High-fidelity heart sound simulation method, device, and storage medium

Abstract

The application discloses a high-fidelity heart sound simulation method, device and storage medium. The method comprises the following steps: determining a standard heart sound signal according to an original heart sound signal; determining pathological characteristic information corresponding to the standard heart sound signal according to the standard heart sound signal based on a standard pathological information template; determining an initial simulation heart sound signal according to the pathological characteristic information based on a preset model; determining first time-frequency information according to the initial simulation heart sound signal; and determining the initial simulation heart sound signal as a target simulation heart sound signal in the case that the first time-frequency information meets a preset condition. The application generates a simulation heart sound signal from the link modeling of physiological parameters, mechanical responses and acoustic outputs, can directly output a simulation result conforming to auscultation characteristics, improves the physiological authenticity and engineering usability, meanwhile, conforms to the mapping relationship between physiological parameters and heart sounds, makes the simulation result have strong explainability, and can also generate heart sound labeling data in a large scale, and solves the current situation of the scarcity of heart sound data.

Description

Technical Field

[0001] This application relates to a high-fidelity audio simulation method, electronic device, and computer-readable storage medium, belonging to the field of pathological feature simulation technology. Background Technology

[0002] The phonocardiogram (PCG) is the acoustic representation of the opening and closing of heart valves, ventricular vibrations, and mechanical vibrations generated by blood flow turbulence in the chest wall. The first heart sound (S1) is mainly associated with the closure of atrioventricular valves (such as the mitral valve), and the second heart sound (S2) is mainly associated with the closure of aortic / pulmonary valves (semilunar valves).

[0003] Currently, there are very few labeled datasets for heart sounds. The distribution of different pathological categories in known datasets is unbalanced, and they mostly come from different devices or environments. This results in insufficient publicly available / labeled data and heterogeneous distribution, making deep models in the field of heart sound recognition prone to overfitting. Consequently, it is difficult to achieve cross-device or cross-population generalization of heart sound data.

[0004] Currently, the heart sound simulation models in related technologies often use simplified 1DOF / 2DOF mechanical vibration models and use ideal pulses or simplified pressure differences as excitations. These models cannot reflect the real hemodynamic situation, lack physical models for generating sound pressure level heart sounds, and lack systematic generation of large-scale pathological variant data of tens of thousands of records that can be used for deep learning.

[0005] Furthermore, current time-frequency methods and deep models in related technologies lack stability under noise, sampling rate, and sensor differences. Fine-grained features of time-frequency maps (such as high-frequency components) can be distorted due to sampling rate, sensor frequency response, and environmental noise. They are also sensitive to differences in parameter selection, resulting in significant differences in the same pathological features on different devices. Summary of the Invention

[0006] This application discloses a high-fidelity audio simulation method, an electronic device, and a computer-readable storage medium.

[0007] The high-fidelity audio simulation method in this application includes the following steps: Based on the pre-acquired raw heart sound signals, a filtering and noise reduction process is performed to determine the standard heart sound signals; Based on a preset standard pathological information template, time-frequency analysis and pathological matching are performed according to the standard heart sound signal to determine the pathological feature information corresponding to the standard heart sound signal; Based on the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, the initial simulated heart sound signal is determined according to the pathological feature information. Based on the initial simulated heart sound signal, perform time-frequency analysis to determine the first time-frequency information; If the first time-frequency information meets the preset conditions, the initial simulated heart sound signal is determined as the target simulated heart sound signal.

[0008] In some implementations, the step of performing filtering and noise reduction processing based on the pre-acquired raw heart sound signals to determine the standard heart sound signals includes: Based on the original heart sound signal, effective signal filtering is performed under a preset signal sampling rate to determine the original effective heart sound signal, so as to retain the signals corresponding to the first heart sound and the second heart sound; Based on the Butterworth filter, bandpass filtering is performed according to the original valid heart sound signal to determine the responsive valid heart sound signal in order to preserve the physiological characteristics of the heart sound; Based on the effective heart sound signal in the response, spike denoising is performed to determine the standard heart sound signal in order to remove instantaneous spike noise.

[0009] In some implementations, the step of determining the pathological feature information corresponding to the standard heart sound signal by performing time-frequency analysis and pathological matching based on the preset standard pathological information template and the standard heart sound signal includes: Based on the standard heart sound signal, under preset frequency resolution and preset time resolution conditions, a synchronous compression transformation is performed to determine the second time-frequency information corresponding to the standard heart sound signal; Based on the second time-frequency information and the standard pathological information template, normalized cross-correlation calculation is performed to determine the pathological feature information, wherein the standard pathological information template includes multiple different pathological features, and the pathological feature information includes classification confidence information corresponding to the multiple different pathological features.

[0010] In some implementations, the initial simulated heart sound signal is determined based on the pathological feature information, according to the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, including: Based on multiple pre-input physiological parameters and the CircAdapt model, hemodynamic simulation variables and equivalent mechanical driving force are determined. The hemodynamic simulation variables include at least left ventricular pressure, aortic pressure, and blood flow. The equivalent mechanical driving force is proportional to the left ventricular pressure, aortic pressure, and blood flow.

