Heart mechanical motion signal processing system
By performing unified scaling and dimensionality reduction feature extraction on time-frequency domain data of cardiac mechanical motion signals, the problem of assessing cardiac mechanical motion status in existing technologies has been solved, enabling interpretable and accurate assessment and visualization analysis.
Patent Information
- Application Number
- CN202511654083.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2025-09-17
- Filing Date
- 2025-11-12
- Publication Date
- 2026-01-16
AI Technical Summary
Existing technologies struggle to accurately assess cardiac mechanical motion, and neural network-based analysis methods lack interpretability.
By acquiring time-frequency domain data of cardiac mechanical motion signals, time, energy, and frequency normalization is performed to form a time-frequency spatial distribution pattern of uniform scale. Dimensionality reduction and feature extraction are then performed, and combined with visualization and threshold filtering, an interpretable evaluation method is provided.
It enables precise quantitative assessment of cardiac mechanical motion, with visualized and clinically interpretable results. It allows for comparisons across cardiac cycles and individuals, reduces noise interference, and improves the interpretability of the analysis.
Smart Images
Figure CN121337368A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the field of time-frequency domain analysis of non-invasive body surface signals of cardiac mechanical movement, and particularly relates to a cardiac mechanical movement signal processing system, and especially to a heart shock and heart sound signal processing system reflecting the state of cardiac mechanical movement. BACKGROUND
[0002] The heart is essentially a biological pump composed of atrium, ventricle and valve structure, which contracts and relaxes under the stimulation of an electrocardiosignal to realize the function of one-way blood pumping. The mechanical movement of the heart in contraction and relaxation has a high spatiotemporal coordination. Cardiac ultrasound and nuclear magnetic resonance are common means for evaluating the structure and mechanical function of the heart. For example, the application of echocardiography technology to analyze the characteristics of the heart wall movement is an important examination method for myocardial contraction function and ventricular wall coordination, but it is complex or expensive and not easy to obtain. So far, there is no simple method to accurately evaluate the state of cardiac mechanical movement.
[0003] Heart shock signals and heart sound signals, as non-invasive body surface signals representing cardiac mechanical movement, have always attracted research attention. Time domain analysis and frequency domain analysis of heart shock and heart sound have been extensively studied.
[0004] As shown in the lower half of FIG. 1, Figure 1 , Figure 2 , Figure 3 Time domain analysis of heart shock is shown in the lower half of FIG. 1. Specifically, the heart shock signal and the electrocardiosignal are synchronously collected, and after basic filtering (the heart shock signal is taken as 2-100 Hz, and the electrocardiosignal is taken as 1.5-30 Hz), the time domain graph is directly drawn, in which the red color is the electrocardiogram corresponding to the electrocardiosignal, and the blue color is the heart shock graph corresponding to the heart shock signal. However, the time domain graph is only a kind of intuitive and simple data visualization of the original collected signal, and cannot provide deeper signal nature characteristics of the heart shock signal in the time-frequency domain. Moreover, the heart shock signal in the time domain graph is highly complex, and only looking at the time domain graph or combining the electrocardiosignal cannot obtain scientific and effective information of the heart state.
[0005] In the prior art, researchers use neural networks, deep learning and other methods to train classification models of time-frequency data in different heart health states (such as healthy people, heart failure patients, etc.). The biggest defect of this route is that the rules and characteristics are completely handed over to the "black box" of the neural network, which cannot be controlled and is not interpretable.
[0006] For example, IEEE document "A Binary Classification of Cardiovascular Abnormality Using Time Frequency Features of Cardio-mechanical Signals" mentioned the time-frequency domain analysis of heart vibration signals, but its specific analysis method is to analyze through a neural network, which has the disadvantages of needing a large number of samples to pre-train the model and the unexplainability of the neural network analysis process.
[0007] For another example, IEEE document "Time–Frequency-Domain Deep Learning Framework for the Automated Detection of Heart Valve Disorders Using PCG Signals" mentioned that time-frequency domain analysis of heart sound signals through a deep neural network can reveal the condition of heart valve diseases. However, this method also lacks explainability and cannot establish an explainable logic with the physiological and pathological processes of the heart. SUMMARY
[0008] In view of the defects in the prior art, the purpose of the present application is to provide a heart mechanical movement signal processing system.
[0009] According to the heart mechanical movement signal processing system provided by the present application, the following technical scheme is adopted: Module M1: Obtain time-frequency domain data of a plurality of set periods of heart mechanical movement signals; divide the time-frequency domain data of the heart mechanical movement signals according to the set periods, and perform normalization processing according to time and energy normalization processing to form one or more sets of uniform scale time-frequency domain data; The heart mechanical movement signals include heart vibration signals and / or heart sound signals.
[0010] Preferably, the system further comprises: Module M2: Fuse the plurality of sets of uniform scale time-frequency domain data to obtain a distribution pattern of the energy of the heart mechanical movement in the uniform scale time-frequency space; The distribution pattern of the energy of the heart mechanical movement in the uniform scale time-frequency space includes a visualized time-frequency space distribution pattern. The fusion includes averaging or superimposing and then filtering through a threshold value. Module M3: Perform dimension reduction and / or feature extraction on the distribution pattern of the energy of the heart mechanical movement in the uniform scale time-frequency space to obtain a processed distribution pattern of the energy of the heart mechanical movement in the uniform scale time-frequency space. The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network; Alternatively, the cardiac mechanical motion signal processing system, in module M1, fuses the time-frequency domain data of the plurality of set-period cardiac mechanical motion signals to obtain a set of uniform-scale time-frequency domain data before the energy is normalized, wherein the fusion includes averaging or superimposing and then filtering through a threshold; and further comprising: Module M200: performing dimension reduction and / or feature extraction on the distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space formed by the uniform-scale time-frequency domain data, to obtain a processed distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space; The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
[0011] Preferably, further comprising: Module M3': performing dimension reduction and / or feature extraction on the plurality of sets of uniform-scale time-frequency domain data, respectively, to obtain the distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space; The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network. Module M2': fusing the plurality of sets of uniform-scale time-frequency domain data after the dimension reduction and / or feature extraction, to obtain a processed distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space; The distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space includes a visualized time-frequency space distribution pattern. The fusion includes averaging or superimposing and then filtering through a threshold.
[0012] Preferably, further comprising: Module M4: evaluating the cardiac function status according to the distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space or according to the processed distribution pattern of the cardiac mechanical motion energy in the uniform-scale time-frequency space; or further performing data analysis after the dimension reduction and / or feature extraction to evaluate the cardiac function status.
