Method for evaluating mechanical efficiency of heart based on strain-blood flow coupling model

Abstract

The application relates to the technical field of medical artificial intelligence, and particularly discloses a heart mechanical efficiency evaluation method based on a strain-blood flow coupling model, which comprises the following steps: synchronously collecting electrocardio, heart sound and photoelectric volume pulse wave signals, adopting master-slave clock synchronization and R wave triggering alignment to realize multi-modal signal timing calibration; extracting myocardial strain rate, stroke volume and systolic energy consumption coefficient based on the calibrated signals, and constructing a strain-blood flow coupling efficiency index; dynamically calibrating the coupling efficiency index through a double-parameter self-calibration mechanism of heart rate drift and heart rate change rate; and finally combining multi-scale feature extraction and a gradient boosting decision tree model to output a heart mechanical efficiency grade and an ischemia risk probability; the application solves the problems of missing dynamic load evaluation, isolated multi-modal parameters and insufficient individual difference adaptability in the prior art, and realizes early myocardial ischemia detection sensitivity and wearable device-level real-time monitoring capability.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence technology, specifically to a method for evaluating cardiac mechanical efficiency based on a strain-blood flow coupling model. Background Technology

[0002] Current assessments of cardiac mechanical efficiency primarily rely on static physiological parameters and single-modal signal analysis. In clinical practice, echocardiography and static electrocardiogram (ECG) monitoring are commonly used to obtain cardiac function indicators. Existing technologies largely depend on population statistical thresholds and fixed calibration modes, and in the field of wearable devices, they have initially achieved the acquisition of ECG, photoplethysmography (PPG), and pulse wave signals, as well as the calculation of basic parameters. With the development of mobile healthcare technology, cardiac function assessment based on multimodal signal fusion has gradually become a research hotspot. Related methods attempt to enhance the assessment dimensions by combining ECG and heart sound signals, but a complete assessment system has not yet been formed in practical applications.

[0003] The existing technology has the following shortcomings: Traditional static assessment schemes lack multimodal temporal synchronization mechanisms and strain-blood flow coupling models, making it impossible to accurately capture early myocardial ischemia characteristics under dynamic scenarios such as exercise and emotional fluctuations. Existing cardiac mechanical efficiency assessment methods suffer from core technical problems such as insufficient sensitivity under dynamic load conditions and inability to adapt to individual physiological differences. Summary of the Invention

[0004] The purpose of this invention is to provide a method for evaluating cardiac mechanical efficiency based on a strain-blood flow coupling model, in order to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: The method for evaluating cardiac mechanical efficiency based on a strain-blood flow coupling model includes the following steps: S1: Synchronously acquire the subject's electrocardiogram signal, heart sound signal, and photoplethysmography pulse wave signal, construct a multimodal signal, perform time-series calibration on the multimodal signal, and output the time-synchronized multimodal signal sequence; S2: Process the multimodal signal sequence after time synchronization to extract myocardial strain rate, stroke volume and systolic energy consumption coefficient; S3: Calculate the strain-blood flow coupling efficiency index based on myocardial strain rate, stroke volume and systolic energy consumption coefficient, and output the uncalibrated strain-blood flow coupling efficiency index; S4: Perform real-time dynamic calibration of the uncalibrated strain-blood flow coupling efficiency index by measuring heart rate drift and heart rate change rate, and output the calibrated strain-blood flow coupling efficiency index. S5: Extract the multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the photoplethysmography pulse wave signal, input them into the pre-trained strain-flow coupling model, and output the cardiac mechanical efficiency level and ischemia risk probability.

[0006] As a further aspect of the present invention: the extraction process of the myocardial strain rate is as follows: Based on the time-synchronized electrocardiogram (ECG) and heart sound signals, the starting point of myocardial electrical activity is confirmed by detecting whether the slope of the rising edge of the R wave in the ECG exceeds a preset threshold. Wavelet transform is used to analyze the dominant frequency distribution of the mitral and aortic valve closure sounds to locate the start and end points of cardiac systole and output the systolic duration. The change in myocardial displacement is calculated based on the maximum amplitude of the heart sound vibration during systole, and the myocardial strain rate is obtained by combining the systolic duration.

[0007] As a further aspect of the present invention: the process for extracting the stroke volume is as follows: Based on the time-synchronized photoplethysmography (PPG) signal, signal segments are aligned by combining the systolic duration. The pulse wave conduction time, pulse wave amplitude, and pulse wave attenuation coefficient are extracted. The pulse wave attenuation coefficient is determined by the ratio of the amplitude to the peak value at 1 / 2 diastolic position of the PPG signal. Individual age and systolic blood pressure parameters are integrated, and the stroke volume is calculated through a predefined vascular elasticity relationship.