[0011] In some embodiments, the determination of the initial simulated heart sound signal based on the pathological feature information, using a preset CircAdapt model and a two-degree-of-freedom mass-damped-stiffness vibration simulation model, further includes: Based on the pathological features and the preset benchmark parameter adjustment rules, the benchmark parameters of the two-degree-of-freedom mass-damped-stiffness vibration simulation model are determined. Based on the reference parameters and the equivalent mechanical driving force, the simulated displacement vector of the simulated blood clot is determined; Based on the simulated displacement vector, channel filtering and signal merging are performed on the first and second heart sounds to determine the initial simulated heart sound signal.

[0012] In some implementations, the simulated displacement vector satisfies the following system of second-order differential equations:

[0013] in For the quality matrix, Here is the damping matrix. Here is the stiffness matrix. The vector corresponding to the equivalent mechanical driving force. The simulated displacement vector, It is the acceleration vector. It is a velocity vector. For the mass of the proximal mass block, For the mass of the distal mass block, For proximal damping, For distal damping, For coupling damping, For the near-side spring stiffness, For the stiffness of the distal cavity wall, For coupling stiffness, The near-side component of the simulated displacement vector, The far component of the simulated displacement vector, The equivalent mechanical driving force is denoted as .

[0014] In some embodiments, when the initial simulated heart sound signals include multiple signals, the determination of the initial simulated heart sound signals based on the pathological feature information, using a preset CircAdapt model and a two-degree-of-freedom mass-damped-stiffness vibration simulation model, includes: Based on the pathological features and the preset benchmark parameter adjustment rules, the current benchmark parameters of the two-degree-of-freedom mass-damped-stiffness vibration simulation model are determined. Based on the current reference parameters and the equivalent mechanical driving force, the simulated displacement vector of the simulated blood clot is determined, wherein the physiological parameters are set with a preset amplitude of Gaussian random perturbation; Based on the simulated displacement vector, channel filtering and signal merging are performed on the first and second heart sounds to determine the current initial simulated heart sound signal. Multiple current initial simulated heart sound signals form a simulated heart sound signal data set.

[0015] In some implementations, the step of performing time-frequency analysis based on the initial simulated heart sound signal to determine the first time-frequency information includes: Based on the initial simulated heart sound signal, under preset frequency resolution and preset time resolution conditions, a synchronous compression transformation is performed to determine the first time-frequency information; The step of determining the initial simulated heart sound signal as the target simulated heart sound signal when the first time-frequency information meets preset conditions includes: If the comparison result between the first time-frequency information and the preset heart sound standard dataset meets the preset consistency verification expectation, the initial simulated heart sound signal is determined as the target simulated heart sound signal.

[0016] The electronic device in this application includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the high-fidelity audio simulation method described above is implemented.

[0017] The computer-readable storage medium in the embodiments of this application stores a computer program that, when executed by one or more processors, implements the high-fidelity audio simulation method described above.

[0018] The beneficial effects of this application are as follows: The high-fidelity heart sound simulation method in the embodiments of this application utilizes the CircAdapt model, the two-degree-of-freedom mass-damped-stiffness vibration simulation model, and the chest wall acoustic propagation to form a complete coupled framework. It realizes the modeling of the link from physiological parameters, mechanical response to acoustic output and generates simulated heart sound signals on this basis. Compared with the current related technologies that can only generate displacement or vibration signals, this application can directly output sound pressure level heart sound signals that conform to auscultation characteristics, which significantly improves the physiological realism and engineering usability of the simulation results. At the same time, the simulation results follow the clear mapping relationship between physiological parameters and audible heart sounds, making the generated simulated heart sound signals highly interpretable. Furthermore, the high-fidelity heart sound simulation method can be used to generate high-quality heart sound annotation data on a large scale, effectively solving the current situation of scarce heart sound data. Attached Figure Description

[0019] Figure 1 This is one of the flowcharts illustrating the high-fidelity audio simulation method in the embodiments of this application; Figure 2This is the second flowchart of the high-fidelity audio simulation method in the embodiments of this application; Figure 3 This is the third flowchart of the high-fidelity audio simulation method in the embodiments of this application; Figure 4 This is the fourth flowchart of the high-fidelity audio simulation method in the embodiments of this application; Figure 5(a) is the original heart sound waveform of active valvular heart disease in the standard pathological information template of the embodiment of this application; Figure 5(b) is a time-frequency diagram of active valvular heart disease in the standard pathological information template of the embodiments of this application; Figure 6(a) is the original heart sound waveform of mitral regurgitation in the standard pathological information template of the embodiment of this application; Figure 6(b) is a time-frequency diagram of mitral regurgitation in the standard pathological information template of the embodiment of this application; Figure 7(a) is a raw heart sound waveform diagram of mitral stenosis in the standard pathological information template of the embodiment of this application; Figure 7(b) is a time-frequency diagram of mitral stenosis in the standard pathological information template of the embodiment of this application; Figure 8(a) is the original heart sound waveform of mitral valve prolapse in the standard pathological information template of the embodiment of this application; Figure 8(b) is a time-frequency diagram of mitral valve prolapse in the standard pathological information template of the embodiment of this application; Figure 9 This is the fifth flowchart of the high-fidelity audio simulation method in the embodiments of this application. Detailed Implementation

[0020] Please see Figure 1 The high-fidelity audio simulation method in this application includes the following steps: Step 01: Based on the pre-acquired raw heart sound signal, perform filtering and noise reduction processing to determine the standard heart sound signal.