[0013] Preferably, the plurality of sets of uniform-scale time-frequency domain data include visualized time-frequency space distributions. In module M1: dividing the time-frequency domain data of the cardiac mechanical motion signal according to a set period, and performing time normalization processing, energy normalization processing and frequency standardization processing to form a plurality of sets of uniform-scale time-frequency domain data; Or, in the module M1: the time-frequency domain data of the cardiac mechanical movement signal is divided according to a set period and normalized by time and energy, and frequency standardization processing is performed to form a set of uniform scale time-frequency domain data; wherein, before the energy normalization processing, the time-frequency domain data of the cardiac mechanical movement signal of multiple set periods is fused, wherein the fusion includes averaging or superimposing and then filtering through a threshold.
[0014] Preferably, the module M1 comprises: Module M1.1: acquiring the synchronously collected cardiac mechanical movement signal and the electrocardio signal; Module M1.2: processing the cardiac mechanical movement signal into time-frequency domain data of the cardiac mechanical movement signal; Module M1.3: dividing the time-frequency domain data of the cardiac mechanical movement signal according to a set period according to the electrocardio signal; The set period is a cardiac cycle obtained by separating the R peak of the electrocardio signal.
[0015] Preferably, the heart shock signal in the cardiac mechanical movement signal is filtered at 2-100Hz, and the heart sound signal is filtered at 20-1000Hz. Wavelet transform or synchronous compression wavelet transform or S-transform or time redistribution multi-synchronous extrusion transform is used to calculate the cardiac mechanical movement signal to obtain the time-frequency domain data of the cardiac mechanical movement signal.
[0016] Preferably, in the visualized uniform scale time-frequency space distribution pattern, the horizontal axis represents the normalized time, the vertical axis represents the standardized frequency, and the brightness of the point or point cloud reflects the distribution pattern of the normalized energy of the cardiac mechanical movement.
[0017] Preferably, it further comprises: Providing the user with the distribution information of the cardiac mechanical time-frequency energy by time and / or frequency.
[0018] According to the present application, a cardiac mechanical movement signal processing system is provided, characterized by comprising: Obtaining the time-frequency domain data of the cardiac mechanical movement signal of a set period, and providing the user with the distribution information of the cardiac mechanical time-frequency energy by time and / or frequency according to the time-frequency domain data.
[0019] According to the present application, a computer readable storage medium storing a computer program is provided, wherein the computer program is executed by a processor to realize the steps of the cardiac mechanical movement signal processing method, wherein the cardiac mechanical movement signal processing method comprises: Step S1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; performing time normalization and energy normalization on the time-frequency domain data of the cardiac mechanical movement signals according to the set periods, to form one or more groups of time-frequency domain data of uniform scales; the cardiac mechanical movement signals include heart shock signals and / or heart sound signals; Or, comprising: obtaining time-frequency domain data of a set period of cardiac mechanical movement signals, and providing distribution information of cardiac mechanical time-frequency energy according to time and / or frequency to a user according to the time-frequency domain data.
[0020] According to the electronic device provided by the application, the memory, the processor and the computer program stored in the memory and executable on the processor are provided, and the computer program is executed by the processor to realize the steps of the cardiac mechanical movement signal processing method, wherein the cardiac mechanical movement signal processing method comprises: Step S1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; performing time normalization and energy normalization on the time-frequency domain data of the cardiac mechanical movement signals according to the set periods, to form one or more groups of time-frequency domain data of uniform scales; the cardiac mechanical movement signals include heart shock signals and / or heart sound signals; Or, comprising: obtaining time-frequency domain data of a set period of cardiac mechanical movement signals, and providing distribution information of cardiac mechanical time-frequency energy according to time and / or frequency to a user according to the time-frequency domain data.
[0021] Compared with the prior art, the application has the following beneficial effects: 1. The application provides a new cardiac mechanical movement information processing system reflecting the state of cardiac mechanical movement, that is, processing heart shock signals and / or heart sound signals to obtain the distribution mode of the energy of the cardiac mechanical movement signals in the uniform scale time-frequency space: converting the original time-frequency data matrix of the cardiac cycle into the energy distribution in the uniform scale time-frequency space with the normalized time on the x-axis and the standardized frequency on the y-axis, and the brightness of a certain area representing the distribution intensity of the energy of the corresponding specific frequency in the normalized total energy in the corresponding specific time.
[0022] 2、The present application converts the mechanical movement energy of each original, different and non-accurately objectively quantifiable heart cycle into an energy distribution pattern in a unified scale time-frequency space, realizes accurate quantification of the heart mechanical movement state, and the result can be visualized in a heart cycle (systole, diastole) familiar to clinicians, has strong clinical interpretability, and can be used for doctor's assessment of the heart mechanical movement state. The present application obtains the heart mechanical movement energy distribution pattern in a unified scale time-frequency space by a clear and interpretable data processing process, to support further classification of the heart function state.
[0023] 3、The present application eliminates the influence of heart rate fast or slow (i.e. different heart cycle lengths) by time normalization, and ensures that the comparison between different heart cycles and individuals is the energy distribution under the same physiological rhythm percentage.
[0024] 4、The present application eliminates the influence of heartbeat strength (energy) under different people and different movement states by energy normalization, that is, the brightness of each point is converted from the original actual energy value to the proportion of the energy of this point in the total energy. Energy normalization eliminates the influence of heartbeat strength (energy) under different people and different movement states, and ensures that the comparison is the distribution pattern of energy in a unified scale time-frequency space, rather than the strength of the absolute value of energy.
[0025] 5、The present application makes the time-frequency data matrix of each heart cycle have the same dimension frequency axis under the same frequency scale by frequency standardization.
[0026] 6、The present application further realizes dimension reduction, feature extraction and noise filtering by gridding, is more conducive to stable core feature extraction, and has strong interpretability. BRIEF DESCRIPTION OF DRAWINGS
[0027] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings: Figure 1 、 Figure 2 、 Figure 3 The figure is a direct drawing of the original time-frequency domain data of three adjacent and continuous heart cycles (heartbeats) and the corresponding figure of the synchronous time domain signal graph of heart shock and electrocardiogram. Figure 1 、 Figure 2 、 Figure 3 The upper graph in the figure is a direct drawing of the original time-frequency domain data, the vertical axis is frequency [Hz], and the horizontal axis is time [second]. Figure 1 、 Figure 2 、 Figure 3The lower graph in the figure is a synchronous time domain signal graph of heart shock and electrocardio, the left vertical axis is SCG amplitude, the right vertical axis is ECG amplitude, the horizontal axis is time [second], red represents electrocardiogram ECG, and blue represents seismocardiogram SCG.
