Real-time multi-mode physiological signal analysis method for myocardial ischemia

Abstract

The invention discloses a real-time multi-mode physiological signal analysis method for myocardial ischemia, and belongs to the technical field of data analys.The technical scheme includes that electrocardiogram, photoelectric volume pulse waves and thoracic impedance signals are synchronously acquired, and phase difference caused by motion displacement is compensated in real time by the aid of a triaxial accelerometer; constructing a multi-dimensional quality evaluation matrix containing an electrocardiogram signal-to-noise ratio, a photoelectric volume pulse wave perfusion index and a chest impedance variance, and jointly judging whether signal reconstruction is triggered or not based on a dynamic threshold value and accelerometer data; a multi-channel signal block which is judged to be low in quality through the quality evaluation matrix and comprises the electrocardiogram and photoelectric volume pulse waves and chest impedance synchronized with the electrocardiogram is input into a generative adversarial network with electrophysiological constraints to be reconstructed, and a reconstructed high-quality electrocardiogram signal is output; and myocardial ischemia key indexes are extracted from the reconstructed signals, and multi-level risk early warning is carried out. The real-time multi-modal physiological signal analysis method for myocardial ischemia has the beneficial effect that the real-time multi-modal physiological signal analysis method for myocardial ischemia is provided.

Description

Technical Field

[0001] This invention belongs to the field of data analysis technology, and specifically relates to a method for real-time multimodal physiological signal analysis for myocardial ischemia. Background Technology

[0002] Currently, myocardial ischemia detection still mainly relies on electrocardiogram (ECG) monitoring. However, during exercise, ECG and PPG signals are often affected by artifacts, drift, and loss, causing waveform distortion and making misinterpretation highly likely. Conventional methods lack the ability to perform collaborative analysis of multimodal signals and intelligent repair of low-quality signals, and deep models lacking physiological prior constraints are prone to generating unreasonable reconstructed waveforms. Therefore, there is an urgent need for a real-time multimodal physiological signal analysis method for myocardial ischemia to improve the accuracy and reliability of dynamic detection of myocardial ischemia. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time multimodal physiological signal analysis method for myocardial ischemia, thereby improving the early detection capability of myocardial ischemia events.

[0004] This invention is achieved through the following measures: a real-time multimodal physiological signal analysis method for myocardial ischemia, characterized by comprising: Simultaneously acquire electrocardiogram, photoplethysmography pulse wave and chest impedance signals, and use a triaxial accelerometer to compensate for the phase difference caused by motion displacement in real time; A multidimensional quality assessment matrix was constructed, which included ECG signal-to-noise ratio, photoplethysmography perfusion index, and chest impedance variance. Based on dynamic threshold and accelerometer data, it was determined whether signal reconstruction was triggered. Multichannel signal blocks containing electrocardiograms and their synchronized photoplethysmography pulse waves and chest impedance, which are judged to be of low quality by the quality assessment matrix, are input into a generative adversarial network with electrophysiological constraints for reconstruction, and the reconstructed high-quality electrocardiogram signal is output. Key indicators of myocardial ischemia are extracted from the reconstructed signals to provide multi-level risk warnings.

[0005] Furthermore, a four-electrode method was used to acquire thoracic impedance signals between the sternal manubrium and xiphoid process, and a photoplethysmography (PPG) signal was acquired at the radial artery of the left wrist. An electrocardiogram (ECG) signal was also acquired in a standard lead II configuration. A triaxial accelerometer was placed in front of the chest to monitor the X, Y, and Z axis motion at a frequency of at least 100 Hz. Furthermore, when the triaxial accelerometer detected a displacement velocity exceeding a set threshold in any direction, a phase difference calibration procedure was initiated, specifically as follows: Using the chest impedance signal as the reference time axis, calculate the time difference between the peak value of the R wave on the electrocardiogram and the systolic peak of the PPG waveform. Optimize by maximizing mutual information The value is optimized using the principle of maximizing mutual information. The values ​​specifically include: Set time difference The search range and step size; For each candidate Shift the ECG signal ; The mutual information values ​​of the ECG and PPG signals after translation within the same time window were calculated. The mutual information was calculated using the histogram method to estimate the probability distribution. Traverse all Then, select the option that maximizes the mutual information value. As the optimal time difference: Calculate the maximum mutual information value. If the maximum mutual information value is less than the set threshold, the signal quality is determined to be insufficient, the calibration is abandoned, and the segment is marked as low quality; otherwise, the phase calibration is completed.

