals, and a Fourier transform is performed on the phase signal to obtain the time-domain phase characteristics; and

[0011] Based on a nonlinear deformation model of biomechanics, nonrigid displacement components are decomposed from cardiac micromotion signals. The nonrigid stress tensor is calculated based on the nonrigid displacement components and the viscoelastic constitutive equation of myocardial tissue, thereby obtaining motion vector characteristics based on the nonrigid stress tensor.

[0012] In some embodiments of the present invention, after acquiring the cardiac micromotion signal, the method further includes: constructing a cardiac dynamics model based on the cardiac micromotion signal;

[0013] A cardiac dynamics model is constructed based on cardiac micro-motion signals, including:

[0014] The heart surface was discretized into multiple nodes by finite element mesh generation, and the phase change of each node was calculated using the radar system's transmitted signal and the heart's micro-motion signal.

[0015] A cardiac dynamics model is constructed by estimating the displacement vector of each node based on the phase changes of each node and the extended Kalman filter.

[0016] In some embodiments of the present invention, the calculation of the non-rigid stress tensor based on the non-rigid displacement components and the viscoelastic constitutive equation of myocardial tissue includes:

[0017] In the cardiac dynamics model, the difference between the non-rigid displacement components of adjacent nodes on the cardiac surface is calculated to obtain the local strain tensor, and the strain rate tensor is obtained by time differentiation of the local strain tensor.

[0018] The relaxation modulus is determined based on biomechanical experiments or inverse problem solving methods.

[0019] The stress tensor in the time domain is determined using the strain rate tensor, relaxation modulus, and viscoelastic constitutive equation of myocardial tissue.

[0020] In some embodiments of the present invention, the time-domain phase characteristics are obtained in the following manner:

[0021] The cardiac micro-motion signal is divided into two orthogonal signals, and the arctangent function of the ratio of the two orthogonal signals is calculated to obtain the phase signal;

[0022] The first derivative of the phase signal is obtained, and the time-domain phase characteristics are obtained based on the result of the first derivative and the signal sampling rate.

[0023] In some embodiments of the present invention, the input to the pre-trained disease diagnosis model is obtained based on temporal phase features, symbolic dynamics features, and motion vector features, including:

[0024] Local waveform features are obtained from temporal phase features using a one-dimensional convolutional neural network model, temporal dependence features of symbol sequences are obtained from symbol dynamics features using a long short-term memory network model, and multi-scale deformation features are obtained from motion vector features using a graph diffusion network model.

[0025] The system integrates local waveform features, time-dependent features, and multi-scale deformation features, and obtains the input of the pre-trained disease diagnosis model based on the fused mechanical features.

[0026] In some embodiments of the present invention, before fusing local waveform features and time-dependent features, the method further includes: aligning the feature spaces of local waveform features and time-dependent features using JS divergence constraints;

[0027] It integrates local waveform features, time-dependent features, and multi-scale deformation features, including:

[0028] The contribution weights of local waveform features, time-dependent features, and multi-scale deformation features to the detection results are adaptively adjusted based on the signal-to-noise ratio, and a weighted fusion method is used to fuse local waveform features, time-dependent features, and multi-scale deformation features.

[0029] In some embodiments of the present invention, the input to the pre-trained disease diagnosis model is obtained based on the fused mechanical features, including:

[0030] Based on a pre-trained cross-modal diffusion model or myocardial displacement-electrical conduction coupling equation, the fused mechanical features are mapped to electrophysiological signals, and the electrophysiological signals are used as input to a pre-trained disease diagnosis model.

[0031] In some embodiments of the present invention, the cardiac micro-motion signal is obtained by preprocessing the raw signal collected by the radar system. The preprocessing process includes:

[0032] The raw signals acquired by the radar system are input into a high-order anti-aliasing filter and clutter suppressor for filtering.

[0033] The filtered signal is subjected to dynamic beamforming and phase compensation to obtain a noise-reduced signal.

[0034] The cardiac micromotion signal is obtained by optimizing the noise-reduced signal using a beamforming algorithm that maximizes the autocorrelation coefficient and compressed sensing technology.

[0035] Another aspect of the present invention provides a non-contact arrhythmia detection system, including a processor, a memory, and a computer program / instructions stored in the memory. The processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the system implements the steps of the method described in any of the above embodiments.

[0036] Another aspect of the present invention provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method described in any of the above embodiments.

[0037] The non-contact arrhythmia detection method and system proposed in this invention can utilize the non-contact sensing capability of radar systems to detect weak motion signals from the human body, and extract multimodal features of cardiac activity by analyzing the motion signals, thereby realizing the detection and identification of arrhythmias. This application not only captures high-frequency cardiac motion features, but also adopts a non-contact detection method, which has high detection accuracy and user comfort.

