Multi-dimensional cardiac structure function data processing method and processing system

Abstract

The embodiment of the invention relates to the field of multivariable data processing, in particular to a processing method and a processing system for multi-dimensional cardiac structure function data. According to the method, the data of the dynamic electrocardiogram, the photoplethysmography and the seismocardiogram are subjected to noise reduction processing through the feedforward neural network, the data of the dynamic electrocardiogram, the photoplethysmography and the seismocardiogram after noise reduction are obtained, and the credibility of the data is higher; fusing the dynamic electrocardiogram, the photoplethysmography and the seismocardiogram of different time sequence segments, and determining characteristic parameters; combining and splicing the respectively determined characteristic parameters according to a time sequence to obtain a multi-parameter matrix; and processing the multi-parameter matrix by using a multi-modal fusion network to obtain a processed multi-dimensional vector, mutually correcting a plurality of parameters through processing, and splicing the plurality of parameters to obtain a multi-dimensional vector to obtain a high-credibility parameter which better conforms to a real physiological law.

Description

Technical Field

[0001] The embodiments of this application relate to the field of multivariate data processing, and specifically to a method and system for processing multidimensional cardiac structural and functional data. Background Technology

[0002] The statements herein are provided merely as background information in connection with this application and do not necessarily constitute prior art.

[0003] Currently, the clinical diagnosis of structural heart disease mainly relies on echocardiography. Through echocardiography, doctors can obtain information about the patient's cardiac structure and mechanical function non-invasively and directly, and assess the patient's ventricular systolic function, diastolic function, and valvular dynamics.

[0004] However, echocardiography is only suitable for detecting cardiac function in a patient at rest, and it is difficult to perform dynamic monitoring and long-term follow-up of cardiac function. Summary of the Invention

[0005] A brief overview of this application is provided below to offer a basic understanding of certain aspects thereof. It should be understood that this overview is not an exhaustive summary of the application. It is not intended to identify key or essential parts of the application, nor is it intended to limit its scope. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0006] In a first aspect, embodiments of this application provide a method for processing multi-dimensional cardiac structural and functional data, comprising the following steps: obtaining dynamic electrocardiograms, photoplethysmography (PPG), and seismograms of the heart; performing noise reduction processing on the dynamic electrocardiograms, PPG, and seismograms using a feedforward neural network to obtain time-segmented output signals of the dynamic electrocardiograms, PPG, and seismograms after noise reduction; fusing the output signals of different time segments to obtain dynamic electrocardiograms, PPG, and seismograms processed by the feedforward neural network; determining feature parameters of the dynamic electrocardiograms, PPG, and seismograms based on the obtained dynamic electrocardiograms, PPG, and seismograms; combining and concatenating the determined feature parameters in chronological order to obtain a multi-parameter matrix; and processing the multi-parameter matrix using a multimodal fusion network to obtain a processed multi-dimensional vector.

[0007] This method uses a feedforward neural network to denoise dynamic electrocardiogram (ECG), photoplethysmography (PPG), and seismogram (ESG) data, resulting in denoised data with higher reliability. It fuses dynamic ECG, PPG, and ESG data from different time segments and determines feature parameters. The determined feature parameters are then combined and concatenated in chronological order to obtain a multi-parameter matrix. A multimodal fusion network is used to process the multi-parameter matrix, obtaining a processed multi-dimensional vector. This processing allows for mutual correction between multiple parameters, and the resulting multi-dimensional vector concatenates the parameters to obtain highly reliable parameters that better reflect real physiological patterns.

[0008] Secondly, embodiments of this application also provide a multi-dimensional cardiac structural and functional data processing system, comprising the following modules: a data acquisition module configured to acquire dynamic electrocardiograms, photoplethysmography (PPG), and seismograms of the heart; a feedforward neural network configured to perform noise reduction processing on the data from the dynamic electrocardiograms, PPG, and seismograms, respectively obtaining time-segmented output signals of the denoised dynamic electrocardiograms, PPG, and seismograms; and a processor configured to receive the output signals of the feedforward neural network. The processor is configured to: fuse the output signals of different time segments to obtain dynamic electrocardiogram, photoplethysmography (PPG), and seismogram data after processing by a feedforward neural network; determine the feature parameters of the dynamic electrocardiogram, PPG, and seismogram based on the obtained data; combine and concatenate the determined feature parameters in chronological order to obtain a multi-parameter matrix; process the multi-parameter matrix using a multimodal fusion network to obtain a processed multi-dimensional vector; and output the multi-dimensional vector.

