Method for non-contact detection of human myocardial contractility by using millimeter waves

Abstract

The invention discloses a method for non-contact human myocardial contractility detection by using millimeter waves, and relates to the technical field of myocardial contractility detection. The method for performing non-contact human myocardial contractility detection by using millimeter waves comprises the following steps: S1, constructing a millimeter wave radar array and initializing a space model; s2, adaptive adjustment of the dynamic space model; s3, millimeter wave echo signals are collected and preprocessed; s4, high-dimensional signal feature coding and dimension reduction; s5, myocardial contractility index mapping and clinical evaluation; through millimeter wave radar array construction and dynamic adjustment of a space model, non-contact accurate heart monitoring is realized, invasive operation of a traditional technology is avoided, discomfort of a patient is reduced, tiny motion changes of the heart can be captured with high precision, specific algorithm analysis is combined, the contractility of the cardiac muscle can be accurately evaluated, and the cardiac muscle can be accurately monitored. And real-time feedback is provided, and simple and convenient installation and operation can be realized without depending on a high-level technology of an operator.

Description

Technical Field

[0001] The present invention relates to the technical field of myocardial contractility detection, and in particular to a method for non-contact detection of human myocardial contractility using millimeter waves. Background Art

[0002] Myocardial contractility refers to the ability of myocardial fibers to generate tension and shorten under a certain load and is an important indicator for evaluating cardiac function. A decrease in myocardial contractility is often closely related to heart failure. Traditional methods for evaluating myocardial contractility mainly include:

[0003] 1. Echocardiography: It indirectly reflects the myocardial contractility by measuring the left ventricular fractional shortening (FS) and ejection fraction (EF).

[0004] 2. Cardiac catheterization technique: Measure the pressure-volume loop or the rate of rise of left ventricular pressure (dP / dt max) to quantitatively analyze myocardial contractility.

[0005] 3. Magnetic resonance imaging (MRI): Accurately evaluates myocardial structure and function, but the equipment is expensive and the examination process is complicated.

[0006] Although these methods have high accuracy, most of them rely on contact equipment or invasive procedures, are complicated to operate and are not suitable for long-term monitoring. Summary of the Invention

[0007] In view of the shortcomings of the existing technology, the present invention provides a method for non-contact detection of human myocardial contractility using millimeter waves, which solves the problems of existing myocardial contractility assessment methods.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for non-contact detection of human myocardial contractility using millimeter waves, comprising the following steps:

[0009] S1 Millimeter-Wave Radar Array Construction and Spatial Model Initialization: Deploy the millimeter-wave radar array and design a multi-antenna transmitting and receiving system to ensure coverage of the heart monitoring area. Use the time delay and intensity information of the reflected signal to construct an initial three-dimensional spatial model and mark the heart's reference position.

[0010] S2 dynamic spatial model adaptive adjustment: real-time monitoring of human posture changes and motion status, and dynamic update of the heart position prediction model; heart position tracking through camera data collection;

[0011] S3 millimeter-wave echo signal acquisition and preprocessing: Multiple antennas synchronously collect cardiac micro-vibration echo signals to obtain multi-dimensional vibration data; apply the Doppler effect to analyze time-frequency domain characteristics and extract systolic and diastolic phase changes; and use a wavelet denoising algorithm to suppress environmental noise.

[0012] S4 High-Dimensional Signal Feature Coding and Dimensionality Reduction: Construct multi-channel convolutional neural network branches to process time series signals and spatial distribution features respectively; compress high-dimensional data through autoencoders to retain core parameters such as myocardial kinetic energy and stroke volume; optimize network structure to reduce computing power consumption and adapt to the needs of wearable devices;

[0013] S5 Myocardial contractility index mapping and clinical evaluation: Train feature decoding networks to establish nonlinear mapping between millimeter wave signals and myocardial contractility; introduce personalized physiological models; design a dynamic threshold warning system, combined with multi-cycle monitoring data, to identify the risk of heart failure.

[0014] Preferably, the step S1 specifically includes:

[0015] S1.1 Millimeter Wave Radar Array Deployment

[0016] Antenna layout: using uniform linear array (ULA), assuming the number of antennas is N and the antenna spacing is Where λ is the wavelength of millimeter wave, c = 3 × 10 8 m / s is the speed of light, f c =60GHz is the radar operating frequency;

[0017] Signal transmission and reception: The transmission signal adopts sine wave Where A is the signal amplitude, f0 is the transmission frequency, is the initial phase; the received signal is amplified by a low noise amplifier (LNA) and then down-converted;

[0018] S1.2 Initial 3D spatial model construction

[0019] Delay and angle calculation: Assume that the delay of the signal received by the i-th antenna relative to the first antenna is Δt i According to the triangulation positioning principle, the signal incident angle θ can be calculated by the formula calculate;

[0020] Distance calculation: Using the time delay information of multiple antennas, the distance r from the heart to the radar array is calculated by multilateration. If there are three antennas with time delays of Δt1, Δt2, and Δt3, the distance r can be calculated according to the equation group Solve the heart position (x0, y0, z0) based on the geometric relationship, for example, by using the least squares method;

[0021] S1.3 Key Parameter Calibration

[0022] Radar frequency calibration: Use a frequency meter to measure the actual transmission frequency f real , the deviation from the set frequency f is Δf=|f real -f|, set a frequency threshold f′ (such as 1MHz) to ensure Δf<f′;

[0023] Array geometry calibration: measuring the actual antenna spacing d real , the relative error with the theoretical value d And it is guaranteed that ∈ is less than 1%;

[0024] Signal sampling rate calibration: According to Nyquist sampling theorem, the sampling rate f s ≥2f max , where f max is the highest frequency of the signal. If the highest frequency of the signal is 100MHz, then f s ≥200MHz.