[0008] As a further aspect of the present invention: the process for obtaining the energy consumption coefficient during the contraction period is as follows: The R-wave amplitude of the electrocardiogram signal is dynamically compressed, and the S1 intensity of the heart sound signal is logarithmically transformed to obtain the preprocessed R-wave amplitude value and S1 intensity value. The preprocessed R-wave amplitude value and S1 intensity value are directly multiplied and then introduced into a preset normalization factor for dimensional unification to obtain the contraction period energy consumption coefficient.

[0009] As a further aspect of the present invention: the output uncalibrated strain-blood flow coupling efficiency index specifically includes: Multiplying the myocardial strain rate directly by the stroke volume yields a preliminary coupling value between myocardial work and blood flow output. The initial coupling values ​​were normalized for physiological energy consumption by using the energy consumption coefficient during systole. After the normalized values ​​are subjected to a preset physiological rationality check, the uncalibrated strain-blood flow coupling efficiency index is output.

[0010] As a further aspect of the present invention: the physiological energy consumption normalization processing of the preliminary coupling value through the systolic energy consumption coefficient specifically includes: A dynamically sensed normalized benchmark is constructed based on the energy consumption coefficient during the contraction period. The normalization intensity is adjusted by comparing the difference between the current energy consumption coefficient during the contraction period and the preset standard energy consumption value in real time. Input the initial coupling value into the mapping table constructed based on clinical data, and select the corresponding normalization curve according to the numerical range of the systolic energy consumption coefficient; The initial coupling values ​​are subjected to a nonlinear transformation based on the selected normalization curve; The normalized values ​​are physiologically reasonable, and a recalculation process is initiated when an abnormal value is detected.

[0011] As a further aspect of the present invention: the output calibrated strain-blood flow coupling efficiency index specifically includes: Based on a pre-established individual heart rate-efficiency relationship curve, the calibration baseline value is dynamically adjusted according to the current heart rate drift. Simultaneously analyze the interaction between the rate of heart rate change and the amount of heart rate drift, and activate the enhanced calibration mode when the rate of heart rate change exceeds a preset threshold. Based on the output of the enhanced calibration mode, the uncalibrated strain-blood flow coupling efficiency index is dynamically compensated and calculated. The physiological range of the calibrated strain-blood flow coupling efficiency index is verified. When the value exceeds the preset normal physiological range, a correction calculation based on historical data is initiated.

[0012] As a further aspect of the present invention: S5 specifically includes: The multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the photoplethysmography pulse wave signal are obtained to construct a comprehensive evaluation feature vector, which is used as the input of the strain-flow coupling model. The training objective of the strain-flow coupling model is to minimize the difference between the predicted cardiac mechanical efficiency level and the actual clinical diagnosis result. The strain-flow coupling model is trained, and the cardiac mechanical efficiency level and ischemic risk probability are output based on the trained strain-flow coupling model. The strain-flow coupling model is a gradient boosting decision tree model.

[0013] As a further aspect of the present invention: the training process of the strain-blood flow coupling model is as follows: Historical multi-scale features and auxiliary features from photoplethysmography (PPG) signals were collected, and combined with clinically diagnosed cardiac mechanical efficiency levels and ischemic risk probabilities as target labels to construct a historical dataset. The entire dataset was divided into training and validation sets according to a predetermined ratio for training and performance evaluation of the strain-flow coupling model. A gradient boosting decision tree algorithm was used for modeling, with the number of trees, learning rate, and maximum depth set as hyperparameters. During training, each decision tree gradually reduced the prediction error of the previous model using gradient descent, and the optimal split point was selected using second-order derivative information. The prediction results of all decision trees were integrated through a weighted summation to obtain the final result. The final predicted values ​​of cardiac mechanical efficiency and ischemic risk probability are calculated. After each training iteration, the cross-entropy loss between the predicted results and the clinical diagnosis results is calculated as a model performance evaluation index. SHAP feature importance analysis and Bayesian hyperparameter optimization strategy are introduced to automatically tune the model. When the loss function on the validation set tends to stabilize and reach its minimum value, the strain-blood flow coupling model training is completed. The trained gradient boosting decision tree model is deployed to the evaluation system, which receives the comprehensive evaluation feature vector composed of multi-scale features and auxiliary features in real time, outputs the corresponding cardiac mechanical efficiency level and ischemic risk probability, and prevents overfitting through a dynamic early stop mechanism.

[0014] As a further aspect of the present invention, the specific process of constructing the comprehensive evaluation feature vector is as follows: The calibrated strain-blood flow coupling efficiency index was subjected to time-domain statistical analysis and frequency-domain transformation, and its time-domain variation characteristics and main frequency energy distribution characteristics were extracted simultaneously. Based on the strain-blood flow coupling efficiency index sequence within the sliding window, its sample entropy and trend change rate are calculated to quantify the nonlinear dynamic characteristics of the signal. The pulse wave propagation time variation rate of the photoplethysmography pulse wave signal is synchronized and aligned with the dynamic features to form a multidimensional feature set. The multidimensional feature set is subjected to dimensional unification and standardization, and key features are selected by pre-defined feature importance ranking to generate the final comprehensive evaluation feature vector.