[0006] Furthermore, when the triaxial accelerometer detects that the angular velocity of the wearer's torso in any direction exceeds a set threshold, a motion artifact compensation filter is activated to suppress interference in the three channels of electrocardiogram, photoplethysmography pulse wave, and chest impedance, specifically including: The filter is a first-order low-pass filter, and its time constant is dynamically adjusted according to the root mean square value output by the triaxial accelerometer. The time constant τ is calculated by multiplying the root mean square value of acceleration by a preset coefficient. The filter is converted into a digital IIR filter form through bilinear transformation, which supports filtering on edge computing platforms with a delay of no more than 50ms. Different filtering targets are used for different signal channels: High-frequency artifacts above 50Hz are filtered out from the electrocardiogram signal to suppress electromyographic interference and maintain the clarity of the main QRS complex. The baseline drift below 0.5 Hz of the photoplethysmography (PPG) signal is suppressed while preserving the rising pulse waveform. The chest impedance signal is smoothed to prevent abrupt changes while preserving the range of respiratory waveform variations. The filter parameters are updated every 200ms.

[0007] Furthermore, the multidimensional quality assessment matrix consists of three components: electrocardiogram signal-to-noise ratio, photoplethysmography perfusion index, and chest impedance variance. The signal-to-noise ratio of an electrocardiogram (ECG) is obtained by the power ratio of the QRS segment to the isoelectric segment. The photoplethysmography perfusion index is derived from the ratio of the AC component to the DC component of the photoelectric signal. The variance of chest impedance was calculated after removing respiratory modulation.

[0008] Furthermore, a normalized score is calculated based on the quality vector and compared with a personalized dynamic threshold. When the score is lower than the set threshold and the motion intensity exceeds the preset level, the signal reconstruction process is triggered. In extreme motion states, the segment is directly discarded. If key pathological features are detected, reconstruction is forcibly initiated even if the score is close to the threshold. The dynamic threshold is automatically generated by the system based on data collected from the user at rest, and is updated daily to the 25th percentile value of the score distribution. It is coupled with real-time motion data to form a dual-condition judgment mechanism.

[0009] Furthermore, the generative adversarial network is trained end-to-end using a total loss function that includes adversarial loss, L1 reconstruction loss, ST segment cyclic consistency loss, and physiological constraint loss.

[0010] Furthermore, key indicators of myocardial ischemia were extracted from the reconstructed signal, including the ST segment depression index, which was extracted based on the difference between the average amplitude of the ST segment and the isoelectric segment, according to the R-wave localization analysis. The peak value of the time derivative of the impedance signal reflects an indicator of abrupt changes in pleural volume or perfusion. The ratio of the rise time to the period length of the photoelectric pulse wave is calculated as the rise time index. Based on the above indicators, a three-tiered ischemia early warning mechanism is established: When ΔST < -0.1mV and lasts for more than 60 seconds, or the peak value of the Z derivative is abnormal, or RTI > 0.5, it is judged as a Level I warning; When the ST segment is continuously depressed by >0.1mV and the RTI increases by ≥30% simultaneously, a Level II warning is triggered; when the ST segment is depressed by >0.2mV and the ST segment morphology meets the Minnesota I pattern, or when the Z and RTI are severely abnormally synchronized (both >0.6), a Level III warning is immediately triggered.