[0028] Figure 1 The corresponding graph of the direct drawing of the original time-frequency domain data of the first heart cycle and the synchronous time domain signal of heart shock and electrocardio.
[0029] Figure 2 The corresponding graph of the direct drawing of the original time-frequency domain data of the second heart cycle and the synchronous time domain signal of heart shock and electrocardio.
[0030] Figure 3 The corresponding graph of the direct drawing of the original time-frequency domain data of the third heart cycle and the synchronous time domain signal of heart shock and electrocardio.
[0031] Figure 4 、 Figure 5 、 Figure 6 The original time-frequency data of Figure 1 、 Figure 2 、 Figure 3 are converted into energy distribution graphs in a unified scale, specifically, the original, different, and objective and quantifiable original time-frequency domain data in Figure 1 、 Figure 2 、 Figure 3 are converted into energy distribution in a unified scale of time normalization, energy normalization, and frequency standardization in Figure 4 、 Figure 5 、 Figure 6 , obtaining the total energy normalized energy distribution pattern in the time-frequency space of 1000 dimensions of time axis * 256 dimensions of frequency, realizing the comparability and accurate quantification calculation of time-frequency energy in different heart cycles.
[0032] Figure 4 The energy distribution graph of Figure 1 in a unified scale, that is, the visualized time-frequency domain data in a unified scale.
[0033] Figure 5 The energy distribution graph of Figure 2 in a unified scale, that is, the visualized time-frequency domain data in a unified scale.
[0034] Figure 6 The energy distribution graph of Figure 3 in a unified scale, that is, the visualized time-frequency domain data in a unified scale.
[0035] Figure 7 The corresponding graph of the direct drawing of the original time-frequency domain data of the first heart cycle and the synchronous time domain signal of heart shock and electrocardio. Figure 4 、 Figure 5 、 Figure 6The average of the energy distribution over three consecutive cardiac cycles at a uniform scale is calculated to obtain the average plot of the energy distribution pattern of cardiac mechanical motion in a uniform time-frequency space.
[0036] Figure 8 To obtain an average map of the energy distribution over a unified scale space encompassing 20 cardiac cycles, including the three consecutive cardiac cycles mentioned above, that is, to visualize the distribution pattern of cardiac mechanical motion energy in a unified scale time-frequency space, a more stable distribution characteristic of cardiac mechanical motion energy is needed.
[0037] Figure 9 To Figure 8 The energy map obtained by gridding, i.e., the visualized gridded uniform-scale time-frequency spatial distribution pattern, will... Figure 8 The high-resolution energy distribution map of 1000*256 dimensions was reduced to a low-resolution energy distribution map of 100*50 dimensions. The energy distribution features were extracted during the dimensionality reduction. Black represents 0 energy, and the brighter the color, the higher the energy distribution at that point.
[0038] Figure 10 This is a comparison diagram of the energy distribution patterns in the unified-scale time-frequency space of the first group, which consists of healthy individuals and heart failure patients.
[0039] Figure 11 The second group is a comparison of the uniform-scale time-frequency spatial energy distribution patterns of the same healthy person before and after exercise.
[0040] Figure 12 Images are generated from the raw time-frequency domain data of three cardiac cycles, after frequency standardization and time normalization, but without energy normalization.
[0041] Figure 13 To Figure 12 The image is obtained by fusing the three cardiac cycles and then normalizing the energy.
[0042] Figure 14 The original time-frequency domain data of 20 cardiac cycles were first standardized by frequency, normalized by time, fused, and then normalized by energy to obtain the final image. Detailed Implementation
[0043] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.
[0044] likeFigure 1 、 Figure 2 、 Figure 3 The upper half shows a direct plot of the raw time-frequency domain data of the heart sound signal of a healthy person for three consecutive heart cycles. The plot gives us the hint that the frequency energy of each heart cycle is similar in shape in the raw time-frequency space of the direct plot of the raw time-frequency data, but different between each heart cycle, that is, the time-frequency energy distribution pattern of each heart cycle is an original and unique "heart beat snapshot". These "heart beat snapshots" are understandable to clinicians because they can be visualized within a heart cycle and clearly see the shape and duration of the systolic and diastolic frequency energy, which is perfectly understandable from the perspective of cardiac physiology and pathology. Figure 1 、 Figure 2 、 Figure 3 The plot shows more essential time-frequency domain characteristics of the heart sound signal. The horizontal axis is the raw time of a heart cycle (not the same for each heart cycle), and the vertical axis is the raw frequency obtained by time-frequency calculation (not the same for each heart cycle). The light and dark of the color block represents the absolute value of the frequency energy (not the same for each heart cycle, black represents energy 0, and the brighter the color, the higher the energy at that point). It reveals which frequency components exist at each time point of the heart beat and the absolute value of the energy of the frequency component.
[0045] However, Figure 1 、 Figure 2 、 Figure 3 The main drawback of the raw time-frequency space data shown is that the heart sound time-frequency main part looks similar but not exactly the same. The most serious defect of the raw time-frequency data matrix is that it obtains high-resolution time-frequency space information with perfect understandability, but these original and unique "heart beat snapshots" cannot be accurately quantified and compared, and cannot provide accurate quantitative information of various detailed components of the time-frequency space. Doctors can only get subjective impressions and feelings by reading such "heart beat snapshots", but cannot extract accurate and higher-level objective regularities that can be understood in the context of cardiac physiology and pathology. In addition, Figure 1 、 Figure 2 、 Figure 3 Even if the heart sound time-frequency domain data is combined with electrocardiogram, although the division of the heart cycle can be understood, the doctor still cannot clearly understand the similarities and differences of the heart sound energy distribution between different heart cycles, and cannot achieve accurate evaluation. Therefore, further research is needed on the raw time-frequency data matrix obtained from the heart machine movement signal to dig out the "universal regularity, understandability and explainability" of the health and clinical information behind the high-resolution but raw time-frequency data matrix.
[0046] After spending a lot of time observing the original time-frequency energy distribution patterns of the heart mechanical movement signals of a large number of healthy people and heart failure patients, we proposed a scientific hypothesis: the energy distribution of the heart mechanical movement signals in the time-frequency space is regular, and the energy distribution pattern may be related to the heart function status, and the change of the heart function status will be accompanied by the change of the energy distribution in the time-frequency space. The first technical problem to be solved for the proof of this scientific hypothesis is how to extract the energy distribution characteristics in the time-frequency space with universal regularity from many original, different and unique "heart beat snapshots"? The proposal of this scientific hypothesis and its solution have never been reported in the literature.