[0015] The beneficial effects of this invention are: (1) By using multimodal signal temporal synchronization and strain-blood flow coupling modeling, the problem of insufficient adaptability of traditional static assessment methods in dynamic scenarios such as motion and emotional fluctuations is solved. This method establishes a quantitative correlation between myocardial mechanical motion and blood flow output, and can capture the myocardial ischemia characteristics that appear under load when the coronary artery stenosis is less than 50%, effectively reducing the missed detection rate of early lesions.

[0016] (2) A two-parameter self-calibration mechanism based on heart rate drift and rate of change was adopted, combined with an individualized baseline establishment method, to eliminate the interference of individual factors such as age, blood pressure, and vascular elasticity on the assessment indicators. The individual heart rate-efficiency relationship curve established by piecewise cubic polynomial fitting was used to achieve accurate calibration under different heart rate ranges, thereby improving the reliability of the assessment results of the monitored population. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, this invention is a method for evaluating cardiac mechanical efficiency based on a strain-blood flow coupling model, comprising the following steps: S1: Synchronously acquire the subject's electrocardiogram signal, heart sound signal, and photoplethysmography pulse wave signal, construct a multimodal signal, perform time-series calibration on the multimodal signal, and output the time-synchronized multimodal signal sequence; S2: Process the multimodal signal sequence after time synchronization to extract myocardial strain rate, stroke volume and systolic energy consumption coefficient; S3: Calculate the strain-blood flow coupling efficiency index based on myocardial strain rate, stroke volume and systolic energy consumption coefficient, and output the uncalibrated strain-blood flow coupling efficiency index; S4: Perform real-time dynamic calibration of the uncalibrated strain-blood flow coupling efficiency index by measuring heart rate drift and heart rate change rate, and output the calibrated strain-blood flow coupling efficiency index. S5: Extract the multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the photoplethysmography pulse wave signal, input them into the pre-trained strain-flow coupling model, and output the cardiac mechanical efficiency level and ischemia risk probability.

[0021] In S1, the subject's electrocardiogram (ECG), heart sound, and photoplethysmography (PPG) pulse wave signals are simultaneously acquired and constructed into a multimodal signal. The multimodal signal is then time-synchronized, and a time-synchronized multimodal signal sequence is output, specifically including: The signal acquisition process involves using multiple electrode pads placed on the subject's chest to collect electrocardiogram (ECG) signals. These electrodes, made of medical-grade conductive gel, ensure good skin contact and record real-time potential changes caused by myocardial electrical activity. Simultaneously, a heart sound sensor, employing a high-sensitivity accelerometer, is placed at the fourth intercostal space along the left sternal border to directly collect heart sound signals generated by the opening and closing of heart valves through direct contact with the chest wall. A photoelectric sensor, comprising a light-emitting diode and a photosensitive receiver, is installed at the radial artery in the subject's wrist to acquire photoplethysmography (PPG) signals by detecting changes in blood volume in the subcutaneous capillary bed.

[0022] The timing calibration process first employs a master-slave clock synchronization method, setting the ECG signal acquisition unit as the master clock source and the heart sound signal acquisition unit and photoplethysmography (PPG) wave acquisition unit as slave clocks. Initial synchronization of the three signals is achieved through hardware-level clock signal distribution. Subsequently, precise alignment is performed at the software level, using the R-wave peak point in the ECG signal as the time reference point, determined by detecting the rising edge slope of the signal. Based on this reference point, the time offsets of the mitral valve closure sound and the aortic valve closure sound in the heart sound signal are calculated, along with the time offset of the pulse wave peak point in the PPG wave signal.

[0023] The timing correction step calculates the time deviation of each signal channel relative to the ECG reference using cross-correlation analysis, compensating for the time shift of valve closure sounds in the heart sound signal and pulse wave characteristic points in the photoplethysmography (PPG). The compensated multimodal signals are synchronized on the time axis, forming a signal sequence with a unified timestamp. Each time point in this sequence contains data in three dimensions: the amplitude of the synchronously acquired ECG signal, the intensity of the heart sound signal, and the intensity of the PPG signal.

[0024] The signal output step arranges the time-calibrated multimodal signals into a continuous sequence in chronological order, maintaining the correspondence between the three signals in physiological events. The data packet at each sampling moment contains ECG waveform segments, heart sound waveform segments, and photoplethysmography (PPG) waveform segments. These data packets are transmitted to subsequent processing stages in chronological order, providing time-consistent multimodal input data for subsequent parameter extraction.