[0021] Specifically, the high-fidelity heart sound simulation method in this application mainly involves matching various pathological features with pre-acquired raw heart sound signals from reality, and further determining the target simulated heart sound signals for various pathological features using a complete coupled framework formed by hemodynamics (CircAdapt model), physiological structural vibration (two-degree-of-freedom mass-damped-stiffness vibration simulation model), and chest wall acoustic propagation. Since the quality of the raw heart sound signals, which form the data foundation, directly and significantly affects the simulation effect, it is exemplarily necessary to perform filtering and denoising processing on the raw heart sound signals after pre-acquiring them to improve their signal quality and thus ensure the accuracy of the generated target simulated heart sound signals.

[0022] In some implementations, please refer to Figure 2 Step 01 specifically includes: Step 011: Based on the original heart sound signal, perform effective signal filtering under the preset signal sampling rate to determine the original effective heart sound signal, so as to retain the signals corresponding to the first heart sound and the second heart sound; Step 012: Based on the Butterworth filter, perform bandpass filtering according to the original valid heart sound signal to determine the response valid heart sound signal in order to preserve the physiological characteristics of the heart sound; Step 013: Based on the valid heart sound signal in response, perform spike denoising to determine the standard heart sound signal in order to remove instantaneous spike noise.

[0023] Specifically, for the filtering and denoising of the original heart sound signal, for example, it is necessary to first determine the validity and filter the valid signal. In order to ensure the validity of the heart sound signal and the efficiency of the effective signal filtering, for example, this application adopts a step-by-step filtering and denoising process. By setting up multiple levels of processing, each level of processing retains additional information on the basis of the retention effect of the previous level, thereby ensuring the retention of all valid information in the original heart sound features.

[0024] The first stage of processing involves downsampling to filter effective information. The main method is to reduce the original sampling rate to 1000Hz. This method can significantly reduce the amount of data and improve computational efficiency. Experiments have shown that this method can save about 75% of the data processing time. In addition, it can preserve the key components of the heart sounds, especially the first heart sound S1 and the second heart sound S2. The original heart sound signal is processed into the original effective heart sound signal through the above effective information filtering process.

[0025] The second stage of processing involves using a Butterworth filter to perform bandpass filtering on the raw, valid heart sound signal obtained from the first stage. This filters out frequencies outside a preset range, primarily to preserve the core frequency band containing the physiological characteristics of the heart sounds. This core frequency band typically ranges from 25Hz to 400Hz, encompassing key physiological features of the heart sound signal, such as the pitch changes during diastole and systole. Signals below 25Hz may contain noise components such as electromyographic interference and respiratory sounds, while signals above 400Hz often involve high-frequency electronic noise and sensor noise. By filtering out these irrelevant frequency ranges, most external interference can be effectively reduced, and the determined, valid heart sound signal is further purified compared to the raw, valid heart sound signal.

[0026] The bandpass filtering of the Butterworth filter described above is performed according to Formula 1: ... Formula 1 in Let be the transfer function of the filter. The cutoff frequency, Let the filter order be . For Laplace variables.

[0027] The third stage of processing employs the Schmidt spike denoising algorithm, which performs spike denoising on the valid heart sound signals obtained from the second stage of processing. This algorithm dynamically adjusts the adaptive preload by calculating the local statistical characteristics of the signal, thereby accurately detecting and removing instantaneous spike noise, ultimately obtaining a standard heart sound signal with spike noise removed. In this way, while preserving the first heart sound S1, the second heart sound S2, and the key physiological characteristics of the heart sound pattern, it can effectively cope with noise problems in complex environments, significantly improving the reliability and clarity of the signal.

[0028] Please continue reading. Figure 1 The high-fidelity audio simulation method in this application also includes: Step 02: Based on the preset standard pathological information template, perform time-frequency analysis and pathological matching according to the standard heart sound signal to determine the pathological feature information corresponding to the standard heart sound signal.

[0029] Specifically, based on the above implementation method, the next step is to perform time-frequency analysis on the standard heart sound signal obtained from the above steps and perform pathological matching with the preset standard case information template. The ultimate goal is to obtain pathological feature information corresponding to the standard heart sound signal that can be used to guide the execution process of heart sound simulation. The main way in which the pathological feature information guides the execution process of heart sound simulation is to point to one or more pathological feature types with obvious tendencies based on the characteristics of the standard heart sound signal itself. Furthermore, the pathological feature type is used to guide the parameter setting of the two-degree-of-freedom mass-damped-stiffness vibration simulation model, thereby generating a simulated heart sound signal of the corresponding pathological feature type.