[0047] In fact, we did not initially form the concept of "distribution pattern of heart mechanical movement energy in a unified scale time-frequency space" which has never been in the history of research. We chose the technical route that researchers would use in the early research stage, that is, artificial intelligence algorithm, to process the original time-frequency data with intelligent algorithm, and tried methods such as random forest, principal component analysis algorithm (PCA), XGboost, deep learning, etc., and obtained good classification effect. But we encountered the most problems in the process of communicating with clinicians many times, that is, why can we classify? How to establish the correlation with the physiological and pathological process of heart patients in clinical practice? This reflects the contradiction between "powerful but black box intelligent algorithm" and "clinical demand for explainability". For example, we tried the PCA method, trying to extract clear and structured "feature components" from the original time-frequency data (i.e. as shown in Figs. Figure 1 ,2,3). The PCA method can indeed extract many "feature components", although it can achieve good classification effect, but these feature components are completely not interpretable, and cannot establish understandable logic with the physiological and pathological process of heart.
[0048] To solve the strong demand for explainability, we must temporarily abandon the powerful machine learning, deep learning path, and repeatedly study the "understandable but not computable original time-frequency data". We cannot abandon the perfect understandability of the heart cycle. After repeated thinking, we propose whether we can average the time-frequency data of multiple heart cycles, and what will happen? But this is obviously not reported in the literature, and it also faces the key problem of how to solve the inconsistency of several scales between multiple original and unique "heart beat snapshots" that cannot be calculated: the time scale, frequency scale, and energy scale of each heart cycle are inconsistent. As shown in the energy distribution map of the three heart cycles, it can be seen that the time scale of each heart cycle is inconsistent (the R-R interval is absolutely not equal), the frequency scale is also inconsistent (in fact, the number and value of frequency points obtained by wavelet transform of each heart cycle are also different), and the energy scale is also inconsistent (the total energy of each heart cycle is inconsistent, and the total energy between different individuals is also very different).
[0049] After repeated thinking, we believe that this solution must be able to cross the heart cycle and individual differences to extract common features from multiple original and unique "heart beat snapshots".
[0050] As shown in Figure 4 , Figure 5 , Figure 6 To solve the problem of inconsistency of the above three scales, we perform a series of data processing methods on the original time-frequency domain data. The original time-frequency domain data of each specific heart cycle is converted into a standardized energy matrix in a time scale (time normalization), frequency scale (frequency standardization), and energy scale (energy normalization) time-frequency space, so that common features can be extracted and used for fair comparison between different heart states and different individuals.
[0051] The most important point is that after such data processing, the perfect understandability of the original time-frequency data is preserved, but the ability to standardize, unify, and accurately calculate across heart cycles is obtained, which cannot be matched by conventional artificial intelligence technology path.
[0052] As shown in Figure 7 Once the energy distribution in the unified scale time-frequency space (256*1000 dimension matrix) is obtained, the ability to quantitatively calculate multiple heart cycles can be obtained, such as averaging, or summing, or filtering through threshold after summing, etc. Fusion operations, through the accumulation of multiple heart cycle energy in the unified scale space, a stable "energy distribution pattern" is obtained.
[0053] The energy distribution pattern is aligned with human visual cognition and clinical thinking of heart physiology and pathology, and has extremely strong interpretability, which cannot be matched by conventional artificial intelligence technology path. Of course, various dimension reduction and feature extraction operations (such as gridding) can be performed on the energy distribution characteristics, which can ensure the interpretability of subsequent operations from the bottom.
[0054] At this time, the original heart shock time-frequency data that cannot objectively and clearly reflect the mechanical movement state of the heart is converted into an energy distribution pattern in a unified scale time-frequency space through the above series of data processing. The frequency energy distribution pattern in such a unified scale time-frequency space can be directly evaluated by a doctor reading the graph, and can be further quantitatively calculated and feature extracted to realize digital comparison of different heart states, and the whole process of data change has interpretability.
[0055] As shown in Figure 8 , when the number of fused cardiac cycles increases, for example, from 3 cardiac cycles of Figure 7 to 20 cardiac cycles of Figure 8 , the energy time-frequency domain distribution pattern of the mechanical activity of the heart in this state is more embodied than the number of a few unified scale energy distributions. Moreover, averaging multiple cardiac cycles is more helpful to eliminate small noise and physiological jitter and obtain more stable energy distribution characteristics.
[0056] As shown in Figure 9 , the 256,000-dimensional data of the horizontal coordinate 1000*vertical coordinate 256 shown in Figure 8 is reduced and feature extracted by gridding to 5,000-dimensional data of the horizontal coordinate 100*vertical coordinate 50.
[0057] Figure 10 The comparison chart of the gridded time-frequency space distribution patterns of the first group of different people is shown. The left chart is the distribution pattern of the heart shock frequency energy component of a healthy person in a unified scale time-frequency space, and the right chart is the distribution pattern of the heart shock energy component of a heart failure patient in a unified scale time-frequency space. It can be seen very clearly that the distribution patterns of the healthy heart and the heart failure heart are significantly different, and are very easy to understand: The left chart is the healthy heart (25 cardiac cycles averaged). The contraction phase of the shock frequency energy is regular, and the main frequency energy is completed within 15% of the time after the R wave, and the energy distribution is conservative and clear. The diastolic phase of the shock frequency energy occurs in the time window of about 45-50% of the cardiac cycle, and the frequency component is higher, but the energy is low, and the energy distribution is also conservative and clear. It shows that the healthy heart is "clean and efficient", and the overall coordination of myocardial movement is good, without unnecessary abnormal movement.
[0058] The right graph is the energy dispersion of the systolic phase of the heart (31 cardiac cycles average), the main frequency energy appears much later after the R wave, reflecting the delay of myocardial contraction, and the low efficiency of contraction. There is a dispersed mid-frequency energy distribution in the whole cardiac cycle, including the systolic phase and the diastolic phase, indicating the disorder of heart movement, reflecting the poor overall coordination of heart movement in heart failure, and a large amount of abnormal movement that is disorderly and ineffective.
[0059] Obviously, the distribution pattern of the heart shock frequency energy in the unified scale time-frequency space has strong comparability, interpretability, and the ability to distinguish the state of the heart.