[0025] In S2, the multimodal signal sequence after time synchronization is processed to extract myocardial strain rate, stroke volume, and systolic energy consumption coefficient, specifically including: First, myocardial strain rate is extracted. Based on the time-synchronized ECG and heart sound signals, the onset of myocardial electrical activity is confirmed by detecting whether the slope of the rising edge of the R wave in the ECG exceeds a preset threshold of 0.5 mV / ms. Wavelet transform analysis of the heart sound signals using db4 wavelets with 5-level decomposition is employed to identify the dominant peak frequency of the mitral valve closure sound within the range of 100–200 Hz and the aortic valve closure sound within the range of 80–150 Hz. This is used to locate the start and end points of cardiac systole and output the systolic duration. The change in myocardial displacement is calculated based on the maximum amplitude of the heart sound vibration during systole. Combined with the systolic duration, the myocardial strain rate is obtained. The specific calculation formula is: Myocardial strain rate = Change in myocardial displacement / Systolic duration, where the change in myocardial displacement = 0.6 × systolic amplitude of the heart sound signal + 0.4 × R-wave amplitude of the ECG signal.

[0026] Next, stroke volume is extracted. Based on the time-synchronized photoplethysmography (PPG) signal, signal segments are aligned using the systolic duration obtained in the previous steps, and pulse wave conduction time, pulse amplitude, and pulse wave attenuation coefficient are extracted. The pulse wave attenuation coefficient is determined by calculating the ratio of the amplitude to the peak value at the midpoint of the diastolic phase of the PPG signal. Integrating individual age and systolic blood pressure parameters, stroke volume is calculated using a predefined vascular elasticity relationship. The specific calculation formula is: Stroke volume = -0.25 × pulse wave conduction time + 18.6 × pulse amplitude - 12.3 × pulse wave attenuation coefficient + 85.2.

[0027] Then, the systolic energy consumption coefficient was obtained. Dynamic range compression was performed on the R-wave amplitude of the ECG signal, and a base-10 logarithmic transformation was applied to the S1 intensity of the heart sound signal to obtain preprocessed R-wave amplitude and S1 intensity values. The preprocessed R-wave amplitude and S1 intensity values ​​were directly multiplied and then normalized using a normalization factor obtained from clinical data regression to achieve dimensional uniformity, thus obtaining the systolic energy consumption coefficient. The normalization factor, obtained through clinical data regression analysis, can accommodate the intensity differences of ECG and heart sound signals among different individuals. The specific calculation formula is: Systolic energy consumption coefficient = (Dynamically range compressed R-wave amplitude × Logarithmically transformed S1 intensity) / Normalization factor, where dynamic range compression uses a linear transformation to map the original R-wave amplitude value to the 0-1 range, and the logarithmic transformation uses a base-10 logarithmic operation.

[0028] Electrocardiogram (ECG) signals were acquired using multiple electrode pads placed on the subject's chest. These pads were made of medical-grade conductive gel to ensure good skin contact. Heart sound signals were acquired using a high-sensitivity accelerometer placed at the fourth intercostal space along the left sternal border. This sensor directly contacts the chest wall to record the sound signals generated by the opening and closing of the heart valves. Photoplethysmography (PPG) signals were acquired using a photoelectric sensor installed at the radial artery in the subject's wrist. This sensor, comprising a light-emitting diode and a photosensitive receiver, acquires the pulse wave signal by detecting changes in blood volume in the subcutaneous capillary bed.

[0029] For wavelet transform analysis, the db4 wavelet basis function is used in a 5-level decomposition. Extreme points in the characteristic frequency range of the corresponding heart sound signal are found in the detail coefficients of each level. These extreme points are used to pinpoint the precise timing of the mitral and aortic closure sounds. Dynamic range compression employs linear normalization, dividing the original R-wave amplitude by the maximum R-wave amplitude at rest for the subject, compressing the value range to 0-1. Logarithmic transformation uses a base-10 logarithmic operation to convert the S1 intensity value of the heart sound signal to a decibel value. The calculation formula is: S1 intensity decibel value = 20 × log10 (original S1 intensity value / reference intensity value), where the reference intensity value is the average S1 intensity of the normal population.

[0030] In S3, the strain-blood flow coupling efficiency index is calculated based on myocardial strain rate, stroke volume, and systolic energy consumption coefficient. The uncalibrated strain-blood flow coupling efficiency index is output, specifically including: First, a preliminary coupling value is calculated by directly multiplying the myocardial strain rate by the stroke volume to obtain a preliminary coupling value that reflects the degree of matching between myocardial work and blood flow output.

[0031] Physiological energy consumption normalization is performed, which includes the following specific steps: A dynamically sensed normalization benchmark is constructed based on the systolic energy consumption coefficient. The normalization intensity is dynamically adjusted by comparing the current systolic energy consumption coefficient with a preset standard energy consumption value in real time. The preset standard energy consumption value, Estd, is derived from clinical data of healthy subjects and ranges from 20 to 35. When the difference between the current systolic energy consumption coefficient, Ec, and the standard energy consumption value exceeds 5 units (i.e., |Ec-Estd|>5), an enhanced normalization mode is activated.