[0038] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0039] Those skilled in the art will understand that the objectives and advantages achievable with this invention are not limited to those specifically described above, and that the above and other objectives achievable with this invention will become clearer from the following detailed description. Attached Figure Description

[0040] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings:

[0041] Figure 1 This is a flowchart illustrating a non-contact arrhythmia detection method according to an embodiment of the present invention.

[0042] Figure 2 This is a flowchart illustrating a non-contact arrhythmia detection method according to another embodiment of the present invention.

[0043] Figure 3 This is a schematic diagram of the process of constructing a cardiac dynamics model in one embodiment of the present invention.

[0044] Figure 4 This is a schematic diagram of the process of fusing features and obtaining arrhythmia detection results in one embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0046] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0047] 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.

[0048] In the following description, embodiments of the invention 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.

[0049] Most existing methods for detecting cardiac arrhythmias use contact methods, which have drawbacks such as poor comfort and susceptibility to skin impedance. Furthermore, the use of low-frequency probe waves to capture cardiac motion and generate electrocardiogram signals increases the risk of missed or misdiagnosed cases. In short, current arrhythmia detection methods are not only less accurate but also less comfortable.

[0050] Based on this, this application proposes a method for detecting cardiac arrhythmias based on a radar system. For example... Figure 1 As shown, this method can utilize the non-contact sensing capability of radar systems to sense weak motion signals of the human body (such as small chest deformations caused by heartbeats). By analyzing these signals, multimodal features of cardiac activity can be extracted, and artificial intelligence algorithms can be used to achieve automatic detection and identification of arrhythmias.

[0051] Figure 2 This is a schematic flowchart of a non-contact arrhythmia detection method according to one embodiment of this application. Figure 2 As shown, the method includes steps S110 to S120.

[0052] Step S110: Acquire cardiac micromotion signals and extract features. Step S110 can be divided into the following steps: Step S111: Acquire patient cardiac micromotion signals detected by the radar system; Step S112: Extract time-domain, frequency-domain, and spatial-domain features from the cardiac micromotion signals to obtain time-domain phase features, symbolic dynamics features, and motion vector features, respectively.

[0053] Cardiac micromotion signals can be used to characterize minute mechanical movements in the patient's heart region, and the waveform of the signal can be displayed in the form of time versus cardiac displacement on the x and y axes. Because cardiac motion is very weak, to improve resolution, this application uses a radar system to transmit signals for detecting the patient's heart motion and receives cardiac micromotion signals carrying the motion characteristics of the human heart region. This application uses a radar system to detect cardiac micromotion signals, which not only utilizes high-frequency signals for detection but also provides a more comfortable user experience through non-contact detection.

[0054] As an example, traditional ECG relies on contact electrodes to capture myocardial electrical conduction signals (0.05–100 Hz), which suffers from signal attenuation due to changes in skin impedance. This application, however, utilizes frequency-modulated continuous wave (FMCW) millimeter-wave radar (76–81 GHz band) to measure the micro-Doppler effect caused by myocardial mechanical motion (e.g., cardiac displacement determined by millimeter-wave signals can reach an accuracy of 50 μm). This allows for the reconstruction of cardiac dynamics models and the detection of arrhythmias, overcoming the physical limitations of traditional electrophysiological testing. The aforementioned use of millimeter-wave radar to acquire cardiac micro-motion signals is merely an example; this application can also employ other radar systems capable of receiving high-frequency signals, such as lidar or infrared radar, to acquire human motion signals.

[0055] Furthermore, in step S111, the cardiac micro-motion signal can be the original signal detected by the radar system, or the original signal detected by the radar system after signal preprocessing. Considering that there are various interferences during the transmission and reception of signals by the radar system, in addition to common electromagnetic interference, there may also be problems with focusing on the patient's heart region (i.e., in addition to the micro-motion signal of the heart region, it may also include micro-motion signals of other chest regions, causing interference), this application typically requires filtering, noise reduction, and optimization processing of the original signal detected by the radar system, and further feature extraction steps are performed on the cardiac micro-motion signal obtained after signal preprocessing. Moreover, the execution order of filtering, noise reduction, and optimization processing steps is not specifically limited in this application during signal preprocessing; for example, filtering and optimization processing can be performed on the signal after noise reduction processing.