[0011] The beneficial effects of the technical solution provided by this invention are as follows: Through the quality judgment and reconstruction mechanism, low-quality, multi-channel physiological signals can be optimized in a targeted manner. In particular, even in the presence of noise, baseline drift, or short-term missing data in the main channel of electrocardiogram (ECG), high-fidelity reconstructed waveforms can still be generated, thereby preserving key pathological information such as the ST-T segment and improving diagnostic reliability. By introducing photoplethysmography (PPG) and thoracic impedance (Z) as auxiliary input features, and utilizing their information on cardiac synchronicity, perfusion status, and contact quality, the feature extraction accuracy and anti-artifact ability in the ECG reconstruction process are improved, effectively reducing the risk of misjudgment caused by single-channel degradation. A physiological constraint layer is added at the end of the generator to ensure that the reconstructed waveform not only closely resembles the real ECG in morphology but also conforms to normal or pathophysiological laws in biological mechanism, reducing morphological distortion. By calculating the difference between the average potential of the ST segment and the average potential of the isoelectric segment, a mechanism is implemented to force remodeling when ST segment depression is >0.1 mV and directly trigger a Level III warning when ST segment depression is >0.2 mV, enabling rapid response in high-risk situations such as potential myocardial ischemia. Attached Figure Description

[0012] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings listed below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a real-time multimodal physiological signal analysis method for myocardial ischemia according to an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0015] See Figure 1 A real-time multimodal physiological signal analysis method for myocardial ischemia, characterized by comprising: S1. Synchronously acquire electrocardiogram (ECG), photoplethysmography (PPG), and chest impedance Z signals, and use a triaxial accelerometer to compensate for the phase difference caused by motion displacement in real time. The acquisition of various signals includes acquiring thoracic impedance signals between the sternal manubrium and xiphoid process using a four-electrode method, acquiring photoplethysmography (PPG) signals at the radial artery of the left wrist, and acquiring electrocardiogram (ECG) signals in a standard lead II configuration; a triaxial accelerometer monitors the X, Y, and Z axis motion at a frequency of not less than 100 Hz. When the accelerometer detects that the displacement velocity in any direction exceeds a set threshold, a phase difference calibration procedure is initiated, specifically as follows: Using the chest impedance signal as the reference time axis, calculate the time difference between the peak value of the R wave on the electrocardiogram and the systolic peak of the PPG waveform. Optimize by maximizing mutual information Values, specifically including: The search interval was set to [-100ms, +100ms] with a step size of 1ms, and a total of 201 candidate time differences were constructed. ; For each The ECG signal is shifted through a circular buffer. ,get ; Within a 5-second time window (256Hz sampling, 1280 points), for The edge and joint probability distributions of the PPG signal are estimated using the histogram method, and the mutual information value is calculated.

[0016] In the formula, The mutual information value between ECG (electrocardiogram) and PPG (photoplethysmography), expressed in bits. To perform a joint interval traversal and summation for all ECG and PPG amplitude combinations; The first of the ECG signal amplitude Discrete intervals (bins); The first of the PPG signal amplitudes Discrete intervals (bins); For the ECG signal to fall into the first The marginal probabilities of each interval are obtained through histogram statistics. For PPG signal falling into the first The marginal probabilities of each interval; For ECG in the Interval and PPG in the first The joint probability of simultaneous occurrence of intervals is obtained based on two-dimensional histogram statistics. The number of equally wide discrete sub-intervals into which the amplitude ranges of the ECG and PPG signals are divided (histogram binning, recommended value is 32); It is a base-2 logarithmic function used to measure the information gain (in bits) corresponding to the probability ratio.