[0025] Preferably, the step S2 specifically includes:

[0026] S2.1 Real-time monitoring of human posture and motion status

[0027] Image recognition data: The camera collects the image sequence of human motion, and uses the image recognition algorithm to extract the motion features of the human body in the x, y, and z directions, and outputs them as a x 、a y 、a z , after filtering with a low-pass filter, we get The filtering formula is Where α is the filter coefficient;

[0028] S2.2 Dynamically update the heart position prediction model

[0029] Kalman filter:

[0030] Equation of state: x k =Ax k-1 +Bu k +w k ,in is the state vector, B is the control input matrix (if there is no control input, it can be set to zero matrix), u k is the control input vector, w k is the process noise, which obeys N(0,Q) distribution, and Q is the process noise covariance matrix;

[0031] Observation equation: z k =Zx k +v k , where z kis the observation vector (such as image acquisition measurement value), Z is the observation matrix, v k is the observation noise, which obeys N(0, R) distribution, and R is the observation noise covariance matrix;

[0032] Kalman gain calculation: in is the a priori estimated covariance;

[0033] Status Update: is the prior state estimate;

[0034] Covariance update:

[0035] Preferably, the S3 step specifically includes:

[0036] S3.1 Multi-antenna synchronous acquisition

[0037] Assume that the echo signal collected by the i-th antenna is s i (t), the acquisition time of each antenna is ensured to be consistent through the synchronization trigger mechanism;

[0038] Calculate the velocity v(t) and acceleration a(t), and perform the first-order difference of the echo signal s(t) to obtain the velocity The acceleration is obtained by the second-order difference

[0039] S3.2 Doppler effect analysis

[0040] Perform short-time Fourier transform on the echo signal s(t): Where w(τ-t) is the window function, and the time-frequency distribution is obtained. τ represents the integral variable, ω represents the angular frequency, ∈e -jωτ is a complex exponential function, where j is an imaginary unit;

[0041] Extract the phase changes of systole and diastole, and set the phase change range of systole to The diastolic phase range is

[0042] S3.3 Noise Suppression

[0043] Select the wavelet basis function (such as Daubechies wavelet), perform J-layer wavelet decomposition on the signal, and obtain the wavelet coefficient w j，k ; Then use hard threshold or soft threshold processing:

[0044] The hard threshold processing formula is:

[0045] The soft threshold processing formula is:

[0046] Where λ is the threshold, and σ is the standard deviation of noise, N is the signal length;

[0047] Perform wavelet reconstruction to obtain the denoised signal

[0048] Preferably, the S4 step specifically includes:

[0049] S4.1 Construction of multi-channel convolutional neural network (CNN) branches

[0050] Time series signal branch: a one-dimensional convolution layer is used with a convolution kernel size of k t ×1, the number is n t , the step size is s t , the input signal length is L, and the output feature map size is

[0051] Spatial distribution feature branch: using a two-dimensional convolution layer with a convolution kernel size of k s ×k s , the number is n s , the step size is s s , the input feature map size is H×W, and the output feature map size is

[0052] S4.2 Autoencoder Compression

[0053] Encoder: Let the input vector be X, and it is encoded through multiple layers of fully connected layers to obtain a low-dimensional potential vector Z. The encoding function is Z = f enc (X);

[0054] Decoder: Reconstruct the low-dimensional latent vector Z back to a high-dimensional vector The decoding function is

[0055] Loss function: minimize the reconstruction error

[0056] S4.3 Network structure optimization

[0057] Lightweight MobileNet: Using depth-wise separable convolution, decompose the standard convolution into depth-wise convolution and point-wise convolution, assuming the number of input channels is C in , the number of output channels is C out , the convolution kernel size is k×k, and the computational cost of depth-separable convolution is C in ×k×k×H×W+Cin×Cout×H×W, and the computational cost C is less than that of the standard convolution in ×C out ×k×k×H×W.