[0030] Further, please refer to Figure 3 In some implementations, step 02 specifically includes: Step 021: Based on the standard heart sound signal, under the conditions of preset frequency resolution and preset time resolution, perform synchronous compression transformation to determine the second time-frequency information corresponding to the standard heart sound signal; Step 022: Based on the second time-frequency information and the standard pathological information template, perform normalized cross-correlation calculation to determine the pathological feature information. The standard pathology information template includes multiple different pathological features, and the pathological feature information includes classification confidence information corresponding to multiple different pathological features.

[0031] Specifically, the process of determining the pathological features corresponding to standard heart sound signals can be found in the following example: First, based on the standard heart sound signal, a Synchrosqueezing Transform (SST) is performed to conduct time-frequency analysis on the standard heart sound signal, obtaining the second time-frequency information corresponding to the standard heart sound signal. This information can be represented as a time-frequency plot. The SST (Signal-Time Transform) method for time-frequency analysis can convert a one-dimensional standard heart sound signal into a clear time-frequency diagram. It can clearly display transient heart sound features such as the first heart sound, second heart sound, murmurs, opening valve sounds, and clicks. This method has strong time-frequency aggregation and can clearly show pathological features such as diamond-shaped heart sound signals, rectangular heart sound signals, rumbling sounds, and clicks. It is suitable for time-frequency analysis of heart sound signals with non-stationary, transient, and multi-component signal characteristics. In contrast, the STFT method in related technologies is prone to failing to detect subtle murmurs in heart sound time-frequency analysis, while the ordinary wavelet transform method is prone to problems such as unclear feature boundaries and low resolution. For example, the frequency resolution used in the SST process can be set to 1 kHz and the time resolution can be set to 5 ms. Specific values ​​can be adjusted according to actual conditions, and this application does not impose specific limitations.

[0032] Next, based on the second time-frequency information obtained in the above implementation method, the second time-frequency information is combined with the standard pathological information template for calculation to obtain the pathological feature information corresponding to the standard heart sound signal. For example, the above-mentioned standard pathological information template generally includes different pathological features of various heart diseases in terms of time-frequency information. For instance, the standard pathological information template includes heart sound time-frequency maps corresponding to four heart diseases: active valvular heart disease (AS), mitral regurgitation (MR), mitral stenosis (MS), and mitral valve prolapse (MVP).

[0033] For example, please refer to Figure 5. Figure 5(a) shows the original heart sound waveform corresponding to active valvular heart disease (hereinafter referred to as AS), and Figure 5(b) shows the time-frequency diagram corresponding to AS. It can be seen that the pathological features corresponding to AS show a relatively obvious high-frequency component. The typical AS murmur has a diamond-shaped or spindle-shaped waveform, which means that the murmur energy reaches its peak in mid-systole. This feature is usually related to changes in blood flow during the active phase, especially aortic stenosis or incomplete valve closure. In mid-systole, the blood flow velocity increases as it passes through the narrowed aortic valve, resulting in strong turbulence and murmur. At the beginning and end of systole, the intensity of the murmur decreases due to the relatively low blood flow velocity.

[0034] For example, please refer to Figure 6. Figure 6(a) shows the original heart sound waveform corresponding to mitral regurgitation (hereinafter referred to as MR), and Figure 6(b) shows the time-frequency diagram corresponding to MR. MR refers to a condition in which the mitral valve cannot completely close during left ventricular systole, causing some blood to flow back from the left ventricle to the left atrium. From its time-frequency diagram, it can be seen that a typical MR murmur is a rectangular wave, and the first heart sound S1 is weakened or even disappears. Immediately following the first heart sound S1, a high-frequency, uniformly energetic, and strong rectangular bright band appears throughout the entire systolic phase, followed by a normal or potentially disturbed second heart sound S2 bright band, and then, as far as possible, a low-frequency, brief third heart sound S3 wave. This characteristic reflects a holosystolic murmur; the mitral valve cannot completely close, and some blood flows back into the left atrium, causing the volume of S1 to weaken or even disappear. Simultaneously, the backflow of blood during systole creates continuous turbulence, forming a high-frequency rectangular murmur. The weakening of S1 and the appearance of the S3 wave further reflect changes in cardiac pressure and filling.

[0035] For example, please refer to Figure 7. Figure 7(a) shows the original heart sound waveform corresponding to mitral stenosis (hereinafter referred to as MS), and Figure 7(b) shows the time-frequency diagram corresponding to MS. MS refers to the narrowing of the mitral valve opening between the left atrium and the left ventricle, which hinders the flow of blood from the left atrium to the left ventricle during diastole. The time-frequency diagram shows a bright and broad band of the first heart sound S1. This band is mainly caused by the increased pressure difference between the left atrium and left ventricle due to MS, resulting in a more forceful closure of the mitral valve and producing a relatively clear and loud S1 sound. Simultaneously, a high-frequency spike corresponding to the opening sound appears after the second heart sound S2, reflecting mitral valve sclerosis or calcification and loss of valve elasticity. A mid-to-low frequency energy band extending from the opening sound runs throughout diastole and may intensify before the next first heart sound S1. This energy band is caused by a persistent low-frequency murmur due to mitral valve stenosis, which results in slower and turbulent diastolic blood flow into the left ventricle. This low-frequency murmur is usually described as a rumble.