[0017] Traverse all Then, select the option that maximizes the mutual information value. As the optimal time difference: Calculate the maximum mutual information value. If the maximum mutual information value is less than a set threshold (e.g., 0.1 bits), the signal quality is deemed insufficient, calibration is abandoned, and the segment is marked as low quality; otherwise, phase calibration is completed. A higher value indicates a strong statistical correlation between the two (i.e., good timeline alignment). when When the value is less than 0.1 bits, it indicates that the signals are unrelated or that there is significant phase misalignment / artifact interference. This invention achieves dynamic phase alignment compensation between ECG and PPG signals by maximizing this mutual information.

[0018] When the triaxial accelerometer detects that the angular velocity of the wearer's torso in any direction exceeds a set threshold (e.g., when the triaxial accelerometer detects an angular velocity >30° / s), a motion artifact compensation filter is activated to suppress interference in the three channels of ECG, photoplethysmography pulse wave, and chest impedance. Specifically, this includes: The filter is a first-order low-pass filter, and its time constant is dynamically adjusted based on the root mean square value output by the triaxial accelerometer. The time constant τ is calculated by multiplying the root mean square value of the acceleration by a preset coefficient. The continuous domain transfer function is:

[0019] In the formula, s is the Laplace variable, and τ is the filter time constant, according to... Dynamic adjustment; among which The root mean square value of acceleration (unit: g or m / s²). Using the bilinear transform method to Mapping to numbers This yields an IIR filter in the form of a difference equation: ; In the formula, z refers to the complex frequency domain variable of the discrete-time system.

[0020] Differential filtering of different signal channels is performed using IIR filters, specifically as follows: High-frequency artifacts above 50Hz are filtered out from the electrocardiogram signal to suppress electromyographic interference and maintain the clarity of the main QRS complex; the P wave-QRS-T wave morphology is preserved to prevent false detection.

[0021] The baseline drift below 0.5 Hz of the photoplethysmography (PPG) signal is suppressed, while the rising waveform of the pulse is preserved. The positions of the troughs and peaks are stabilized, which is beneficial for the calculation of the PPG perfusion index (PI) and rise time index.

[0022] The abrupt change smoothing process of the thoracic impedance signal preserves the range of respiratory waveform changes and stabilizes the impedance waveform, which is beneficial for extracting ischemic indicators such as the peak value of the first derivative of thoracic impedance.

[0023] S2. Construct a multidimensional quality assessment matrix that includes ECG signal-to-noise ratio, photoplethysmography perfusion index, and chest impedance variance, and determine whether signal reconstruction is triggered based on dynamic threshold and accelerometer data. The ECG signal-to-noise ratio is obtained from the power ratio of the QRS segment to the isoelectric segment of the ECG. The photoplethysmography perfusion index is derived from the ratio of the AC component to the DC component of the photoelectric signal. The chest impedance variance is calculated after removing respiratory modulation. These three factors constitute the signal quality vector within the time window.

[0024] Furthermore, for each 5-second time window (corresponding to 1280 sampling points, sampling rate 256Hz), the following three physiological signal quality indicators were calculated: The power ratio of the QRS complex and isoelectric segment in each cardiac cycle is calculated and used as the signal-to-noise ratio (SNR) of the electrocardiogram signal. The formula is as follows:

[0025] In the formula, SNR: signal-to-noise ratio of electrocardiogram, in decibels (dB), reflects the power ratio of the effective component of the electrocardiogram signal to the background noise; : The standard formula coefficient for converting power ratio to decibels (dB); : ECG signal amplitude at sampling point k (unit: mV); Sampling point k belongs to the set of QRS complex time periods in the electrocardiogram cycle; Sampling point k belongs to the set of equipotential segments (baseline segments); The square of the signal amplitude at sampling point k represents the instantaneous power.

[0026] The perfusion index is obtained by normalizing the AC and DC components of the PPG signal during each cardiac cycle. :

[0027] In the formula, Perfusion index, expressed as a percentage (%), reflects the intensity of peripheral blood flow perfusion; The amplitude of the AC component of the PPG signal within one heartbeat cycle (the difference between the peak and trough values, in mV or relative amplitude). This represents the DC component (average value of the trough points) of the PPG signal within the same heartbeat cycle.