[0058] Preferably, the step S5 specifically includes:

[0059] S5.1 Feature decoding network training

[0060] Using multi-layer perceptron (MLP), the number of neurons in the input layer is n in , the number of hidden layers is l, and the number of neurons in each layer is n h , the number of neurons in the output layer is 1 (output myocardial contractility index);

[0061] The loss function uses mean square error loss where y q is the true value, is the predicted value, P is the total number of samples, and p∈1 to P;

[0062] The optimization algorithm uses the Adam algorithm, and the update parameter formula is m t =β1m t-1 +(1-β1)g t , v t =β2v t-1 +(1-β2)g t 2 , where g t is the gradient, β1, β2, η,∈ are hyperparameters;

[0063] S5.2 Introduction of Personalized Physiological Model

[0064] Establishing an age-myocardial contractility relationship model: Where g and h are fitting parameters, obtained by fitting clinical data; It indicates the maximum rate of change of myocardial contractility, that is, the maximum speed of myocardial contraction; age indicates age;

[0065] Taking weight and medical history into account, the prediction results are modified: where w weight is the weight correction factor, w history is the medical history correction factor, represents the corrected prediction result of myocardial contractility;

[0066] S5.3 Design of dynamic threshold warning system

[0067] Assume that the normal myocardial contractility range is The warning threshold is The maximum rate of change after correction is when Issue early warning.

[0068] Preferably, the S5 step also includes: real-time display of contraction force values, waveforms and health status prompts (normal / warning / dangerous), the contraction force values are displayed in digital form, the waveforms are displayed in dynamic curve form, and the health status prompts are displayed in icons or texts of different colors.

[0069] Preferably, the method further includes step S6: integrating a low-power microcontroller and a wireless transmission module to support long-term continuous operation; developing an adaptive calibration algorithm to regularly correct sensor drift errors; and establishing a cloud database to store individual historical data for trend analysis and personalized medical advice.

[0070] Preferably, the step S6 specifically includes:

[0071] S6.1 low-power microcontroller integrated with wireless transmission module

[0072] The microcontroller adopts ARM Cortex-M series, the operating frequency is f mcu ,Power consumption Pmcu, reduces power consumption through sleep mode and timed wake-up;

[0073] The wireless transmission module uses Bluetooth, and the transmission rate is R bt , transmission distance d bt ;

[0074] S6.2 Adaptive Calibration Algorithm Development

[0075] Assume that the sensor output is y and the true value is x. The least square method is used for calibration. The calibration model is y=ax+b, where a represents the proportional relationship between the sensor output y and the true value x, i.e., the slope. If a is greater than 1, it means that the sensor output amplifies the change of the true value; if a is less than 1, it means that the sensor output reduces the change of the true value; b represents the output value of the sensor when the true value x is 0, i.e., the intercept. Through multiple sets of measurement data (x i ,y i ) Solve for a and b, and output after calibration

[0076] S6.3 Cloud database establishment

[0077] Use the relational database MySQL to store individual historical data, including timestamp t, myocardial maximum contraction force Heart rate HR, etc.

[0078] The stored data were trend analyzed, and the linear regression model Y = dt + o was used to fit the changing trend of myocardial contractility over time and predict the risk of heart failure, where d is the regression coefficient, which represents the rate of change of myocardial contractility over time. If d is positive, it means that myocardial contractility increases over time; if d is negative, it means that myocardial contractility decreases over time; o is the intercept term, which represents the myocardial contractility value when t = 0.

[0079] The present invention provides a method for non-contact detection of human myocardial contractility using millimeter waves. Compared with the existing technology, it has the following advantages:

[0080] 1. This method of using millimeter waves for non-contact detection of human myocardial contractility achieves non-contact precise cardiac monitoring by constructing and dynamically adjusting a spatial model through a millimeter-wave radar array. This avoids the invasive operation of traditional technologies, reduces patient discomfort, and can capture tiny changes in heart movement with high precision. Combined with specific algorithm analysis, it can accurately evaluate myocardial contractility and provide real-time feedback. Compared with traditional cardiac ultrasound and other equipment, the technology of the present invention does not rely on the operator's high-level technology and can achieve relatively simple installation and operation; high-dimensional signal feature encoding and dimensionality reduction are adapted to the needs of wearable devices; a nonlinear mapping between millimeter-wave signals and myocardial contractility is established, and clinical evaluation is performed in combination with personalized physiological models; a low-power microcontroller and a wireless transmission module are integrated to support long-term continuous operation, and a cloud database is established to provide data support for trend analysis and personalized medical advice, thereby improving the ability to identify the risk of heart failure.

[0081] This method utilizes millimeter waves for non-contact detection of human myocardial contractility. It deploys millimeter-wave radars in a uniform linear array, accurately calculates antenna layout parameters, uses sinusoidal wave transmission and reception, and constructs an initial three-dimensional spatial model using triangulation and multilateration. Key parameters are calibrated to ensure the radar system's precise operation, providing a foundation for subsequent accurate monitoring. It can monitor human posture and motion in real time, identify motion using image acquisition data, and dynamically update the heart position prediction model using Kalman filtering. This effectively mitigates interference from human motion and improves the accuracy of heart position tracking.