[0056] In some embodiments of the present invention, the signal preprocessing process includes: inputting the original signal (i.e., the patient's original cardiac micromotion signal) collected by the radar system into a high-order anti-aliasing filter and a clutter suppressor for filtering; wherein, the clutter suppressor can effectively separate the 0.1-5Hz cardiac micromotion signal from the 30-60 breaths / minute respiratory harmonics; performing dynamic beamforming processing (which can improve the signal-to-noise ratio by about 15dB) and phase compensation (used to eliminate carrier frequency offset caused by radar local oscillator leakage and reduce phase distortion) on the filtered signal to obtain a denoised signal; and optimizing the denoised signal through a beamforming algorithm that maximizes the autocorrelation coefficient and compressed sensing technology to obtain the cardiac micromotion signal.

[0057] Specifically, this application employs high-order anti-aliasing filters and clutter suppressors (which can be constructed based on cross-domain diffusion models or empirical mode decomposition) for filtering, combined with dynamic beamforming algorithms and phase compensation techniques for noise reduction, and optimized using beamforming algorithms that maximize autocorrelation coefficients and compressed sensing techniques. That is, this application can preprocess the original signal from multiple dimensions, as follows: Compared to traditional ECG signal preprocessing methods that suppress baseline drift and electromyographic noise, the filtering process in this application needs to match the characteristics of high-frequency signals and enhance anti-interference capabilities. Therefore, for the high-frequency characteristics of radar signals, this application uses high-order anti-aliasing filters to eliminate signal aliasing and utilizes clutter suppression algorithms to suppress environmental electromagnetic interference, thereby reducing the impact of noise from Wi-Fi or Bluetooth bands. During the noise reduction process, in the spatial dimension, an adaptive cardiac motion tracking beam is constructed using dynamic beamforming algorithms to achieve spatial filtering, suppressing multipath effects caused by chest breathing or limb movement, thereby focusing the signal in the cardiac region. Furthermore, in the temporal dimension, phase unwinding techniques can be used to eliminate signal phase ambiguity. During the signal optimization stage, beamforming algorithms that maximize autocorrelation coefficients can be used to accurately locate the cardiac motion region, and compressed sensing technology can be applied to optimize signal reconstruction efficiency.

[0058] As an example, in the filtering process, the high-order anti-aliasing filter can be an existing Butterworth filter, Chebyshev filter, or elliptic filter, or it can be a self-designed filter. For instance, in the noise reduction process, a Long Short-Term Memory (LSTM) network can be used to learn the heart cycle pattern and predict the heart position θ at the next moment. tThis allows for dynamic adjustment of the beam pointing angle, achieving beam tracking accuracy at the 0.1° level. For example, this application can employ a dual-ring phase-locked loop structure to construct a phase compensator, enabling more precise frequency and phase control. The inner ring uses a Costas ring to eliminate carrier frequency offset, while the outer ring compensates for phase ambiguity caused by motion through cross-correlation of the reference signal. This can significantly reduce phase noise (e.g., to 0.05 rad RMS) by correcting phase errors. Furthermore, in signal optimization, this application can utilize a beamforming algorithm that maximizes the autocorrelation coefficient. This optimization method can be similar to existing technologies, including steps such as constructing a signal model, calculating beamforming output, determining the optimization objective, and obtaining the optimal weight vector by solving the optimization problem. However, it can also differ from existing technologies. For instance, in the optimization objective design and solution stages, a cost function can be constructed. (W is the weight vector, and R is the autocorrelation matrix of the desired signal), and the optimal weight vector is solved by maximizing the generalized Rayleigh quotient so that the autocorrelation coefficient of the optimized signal reaches its peak within the heart rhythm cycle (0.5-5Hz).

[0059] In the signal preprocessing process, the filters or algorithms mentioned above are merely examples. Other signal processing devices or algorithms may also be used in this application, and the present invention is not limited thereto.

[0060] In some embodiments of the present invention, after step S111, the method further includes step S113: constructing a cardiac dynamics model based on cardiac micromotion signals. The process of constructing the cardiac dynamics model in step S113 can be implemented using a finite element model, such as... Figure 3 As shown, the specific steps are as follows:

[0061] Step S1131: Discretize the heart surface into multiple nodes by finite element mesh generation.

[0062] Step S1132: Calculate the phase change of each node using the radar system's transmitted signal and the cardiac micro-motion signal. In this step, the phase change of each node is the difference between the phase of the radar system's transmitted signal and the phase of the cardiac micro-motion signal for each node on the heart surface, which can be expressed by the formula Δφ. i =φ i,发射 -φ i,微动信号 , where △φ i φ represents the phase change of node i. i,发射 and φ i,微动信号 These represent the phase of the transmitted signal from node i and the phase of the cardiac micro-motion signal from node i, respectively.

[0063] Step S1133: Estimate the displacement vector di = (dix, diy, diz) of each node based on the phase change of each node and the Extended Kalman Filter (EKF), thereby constructing the cardiac dynamics model.