[0060] Figure 11 The comparison chart of the grid-based time-frequency space distribution pattern of the same healthy person in the second group, the left graph is the distribution pattern of the heart shock frequency energy component of a healthy person in a quiet state in the unified scale time-frequency space, and the right graph is the distribution pattern of the heart shock frequency energy component of the healthy person after walking fast in the unified scale time-frequency space. It can be seen obviously that: The left graph is the distribution pattern of the heart shock frequency energy component in the unified scale time-frequency space in a quiet state (54 cardiac cycles average), which also conforms to the characteristics of the healthy person mentioned above. The right graph (40 cardiac cycles average) is the distribution pattern of the heart shock frequency energy component in the unified scale time-frequency space in the "forced" working state after walking fast (increased heart movement load). The distribution pattern still maintains the regular, clear, and conservative characteristics of the healthy person, but the frequency energy distribution of the systolic phase is significantly increased, indicating that under the movement load, the heart needs to work "hard" to meet the movement load demand of the body.
[0061] The processing method of the present application will be described in more detail below.
[0062] According to the heart mechanical movement signal processing system provided by the present application, the system comprises: Module M1: Obtain the time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals, for example, obtain the time-frequency domain data of a heart shock signal of 1 cardiac cycle: the time-frequency domain data of an example is a data table of 451*242 ("individual snapshot"), in which the column represents "time", and each column corresponds to an extremely short instant, for example, the first column is 0.002 seconds, the second column is 0.004 seconds, the third column is 0.006 seconds, and so on until the end of the time window of the cardiac cycle, a total of 451 time points; the row of the data table represents "frequency", and each row corresponds to a specific vibration frequency, for example, the first row is 5.1 Hz, the second row is 5.8 Hz, the third row is 6.7 Hz, and so on until the highest frequency (such as 100 Hz) of the filter cutoff, a total of 242 frequency points; the value in the data table represents "energy", for example, the number in each cell in the table represents how strong the energy (or amplitude) of the heart shock signal is at the time point corresponding to the column and the frequency corresponding to the row. The time-frequency data visualization is shown in the upper picture of Figure 1 The horizontal axis reflects the time points in the time-frequency data table, the vertical axis reflects the frequency points in the time-frequency data table, and the color reflects the size of the real energy value in the time-frequency data table); the time points, frequency points and energy values in the time-frequency data table of each cardiac cycle are all different. The time-frequency domain data of the cardiac mechanical movement signal is divided according to the set period, and is normalized in time and energy, and is frequency standardized, forming a plurality of sets of unified scale time-frequency domain data; the cardiac mechanical movement signal includes a heart shock signal and / or a heart sound signal. The plurality of sets of unified scale time-frequency domain data are visualized data. In the distribution pattern of the visualized unified scale time-frequency space, the horizontal axis represents the normalized time, the vertical axis represents the standardized frequency, and the brightness of the point or point cloud reflects the distribution pattern of the normalized energy of the cardiac mechanical movement.
[0063] In the time normalization process, for example, the time axis of 451 time points of the aforementioned data matrix of 451*242 is "stretched" to a preset standard length of 1000 points. Using one-dimensional interpolation method, 549 new points are calculated and "inserted" between the 451 points of each row, and finally 1000 points are obtained. Each row is operated in the same way, so that the time axis of each cardiac cycle has been normalized, and the "length" is now 1000 points regardless of the original length.
[0064] In the frequency standardization process, for example, the aforementioned 242 non-uniformly distributed frequency points are mapped onto a pre-set, standard, uniformly distributed frequency axis, for example, 256 points. Using a one-dimensional interpolation method, according to the values of the 242 frequency points in each column, the energy values at the new 256 standard frequency points are estimated. Each column is operated in the same way, so that the frequency axis of each cardiac cycle is standardized, regardless of the original number of frequency points, which is now 256 standard frequency points.
[0065] In this way, the original time-frequency data of each different cardiac cycle is converted into a new 256*1000 matrix.
[0066] In the energy normalization process, for example, the sum of all 256,000 energy values in the obtained 256*1000 matrix is E_total, and then each energy value in the matrix is divided by E_total to obtain the final energy distribution in the uniform scale time-frequency space. Now this matrix has a very important property: the sum of the energy of all 256,000 points is equal to 1, that is, the brightness of each point is converted from the original actual energy value it represents to the proportion of the total energy sum of 1. Now each value in the matrix can be understood as an "energy percentage", that is, what proportion of the total energy of this cardiac cycle is distributed at this particular time-frequency point.
[0067] In this way, the original unique 271*541 dimensional "individual snapshot" has been converted into an energy distribution in a 256*1000 dimensional uniform scale time-frequency space, which is a "standard digital fingerprint" that can be fairly compared with any other heart beat.
[0068] Module M2: fuse multiple sets of uniform scale time-frequency domain data to obtain a distribution pattern of cardiac mechanical movement energy in a uniform scale time-frequency space; the distribution pattern of cardiac mechanical movement energy in a uniform scale time-frequency space includes a visualized time-frequency space distribution pattern; the fusion includes averaging or superimposing and then filtering by threshold.
[0069] Module M3: dimension reduction and / or feature extraction is performed on the distribution pattern of cardiac mechanical movement energy in a uniform scale time-frequency space to obtain a processed distribution pattern of cardiac mechanical movement energy in a uniform scale time-frequency space; the dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
[0070] One instance of the grid, after obtaining the aforementioned 1000*256 dimension unified scale time-frequency space energy distribution, further virtually covers a 5000 dimension grid with a time axis of 100 and a frequency axis of 50, each grid in the 5000 dimension grid covers a corresponding number of small grids in the 256*1000 matrix, and the energy values of the number of small grids are averaged to obtain the energy value of each grid in the 5000 dimension grid. Each grid no longer represents a microscopic, high-resolution energy value, but represents a macroscopic average energy in a time-frequency region.
[0071] Module M4: according to the energy of the heart mechanical movement in the unified scale time-frequency space distribution mode or according to the energy of the processed heart mechanical movement in the unified scale time-frequency space distribution mode, the heart function state is evaluated; or, after dimension reduction and / or feature extraction, further data analysis is performed to evaluate the heart function state. Among them, the data analysis, for example, inputting the data after dimension reduction and / or feature extraction into the trained deep neural network model to obtain the heart function state; again, for example, comparing the data after dimension reduction and / or feature extraction with the template, the heart function state can also be obtained. Further, the samples for training the deep neural network model can come from healthy people, disease people, and their own history. The template comparison method can be compared with the unified scale time-frequency space distribution mode and / or data after dimension reduction and / or feature extraction of healthy people, disease people or their own history.