[0032] The initial coupling values ​​are input into a mapping table constructed based on clinical data. A corresponding normalized curve is selected based on the numerical range of the systolic energy expenditure coefficient. The mapping table divides the systolic energy expenditure coefficient into three numerical ranges: 15-25, 25-35, and 35-45, each corresponding to different normalized curve parameters. A nonlinear transformation is then applied to the initial coupling values ​​based on the selected normalized curve. The transformation formula is as follows: ;in Represents the normalized value. Indicates the initial coupling value. Indicates the energy consumption coefficient during the contraction period; , , For the normalized curve parameters, the parameter values ​​corresponding to different numerical ranges are: When 15≤ <25 o'clock, =0.002, =0.15, =2.8; When 25≤ <35 hours, =0.0015, =0.12, =3.2; When 35≤ <45 hours, =0.001, =0.08, =3.6; Then, physiological rationality verification was performed, and the normalized values ​​were validated. A reasonable lower limit for the physiological range of the strain-blood flow coupling efficiency index was set. upper limit Outputs uncalibrated strain-blood flow coupling efficiency index. The rules for determining it are as follows: ; During the verification process, if the output values ​​for three consecutive calculation cycles exceed the reasonable range, the recalculation process is initiated, and the current ECG signal, heart sound signal, and photoplethysmography pulse wave signal are reacquired, and all the aforementioned calculation steps are repeated.

[0033] Finally, the uncalibrated strain-flow coupling efficiency index is output, reflecting the cardiac mechanical efficiency without considering the effects of heart rate variations. The output value is rounded to two decimal places and temporarily stored in a data buffer to provide input data for subsequent individualized self-calibration steps. Throughout the calculation process, all intermediate results are stored in real time. If any calculation step encounters an anomaly, the calculation can be restarted from the previous correct node, ensuring the reliability and continuity of the calculation process.

[0034] All parameters and thresholds used in this calculation process were derived from statistical analysis of clinical data. The initial coupling value was calculated using direct multiplication, reflecting the direct coupling relationship between myocardial strain and blood flow output. The mapping table used in the normalization process was constructed based on clinical data, covering a broad sample from healthy individuals to patients with cardiovascular disease. The upper and lower limits for physiological rationality verification were determined by statistically analyzing measurement data under different physiological states, covering the physiological variation range of most populations. The triggering conditions for the recalculation process were set based on stability requirements in clinical practice, ensuring reliable calculation results even with fluctuations in signal quality.

[0035] In S4, the uncalibrated strain-blood flow coupling efficiency index is dynamically calibrated in real time using heart rate drift and heart rate change rate, and the calibrated strain-blood flow coupling efficiency index is output, specifically including: Individual heart rate-efficiency relationship curves were established. Basal physiological signals were collected for 5 minutes at rest, during which heart rate fluctuations were required to be less than 5 beats / min. Individualized heart rate-efficiency relationship curves were established through three active heart rate perturbation tests: heart rate was adjusted to resting heart rate +10 beats / min, resting heart rate -10 beats / min, and resting heart rate +15 beats / min through stepping exercises, with each perturbation lasting 30 seconds. Heart rate values ​​and the corresponding strain-flow coupling efficiency index were recorded simultaneously. A piecewise cubic polynomial was used to fit the collected data to the curve, with a goodness of fit greater than 0.98. This curve established an individualized benchmark reference for the strain-flow coupling efficiency index at different heart rate levels.

[0036] Determine the dynamic calibration baseline value. Calculate the heart rate drift based on the difference between the current real-time heart rate and the resting heart rate, using the formula: Heart rate drift equals current heart rate minus resting heart rate. Based on this heart rate drift, look up the corresponding calibration baseline value on a pre-established individual heart rate-efficiency relationship curve. Simultaneously, calculate the heart rate variability rate, using the formula: Heart rate variability rate equals the change in heart rate per unit time, in beats per minute per second. When the heart rate variability rate exceeds a preset threshold of 2 beats per minute per second, activate the enhanced calibration mode.

[0037] Dynamic compensation calculations are performed. Based on the calibration baseline and the current heart rate variability, the uncalibrated strain-flow coupling efficiency index is dynamically compensated. The compensation formula is: the calibrated strain-flow coupling efficiency index equals the uncalibrated strain-flow coupling efficiency index multiplied by the compensation coefficient. The compensation coefficient is calculated as follows: the compensation coefficient equals 1 minus the individual calibration coefficient multiplied by the ratio of heart rate drift to resting heart rate, then minus 0.1 multiplied by the difference between the heart rate variability rate and 2 (when the heart rate variability rate is greater than 2) or 0 (when the heart rate variability rate is less than or equal to 2). The individual calibration coefficient is obtained through fitting three heart rate perturbation experiments, with a value ranging from 0.3 to 0.8.