[0064] As an example, after step S1133, the method further includes step S1134: iteratively updating the cardiac dynamics model, specifically: using an extended Kalman filter to fuse displacement observations (the displacement values ​​of each node on the cardiac surface are determined by the signals received during continuous radar detection) and model predictions (the displacement predictions of the EKF) in real time, iteratively optimizing the displacement vectors of each node, realizing real-time tracking of the deformation process, and improving the accuracy of dynamic response.

[0065] In isolated porcine heart experiments, the cardiac dynamics model constructed in this application can detect localized abnormal myocardial bulging at the 0.2 mm level (simulating atrial fibrillation mechanical activity), which is superior to the minimum detection threshold (0.5 mm) of traditional methods. Furthermore, clinical test data show that for atrial fibrillation, the success rate of capturing cardiac micromotion signals is increased from 68% with traditional radar to 93%, and the false alarm rate is reduced to 2.1%.

[0066] Step S112: Extract the mechanical features of cardiac activity from the cardiac micromotion signal, including temporal phase features, symbolic dynamics features, and motion vector features.

[0067] Traditional methods often employ time-frequency analysis (such as wavelet transform) or morphological features (such as the standard deviation of the RR interval), which have the following limitations: limited time-frequency resolution, making it difficult to capture microsecond-level electrophysiological abnormalities; sensitivity to noise (such as electromyographic interference); neglect of phase information of the original IQ domain signal (such as orthogonal phase nonlinear changes); and the use of single features (such as the RR interval), failing to achieve multimodal fusion. Therefore, this application not only proposes fusing multimodal features but also introduces new feature extraction methods in the time-frequency and spatial dimensions to improve resolution. Furthermore, this application introduces a deep learning model to extract weak cardiac mechanical motion features from complex signals, overcoming the limitations of traditional ECG methods that rely on wavelet transform or empirical mode decomposition to separate motion artifacts.

[0068] (1) Regarding the time domain dimension: This application can decompose the cardiac micro-motion signal received by the radar system into I-channel and Q-channel (I-channel and Q-channel signals are orthogonal signals), and use dual-channel differential phase analysis and short-time fractional Fourier transform (STFRFT) to achieve time-frequency analysis, thereby extracting the chaotic phase features caused by arrhythmia. Specifically, the formula for extracting the time-domain phase features from the cardiac micro-motion signal can be expressed as:

[0069]

[0070] Where, φ d (t) represents the time-domain phase characteristic at time t, f s This represents the signal sampling rate.

[0071] The above-mentioned formula for extracting time-domain phase features is based on complex signals (I+jQ) and Fourier transform. Through Fourier transform, the time-domain I / Q signal can be converted to the frequency domain, allowing analysis of the signal's spectral characteristics (such as Doppler shift). Therefore, this application can also be considered as extracting time-domain phase features through the following process: dividing the cardiac micro-motion signal into two orthogonal signals, obtaining the phase signal based on the phase information of the two orthogonal signals, and performing a Fourier transform on the phase signal to obtain the time-domain phase features. The process involves obtaining a phase signal based on the phase information of two orthogonal signals, and then performing a Fourier transform on the phase signal to obtain time-domain phase features. This includes: dividing the cardiac micromotion signal into two orthogonal signals, calculating the arctangent function of the ratio of the two orthogonal signals to obtain the phase signal; performing a short-time fractional Fourier transform on the phase signal, and obtaining a time-domain representation by adjusting the fractional-order parameter α (e.g., analyzing the signal by rotating it on the time-frequency plane and separating different motion modes); extracting short-time phase change features from the time-domain representation (e.g., quantifying the instantaneous frequency shift of cardiac motion through peak detection or energy focusing), and using these features as time-domain phase features.

[0072] Considering the non-stationary characteristics of cardiac micro-motion signals (such as transient abnormalities in atrial fibrillation), and that STFRFT can more flexibly match the time-frequency distribution of non-stationary signals and improve the detection accuracy of weak phase changes, the signal processing scheme proposed in this application in the time domain dimension can directly extract phase change rate features, significantly improving the processing accuracy of complex non-stationary signals. Its technical complexity far exceeds that of the traditional linear filtering process of ECG.

[0073] (2) Regarding the frequency domain dimension: Multi-scale symbolic dynamics operations are used to extract symbolic dynamics features from cardiac micromotion signals to capture nonlinear dynamic characteristics. The process of symbolic dynamics operations in this application is the same as that in the prior art, which includes three steps: symbolization, multi-scale decomposition, and feature extraction. Specifically, the cardiac micromotion signal can be decomposed into orthogonal I-channel and Q-channel signals; the amplitude of the I-channel or Q-channel signal is quantized into multiple symbols to form a symbol sequence; the transfer rules of the symbol sequence are statistically analyzed to generate a transfer probability matrix. For example, quantizing the amplitude of the I-channel or Q-channel sub-signals yields four symbols {A,B,C,D}, and the transfer rules of a symbol sequence of length 3 (e.g., the symbol sequence “A→B→C”) are statistically analyzed to form a 16×4-dimensional transfer probability matrix.