[0036] For example, please refer to Figure 8. Figure 8(a) shows the original heart sound waveform corresponding to mitral valve prolapse (hereinafter referred to as MVP), and Figure 8(b) shows the time-frequency diagram corresponding to MVP. MVP refers to a disease in which the mitral valve leaflets are structurally abnormal, causing them to bulge or prolapse into the left atrium during ventricular systole. From its time-frequency diagram, it can be seen that a normal bright band of the first heart sound S1 is first presented. Between the first heart sound S1 and the second heart sound S2, a very high and very narrow bright line of a click appears in mid-systole. The main cause of the click is that the mitral valve fails to close completely and rebounds rapidly during mid-systole. Immediately following the click line to the second heart sound S2 is a late systolic murmur with a higher frequency. This is mainly caused by the mitral valve failing to close completely, causing blood to partially flow back into the left atrium during systole, creating turbulence. As the mitral valve opens and closes, the frequency of the murmur extends from the mid-frequency range to the high-frequency range.

[0037] In summary, the above standard pathology information template can be briefly summarized as follows: AS: Mid-contraction 60~200Hz diamond-shaped high-frequency band; MR: Rectangular full-shrinkage noise band of 40–180 Hz during full-shrinkage period; MS: S1 enhancement + open-lobe tone + diastolic 20~80Hz rumble low-frequency band; MVP: Mid-contraction click spike + late-contraction increasing high-frequency noise band.

[0038] Based on the standard pathological information template shown in the example above, and in conjunction with the time-frequency diagram corresponding to the standard heart sound signal... By performing normalized cross-correlation calculations according to Formula 2, the time-frequency diagram can be obtained. Relating the four sets of classification confidence information for the four diseases mentioned above, these four sets of classification confidence information are finally combined into a classification confidence vector, thereby obtaining the pathological feature information corresponding to the standard heart sound signal in the above implementation method: ... Formula 2 in This indicates that the standard heart sound signal corresponds to a disease. Classification confidence information, Indicates time, Indicates frequency, Indicates disease The corresponding time-frequency diagram, You can use AS, MR, MS, and MVP. Then, based on the obtained classification confidence information... The classification confidence vector is obtained. As shown in Formula 3: ... Formula 3 The obtained classification confidence vector It can be used to guide parameter setting for two-degree-of-freedom mass-damped-stiffness vibration simulation models. (Classification confidence information) The similarity between the physiological characteristics of standard heart sound signals and the pathological characteristics of various diseases in the standard pathological information template is described.

[0039] Please continue reading. Figure 1 The high-fidelity audio simulation method in this application also includes: Step 03: Based on the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, determine the initial simulated heart sound signal according to the pathological feature information.

[0040] Specifically, based on the above implementation method, and having obtained the pathological feature information corresponding to the standard heart sound signal, the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model (hereinafter referred to as the two-degree-of-freedom MCK vibration model) are invoked as guidance to perform simulation, thereby achieving high-fidelity simulation of the heart sound signal based on the standard heart sound signal collected in reality.

[0041] In some implementations, step 03 first includes: Based on multiple pre-input physiological parameters, the hemodynamic simulation variables and equivalent mechanical driving force are determined using the CircAdapt model. The hemodynamic simulation variables include at least left ventricular pressure, aortic pressure, and blood flow. The equivalent mechanical driving force is proportional to left ventricular pressure, aortic pressure, and blood flow.

[0042] Specifically, based on hemodynamic principles, the CircAdapt model, a current technology, is used to calculate hemodynamic simulation variables and equivalent mechanical driving forces in real time based on the physiological parameters of the heart chambers and various blood vessels. These physiological parameters generally include at least valve area, vascular compliance, cardiac chamber resistance, and myocardial elasticity parameters. The calculated equivalent mechanical driving force can be used as the mechanical vibration excitation in the subsequent two-degree-of-freedom MCK vibration model to simulate heart sound signals, thus providing physical constraints for the timing and intensity of simulated heart sounds. Due to the existence of hemodynamic principles, the equivalent mechanical driving force can automatically reflect the opening and closing timing of heart valves, and different pathological conditions can be simulated by adjusting the hemodynamic parameters.

[0043] For example, the hemodynamic simulation variables mentioned above can be left ventricular pressure, aortic pressure, and blood flow. The relationship between the equivalent mechanical driving force and the above three is shown in Equation 4: ... Formula 4 in For equivalent mechanical driving force, Left ventricular pressure, Aortic pressure, For blood flow.

[0044] Furthermore, in some implementations, please refer to Figure 4 Step 03 further includes: Step 0321: Determine the baseline parameters of the two-degree-of-freedom mass-damped-stiffness vibration simulation model based on pathological characteristics and preset baseline parameter adjustment rules. Step 0322: Determine the simulation displacement vector of the simulated blood clot based on the reference parameters and the equivalent mechanical driving force; Step 0323: Based on the simulated displacement vector, perform channel filtering and signal merging for the first and second heart sounds to determine the initial simulated heart sound signals.