[0028] To eliminate the influence of respiratory modulation, the chest impedance signal Z(t) is first passed through a 0.1–0.3 Hz bandpass filter to extract the respiratory baseline component Z_resp(t), and then this component is removed from the original signal to obtain the net impedance signal:

[0029] Within a sliding time window (e.g., 5 seconds), calculate its standard deviation or variance:

[0030] The amplitude of the original chest impedance signal at time t (unit: Ω); From Extracted respiratory components (obtained through a 0.1–0.3 Hz bandpass filter); The signal is the chest impedance signal, after removing the influence of respiration; N is the total number of sampling points when calculating the variance (e.g., 1280 points / 5 seconds); i is the index number of the sampling point (i=1,2,…,N); This represents the average value of the net chest impedance signal within the calculation window; The variance of the net chest impedance signal (unit: Ω²) reflects electrode contact and signal stability.

[0031] After normalizing the above three indicators, a quality assessment vector Q is constructed:

[0032] This vector is updated every 5 seconds and is used to assess the overall quality of the current signal segment.

[0033] When the qualification criteria are met >15dB: ECG signal quality is acceptable. PI>1.5%: Photoelectric signal perfusion quality is acceptable. <0.2 Ω: The chest impedance signal is in a stable contact state. If any index in the Q vector does not meet the above threshold, it is marked as a low-quality segment and enters the signal reconstruction process.

[0034] Furthermore, a normalized score is calculated based on the quality vector and compared with a personalized dynamic threshold. When the score is lower than the set threshold and the motion intensity exceeds the preset level, the signal reconstruction process is triggered. In extreme motion states, the segment is directly discarded. If key pathological features are detected, reconstruction is forcibly initiated even if the score is close to the threshold. The dynamic threshold is automatically generated by the system based on data collected from the user at rest, and is updated daily to the 25th percentile value of the score distribution. It is coupled with real-time motion data to form a dual-condition judgment mechanism.

[0035] The overall quality score for vector Q is calculated using a weighted normalization method. The normalization function is Sigmoid, and the weighted scoring formula is as follows:

[0036] Among them, weight =0.5, =0.3, =0.2; satisfies .

[0037] The intensity of the exercise is based on the root mean square value of the acceleration. It is determined that the joint overall quality score adopts a dual-threshold trigger mechanism: overall quality score Below the 25th percentile of the rating distribution and Signal reconstruction is triggered when the g exceeds 1.5g.

[0038] In addition, if If the amount is greater than 3.0g, it can be considered that the fragment should be discarded directly under extreme exercise conditions; If a ST segment voltage drop >0.1mV is detected, even if Approaching the threshold also forces reconstruction; ST segment depression >0.1mV (i.e. >1mm) is considered an early sign of possible myocardial ischemia in clinical electrocardiogram diagnostic criteria; even if the amplitude change is not large, it may correspond to the early stage of subepicardial ischemia, coronary artery stenosis or insufficient blood supply; if early ischemia is treated in time, there is a chance to avoid myocardial necrosis, so it is necessary to preserve and analyze signal details as much as possible and force the reconstruction process to restore the waveform authenticity to the greatest extent.

[0039] If ST segment depression is >0.2mV, skip remodeling and directly trigger a Level III warning. ST segment depression >0.2mV is an important diagnostic criterion for severe cardiac emergencies; this ECG morphological change should immediately trigger a Level III warning without waiting for further verification.

[0040] S3. The multi-channel signal block containing ECG and its synchronous photoplethysmogram and chest impedance, which is judged to be of low quality by the quality assessment matrix, is input into a generative adversarial network with electrophysiological constraints for reconstruction, and the reconstructed high-quality ECG signal is output.