[0074] (3) For the spatial dimension: Based on the nonlinear deformation model of biomechanics, non-rigid displacement components are decomposed from the cardiac micromotion signal, and non-rigid stress tensor (also known as stress distribution tensor) is calculated based on the non-rigid displacement components and the viscoelastic constitutive equation of myocardial tissue, thereby obtaining motion vector characteristics based on the stress tensor.

[0075] More specifically, traditional ECG relies on 12-lead potential difference analysis, which cannot detect mechanical asynchrony (such as left bundle branch block accompanied by ventricular asynchrony). This application, however, can quantify interventricular mechanical delay through three-dimensional motion vector field analysis. For example, it can identify local contraction abnormalities in premature ventricular contractions (typically, abrupt changes in cardiac motion curvature >35%) through motion trajectory curvature analysis. The feature extraction process in the spatial dimension of this application is as follows: Based on a biomechanical nonlinear deformation model, cardiac motion can be decomposed into rigid displacement components and non-rigid deformation components. Rigid displacement components can represent overall positional changes in the patient's body, such as chest cavity movement caused by respiration, while non-rigid displacement components can represent positional changes caused by local myocardial contraction or expansion. Filtering out rigid displacement components can improve the signal-to-noise ratio. After calculating the non-rigid stress tensor using the non-rigid displacement components, the non-rigid stress tensor can be used as motion vector features. Alternatively, a deep learning network can be used to extract motion vector features from the non-rigid stress tensor for subsequent joint feature analysis. Furthermore, the nonlinear deformation model of biomechanics mentioned in this application can be a nonlinear anisotropic material model, a nonlinear finite element model, or a nonlinear elastic model, etc., and the viscoelastic constitutive equation of myocardial tissue can also be constructed based on existing nonlinear viscoelastic models such as the generalized Maxwell model or the distributed viscoelastic model. This invention does not specifically limit the nonlinear deformation model of biomechanics and the viscoelastic constitutive equation of myocardial tissue, and can be selected according to needs.

[0076] As an example, this application establishes the following viscoelastic constitutive equation for myocardial tissue to describe the dynamic deformation response of myocardium under millimeter wave action:

[0077]

[0078] Wherein, ∈(τ) is the strain rate tensor, σ(t) is the stress tensor, and G(t) is the relaxation modulus, used to describe the viscoelastic properties of myocardial materials.

[0079] In some embodiments of the present invention, the calculation of the non-rigid stress tensor based on the non-rigid displacement components and the viscoelastic constitutive equation of myocardial tissue includes: in a cardiac dynamics model, calculating the difference between the non-rigid displacement components of adjacent nodes on the cardiac surface to obtain the local strain tensor, and performing time differentiation on the local strain tensor to obtain the strain rate tensor; determining the relaxation modulus based on biomechanical experiments or inverse problem solving; and determining the stress tensor in the time domain using the strain rate tensor, relaxation modulus, and the viscoelastic constitutive equation of myocardial tissue. Specifically, the process of determining G(t) through experimental calibration is as follows: based on biomechanical experimental data (such as in vitro myocardial stretching tests), a mathematical form of G(t) is preset (such as an exponential decay model); the process of determining G(t) through inverse problem solving is as follows: the cardiac micro-motion signal containing displacement data measured by a radar system is input into the constitutive equation, and the parameters of G(t) are back-derived using an optimization algorithm (such as the least squares method). That is, after calculating the strain rate tensor and relaxation modulus based on the cardiac micro-motion signal, the non-rigid stress tensor can be calculated using the above-mentioned constitutive equation. The above-mentioned methods for calculating strain rate tensor and relaxation modulus are merely examples. Other existing calculation methods can also be used to calculate strain rate tensor and relaxation modulus. This invention is not limited thereto.

[0080] Step S120: This application can combine multimodal information such as time-domain phase changes and frequency-domain Doppler features for analysis to achieve accurate detection of the patient's cardiac motion. Specifically, the input to the pre-trained disease diagnosis model is obtained based on time-domain phase features, symbol dynamics features, and motion vector features; inputting this into the disease diagnosis model can output the patient's arrhythmia detection results.