[0082] 3. This method for non-contact human myocardial contractility detection using millimeter waves combines multi-channel input data and, through the collaborative work of encoders and decoders, achieves efficient representation and information extraction of physiological signals. First, we define multi-channel inputs, including time series data of cardiac micro-motion signals from multiple angles, as well as personal information, including gender, age, height, weight, etc., and normalize the personal information to fully utilize each user's physiological information. These inputs are processed by a specially designed encoder. The encoder's main task is to map the raw time series data into a latent space with high-level abstract features, which is then concatenated with joint features and imported into a feature decoding network to ultimately output a prediction of cardiac function. Through this end-to-end processing flow, we are able to generate more accurate prediction results for each user, organically combining personal information with physiological signals, and improving the overall performance and applicability of the prediction model.

[0083] 4. This method for non-contact human myocardial contractility detection using millimeter waves integrates a low-power microcontroller with a wireless transmission module. It utilizes an ARM Cortex-M series microcontroller and uses sleep mode and timed wakeup to reduce power consumption. It utilizes Bluetooth for data transmission, effectively extending the device's operating time while ensuring data transmission, facilitating long-term continuous monitoring. An adaptive calibration algorithm was developed, using the least squares method to calibrate the sensor. By solving the calibration model parameters, sensor output errors were corrected, improving the accuracy and reliability of the measured data. A cloud-based database was established to store individual historical data for subsequent trend analysis. A linear regression model was used to fit the trend of myocardial contractility changes and predict the risk of heart failure, providing data support and decision-making basis for personalized medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 This is a schematic diagram of the millimeter wave radar layout of the present invention;

[0085] Figure 2 Schematic diagram of myocardial force prediction according to the present invention;

[0086] Figure 3 A schematic diagram of the present invention for simultaneously extracting multiple micro-motion signals using millimeter waves;

[0087] Figure 4 This is a schematic diagram of the present invention combining multi-angle cardiac micro-motion signals with personal information to obtain myocardial force. DETAILED DESCRIPTION

[0088] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0089] The present invention provides four technical solutions:

[0090] Figure 1-3 A first embodiment is shown: a method for non-contact detection of human myocardial contractility using millimeter waves, comprising the following steps:

[0091] S1 Millimeter-Wave Radar Array Construction and Spatial Model Initialization: Deploy the millimeter-wave radar array and design a multi-antenna transmitting and receiving system to ensure coverage of the heart monitoring area. Use the time delay and intensity information of the reflected signal to construct an initial three-dimensional spatial model and mark the heart's reference position.

[0092] S2 dynamic spatial model adaptive adjustment: real-time monitoring of human posture changes and motion status, and dynamic update of the heart position prediction model; heart position tracking through camera data collection;

[0093] S3 millimeter-wave echo signal acquisition and preprocessing: Multiple antennas synchronously collect cardiac micro-vibration echo signals to obtain multi-dimensional vibration data; apply the Doppler effect to analyze time-frequency domain characteristics and extract systolic and diastolic phase changes; and use a wavelet denoising algorithm to suppress environmental noise.

[0094] S4 High-Dimensional Signal Feature Coding and Dimensionality Reduction: Construct multi-channel convolutional neural network branches to process time series signals and spatial distribution features respectively; compress high-dimensional data through autoencoders to retain core parameters such as myocardial kinetic energy and stroke volume; optimize network structure to reduce computing power consumption and adapt to the needs of wearable devices;

[0095] S5 Myocardial Contractility Index Mapping and Clinical Assessment: Train feature decoding networks to establish nonlinear mapping between millimeter wave signals and myocardial contractility; introduce personalized physiological models; design a dynamic threshold warning system that combines multi-cycle monitoring data to identify heart failure risks;

[0096] Step S6: Integrate a low-power microcontroller and a wireless transmission module to support long-term continuous operation; develop an adaptive calibration algorithm to regularly correct sensor drift errors; establish a cloud database to store individual historical data for trend analysis and personalized medical advice.

[0097] By constructing and dynamically adjusting the spatial model through the millimeter-wave radar array, non-contact precise cardiac monitoring is achieved, avoiding the invasive operation of traditional technologies, reducing patient discomfort, and being able to capture tiny changes in the heart's movement with high precision. Combined with specific algorithm analysis, it can accurately evaluate the contractility of the myocardium and provide real-time feedback. Compared with traditional cardiac ultrasound and other equipment, the technology of the present invention does not rely on the operator's high-level technology and can achieve relatively simple installation and operation; high-dimensional signal feature encoding and dimensionality reduction are adapted to the needs of wearable devices; a nonlinear mapping of millimeter-wave signals and myocardial contractility is established, and clinical evaluation is combined with personalized physiological models; a low-power microcontroller and a wireless transmission module are integrated to support long-term continuous operation, and a cloud database is established to provide data support for trend analysis and personalized medical advice, thereby improving the ability to identify the risk of heart failure.