[0041] The low-quality signal segments are input into the adversarial network with a 5-second time window. The input signal is a 1280×3 matrix, and the input signals are electrocardiogram (ECG), photoplethysmography (PPG) and chest impedance (Z) signals, respectively. The generative adversarial network (GAN) employs an improved U-Net structure. The encoder includes multi-scale one-dimensional convolutional downsampling units and channel attention modules, while the bottleneck layer contains bidirectional long short-term memory (Bi-LSTM) units. The decoder includes upsampling and skip-connection fusion units. A physiological constraint layer is set at the decoder output, taking the reconstructed ECG signal as input and calculating the partial differential equation residuals based on a simplified action potential model, as well as medical rule penalty terms based on the PR interval, QRS duration, T-wave asymmetry, and ST-segment shift. The discriminator of the GAN is a one-dimensional PatchGAN structure, containing multi-layer one-dimensional convolutions and LeakyReLU activation, outputting local true / false discrimination scores. The GAN is trained end-to-end using a total loss function comprising adversarial loss, L1 reconstruction loss, ST-segment cyclic consistency loss, and physiological constraint loss, with the loss weights adjusted at different training stages. After training, the GAN is used to infer the input multi-channel signal, outputting only the reconstructed high-quality ECG waveform.

[0042] Specifically, the residual of the partial differential equation is:

[0043] Where V is the transmembrane potential, representing the instantaneous potential difference inside and outside the cardiomyocyte. It is not a sampled signal, but only serves as a physiological boundary condition for the feasible region of the generated signal; D is the diffusion coefficient, which controls the spatial smoothness of the electrical signal in the tissue and avoids physiologically inconsistent abrupt changes in the reconstructed waveform at adjacent sampling points. The membrane capacitance affects the rate of potential change, which in turn determines the duration of the ST segment and the morphology of the T wave. The sodium ion flow determines the rapid rise phase of the QRS wave; The flow of calcium ions determines the morphology of the plateau phase (ST segment); Potassium ion flow controls the repolarization process (T wave).

[0044] Specifically, the total loss function is:

[0045] In the formula, To combat the loss, Wasserstein distance optimization is used to ensure that the overall distribution of the reconstructed signal closely approximates the real signal.

[0046] To reconstruct the loss and ensure accurate recovery of the time-domain waveform. , This is the reconstructed signal of the low-quality signal x by the generator; This is the corresponding high-quality reference signal; Let L1 be the norm, representing the sum of the absolute values ​​of the errors at all sampling points.

[0047] For cycle consistency loss, Focusing on ST segment consistency, it is defined as:

[0048] The constraints are: .

[0049] Where T is the total duration of the integration interval (80 milliseconds). The end position of the displacement QRS composite wave (start of ST segment); , These are the amplitudes of the reconstructed signal and the reference signal at time t, respectively; 0.02mV is the maximum allowable error for the ST segment.

[0050] To mitigate physiological constraints and ensure that the generated signal conforms to medical electrophysiological principles, three time-domain constraints are included, specifically:

[0051] These are medical time-domain constraints for PR interval, QRS width, and T-wave asymmetry, respectively.

[0052] : These are the weighting coefficients for each loss component, used to adjust the relative importance of different loss terms in the total loss function.

[0053] Weights that counteract loss are used to control the degree of optimization the generator makes in terms of realism.

[0054] The weights of the reconstruction loss are used to control the consistency of the generated signal with the target "clean" signal in the time domain amplitude. : The weight of the cycle consistency loss, used to constrain the temporal morphology preservation of the ST-T waveform under medical standards.

[0055] The weights of the physiological constraint loss are used to ensure that the generated signal conforms to the physiological boundary conditions and dynamic model of myocardial electrical activity.

[0056] in , , , The value is dynamically adjusted based on the real-time quality vector Q and the pathological risk level, especially in the early stages of training. Generally equal to 0, Smaller, improves in the later stages of training , To enhance the constraints of medical morphology.