[0081] The input to the pre-trained disease diagnosis model is obtained based on temporal phase features, symbolic dynamics features, and motion vector features, including: using temporal phase features, symbolic dynamics features, and motion vector features as input datasets to input the pre-trained disease diagnosis model; or, fusing temporal phase features, symbolic dynamics features, and motion vector features, and using the fused features as input to the pre-trained disease diagnosis model.

[0082] In some embodiments of the present invention, since the three different dimensions of features extracted from cardiac micromotion signals have low information abundance, this application can utilize a multi-branch encoder to achieve the condensation and fusion of multimodal features. In the temporal branch, symbolic dynamics branch, and motion branch, temporal phase features, symbolic dynamics features, and motion vector features can be encoded respectively. The fused encoded features yield fused mechanical features. The specific process of encoding and fusing multimodal features using a multi-branch encoder is as follows: a one-dimensional convolutional neural network model (i.e., 1D CNN model) is used to obtain local waveform features (also called local waveform encoded features) from temporal phase features; a long short-term memory network model (i.e., LSTM model) is used to obtain temporal dependence features of symbol sequences (also called temporal dependence encoded features) from symbolic dynamics features; and a graph diffusion network model is used to obtain multi-scale deformation features (also called multi-scale deformation encoded features) from motion vector features; the local waveform features, temporal dependence features, and multi-scale deformation features are fused, and the input of a pre-trained disease diagnosis model is obtained based on the fused mechanical features.

[0083] As an example, the layer structure of a 1D CNN model can be a one-dimensional convolutional layer (64, k=7) - a max pooling layer - a batch normalization layer, and the hidden units of an LSTM model can be 128. The structure and parameters of the one-dimensional convolutional neural network model, long short-term memory network model, and graph diffusion network model mentioned in this application can be set according to requirements, and this invention is not limited thereto. In addition, motion vector features can also be directly fused with local waveform features and temporal dependency features to obtain fused mechanical features without encoding.

[0084] Furthermore, local waveform features, time-dependent features, and multi-scale deformation features are integrated, including: adaptively adjusting the contribution weights of local waveform features, time-dependent features, and multi-scale deformation features to the detection results based on the signal-to-noise ratio, and using a weighted fusion method to integrate local waveform features, time-dependent features, and multi-scale deformation features.

[0085] Based on the signal-to-noise ratio (SNR), the contribution of each branch's features to the diagnostic result can be adaptively adjusted through dynamic weights. For example, the weights of each branch can be dynamically adjusted using learnable parameters α∈[0,1] and β∈[0,1], where α and β can be generated from the current signal's SNR value using a sigmoid function, thereby achieving noise adaptive fusion. The formula for adjusting the fused features through weights is as follows:

[0086]

[0087] in, This represents the local waveform features obtained by encoding the time-domain phase features. This represents the temporal dependency features obtained from symbolic dynamics feature encoding. Motion vector feature encoding yields multi-scale deformation features. Mechanical characteristics of fusion coding.

[0088] As an example, such as Figure 4 As shown, since the time-domain phase characteristics and symbol dynamics characteristics are obtained based on the IQ-channel signals, the mechanical characteristics of the fused encoding can be obtained by first coupling the time-domain and frequency-domain characteristics based on the in-phase encoder or quadrature encoder, and then coupling the motion vector characteristics. Will Input decoder (the decoder is used to input decoder) (The information is converted into information that a pre-trained disease diagnosis model can recognize), and finally, the patient's arrhythmia detection results can be obtained through the disease diagnosis model.

[0089] Furthermore, time-domain and frequency-domain features may suffer from feature space misalignment. Therefore, before fusing local waveform features and time-dependent features, the method further includes aligning the feature spaces of local waveform features and time-dependent features using JS divergence constraints. JS divergence (Jensen-Shannon Divergence) is an index used to measure the similarity between two probability distributions. Adding JS divergence constraints to the loss function means constructing the loss function based on JS divergence. Based on the original formula of JS divergence and the branch features in this application, the following formula for the loss function can be obtained:

[0090]

[0091] Where KL represents the KL divergence,

[0092] In tests on the MIT-BIH AF database, the fused multimodal coding features achieved an F1-score of 94.7%, a significant improvement over single time-domain features (89.2%) and sign features (91.1%). Furthermore, it achieved a detection sensitivity of 98.3% for the f-wave specific to atrial fibrillation, effectively identifying low-frequency oscillations in the microvolt range (<50μV) that are missed by traditional methods.