[0098] The second embodiment differs from the first embodiment mainly in that step S1 specifically includes:

[0099] S1.1 Millimeter Wave Radar Array Deployment

[0100] Antenna layout: using uniform linear array (ULA), assuming the number of antennas is N and the antenna spacing is Where λ is the wavelength of millimeter wave, c = 3 × 10 8 m / s is the speed of light, f c =60GHz is the radar operating frequency;

[0101] Signal transmission and reception: The transmission signal adopts sine wave Where A is the signal amplitude, f0 is the transmission frequency, is the initial phase; the received signal is amplified by a low noise amplifier (LNA) and then down-converted;

[0102] S1.2 Initial 3D spatial model construction

[0103] Delay and angle calculation: Assume that the delay of the signal received by the i-th antenna relative to the first antenna is Δt i According to the triangulation positioning principle, the signal incident angle θ can be calculated by the formula calculate;

[0104] Distance calculation: Using the time delay information of multiple antennas, the distance r from the heart to the radar array is calculated by multilateration. If there are three antennas with time delays of Δt1, Δt2, and Δt3, the distance r can be calculated according to the equation group Solve the heart position (x0, y0, z0) based on the geometric relationship, for example, by using the least squares method;

[0105] S1.3 Key Parameter Calibration

[0106] Radar frequency calibration: Use a frequency meter to measure the actual transmission frequency f real , the deviation from the set frequency f is Δf=|f real -f|, set a frequency threshold f′ (such as 1MHz) to ensure Δf<f′;

[0107] Array geometry calibration: measuring the actual antenna spacing d real , the relative error with the theoretical value d And it is guaranteed that ∈ is less than 1%;

[0108] Signal sampling rate calibration: According to Nyquist sampling theorem, the sampling rate f s ≥2f max , where f max is the highest frequency of the signal. If the highest frequency of the signal is 100MHz, then f s ≥200MHz.

[0109] The S2 step specifically includes:

[0110] S2.1 Real-time monitoring of human posture and motion status

[0111] Image recognition data: The camera collects the image sequence of human motion, and uses the image recognition algorithm to extract the motion features of the human body in the x, y, and z directions, and outputs them as a x 、a y 、a z , after low-pass filtering, we get The filtering formula is Where α is the filter coefficient;

[0112] S2.2 Dynamically update the heart position prediction model

[0113] Kalman filter:

[0114] Equation of state: x k =Ax k-1 +Bu k +w k ,in is the state vector, B is the control input matrix (if there is no control input, it can be set to zero matrix), u k is the control input vector, w k is the process noise, which obeys N(0,Q) distribution, and Q is the process noise covariance matrix;

[0115] Observation equation: z k =Zx k +v k , where z k is the observation vector (such as image acquisition measurement value), Z is the observation matrix, vk is the observation noise, which obeys N(0, R) distribution, and R is the observation noise covariance matrix;

[0116] Kalman gain calculation: in is the a priori estimated covariance;

[0117] Status Update: is the prior state estimate;

[0118] Covariance update:

[0119] The system deploys millimeter-wave radar in a uniform linear array, precisely calculates antenna layout parameters, uses sinusoidal wave transmission and reception, and constructs an initial three-dimensional spatial model using triangulation and multilateration. Key parameters are calibrated to ensure the radar system's precise operation, providing a foundation for subsequent accurate monitoring. It can monitor human posture and motion in real time, using image acquisition data to identify motion and dynamically update the heart position prediction model through Kalman filtering, effectively addressing human motion interference and improving the accuracy of heart position tracking.

[0120] The third embodiment differs from the first embodiment mainly in that step S3 specifically includes:

[0121] S3.1 Multi-antenna synchronous acquisition

[0122] Assume that the echo signal collected by the i-th antenna is s i (t), the acquisition time of each antenna is ensured to be consistent through the synchronization trigger mechanism;

[0123] Calculate the velocity v(t) and acceleration a(t), and perform the first-order difference of the echo signal s(t) to obtain the velocity The acceleration is obtained by the second-order difference

[0124] S3.2 Doppler effect analysis

[0125] Perform short-time Fourier transform on the echo signal s(t): Where w(τ-t) is the window function, and the time-frequency distribution is obtained. τ represents the integral variable, ω represents the angular frequency, and e -jωτ is a complex exponential function, where j is an imaginary unit;

[0126] Extract the phase changes of systole and diastole, and set the phase change range of systole to The diastolic phase range is

[0127] S3.3 Noise Suppression

[0128] Select the wavelet basis function (such as Daubechies wavelet), perform J-layer wavelet decomposition on the signal, and obtain the wavelet coefficient w j，k ; Then use hard threshold or soft threshold processing:

[0129] The hard threshold processing formula is:

[0130] The soft threshold processing formula is:

[0131] Where λ is the threshold, and σ is the standard deviation of noise, N is the signal length;

[0132] Perform wavelet reconstruction to obtain the denoised signal

[0133] The S4 step specifically includes:

[0134] S4.1 Multi-way Convolutional Neural Network (CNN) Branch Construction

[0135] Time series signal branch: a one-dimensional convolution layer is used with a convolution kernel size of k t ×1, the number is n t , the step size is s t , the input signal length is L, and the output feature map size is

[0136] Spatial distribution feature branch: using a two-dimensional convolution layer with a convolution kernel size of k s ×k s , the number is n s , the step size is s s , the input feature map size is H×W, and the output feature map size is