[0038] Perform physiological range verification. The normal physiological range for the calibrated strain-flow coupling efficiency index is set to 0.5 to 8.0. When the calculated value is below the lower limit, the average of the five most recent valid historical data points is used as the output; when it is above the upper limit, the median of the three most recent valid historical data points is used as the output. If the output value exceeds the normal physiological range for three consecutive calculation cycles, the signal re-acquisition process is initiated to reacquire the current ECG signal, heart sound signal, and photoplethysmography (PPG) pulse wave signal, repeating the entire calculation process.

[0039] In S5, the multi-scale features of the calibrated strain-flow coupling efficiency index and auxiliary features of the photoplethysmography pulse wave signal are extracted and input into the pre-trained strain-flow coupling model. The output is the cardiac mechanical efficiency level and the ischemic risk probability, specifically including: The multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the photoplethysmography pulse wave signal are obtained to construct a comprehensive evaluation feature vector, which is used as the input of the strain-flow coupling model. The training objective of the strain-flow coupling model is to minimize the difference between the predicted cardiac mechanical efficiency level and the actual clinical diagnosis result. The strain-flow coupling model is trained, and the cardiac mechanical efficiency level and ischemic risk probability are output based on the trained strain-flow coupling model. The strain-flow coupling model is a gradient boosting decision tree model.

[0040] The training process of the strain-blood flow coupling model is as follows: Historical multi-scale features and auxiliary features from photoplethysmography (PPG) signals were collected, and combined with clinically diagnosed cardiac mechanical efficiency levels and ischemic risk probabilities as target labels to construct a historical dataset. The entire dataset was divided into training and validation sets according to a predetermined ratio for training and performance evaluation of the strain-flow coupling model. A gradient boosting decision tree algorithm was used for modeling, with the number of trees, learning rate, and maximum depth set as hyperparameters. During training, each decision tree gradually reduced the prediction error of the previous model using gradient descent, and the optimal split point was selected using second-order derivative information. The prediction results of all decision trees were integrated through a weighted summation to obtain the final result. The final predicted values ​​of cardiac mechanical efficiency and ischemic risk probability are calculated. After each training iteration, the cross-entropy loss between the predicted results and the clinical diagnosis results is calculated as a model performance evaluation index. SHAP feature importance analysis and Bayesian hyperparameter optimization strategy are introduced to automatically tune the model. When the loss function on the validation set tends to stabilize and reach its minimum value, the strain-blood flow coupling model training is completed. The trained gradient boosting decision tree model is deployed to the evaluation system, which receives the comprehensive evaluation feature vector composed of multi-scale features and auxiliary features in real time, outputs the corresponding cardiac mechanical efficiency level and ischemic risk probability, and prevents overfitting through a dynamic early stop mechanism.

[0041] In constructing the comprehensive evaluation feature vector, the time-domain and frequency-domain features of the calibrated strain-blood flow coupling efficiency index are first extracted. Time-domain statistical analysis calculates the mean and coefficient of variation of the index within a 30-second time window, where the coefficient of variation is equal to the ratio of the standard deviation to the mean. Frequency-domain transformation uses Fast Fourier Transform to analyze the signal energy distribution in the 0.01-0.1 Hz frequency band and calculates the dominant frequency energy proportion, i.e., the ratio of the energy in this frequency band to the total energy.

[0042] Next, the nonlinear dynamic characteristics are calculated. Based on the strain-blood flow coupling efficiency index sequence within the sliding window, the sample entropy value is calculated. This method quantifies signal complexity by statistically analyzing the probability of new patterns appearing within the window. Simultaneously, the trend change rate, i.e., the average change amplitude of the index per unit time, is calculated, and the slope value is obtained by analyzing the trend of data points within the window using linear regression.

[0043] Then, multi-feature synchronous integration is performed. The pulse wave conduction time variability of the photoplethysmography (PPG) signal is time-aligned with the aforementioned dynamic features. The PPG conduction time variability is equal to the ratio of the standard deviation to the mean of the pulse wave conduction time over 10 consecutive cardiac cycles. The aligned features constitute a multi-dimensional feature set including time-domain features, frequency-domain features, nonlinear features, and pulse wave features.

[0044] Finally, feature standardization and selection are performed. The multidimensional feature set is standardized using z-score normalization, ensuring that the mean of each feature is 0 and the standard deviation is 1. Based on a feature importance ranking table established in advance using clinical data, the top 6 features by importance are selected to form the final comprehensive evaluation feature vector. This feature vector serves as the input data for subsequent intelligent evaluation, ensuring that the evaluation process includes comprehensive physiological information while avoiding feature redundancy.

[0045] In the output step, the constructed comprehensive evaluation feature vector is first input into the pre-trained gradient boosting decision tree model. This model contains 100 decision trees, each with a maximum depth of 6 layers, and a learning rate of 0.1. The model receives a comprehensive evaluation feature vector containing time-domain features, frequency-domain features, nonlinear dynamic features, and pulse wave features as input.