[0093] In some embodiments of the present invention, the disease diagnosis model can perform multi-task diagnosis, determine whether a patient has arrhythmia, and output arrhythmia classification results, thereby achieving arrhythmia identification and classification. For example, the output results of the disease diagnosis model include: normal heart rhythm, or arrhythmia type such as atrial fibrillation, premature ventricular contraction classification, or conduction block localization. Based on the output results of the disease diagnosis model, alarms can be triggered or abnormal event data can be stored simultaneously. Depending on the function of the disease diagnosis model, it can be constructed based on a classification model, wherein the classification model used to construct the disease diagnosis model can be a vector machine, decision tree, or CardioNet neural network, etc. The disease diagnosis model mentioned in this application can be an existing model or a self-designed disease diagnosis model; the present invention does not specifically limit it.

[0094] As an example, the CardioNet neural network can be used for the diagnosis of cardiovascular diseases. For instance, based on deep learning models (such as CNN and LSTM) and the CardioNet neural network, feature extraction and classification of electrocardiogram (ECG) signals can be performed. However, the input data of existing disease diagnosis models is mostly ECG signals. To make it suitable for existing disease diagnosis models, after obtaining the fused encoded mechanical features, a pre-trained cross-modal diffusion model or myocardial displacement-electrical conduction coupling equation can be used to map the fused mechanical features to electrophysiological signals, and then use the electrophysiological signals as the input to the pre-trained disease diagnosis model. That is, the fused mechanical features can be used as the input to the disease diagnosis model, or the fused mechanical features can be converted into electrophysiological signals and then used as the input to the disease diagnosis model. For example, this application constructs a myocardial displacement-electrical conduction coupling equation to convert the mechanical displacement of the heart into the phase change of myocardial electrical activity, and its formula is as follows:

[0095]

[0096] Where λ is the wavelength of the electromagnetic wave emitted by the radar system, and vmyocardium(t) is the velocity of the ventricular wall at time t. This refers to the electrocardiogram (ECG) signal corresponding to the heart's micro-movement signals.

[0097] The pre-trained cross-modal diffusion model or myocardial displacement-electrical conduction coupling equation mentioned above is only an example, and this application does not limit the specific way of mapping mechanical features to electrophysiological signals.

[0098] Furthermore, considering individual differences among patients, disease diagnostic models can be constructed by combining classification models and adaptive learning mechanisms. For example, an individual physiological feature database can be built, and real-time calibration of individual physiological features can be achieved through online adaptive learning mechanisms (such as establishing a user baseline using cardiac micromotion signals collected in the first 30 seconds). An environmental interference feature database can also be built, and preprocessing of raw signals can be achieved through online transfer learning to identify environmental electromagnetic interference features in real time. Moreover, since disease diagnostic models are used to identify arrhythmias caused by factors such as heart rate variability and differences between atrial and ventricular activity, prior knowledge of myocardial kinematics can be embedded in the disease diagnostic model as constraints on the diagnostic results, such as displacement continuity constraints or periodic symmetry constraints.

[0099] As an example, a high-resolution radar vital signs imaging system can be developed based on the aforementioned arrhythmia detection methods to achieve non-contact cardiac function assessment, overcoming the functional limitations of traditional ECG. For instance, the spatial resolution of the radar vital signs imaging system can reach 3mm (axial) × 8° (azimuth), and the temporal resolution can reach a frame interval of 50ms, enabling non-contact cardiac function assessment (ejection fraction calculation error can be less than 5%), thus overcoming the functional limitations of traditional ECG.

[0100] The arrhythmia detection method proposed in this application can be applied to various scenarios such as home, hospital, and nursing home, and has the following significant advantages:

[0101] (1) Non-contact, non-invasive detection: No sensors need to be worn, avoiding the problems of poor comfort and poor compliance of traditional detection methods, thus improving the user experience. It can also achieve 24-hour continuous monitoring, improving the detection rate of atrial fibrillation.

[0102] (2) A novel data processing method: a novel signal processing method is proposed based on the high frequency characteristics of the signal using high frequency probe waves; a new method for extracting time-domain phase features is proposed; and based on the current consideration of time-domain or frequency-domain analysis, time-domain phase, frequency-domain Doppler and spatial motion vector field features are combined to overcome the problem of feature singularity; and mechanical motion features are mapped to ECG signals to break through the limitations of ECG electrical signals.

[0103] (3) Adopt a dynamic disease diagnosis model: Use an online learning mechanism to calibrate individual physiological characteristics in real time, improve individual adaptability, and reduce false alarm rate.

[0104] (4) It has a high detection accuracy: Through clinical trials, the accuracy of atrial fibrillation identification has reached 97.32%.

[0105] Corresponding to the above method, the present invention also provides a non-contact arrhythmia detection system, the system including a computer device, the computer device including a processor and a memory, the memory storing computer programs / instructions, the processor being used to execute the computer programs / instructions stored in the memory, and when the computer programs / instructions are executed by the processor, the system implements the steps of the method as described above.