[0045] Specifically, based on the above implementation method, and after determining the equivalent mechanical driving force based on the CircAdapt model, the two-degree-of-freedom MCK vibration model can be invoked to simulate the heartbeat process under the guidance of pathological feature information, thereby realizing the simulation process for heart sound signals.

[0046] Specifically, firstly, based on the pathological feature information corresponding to the standard heart sound signal, and based on the preset baseline parameter adjustment rules of the two-degree-of-freedom MCK vibration model, the baseline parameters of the two-degree-of-freedom MCK vibration model are determined. For example, based on the above example, according to the classification confidence vector corresponding to the standard heart sound signal... Based on the disease type corresponding to the component with the highest value, the corresponding rule entry is selected in the baseline parameter adjustment rules to determine the baseline parameters of the two-degree-of-freedom MCK vibration model.

[0047] Furthermore, the baseline parameters of the two-degree-of-freedom MCK vibration model include: the mass of the proximal mass block. Mass of the distal mass block Proximal damping Distal damping Coupled damping Proximal spring stiffness Distal cavity wall stiffness and coupling stiffness The above-mentioned baseline parameter adjustment rules are shown in Table 1. It should be noted that the baseline parameter adjustment rules include not only adjusting the baseline parameters themselves, but also restricting the adjustment of hemodynamic simulation variables in the CircAdapt model. Generally, the normal cardiac operating characteristics are used as the baseline, and the baseline parameters are adjusted in different directions according to different pathological characteristics.

[0048] Table 1. Baseline Parameter Adjustment Rules

[0049] Next, with the baseline parameters determined, the second-order differential equations of the dynamic behavior based on the two-degree-of-freedom MCK vibration model can be used to solve for the simulated displacement vector of the simulated blood clot during the simulated heartbeat. Please refer to Formula 5 for details: ... Formula 5 in For the quality matrix, Here is the damping matrix. Here is the stiffness matrix. The vector corresponding to the equivalent mechanical driving force. To simulate the displacement vector, It is the acceleration vector. It is a velocity vector. For the mass of the proximal mass block, For the mass of the distal mass block, For proximal damping, For distal damping, For coupling damping, For the near-side spring stiffness, For the stiffness of the distal cavity wall, For coupling stiffness, To simulate the proximal component of the displacement vector, To simulate the far component of the displacement vector, It is the equivalent mechanical driving force.

[0050] For the specific solution of the second-order differential equation system shown in Equation 5, for example, it can be implemented in the CircAdapt model during valve closure using the Newmark-β integration method currently in related technologies. The specific solution details can be adjusted according to the actual situation, and this application does not impose any specific limitations.

[0051] Finally, based on the simulated displacement vector obtained from the solution... Using digital filters to simulate displacement vectors This mechanical vibration is mapped to a simulated chest wall recording of heart sound signals, thus ensuring consistency between the simulation data and real electronic auscultation results. Specifically, for the first heart sound S1, which includes the mitral valve component M1 and the tricuspid valve component T1, a first-channel digital filter is used to attenuate M1 and T1 by 30 dB within the 20–100 Hz range, and attenuate them by -12 dB per octave above 100 Hz. For the second heart sound S2, which includes the aortic valve component A2 and the pulmonary valve component P2, a second-channel digital filter is used to attenuate A2 and P2 by 46 dB within the 20–100 Hz range, and attenuate them by -6 dB per octave above 100 Hz. After filtering these different components of the heart sound signal using the first and second-channel digital filters respectively, the signals are combined to generate the initial simulated heart sound signal corresponding to the standard heart sound signal.

[0052] In particular, in some implementations, please refer to Figure 9 Step 03 further includes: Step 03201: Determine the current baseline parameters of the two-degree-of-freedom mass-damped-stiffness vibration simulation model based on pathological feature information and preset baseline parameter adjustment rules; Step 03202: Determine the simulation displacement vector of the simulated blood clot based on the reference parameters and the equivalent mechanical driving force, wherein the physiological parameters are set with a preset amplitude of Gaussian random perturbation; Step 03203: Based on the simulated displacement vector, perform channel filtering and signal merging for the first and second heart sounds to determine the initial simulated heart sound signal corresponding to the current standard heart sound signal. Multiple initial simulated heart sound signals form a simulated heart sound signal data set.

[0053] Specifically, based on the above-described embodiments, the high-fidelity heart sound simulation method in this application can not only generate corresponding initial simulated heart sound signals one-to-one based on standard heart sound signals, but also create different model conditions by making different adjustments to the physiological parameters in the CircAdapt model, generating a series of different initial simulated heart sound signals under diverse conditions. Thus, the high-fidelity heart sound simulation method described above can generate a large number of simulated heart sound signal datasets to solve the problem of the extremely small number of labeled datasets for heart sounds in current related technologies.