[0072] The module M1 comprises: Module M1.1: acquiring synchronous acquisition of heart mechanical movement signal and electrocardio signal; Module M1.2: processing the heart mechanical movement signal into time-frequency domain data of the heart mechanical movement signal; Module M1.3: dividing the time-frequency domain data of the heart mechanical movement signal according to the set period according to the electrocardio signal; The set period is the heart cycle obtained by separating the R peak of the electrocardio signal. The heart vibration signal in the heart mechanical movement signal is filtered at 2-100Hz, and the heart sound signal is filtered at 20-1000Hz; wavelet transform or synchronous compression wavelet transform or S-transform or time redistribution multi-synchronous extrusion transform is used to calculate the heart mechanical movement signal to obtain the time-frequency domain data of the heart mechanical movement signal.
[0073] In one variant, the change is that module M3' and module M2' replace module M2 and module M3. In this variant, the heart mechanical movement signal processing system comprises: Module M1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; dividing the time-frequency domain data of the cardiac mechanical movement signals according to the set periods and performing normalization processing according to time and energy to form a plurality of sets of time-frequency domain data of uniform scales; Module M3': performing dimension reduction and / or feature extraction on the plurality of sets of time-frequency domain data of uniform scales to obtain a distribution pattern of energy of cardiac mechanical movement in a time-frequency space of uniform scales; The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network. Module M2': fusing the plurality of sets of time-frequency domain data of uniform scales after the dimension reduction and / or feature extraction to obtain a distribution pattern of processed energy of cardiac mechanical movement in a time-frequency space of uniform scales. Module M4: evaluating a cardiac function state according to the distribution pattern of energy of cardiac mechanical movement in a time-frequency space of uniform scales or according to the distribution pattern of processed energy of cardiac mechanical movement in a time-frequency space of uniform scales; or further performing data analysis after the dimension reduction and / or feature extraction to evaluate the cardiac function state.
[0074] The application also provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the steps of the cardiac mechanical movement signal processing method, wherein the cardiac mechanical movement signal processing method comprises: Step S1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; dividing the time-frequency domain data of the cardiac mechanical movement signals according to the set periods and performing normalization processing according to time and energy to form a plurality of sets of time-frequency domain data of uniform scales; the cardiac mechanical movement signals include heart shock signals and / or heart sound signals.
[0075] Step S2: fusing the plurality of sets of time-frequency domain data of uniform scales to obtain a distribution pattern of energy of cardiac mechanical movement in a time-frequency space of uniform scales.
[0076] Step S3: performing dimension reduction and / or feature extraction on the distribution pattern of energy of cardiac mechanical movement in a time-frequency space of uniform scales to obtain a distribution pattern of processed energy of cardiac mechanical movement in a time-frequency space of uniform scales, and the dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
[0077] Step S4: evaluating the cardiac function state according to the energy distribution pattern of the cardiac mechanical movement in the uniform scale time-frequency space or according to the energy distribution pattern of the processed cardiac mechanical movement in the uniform scale time-frequency space; or further performing data analysis after dimension reduction and / or feature extraction to evaluate the cardiac function state.
[0078] The step S1 divides the time-frequency domain data of the cardiac mechanical movement signal according to the set period, and performs time normalization processing and energy normalization processing, and frequency standardization processing to form multiple sets of uniform scale time-frequency domain data.
[0079] The step S1 includes: step S1.1: acquiring the synchronously collected cardiac mechanical movement signal and electrocardiogram signal; step S1.2: processing the cardiac mechanical movement signal into time-frequency domain data of the cardiac mechanical movement signal; and step S1.3: dividing the time-frequency domain data of the cardiac mechanical movement signal according to the set period according to the electrocardiogram signal.
[0080] In one variant, steps S3' and S2' are used to replace steps S2 and S3. In this variant, the cardiac mechanical movement signal processing method includes: Step S1: acquiring time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; dividing the time-frequency domain data of the cardiac mechanical movement signals according to the set period, and performing time normalization processing and energy normalization processing to form multiple sets of uniform scale time-frequency domain data; Step S3': respectively performing dimension reduction and / or feature extraction on the multiple sets of uniform scale time-frequency domain data to obtain an energy distribution pattern of the cardiac mechanical movement in the uniform scale time-frequency space; The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network; Step S2': fusing the multiple sets of uniform scale time-frequency domain data after dimension reduction and / or feature extraction to obtain a processed energy distribution pattern of the cardiac mechanical movement in the uniform scale time-frequency space; Step S4: evaluating the cardiac function state according to the energy distribution pattern of the cardiac mechanical movement in the uniform scale time-frequency space or according to the energy distribution pattern of the processed cardiac mechanical movement in the uniform scale time-frequency space; or further performing data analysis after dimension reduction and / or feature extraction to evaluate the cardiac function state.
[0081] The application further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for processing a cardiac mechanical motion signal, wherein the method for processing a cardiac mechanical motion signal comprises: Step S1: acquiring time-frequency domain data of a plurality of cardiac mechanical motion signals in a set period; dividing the time-frequency domain data of the cardiac mechanical motion signals according to the set period, and performing normalization processing according to time and energy normalization processing to form a plurality of sets of time-frequency domain data of uniform scale; the cardiac mechanical motion signals comprise heart shock signals and / or heart sound signals.
[0082] Step S2: fusing the plurality of sets of time-frequency domain data of uniform scale to obtain a distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale.
[0083] Step S3: performing dimension reduction and / or feature extraction on the distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale to obtain a processed distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale, wherein the dimension reduction and / or feature extraction comprises gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
[0084] Step S4: evaluating a cardiac function state according to the distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale or according to the processed distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale; or, after the dimension reduction and / or feature extraction, further performing data analysis to evaluate the cardiac function state.
[0085] In the step S1, the time-frequency domain data of the cardiac mechanical motion signals are divided according to the set period, and normalized according to time and energy normalization processing, and frequency standardization processing is performed to form a plurality of sets of time-frequency domain data of uniform scale.
[0086] The step S1 comprises: step S1.1: acquiring synchronously collected cardiac mechanical motion signals and electrocardiogram signals; step S1.2: processing the cardiac mechanical motion signals into time-frequency domain data of the cardiac mechanical motion signals; and step S1.3: dividing the time-frequency domain data of the cardiac mechanical motion signals according to the set period based on the electrocardiogram signals.