[0057] S4. Extract key indicators of myocardial ischemia from the reconstructed signal and conduct multi-level risk warning.

[0058] Key indicators of myocardial ischemia were extracted from the reconstructed signal, including the ST segment depression index, which was extracted by analyzing the difference between the average ST segment amplitude and the isoelectric segment based on R-wave localization. The ST segment depression index was calculated by detecting the 80ms segment following each R-wave position in the electrocardiogram (ECG) as the ST segment and calculating the average potential of this segment. Average potential of the equipotential section The difference:

[0059] The average potential of this segment refers to the arithmetic mean of the potential values ​​of the sampling points within the target analysis segment (including but not limited to the ST segment, QRS segment, and T wave segment) determined based on the feature point detection results in the electrocardiogram signal. The average potential of the isoelectric segment is also included. The isoelectric average value is obtained by arithmetically averaging the potential values ​​of sampling points within the same cardiac cycle, selecting an isoelectric reference segment (including but not limited to the PR segment or TP segment) with no obvious depolarization or repolarization activity.

[0060] The peak value of the time derivative of the impedance signal reflects an abrupt change in pleural volume or perfusion; the first-order time derivative of the pleural impedance signal (Z) is calculated, and the extreme value of the derivative is extracted in each respiratory cycle. As a perfusion response index for measuring sudden changes in thoracic impedance.

[0061] The rise time index (RTI) of photoplethysmography (PPG) is extracted. The ratio of the rise time of each pulse to the total length of the cycle is used as an exponential form to calculate the ratio of the rise time of the photoplethysmography to the length of the cycle. Based on the above indicators, a three-tiered ischemia early warning mechanism is established: When ΔST < -0.1mV and lasts for more than 60 seconds, or the peak value of the Z derivative is abnormal, or RTI > 0.5, it is judged as a Level I warning; When the ST segment continues to be depressed by more than 0.1mV and the RTI increases by ≥30% simultaneously, a Level II warning is triggered. When the ST segment depression is >0.2mV and the ST segment morphology meets the Minnesota I type, or when the Z and RTI synchronization is severely abnormal, and both are >0.6, a Level III warning is immediately triggered.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 method for real-time multimodal physiological signal analysis in myocardial ischemia, characterized in that, include: Simultaneously acquire electrocardiogram, photoplethysmography pulse wave and chest impedance signals, and use a triaxial accelerometer to compensate for the phase difference caused by motion displacement in real time; A multidimensional quality assessment matrix was constructed, which included ECG signal-to-noise ratio, photoplethysmography perfusion index, and chest impedance variance. Based on dynamic threshold and accelerometer data, it was determined whether signal reconstruction was triggered. Multichannel signal blocks containing electrocardiograms and their synchronized photoplethysmography pulse waves and chest impedance, which are judged to be of low quality by the quality assessment matrix, are input into a generative adversarial network with electrophysiological constraints for reconstruction, and the reconstructed high-quality electrocardiogram signal is output. Key indicators of myocardial ischemia are extracted from the reconstructed signals to provide multi-level risk warnings.

2. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 1, characterized in that, The thoracic impedance signal was acquired between the manubrium of the sternum and the xiphoid process using the four-electrode method, the photoplethysmography pulse wave signal was acquired at the radial artery of the left wrist, and the electrocardiogram signal was acquired under the standard II lead configuration; the three-axis accelerometer was placed in front of the chest to monitor the X, Y, and Z axis motion status at a frequency of not less than 100Hz.

3. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 1, characterized in that, When the triaxial accelerometer detects a displacement velocity exceeding a set threshold in any direction, the phase difference calibration process is initiated, specifically as follows: Using the chest impedance signal as the reference time axis, calculate the time difference between the peak value of the R wave on the electrocardiogram and the systolic peak of the PPG waveform. Optimize by maximizing mutual information The value is optimized using the principle of maximizing mutual information. The values ​​specifically include: Set time difference The search range and step size; For each candidate Shift the ECG signal ; The mutual information values ​​of the ECG and PPG signals after translation within the same time window were calculated. The mutual information was calculated using the histogram method to estimate the probability distribution. Traverse all Then, select the option that maximizes the mutual information value. As the optimal time difference: Calculate the maximum mutual information value. If the maximum mutual information value is less than the set threshold, the signal quality is determined to be insufficient, the calibration is abandoned, and the segment is marked as low quality; otherwise, the phase calibration is completed.

4. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 3, characterized in that, When the triaxial accelerometer detects that the angular velocity of the wearer's torso in any direction exceeds a set threshold, a motion artifact compensation filter is activated to suppress interference in the three channels of electrocardiogram, photoplethysmography pulse wave, and chest impedance. Specifically, this includes: The filter is a first-order low-pass filter, and its time constant is dynamically adjusted according to the root mean square value output by the triaxial accelerometer. The time constant τ is calculated by multiplying the root mean square value of acceleration by a preset coefficient. The filter is converted into a digital IIR filter form through bilinear transformation, which supports filtering on edge computing platforms with a delay of no more than 50ms. Different filtering targets are used for different signal channels: High-frequency artifacts above 50Hz are filtered out from the electrocardiogram signal to suppress electromyographic interference and maintain the clarity of the main QRS complex. The baseline drift below 0.5 Hz of the photoplethysmography (PPG) signal is suppressed while preserving the rising pulse waveform. The chest impedance signal is smoothed to prevent abrupt changes while preserving the range of respiratory waveform variations. The filter parameters are updated every 200ms.

5. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 4, characterized in that, The multidimensional quality assessment matrix consists of three components: ECG signal-to-noise ratio, photoplethysmography perfusion index, and chest impedance variance. The signal-to-noise ratio of an electrocardiogram (ECG) is obtained by the power ratio of the QRS segment to the isoelectric segment. The photoplethysmography perfusion index is derived from the ratio of the AC component to the DC component of the photoelectric signal. The variance of chest impedance was calculated after removing respiratory modulation.

6. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 5, characterized in that, A normalized score is calculated based on the quality vector and compared with a personalized dynamic threshold. When the score is lower than the set threshold and the motion intensity exceeds the preset level, the signal reconstruction process is triggered. The segment is discarded directly under extreme motion conditions. If key pathological features are detected, reconstruction is forcibly started even if the score is close to the threshold. The dynamic threshold is automatically generated by the system based on data collected from the user at rest, and is updated daily to the 25th percentile value of the score distribution. It is coupled with real-time motion data to form a dual-condition judgment mechanism.

7. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 1, characterized in that, The generative adversarial network is trained end-to-end using a total loss function that includes adversarial loss, L1 reconstruction loss, ST segment cycle consistency loss, and physiological constraint loss.

8. The real-time multimodal physiological signal analysis method for myocardial ischemia according to claim 1, characterized in that, Key indicators of myocardial ischemia were extracted from the reconstructed signal, including the ST segment depression index, which was extracted from the ST segment depression index based on the difference between the average ST segment amplitude and the isoelectric segment based on R-wave localization analysis. The peak value of the time derivative of the impedance signal reflects an indicator of abrupt changes in pleural volume or perfusion. The ratio of the rise time to the period length of the photoelectric pulse wave is calculated as the rise time index. Based on the above indicators, a three-tiered ischemia early warning mechanism is established: When ΔST < -0.1mV and lasts for more than 60 seconds, or the peak value of the Z derivative is abnormal, or RTI > 0.5, it is judged as a Level I warning; When the ST segment continues to be depressed by more than 0.1mV and the RTI increases by ≥30% simultaneously, a Level II warning is triggered. When the ST segment depression is >0.2mV and the ST segment morphology meets the Minnesota I type, or when the Z and RTI synchronization is severely abnormal, and both are >0.6, a Level III warning is immediately triggered.

Images