[0054] For example, the specific method for creating different model conditions mainly involves applying Gaussian perturbations within a range of ±20% to physiological parameters such as valve area, vascular compliance, intracardiac resistance, and myocardial elasticity in the CircAdapt model. By arranging and combining these perturbations, a large number of different CircAdapt model conditions are formed to simulate the physiological differences existing in different individuals. Under these diverse physiological differences, the high-fidelity heart sound simulation method described above can simulate five major categories of heart sound signals—normal, AS, MR, MS, and MVP—under physiological conditions with subtle differences. Specifically, based on these physiological differences, when the duration of the initial simulated heart sound signal is set to approximately 3 seconds, at least 10,000 initial simulated heart sound signals with different physiological conditions can be generated in batches, thus forming a simulated heart sound signal dataset.

[0055] Please continue reading. Figure 1 The high-fidelity audio simulation method in this application also includes: Step 04: Based on the initial simulated heart sound signal, perform time-frequency analysis to determine the first time-frequency information; Step 05: If the first time-frequency information meets the preset conditions, the initial simulated heart sound signal is determined as the target simulated heart sound signal.

[0056] Specifically, based on the above implementation method, and given the initial simulated heart sound signal obtained through the simulation process described above, it is still necessary to evaluate and verify the effectiveness of the initial simulated heart sound signal. Therefore, exemplarily, the initial simulated heart sound signal is first treated as a real heart sound signal obtained through auscultation. Referring to the time-frequency analysis method used for standard heart sound signals in the above implementation method, time-frequency analysis is performed on the initial simulated heart sound signal to convert it into a corresponding time-frequency graph (corresponding to the first time-frequency information). Then, feature comparison analysis is performed based on the time-frequency graph corresponding to the initial simulated heart sound signal to achieve the aforementioned evaluation and verification.

[0057] Furthermore, in some embodiments, step 04 further includes: Based on the initial simulated heart sound signal, under the conditions of preset frequency resolution and preset time resolution, a synchronous compression transformation is performed to determine the first time-frequency information; Step 05 further includes: If the comparison results between the first time-frequency information and the preset heart sound standard dataset meet the preset consistency verification expectations, the initial simulated heart sound signal is determined as the target simulated heart sound signal.

[0058] Specifically, the method of performing time-frequency analysis on the initial simulated heart sound signal to obtain the first time-frequency information can be implemented by referring to the method of performing time-frequency analysis on the standard heart sound signal using SST in the above embodiment, which will not be repeated here. The frequency resolution used in the SST process can be set to 1kHz and the time resolution can be set to 5ms. The specific values ​​can be adjusted according to the actual situation, and this application does not make specific limitations.

[0059] Next, for the evaluation and verification method of the first time-frequency information, for example, the heart sound signals included in the currently publicly available standard heart sound dataset are used as the comparison object. Based on five different pathological features, namely normal, AS, MR, MS, and MVP, the heart sound signals included in the currently publicly available standard heart sound dataset are divided into five categories. During the evaluation and verification, according to the pathological feature information corresponding to the initial simulated heart sound signal, the heart sound signal of the corresponding category is called from the heart sound signals included in the currently publicly available standard heart sound dataset to perform signal feature comparison. If the consistency of the initial simulated heart sound signal and the heart sound signals included in the currently publicly available standard heart sound dataset in terms of signal features meets the verification expectation, then the initial simulated heart sound signal is considered to meet the usage requirements, and thus the initial simulated heart sound signal is determined as the target simulated heart sound signal, and the heart sound simulation process is finally completed.

[0060] Thus, the high-fidelity heart sound simulation method in this application utilizes the CircAdapt model, the two-degree-of-freedom mass-damped-stiffness vibration simulation model, and the chest wall acoustic propagation to form a complete coupled framework. It models the link from physiological parameters and mechanical response to acoustic output and generates simulated heart sound signals based on this. Compared with the current related technologies that can only generate displacement or vibration signals, this application can directly output sound pressure level heart sound signals that conform to auscultation characteristics, significantly improving the physiological realism and engineering usability of the simulation results. At the same time, the simulation results follow a clear mapping relationship between physiological parameters and audible heart sounds, making the generated simulated heart sound signals highly interpretable. Furthermore, the high-fidelity heart sound simulation method can be used to generate high-quality heart sound annotation data on a large scale, effectively solving the current problem of scarce heart sound data.

[0061] The electronic device in this application includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the high-fidelity audio simulation method described above is implemented.

[0062] The computer-readable storage medium in the embodiments of this application stores a computer program that, when executed by one or more processors, implements the high-fidelity audio simulation method described above.

[0063] The above description is merely a preferred embodiment of this application and is not intended to limit this application in any way. Although this application has disclosed the preferred embodiment as above, it is not intended to limit this application. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the technical solution of this application. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the technical solution of this application, based on the technical essence of this application and within the spirit and principles of this application, shall still fall within the protection scope of the technical solution of this application.