[0137] S4.2 Autoencoder Compression

[0138] Encoder: Let the input vector be X, and it is encoded through multiple layers of fully connected layers to obtain a low-dimensional potential vector Z. The encoding function is Z = f enc (X);

[0139] Decoder: Reconstruct the low-dimensional latent vector Z back to a high-dimensional vector The decoding function is

[0140] Loss function: minimize the reconstruction error

[0141] S4.3 Network structure optimization

[0142] Lightweight MobileNet: Using depth-wise separable convolution, decompose the standard convolution into depth-wise convolution and point-wise convolution, assuming the number of input channels is C in , the number of output channels is C out , the convolution kernel size is k×k, and the computational cost of depth-separable convolution is C in ×k×k×H×W+Cin×Cout×H×W, and the computational cost C is less than that of the standard convolution in ×C out ×k×k×H×W.

[0143] The S5 step specifically includes:

[0144] S5.1 Feature decoding network training

[0145] Using multi-layer perceptron (MLP), the number of neurons in the input layer is n in , the number of hidden layers is l, and the number of neurons in each layer is n h , the number of neurons in the output layer is 1 (output myocardial contractility index);

[0146] The loss function uses mean square error loss where y q is the true value, is the predicted value, P is the total number of samples, and p∈1 to P;

[0147] The optimization algorithm uses the Adam algorithm, and the update parameter formula is m t =β1m t-1 +(1-β1)g t , v t =β2v t-1 +(1-β2)g t 2 , where g t is the gradient, β1, β2, η,∈ are hyperparameters;

[0148] S5.2 Introduction of Personalized Physiological Model

[0149] Establishing an age-myocardial contractility relationship model: Where g and h are fitting parameters, obtained by fitting clinical data; It indicates the maximum rate of change of myocardial contractility, that is, the maximum speed of myocardial contraction; age indicates age;

[0150] Taking weight and medical history into account, the prediction results are modified: where w weight is the weight correction factor, w history is the medical history correction factor, represents the corrected prediction result of myocardial contractility;

[0151] S5.3 Design of dynamic threshold warning system

[0152] Assume that the normal myocardial contractility range is The warning threshold is The maximum rate of change after correction is when Issue early warning.

[0153] The S5 step also includes: real-time display of contraction force values, waveforms and health status prompts (normal / warning / dangerous), the contraction force values are displayed in digital form, the waveforms are displayed in dynamic curve form, and the health status prompts are displayed in icons or texts of different colors.

[0154] This technology combines multi-channel input data and achieves efficient representation and information extraction of physiological signals through the collaborative work of encoders and decoders. First, we define multi-channel input, including time series data of multi-angle cardiac micro-motion signals, and personal information, including gender, age, height, weight, etc., and normalize the personal information to make full use of each user's physiological information. These inputs are processed by a specially designed encoder. The main task of the encoder is to map the original time series data into a latent space with high-level abstract features, and then splice it with the joint features and import it into the feature decoding network, and finally output a prediction of cardiac function. Through this end-to-end processing flow, we can generate more accurate prediction results for each user, organically combine personal information with physiological signals, and improve the overall performance and applicability of the prediction model.

[0155] The fourth embodiment differs from the first embodiment mainly in that the step S6 specifically includes:

[0156] S6.1 low-power microcontroller integrated with wireless transmission module

[0157] The microcontroller adopts ARM Cortex-M series, with an operating frequency of f mcu ,Power consumption Pmcu, reduces power consumption through sleep mode and timed wake-up;

[0158] The wireless transmission module uses Bluetooth, and the transmission rate is R bt , transmission distance d bt ;

[0159] S6.2 Adaptive Calibration Algorithm Development

[0160] Assume that the sensor output is y and the true value is x. The least square method is used for calibration. The calibration model is y=ax+b, where a represents the proportional relationship between the sensor output y and the true value x, i.e., the slope. If a is greater than 1, it means that the sensor output amplifies the change of the true value; if a is less than 1, it means that the sensor output reduces the change of the true value; b represents the output value of the sensor when the true value x is 0, i.e., the intercept. Through multiple sets of measurement data (x i ,y i ) Solve for a and b, and output after calibration

[0161] S6.3 Cloud database establishment

[0162] Use the relational database MySQL to store individual historical data, including timestamp t, myocardial maximum contraction force Heart rate HR, etc.

[0163] The stored data were trend analyzed, and the linear regression model Y = dt + o was used to fit the changing trend of myocardial contractility over time and predict the risk of heart failure, where d is the regression coefficient, which represents the rate of change of myocardial contractility over time. If d is positive, it means that myocardial contractility increases over time; if d is negative, it means that myocardial contractility decreases over time; o is the intercept term, which represents the myocardial contractility value when t = 0.

[0164] The present invention integrates a low-power microcontroller with a wireless transmission module, employing an ARM Cortex-M series microcontroller and reducing power consumption through sleep mode and timed wake-up. It utilizes Bluetooth for data transmission, effectively extending the device's operating time while ensuring data transmission, facilitating long-term continuous monitoring. An adaptive calibration algorithm is developed, employing the least squares method to calibrate the sensor. By solving the calibration model parameters, sensor output errors are corrected, improving the accuracy and reliability of the measured data. A cloud-based database is established to store individual historical data for subsequent trend analysis. A linear regression model is used to fit the changing trend of myocardial contractility and predict the risk of heart failure, providing data support and decision-making basis for personalized medicine.