[0087] In one variant, the change is that steps S3' and S2' are used to replace steps S2 and S3. In this variant, the method for processing a cardiac mechanical motion signal comprises: Step S1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; dividing the time-frequency domain data of the cardiac mechanical movement signals according to the set periods and performing normalization processing according to time and energy, to form a plurality of sets of time-frequency domain data of uniform scales; Step S3': performing dimension reduction and / or feature extraction on the plurality of sets of time-frequency domain data of uniform scales, respectively, to obtain a distribution pattern of energy of cardiac mechanical movement in a time-frequency space of uniform scales; The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network. Step S2': fusing the plurality of sets of time-frequency domain data of uniform scales after dimension reduction and / or feature extraction, to obtain a distribution pattern of energy of processed cardiac mechanical movement in a time-frequency space of uniform scales; Step S4: evaluating a cardiac function state according to the distribution pattern of energy of cardiac mechanical movement in a time-frequency space of uniform scales or according to the distribution pattern of energy of processed cardiac mechanical movement in a time-frequency space of uniform scales; or, further performing data analysis after dimension reduction and / or feature extraction, to evaluate the cardiac function state.
[0088] The cardiac mechanical movement signal processing system can be implemented by performing the flow steps of the cardiac mechanical movement signal processing method, that is, the cardiac mechanical movement signal processing method can be understood by those skilled in the art as a preferred embodiment of the cardiac mechanical movement signal processing system.
[0089] In more embodiments, the cardiac mechanical movement signal processing system further comprises providing cardiac mechanical time-frequency energy distribution information according to time and / or frequency to a user. Alternatively, the cardiac mechanical movement signal processing system comprises obtaining time-frequency domain data of a set period of cardiac mechanical movement signals, and providing cardiac mechanical time-frequency energy distribution information according to time and / or frequency to a user according to the time-frequency domain data.
[0090] For example, the cardiac mechanical time-frequency energy distribution information according to time and / or frequency can be: calculating an original energy value or an energy normalized percentage T200 in the first 20% of a cardiac cycle, calculating an original energy value or an energy normalized percentage T300 in the first 30% of a cardiac cycle, obtaining T200 / T300, and using the ratio to represent the "time concentration degree" of the energy in the systolic period, that is, what percentage of the energy in the systolic period occurs in the early and middle systolic period (the first 20% of a cardiac cycle).
[0091] For example, the distribution information of the mechanical heart time-frequency domain energy according to time and / or frequency can be: calculating the original energy value or the percentage of energy normalization in the time-frequency region surrounded by 30%-100% of the cardiac cycle time and 2-10Hz frequency, or further calculating the ratio of the original energy value or the percentage of energy normalization to the number of pixels occupied by the energy, to represent the low-frequency energy distribution characteristics at the end of systole and diastole in the cardiac cycle.
[0092] For example, the distribution information of the mechanical heart time-frequency domain energy according to time and / or frequency can be: calculating the original energy value or the percentage of energy normalization in the time-frequency region surrounded by 30-60% of the cardiac cycle time and 10-100Hz frequency, or further calculating the ratio of the original energy value or the percentage of energy normalization to the number of pixels occupied by the energy, to represent the high-frequency energy distribution characteristics in the diastolic period in the cardiac cycle.
[0093] The distribution information of the mechanical heart time-frequency domain energy according to time and / or frequency can be provided by the data itself, or by a graphical image based on the data.
[0094] The present application can provide energy distribution information by ratio, or by gradient, neural network, etc. The processing object can be data in a time period and / or a frequency period, or a graphical image corresponding to the data. The processing object can also be a certain time and / or frequency, such as the ratio of the energy of two time frequencies.
[0095] Those skilled in the art know that, in addition to implementing the system and each device, module and unit thereof provided by the present application in the form of pure computer readable program code, the same functions can also be achieved by logically programming the method steps to form a system and each device, module and unit thereof provided by the present application in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers, etc. Therefore, the system and each device, module and unit thereof provided by the present application can be considered as a hardware component, and the devices, modules and units included therein for achieving various functions can also be considered as structures within the hardware component. The devices, modules and units for achieving various functions can also be considered as both software modules for implementing methods and structures within hardware components.
[0096] More variations of the present application will be described below.
[0097] In one variation, the mechanical heart motion signal processing system comprises: Module M1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals; dividing the time-frequency domain data of the cardiac mechanical motion signals according to the set periods and performing normalization processing according to time, fusing and then performing energy normalization processing to form a set of time-frequency domain data of uniform scale, such as Figure 13 This set of data shows, or as Figure 14 This set of data shows; Module M200: performing dimension reduction and / or feature extraction on the distribution pattern of the energy of the cardiac mechanical motion constituted by the time-frequency domain data of uniform scale in the time-frequency space of uniform scale to obtain the processed distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale. The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
[0098] In module M1, before the energy normalization processing, the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals are fused, wherein the fusion includes averaging or superimposing and then filtering by threshold. Specifically, in module M1: the time-frequency domain data of the cardiac mechanical motion signals are divided according to the set periods and normalized according to time and energy normalization processing, and frequency standardization processing to form a set of time-frequency domain data of uniform scale; wherein before the energy normalization processing, the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals are fused, wherein the fusion includes averaging or superimposing and then filtering by threshold.
[0099] More specifically, the fusion is performed in module M1. For example, the first kind is to obtain the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals normalized according to time, then fuse, and then perform energy normalization processing. For example, the second kind is to obtain the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals normalized according to time, then perform energy normalization processing, and then fuse. Whether it is the first kind or the second kind, a set of time-frequency domain data of uniform scale is obtained as the distribution pattern of the energy of the cardiac mechanical motion in the time-frequency space of uniform scale.
[0100] As shown in Figure 12 , the time-frequency domain data of 3 cardiac cycles of cardiac mechanical motion signals normalized according to time are obtained first. Then, the time-frequency domain data of the 3 cardiac cycles of cardiac mechanical motion signals normalized according to time are fused, for example, averaged, and then energy normalized to obtain the image as shown in Figure 13 As shown in Figure 14 , the processing method is the same as Figure 12The processing method of the cardiac mechanical motion signal is the same, except that the time-frequency domain data of the cardiac mechanical motion signal of 20 cardiac cycles is fused.
[0101] In another variant, the processing method of the cardiac mechanical motion signal comprises: Step S1: Obtain time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals; divide the time-frequency domain data of the cardiac mechanical motion signals according to the set period and normalize according to time, fuse, and then normalize the energy to form a set of uniform scale time-frequency domain data; Step S200: Dimension reduction and / or feature extraction are performed on the distribution pattern of the energy of the cardiac mechanical motion in the uniform scale time-frequency space formed by the uniform scale time-frequency domain data to obtain the distribution pattern of the processed energy of the cardiac mechanical motion in the uniform scale time-frequency space. The dimension reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
[0102] In step S1, the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals are fused before energy normalization, wherein the fusion includes averaging or superimposing and then filtering by threshold.