Claims

1. A high-fidelity audio simulation method, characterized in that, The method includes: Based on the pre-acquired raw heart sound signals, a filtering and noise reduction process is performed to determine the standard heart sound signals; Based on a preset standard pathological information template, time-frequency analysis and pathological matching are performed according to the standard heart sound signal to determine the pathological feature information corresponding to the standard heart sound signal; Based on the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, the initial simulated heart sound signal is determined according to the pathological feature information. Based on the initial simulated heart sound signal, perform time-frequency analysis to determine the first time-frequency information; If the first time-frequency information meets the preset conditions, the initial simulated heart sound signal is determined as the target simulated heart sound signal.

2. The method according to claim 1, characterized in that, The step of performing filtering and noise reduction processing on the pre-acquired raw heart sound signals to determine the standard heart sound signals includes: Based on the original heart sound signal, effective signal filtering is performed under a preset signal sampling rate to determine the original effective heart sound signal, so as to retain the signals corresponding to the first heart sound and the second heart sound; Based on the Butterworth filter, bandpass filtering is performed according to the original valid heart sound signal to determine the responsive valid heart sound signal in order to preserve the physiological characteristics of the heart sound; Based on the effective heart sound signal in the response, spike denoising is performed to determine the standard heart sound signal in order to remove instantaneous spike noise.

3. The method according to claim 1, characterized in that, The method, based on a preset standard pathological information template, performs time-frequency analysis and pathological matching according to the standard heart sound signal to determine the pathological feature information corresponding to the standard heart sound signal, including: Based on the standard heart sound signal, under preset frequency resolution and preset time resolution conditions, a synchronous compression transformation is performed to determine the second time-frequency information corresponding to the standard heart sound signal; Based on the second time-frequency information and the standard pathological information template, normalized cross-correlation calculation is performed to determine the pathological feature information, wherein the standard pathological information template includes multiple different pathological features, and the pathological feature information includes classification confidence information corresponding to the multiple different pathological features.

4. The method according to claim 3, characterized in that, The initial simulated heart sound signal, based on the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, is determined according to the pathological feature information, including: Based on multiple pre-input physiological parameters and the CircAdapt model, hemodynamic simulation variables and equivalent mechanical driving force are determined. The hemodynamic simulation variables include at least left ventricular pressure, aortic pressure, and blood flow. The equivalent mechanical driving force is proportional to the left ventricular pressure, aortic pressure, and blood flow.

5. The method according to claim 4, characterized in that, The method based on the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, determining the initial simulated heart sound signal according to the pathological feature information, further includes: Based on the pathological features and the preset benchmark parameter adjustment rules, the benchmark parameters of the two-degree-of-freedom mass-damped-stiffness vibration simulation model are determined. Based on the reference parameters and the equivalent mechanical driving force, the simulated displacement vector of the simulated blood clot is determined; Based on the simulated displacement vector, channel filtering and signal merging are performed on the first and second heart sounds to determine the initial simulated heart sound signal.

6. The method according to claim 5, characterized in that, The simulated displacement vector satisfies the following system of second-order differential equations: in For the quality matrix, Here is the damping matrix. Here is the stiffness matrix. The vector corresponding to the equivalent mechanical driving force. The simulated displacement vector, It is the acceleration vector. It is a velocity vector. For the mass of the proximal mass block, For the mass of the distal mass block, For proximal damping, For distal damping, For coupling damping, For the near-side spring stiffness, For the stiffness of the distal cavity wall, For coupling stiffness, The near-side component of the simulated displacement vector, The far component of the simulated displacement vector, The equivalent mechanical driving force is denoted as .

7. The method according to claim 4, characterized in that, When the initial simulated heart sound signals include multiple signals, the initial simulated heart sound signals are determined based on the preset CircAdapt model and the two-degree-of-freedom mass-damped-stiffness vibration simulation model, according to the pathological feature information, including: Based on the pathological features and the preset benchmark parameter adjustment rules, the current benchmark parameters of the two-degree-of-freedom mass-damped-stiffness vibration simulation model are determined. Based on the reference parameters and the equivalent mechanical driving force, the simulated displacement vector of the simulated blood clot is determined, wherein the physiological parameters are set with a Gaussian random perturbation of a preset amplitude. Based on the simulated displacement vector, channel filtering and signal merging are performed on the first and second heart sounds to determine the initial simulated heart sound signal corresponding to the current standard heart sound signal. Multiple initial simulated heart sound signals form a simulated heart sound signal data set.

8. The method according to any one of claims 1-7, characterized in that, The step of performing time-frequency analysis based on the initial simulated heart sound signal to determine the first time-frequency information includes: Based on the initial simulated heart sound signal, under preset frequency resolution and preset time resolution conditions, a synchronous compression transformation is performed to determine the first time-frequency information; The step of determining the initial simulated heart sound signal as the target simulated heart sound signal when the first time-frequency information meets preset conditions includes: If the comparison result between the first time-frequency information and the preset heart sound standard dataset meets the preset consistency verification expectation, the initial simulated heart sound signal is determined as the target simulated heart sound signal.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the high-fidelity audio simulation method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by one or more processors, implements the high-fidelity audio simulation method as described in any one of claims 1-8.

Images