[0165] Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0166] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0167] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for non-contact detection of human myocardial contractility using millimeter waves, characterized in that: The following steps are involved: S1 Millimeter-Wave Radar Array Construction and Spatial Model Initialization: Deploy the millimeter-wave radar array and design a multi-antenna transmitting and receiving system to ensure coverage of the heart monitoring area. Use the time delay and intensity information of the reflected signal to construct an initial three-dimensional spatial model and mark the heart's reference position. S2 dynamic spatial model adaptive adjustment: real-time monitoring of human posture changes and motion status, and dynamic update of the heart position prediction model; heart position tracking through camera data collection; S3 millimeter-wave echo signal acquisition and preprocessing: Multiple antennas synchronously collect cardiac micro-vibration echo signals to obtain multi-dimensional vibration data; apply the Doppler effect to analyze time-frequency domain characteristics and extract systolic and diastolic phase changes; and use a wavelet denoising algorithm to suppress environmental noise. S4 High-Dimensional Signal Feature Coding and Dimensionality Reduction: Construct multi-channel convolutional neural network branches to process time series signals and spatial distribution features respectively; compress high-dimensional data through autoencoders to retain core parameters such as myocardial kinetic energy and stroke volume; optimize network structure to reduce computing power consumption and adapt to the needs of wearable devices; S5 Myocardial Contractility Index Mapping and Clinical Assessment: Train the feature decoding network to establish a nonlinear mapping between millimeter wave signals and myocardial contractility; Introduce personalized physiological models; design a dynamic threshold warning system, combine multi-cycle monitoring data, and identify the risk of heart failure.

2. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The S1 step specifically includes: S1.1 Millimeter Wave Radar Array Deployment Antenna layout: using uniform linear array, the number of antennas is N, the antenna spacing is Where λ is the wavelength of millimeter wave, c = 3 × 10 8 m / s is the speed of light, f c =60GHz is the radar operating frequency; Signal transmission and reception: The transmission signal adopts sine wave Where A is the signal amplitude, f0 is the transmission frequency, is the initial phase; the received signal is amplified by a low noise amplifier and then down-converted; S1.2 Initial 3D spatial model construction Delay and angle calculation: Assume that the delay of the signal received by the i-th antenna relative to the first antenna is Δt i According to the triangulation positioning principle, the signal incident angle θ can be calculated by the formula calculate; Distance calculation: Using the time delay information of multiple antennas, the distance r from the heart to the radar array is calculated by multilateration. If there are three antennas with time delays of Δt1, Δt2, and Δt3, the distance r can be calculated according to the equation group Solve the heart position (x0, y0, z0) using geometric relationships; S1.3 Key Parameter Calibration Radar frequency calibration: Use a frequency meter to measure the actual transmission frequency f real , the deviation from the set frequency f is Δf=|f real -f|, set a frequency threshold f′ to ensure Δf<f′; Array geometry calibration: measuring the actual antenna spacing d real , the relative error with the theoretical value d And it is guaranteed that ∈ is less than 1%; Signal sampling rate calibration: According to Nyquist sampling theorem, the sampling rate f s ≥2f max , where f max is the highest frequency of the signal.

3. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The S2 step specifically includes: S2.1 Real-time monitoring of human posture and motion status Image recognition data: The camera collects the image sequence of human motion, and uses the image recognition algorithm to extract the motion features of the human body in the x, y, and z directions, and outputs them as a x 、a y 、a z , after low-pass filtering, we get The filtering formula is Where α is the filter coefficient; S2.2 Dynamically update the heart position prediction model Kalman filter: Equation of state: x k =Ax k-1 +Bu k +w k ,in is the state vector, B is the control input matrix, u k is the control input vector, w k is the process noise, which obeys N(0,Q) distribution, and Q is the process noise covariance matrix; Observation equation: z k =Zx k +v k , where z k is the observation vector, Z is the observation matrix, v k is the observation noise, which obeys N(0, R) distribution, and R is the observation noise covariance matrix; Kalman gain calculation: in is the a priori estimated covariance; Status Update: is the prior state estimate; Covariance update:

4. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The S3 step specifically includes: S3.1 Multi-antenna synchronous acquisition Assume that the echo signal collected by the i-th antenna is s i (t), the acquisition time of each antenna is ensured to be consistent through the synchronization trigger mechanism; Calculate the velocity v(t) and acceleration a(t), and perform the first-order difference of the echo signal s(t) to obtain the velocity The acceleration is obtained by the second-order difference S3.2 Doppler effect analysis Perform short-time Fourier transform on the echo signal s(t): Where w(τ-t) is the window function, and the time-frequency distribution is obtained. τ represents the integral variable, ω represents the angular frequency, and e -jωτ is a complex exponential function, where j is an imaginary unit; Extract the phase changes of systole and diastole, and set the phase change range of systole to The diastolic phase range is S3.3 Noise Suppression Select the wavelet basis function and perform J-layer wavelet decomposition on the signal to obtain the wavelet coefficient w j，k ; Then use hard threshold or soft threshold processing: The hard threshold processing formula is: The soft threshold processing formula is: Where λ is the threshold, and σ is the standard deviation of noise, N is the signal length; Perform wavelet reconstruction to obtain the denoised signal 5. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The S4 step specifically includes: S4.1 Multi-way Convolutional Neural Network (CNN) Branch Construction Time series signal branch: a one-dimensional convolution layer is used with a convolution kernel size of k t ×1, the number is n t , the step size is s t , the input signal length is L, and the output feature map size is Spatial distribution feature branch: using a two-dimensional convolution layer with a convolution kernel size of k s ×k s , the number is n s , the step size is s s , the input feature map size is H×W, and the output feature map size is S4.2 Autoencoder Compression Encoder: Let the input vector be X, and it is encoded through multiple layers of fully connected layers to obtain a low-dimensional potential vector Z. The encoding function is Z = f enc (X); Decoder: Reconstruct the low-dimensional latent vector Z back to a high-dimensional vector The decoding function is Loss function: minimize the reconstruction error S4.3 Network structure optimization Lightweight MobileNet: Using depth-wise separable convolution, decompose the standard convolution into depth-wise convolution and point-wise convolution, assuming the number of input channels is C in , the number of output channels is C out , the convolution kernel size is k×k, and the computational cost of depth-separable convolution is C in ×k×k×H×W+Cin×Cout×H×W, and the computational cost C is less than that of the standard convolution in ×C out ×k×k×H×W.

6. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The S5 step specifically includes: S5.1 Feature decoding network training Using multi-layer perceptron (MLP), the number of neurons in the input layer is n in , the number of hidden layers is l, and the number of neurons in each layer is n h , the number of neurons in the output layer is 1 (output myocardial contractility index); The loss function uses mean square error loss where y q is the true value, is the predicted value, P is the total number of samples, and p∈1 to P; The optimization algorithm uses the Adam algorithm, and the update parameter formula is m t =β1m t-1 +(1-β1)g t , v t =β2v t-1 +(1-β2)g t 2 , where g t is the gradient, β1, β2, η,∈ are hyperparameters; S5.2 Introduction of Personalized Physiological Model Establishing an age-myocardial contractility relationship model: Where g and h are fitting parameters, obtained by fitting clinical data; It indicates the maximum rate of change of myocardial contractility, that is, the maximum speed of myocardial contraction; age indicates age; Taking weight and medical history into account, the prediction results are modified: where w weight is the weight correction factor, w history is the medical history correction factor, represents the corrected prediction result of myocardial contractility; S5.3 Design of dynamic threshold warning system Assume that the normal myocardial contractility range is The warning threshold is The maximum rate of change after correction is when Issue early warning.

7. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The step S5 also includes: real-time display of contraction force values, waveforms and health status prompts, the contraction force values are displayed in digital form, the waveforms are displayed in dynamic curve form, and the health status prompts are displayed in icons or texts of different colors.

8. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 1, characterized in that: The method further includes step S6: integrating a low-power microcontroller with a wireless transmission module to support long-term continuous operation; Develop an adaptive calibration algorithm to regularly correct sensor drift errors; Establish a cloud database to store individual historical data for trend analysis and personalized medical recommendations.

9. The method for non-contact detection of human myocardial contractility using millimeter waves according to claim 8, characterized in that: The S6 step specifically includes: S6.1 low-power microcontroller integrated with wireless transmission module The microcontroller adopts ARM Cortex-M series, with an operating frequency of f mcu ,Power consumption Pmcu, reduces power consumption through sleep mode and timed wake-up; The wireless transmission module uses Bluetooth, and the transmission rate is R bt , transmission distance d bt ; S6.2 Adaptive Calibration Algorithm Development Assume that the sensor output is y and the true value is x. The least square method is used for calibration. The calibration model is y=ax+b, where a represents the proportional relationship between the sensor output y and the true value x, i.e., the slope. If a is greater than 1, it means that the sensor output amplifies the change of the true value; if a is less than 1, it means that the sensor output reduces the change of the true value; b represents the output value of the sensor when the true value x is 0, i.e., the intercept. Through multiple sets of measurement data (x i ,y i ) Solve for a and b, and output after calibration S6.3 Cloud database establishment Use the relational database MySQL to store individual historical data, including timestamp t, myocardial maximum contraction force Heart rate HR; The stored data were trend analyzed, and the linear regression model Y = dt + o was used to fit the changing trend of myocardial contractility over time and predict the risk of heart failure, where d is the regression coefficient, which represents the rate of change of myocardial contractility over time. If d is positive, it means that myocardial contractility increases over time; if d is negative, it means that myocardial contractility decreases over time; o is the intercept term, which represents the myocardial contractility value when t = 0.

Images