[0103] More specifically, the fusion is performed in step S1. For example, the first kind is to obtain the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals normalized according to time, then fuse, and then normalize the energy. The second kind is to obtain the time-frequency domain data of a plurality of set periods of cardiac mechanical motion signals normalized according to time, then normalize the energy, and then fuse. Whether it is the first kind or the second kind, a set of uniform scale time-frequency domain data is obtained as the distribution pattern of the energy of the cardiac mechanical motion in the uniform scale time-frequency space.
[0104] The specific embodiments of the present application are described above. It should be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essential content of the present application. In the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.
Claims
1. A cardiac mechanical motion signal processing system, characterized in that, Comprising: Module M1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; normalizing the time-frequency domain data of the cardiac mechanical movement signals according to a set period and time and energy to form one or more sets of uniform scale time-frequency domain data; The cardiac mechanical movement signals include heart shock signals and / or heart sound signals.
2. The cardiac mechanical motion signal processing system of claim 1, wherein, Further comprising: Module M2: fusing the plurality of sets of uniform scale time-frequency domain data to obtain a distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space; The distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space includes a visualized time-frequency space distribution pattern; The fusion includes averaging or superimposing and then filtering through a threshold; Module M3: reducing the dimensionality and / or feature extraction of the distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space to obtain a processed distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space; The dimensionality reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network; Alternatively, the cardiac mechanical movement signal processing system, in module M1, fuses the plurality of set periods of cardiac mechanical movement signal time-frequency domain data to obtain a set of uniform scale time-frequency domain data before energy normalization, wherein the fusion includes averaging or superimposing and then filtering through a threshold; and further comprising: Module M200: reducing the dimensionality and / or feature extraction of the distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space formed by the uniform scale time-frequency domain data to obtain a processed distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space; The dimensionality reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network.
3. The cardiac mechanical motion signal processing system of claim 1, wherein, Further comprising: Module M3': reducing the dimensionality and / or feature extraction of the plurality of sets of uniform scale time-frequency domain data to obtain a distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space; The dimensionality reduction and / or feature extraction includes gridding, non-negative matrix factorization, wavelet scattering transform, convolutional autoencoder, neural network or deep neural network; Module M2': fusing the plurality of sets of uniform scale time-frequency domain data after dimensionality reduction and / or feature extraction to obtain a processed distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space; The distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space includes a visualized time-frequency space distribution pattern; The fusion includes averaging or superimposing and then filtering through a threshold.
4. The cardiac mechanical motion signal processing system of claim 2 or 3, wherein, Further comprising: Module M4: evaluating the cardiac function state according to the distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space or according to the processed distribution pattern of the energy of the cardiac mechanical movement in the uniform scale time-frequency space; or further performing data analysis after dimensionality reduction and / or feature extraction to evaluate the cardiac function state.
5. The cardiac mechanical motion signal processing system of claim 1, wherein, The multiple groups of time-frequency domain data of uniform scale include visualized time-frequency space distribution; In module M1: the time-frequency domain data of the cardiac mechanical movement signal is divided according to the set period, normalized by time and energy, and frequency standardized, forming multiple groups of time-frequency domain data of uniform scale; Or, in module M1: the time-frequency domain data of the cardiac mechanical movement signal is divided according to the set period, normalized by time and energy, and frequency standardized, forming a group of time-frequency domain data of uniform scale; wherein, before the energy normalization processing, the time-frequency domain data of the cardiac mechanical movement signal of multiple set periods is fused, wherein the fusion includes averaging or superimposing and filtering by threshold.
6. The cardiac mechanical motion signal processing system of claim 1, wherein, The module M1 includes: Module M1.1: acquiring the synchronously collected cardiac mechanical movement signal and electrocardio signal; Module M1.2: processing the cardiac mechanical movement signal into time-frequency domain data of the cardiac mechanical movement signal; Module M1.3: dividing the time-frequency domain data of the cardiac mechanical movement signal according to the set period according to the electrocardio signal; The set period is the cardiac cycle separated by the R peak of the electrocardio signal.
7. The cardiac mechanical motion signal processing system of claim 1, wherein, The cardiac mechanical movement signal is filtered at 2-100Hz for the heart shock signal and at 20-1000Hz for the heart sound signal; Wavelet transform or synchronous compression wavelet transform or S-transform or time redistribution multi-synchronous extrusion transform is used to calculate the cardiac mechanical movement signal to obtain the time-frequency domain data of the cardiac mechanical movement signal.
8. The cardiac mechanical motion signal processing system of claim 2 or 3, wherein, In the visualized distribution mode of the time-frequency space of uniform scale, the horizontal axis represents the normalized time, the vertical axis represents the standardized frequency, and the brightness of the point or point cloud reflects the distribution mode of the normalized energy of the cardiac mechanical movement.
9. The cardiac mechanical motion signal processing system of claim 1, wherein, Further comprising: Providing the user with the distribution information of the cardiac mechanical time-frequency energy by time and / or frequency.
10. A system for processing a mechanical cardiac motion signal, the system comprising: Comprising: Acquiring the time-frequency domain data of the cardiac mechanical movement signal of the set period, and providing the user with the distribution information of the cardiac mechanical time-frequency energy by time and / or frequency according to the time-frequency domain data.
11. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the cardiac mechanical movement signal processing method, wherein the cardiac mechanical movement signal processing method includes: Step S1: acquiring multiple groups of time-frequency domain data of the cardiac mechanical movement signal of the set period; dividing the time-frequency domain data of the cardiac mechanical movement signal according to the set period, normalizing by time and energy, and forming one or more groups of time-frequency domain data of uniform scale; the cardiac mechanical movement signal includes heart shock signal and / or heart sound signal; Or, comprising: acquiring the time-frequency domain data of the cardiac mechanical movement signal of the set period, and providing the user with the distribution information of the cardiac mechanical time-frequency energy by time and / or frequency according to the time-frequency domain data.
12. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is executed by the processor to realize the steps of the cardiac mechanical movement signal processing method, wherein the cardiac mechanical movement signal processing method includes: Step S1: obtaining time-frequency domain data of a plurality of set periods of cardiac mechanical movement signals; dividing the time-frequency domain data of the cardiac mechanical movement signals according to the set periods, and performing normalization processing according to time and energy, to form one or more groups of time-frequency domain data of uniform scales; the cardiac mechanical movement signals include heart shock signals and / or heart sound signals; Or, comprising: obtaining time-frequency domain data of a set period of cardiac mechanical movement signals, and providing distribution information of cardiac mechanical time-frequency energy according to time and / or frequency to a user according to the time-frequency domain data.