[0106] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the aforementioned edge computing server deployment method. 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.

[0107] 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 invention. 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 invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0108] It should be clarified that the present invention 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 the present invention 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 the present invention.

[0109] In this invention, 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.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-contact cardiac arrhythmia detection system, characterized in that, This system is used to perform the following steps: Acquire patient cardiac micromotion signals detected by a radar system, and obtain time-domain phase features, symbolic dynamics features, and motion vector features based on the cardiac micromotion signals; The input to the pre-trained disease diagnosis model is obtained based on time-domain phase features, symbolic dynamics features, and motion vector features. The input is then fed into the disease diagnosis model, and the output is the patient's arrhythmia detection result. The time-domain phase features and motion vector features are obtained through the following methods: The cardiac micro-motion signal is divided into two orthogonal signals. A phase signal is obtained based on the phase information of the two orthogonal signals, and a Fourier transform is performed on the phase signal to obtain the time-domain phase characteristics; and Based on a nonlinear deformation model of biomechanics, nonrigid displacement components are decomposed from the cardiac micromotion signal, and nonrigid stress tensors are calculated based on the nonrigid displacement components and the viscoelastic constitutive equation of myocardial tissue, thereby obtaining motion vector characteristics based on the nonrigid stress tensors. The system receives input to the pre-trained disease diagnosis model based on temporal phase features, symbolic dynamics features, and motion vector features in the following manner: Local waveform features are obtained from temporal phase features using a one-dimensional convolutional neural network model, temporal dependence features of symbol sequences are obtained from symbol dynamics features using a long short-term memory network model, and multi-scale deformation features are obtained from motion vector features using a graph diffusion network model. The system integrates local waveform features, time-dependent features, and multi-scale deformation features, and obtains the input of the pre-trained disease diagnosis model based on the fused mechanical features.

2. The system according to claim 1, characterized in that, After acquiring cardiac micro-motion signals, the system is also used to construct cardiac dynamics models based on these signals. The cardiac dynamics model was constructed in the following manner: The heart surface was discretized into multiple nodes by finite element mesh generation, and the phase change of each node was calculated using the radar system's transmitted signal and the heart's micro-motion signal. A cardiac dynamics model is constructed by estimating the displacement vector of each node based on the phase changes of each node and the extended Kalman filter.

3. The system according to claim 2, characterized in that, The system calculates the non-rigid stress tensor in the following way: In the cardiac dynamics model, the difference between the non-rigid displacement components of adjacent nodes on the cardiac surface is calculated to obtain the local strain tensor, and the strain rate tensor is obtained by time differentiation of the local strain tensor. The relaxation modulus is determined based on biomechanical experiments or inverse problem solving methods. The stress tensor in the time domain is determined using the strain rate tensor, the relaxation modulus, and the viscoelastic constitutive equation of myocardial tissue.

4. The system according to claim 1, characterized in that, The time-domain phase characteristics are obtained through the following method: The cardiac micro-motion signal is divided into two orthogonal signals, and the arctangent function of the ratio of the two orthogonal signals is calculated to obtain the phase signal; The first derivative of the phase signal is calculated, and the time-domain phase characteristics are obtained based on the result of the first derivative and the signal sampling rate.

5. The system according to claim 1, characterized in that, Before fusing local waveform features and temporal dependency features, the system also uses JS divergence constraints to align the feature spaces of local waveform features and temporal dependency features. This system integrates local waveform features, time-dependent features, and multi-scale deformation features in the following manner: The contribution weights of local waveform features, time-dependent features, and multi-scale deformation features to the detection results are adaptively adjusted based on the signal-to-noise ratio, and a weighted fusion method is used to fuse local waveform features, time-dependent features, and multi-scale deformation features.

6. The system according to claim 1, characterized in that, This system obtains the input for a pre-trained disease diagnosis model based on the fused mechanical features in the following way: Based on a pre-trained cross-modal diffusion model or myocardial displacement-electrical conduction coupling equation, the fused mechanical features are mapped to electrophysiological signals, and the electrophysiological signals are used as input to a pre-trained disease diagnosis model.

7. The system according to claim 1, characterized in that, The cardiac micro-motion signal is obtained by preprocessing the raw signal acquired by the radar system. The preprocessing includes: The raw signals acquired by the radar system are input into a high-order anti-aliasing filter and clutter suppressor for filtering. The filtered signal is subjected to dynamic beamforming and phase compensation to obtain a noise-reduced signal. The cardiac micromotion signal is obtained by optimizing the noise-reduced signal using a beamforming algorithm that maximizes the autocorrelation coefficient and compressed sensing technology.

Images