[0046] During model processing, each decision tree sequentially splits the input features, assigning samples to different leaf nodes based on the comparison between feature values ​​and node thresholds. The predictions from each tree are integrated using a weighted summation method, where weights are obtained by minimizing the cross-entropy loss function during training. For judging cardiac mechanical efficiency levels, the model outputs an integer value between 0 and 3, corresponding to four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal. Specifically, the criteria are: a comprehensive score greater than or equal to 2.4 is considered normal; between 1.6 and 2.4 is mildly abnormal; between 0.8 and 1.6 is moderately abnormal; and less than 0.8 is severely abnormal.

[0047] For calculating the probability of ischemic risk, the model first outputs a raw probability value between 0 and 1, then converts it to a percentage form from 0% to 100% using a sigmoid function. During the probability calculation, the model pays particular attention to the sample entropy feature and the pulse wave transit time variation rate feature, with weight coefficients of 0.35 and 0.25 respectively, higher than the weight coefficients of other features. Finally, the model simultaneously outputs two assessment results: cardiac mechanical efficiency grade and ischemic risk probability, providing quantitative reference for clinical diagnosis.

[0048] Before outputting the results, the system performs a validity check. If there is a significant discrepancy between the output ischemic risk probability and the cardiac mechanical efficiency grade (e.g., the efficiency grade is normal but the risk probability is greater than 50%), the system will initiate a review mechanism and recalculate using historical data from the most recent 5 minutes. All output results are timestamped and linked to the original signal data to ensure the traceability of the assessment results.

[0049] The working principle of this invention is as follows: By synchronously acquiring electrocardiogram (ECG) signals, heart sound signals, and photoplethysmography (PPG) pulse wave signals, myocardial strain rate, stroke volume, and systolic energy consumption coefficient are extracted after time-series calibration. The strain-flow coupling efficiency index is calculated, and then real-time dynamic calibration is performed using heart rate drift and heart rate change rate. Finally, based on the multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the PPG pulse wave signal, a pre-trained gradient boosting decision tree model is used to output the cardiac mechanical efficiency level and ischemic risk probability, thus achieving a precise quantitative assessment of cardiac mechanical efficiency.

[0050] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for evaluating cardiac mechanical efficiency based on a strain-blood flow coupling model, characterized in that, Includes the following steps: S1: Synchronously acquire the subject's electrocardiogram signal, heart sound signal, and photoplethysmography pulse wave signal, construct a multimodal signal, perform time-series calibration on the multimodal signal, and output the time-synchronized multimodal signal sequence; S2: Process the multimodal signal sequence after time synchronization to extract myocardial strain rate, stroke volume and systolic energy consumption coefficient; S3: Calculate the strain-blood flow coupling efficiency index based on myocardial strain rate, stroke volume and systolic energy consumption coefficient, and output the uncalibrated strain-blood flow coupling efficiency index; S4: Perform real-time dynamic calibration of the uncalibrated strain-blood flow coupling efficiency index by measuring heart rate drift and heart rate change rate, and output the calibrated strain-blood flow coupling efficiency index. S5: Extract the multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the photoplethysmography pulse wave signal, input them into the pre-trained strain-flow coupling model, and output the cardiac mechanical efficiency level and ischemia risk probability.

2. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 1, characterized in that, The extraction process of the myocardial strain rate is as follows: Based on the time-synchronized electrocardiogram (ECG) and heart sound signals, the starting point of myocardial electrical activity is confirmed by detecting whether the slope of the rising edge of the R wave in the ECG exceeds a preset threshold. Wavelet transform is used to analyze the dominant frequency distribution of the mitral and aortic valve closure sounds to locate the start and end points of cardiac systole and output the systolic duration. The change in myocardial displacement is calculated based on the maximum amplitude of the heart sound vibration during systole, and the myocardial strain rate is obtained by combining the systolic duration.

3. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 1, characterized in that, The process for extracting the stroke volume is as follows: Based on the time-synchronized photoplethysmography (PPG) signal, signal segments are aligned by combining the systolic duration. The pulse wave conduction time, pulse wave amplitude, and pulse wave attenuation coefficient are extracted. The pulse wave attenuation coefficient is determined by the ratio of the amplitude to the peak value at 1 / 2 diastolic position of the PPG signal. Individual age and systolic blood pressure parameters are integrated, and the stroke volume is calculated through a predefined vascular elasticity relationship.

4. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 1, characterized in that, The process for obtaining the energy consumption coefficient during the contraction period is as follows: The R-wave amplitude of the electrocardiogram signal is dynamically compressed, and the S1 intensity of the heart sound signal is logarithmically transformed to obtain the preprocessed R-wave amplitude value and S1 intensity value. The preprocessed R-wave amplitude value and S1 intensity value are directly multiplied and then introduced into a preset normalization factor for dimensional unification to obtain the contraction period energy consumption coefficient.

5. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 1, characterized in that, The output uncalibrated strain-blood flow coupling efficiency index specifically includes: Multiplying the myocardial strain rate directly by the stroke volume yields a preliminary coupling value between myocardial work and blood flow output. The initial coupling values ​​were normalized for physiological energy consumption by using the energy consumption coefficient during systole. After the normalized values ​​are subjected to a preset physiological rationality check, the uncalibrated strain-blood flow coupling efficiency index is output.

6. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 5, characterized in that, The process of normalizing the initial coupling value using the systolic energy consumption coefficient specifically includes: A dynamically sensed normalized benchmark is constructed based on the energy consumption coefficient during the contraction period. The normalization intensity is adjusted by comparing the difference between the current energy consumption coefficient during the contraction period and the preset standard energy consumption value in real time. Input the initial coupling value into the mapping table constructed based on clinical data, and select the corresponding normalization curve according to the numerical range of the systolic energy consumption coefficient; The initial coupling values ​​are subjected to a nonlinear transformation based on the selected normalization curve; The normalized values ​​are physiologically reasonable, and a recalculation process is initiated when an abnormal value is detected.

7. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 1, characterized in that, The output-calibrated strain-blood flow coupling efficiency index specifically includes: Based on a pre-established individual heart rate-efficiency relationship curve, the calibration baseline value is dynamically adjusted according to the current heart rate drift. Simultaneously analyze the interaction between the rate of heart rate change and the amount of heart rate drift, and activate the enhanced calibration mode when the rate of heart rate change exceeds a preset threshold. Based on the output of the enhanced calibration mode, the uncalibrated strain-blood flow coupling efficiency index is dynamically compensated and calculated. The physiological range of the calibrated strain-blood flow coupling efficiency index is verified. When the value exceeds the preset normal physiological range, a correction calculation based on historical data is initiated.

8. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 1, characterized in that, S5 specifically includes: The multi-scale features of the calibrated strain-flow coupling efficiency index and the auxiliary features of the photoplethysmography pulse wave signal are obtained to construct a comprehensive evaluation feature vector, which is used as the input of the strain-flow coupling model. The training objective of the strain-flow coupling model is to minimize the difference between the predicted cardiac mechanical efficiency level and the actual clinical diagnosis result. The strain-flow coupling model is trained, and the cardiac mechanical efficiency level and ischemic risk probability are output based on the trained strain-flow coupling model. The strain-flow coupling model is a gradient boosting decision tree model.

9. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 8, characterized in that, The training process of the strain-blood flow coupling model is as follows: Historical multi-scale features and auxiliary features of photoplethysmography pulse wave signals were collected and combined with clinically diagnosed cardiac mechanical efficiency level and ischemic risk probability as target labels to construct a historical dataset. The entire dataset was divided into training set and validation set according to a preset ratio for training and performance evaluation of strain-blood flow coupling model. The gradient boosting decision tree algorithm is used for modeling, and the number of trees, learning rate, and maximum depth are set as hyperparameters. During training, each decision tree gradually reduces the prediction error of the previous model through gradient descent, and selects the optimal split point using second derivative information. The prediction results of all decision trees are integrated through weighted summation to obtain the final predicted values ​​of cardiac mechanical efficiency level and ischemic risk probability. After each training iteration, the cross-entropy loss between the prediction results and the clinical diagnosis results is calculated as a model performance evaluation index. SHAP feature importance analysis and Bayesian hyperparameter optimization strategy are introduced to automatically tune the model. When the loss function on the validation set tends to stabilize and reach its minimum value, the strain-blood flow coupling model training is completed. The trained gradient boosting decision tree model is deployed to the evaluation system, which receives the comprehensive evaluation feature vector composed of multi-scale features and auxiliary features in real time, outputs the corresponding cardiac mechanical efficiency level and ischemic risk probability, and prevents overfitting through a dynamic early stop mechanism.

10. The method for evaluating cardiac mechanical efficiency based on a strain-flow coupling model according to claim 8, characterized in that, The specific process for constructing the comprehensive evaluation feature vector is as follows: The calibrated strain-blood flow coupling efficiency index was subjected to time-domain statistical analysis and frequency-domain transformation, and its time-domain variation characteristics and main frequency energy distribution characteristics were extracted simultaneously. Based on the strain-blood flow coupling efficiency index sequence within the sliding window, its sample entropy and trend change rate are calculated to quantify the nonlinear dynamic characteristics of the signal. The pulse wave propagation time variation rate of the photoplethysmography pulse wave signal is synchronized and aligned with the dynamic features to form a multidimensional feature set. The multidimensional feature set is subjected to dimensional unification and standardization, and key features are selected by pre-defined feature importance ranking to generate the final comprehensive evaluation feature vector.

Images