[0009] This processing system acquires dynamic electrocardiograms (ECGs), photoplethysmography (PPGs), and seismograms through a data acquisition module. A feedforward neural network denoises these data, resulting in denoised ECGs, PPGs, and seismograms with higher reliability. The system then fuses ECGs, PPGs, and seismograms from different time segments and determines feature parameters. These feature parameters are combined and concatenated in chronological order to obtain a multi-parameter matrix. A multimodal fusion network processes this matrix to obtain a multi-dimensional vector, which is then output by the processor. This processing system enables mutual correction between multiple parameters and concatenates them to obtain a multi-dimensional vector, resulting in highly reliable parameters that better reflect real physiological patterns. Attached Figure Description

[0010] Other objects and advantages of this application will become apparent from the following description of embodiments of this application with reference to the accompanying drawings, and will help to provide a comprehensive understanding of this application.

[0011] Figure 1 This is a flowchart illustrating the method for processing multi-dimensional cardiac structural and functional data provided in the embodiments of this application.

[0012] Figure 2 This is a schematic diagram of the device for processing multi-dimensional cardiac structural and functional data provided in the embodiments of this application.

[0013] Figure 3 This is a schematic diagram of the device for processing multi-dimensional cardiac structural and functional data provided in the embodiments of this application.

[0014] Figure 4 This is a schematic diagram of the device for using the multi-dimensional cardiac structural and functional data processing system provided in the embodiments of this application.

[0015] Figure 5 These are noise-reduced dynamic electrocardiograms, photoplethysmograms, and seismograms obtained by applying the multi-dimensional cardiac structural and functional data processing method provided in the embodiments of this application.

[0016] It should be noted that the accompanying drawings are not necessarily drawn to scale, but are shown only in a schematic manner without affecting the reader's understanding.

[0017] Explanation of reference numerals in the attached figures: 10. An apparatus for processing multi-dimensional cardiac structural and functional data using the embodiments of this application; 11. Main body; 111. Data acquisition module; 1111. Dynamic electrocardiogram sensor; 1112. Photoplethysmography sensor; 1113. Ectocardiogram sensor; 112. Data processing module; 113. Data storage module; 114. Power supply module; 1141. Battery; 12. Patch panel. Detailed Implementation

[0018] Exemplary embodiments of this application will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of actual implementations are described in the specification. However, it should be understood that many implementation-specific decisions must be made in the development of any such actual embodiment to achieve the developer's specific goals, such as complying with constraints related to the system and business, and these constraints may vary depending on the implementation. Furthermore, it should be understood that while development work can be very complex and time-consuming, such development work is merely a routine task for those skilled in the art who benefit from the content of this application.

[0019] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the equipment structure and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0020] Existing methods or systems for cardiac function testing can only detect one or two parameters, making it difficult to directly reflect the cardiac function of the monitored individual. For example, while Holter monitoring can monitor a user for more than 24 hours, it can only monitor heart rate and cannot quantitatively reflect the heart's pumping function; photoplethysmography (PPG) can monitor blood oxygenation and pulse rate, but cannot monitor the heart's mechanical contractile function; wearable devices for phonocardiography and seismography are sensitive to noise, resulting in poor data quality when measured in a non-resting state.

[0021] Embodiments of this application provide a method for processing multi-dimensional cardiac structural and functional data, see [link to relevant documentation]. Figure 1 , Figure 1 The diagram illustrates a flowchart of a method for processing multi-dimensional cardiac structural and functional data according to an embodiment of this application. The method includes the following steps: S10: Obtaining dynamic electrocardiogram (ECG), photoplethysmography (PPG), and seismogram of the heart, respectively; S20: Using a feedforward neural network to denoise the data from the dynamic ECG, PPG, and seismogram, obtaining the time-segmented output signals of the denoised ECG, PPG, and seismogram, respectively; S30: Fusing the output signals from different time segments. The signal is processed by a feedforward neural network to obtain dynamic electrocardiogram, photoplethysmography (PPG), and seismogram data; S40: Based on the dynamic electrocardiogram, PPG, and seismogram obtained in step S30, the feature parameters of the dynamic electrocardiogram, PPG, and seismogram are determined respectively; S50: The feature parameters determined in step S40 are combined and concatenated in chronological order to obtain a multi-parameter matrix; S60: The multi-parameter matrix is ​​processed using a multimodal fusion network to obtain a processed multidimensional vector.

[0022] The embodiments of this application employ a feedforward neural network to denoise dynamic electrocardiogram (ECG), photoplethysmography (PPG), and seismogram (ESG) data, obtaining denoised dynamic ECG, PPG, and ESG data, thus increasing the reliability of the data. Dynamic ECG, PPG, and ESG data from different time segments are fused and feature parameters are determined. The determined feature parameters are then combined and concatenated in chronological order to obtain a multi-parameter matrix. A multi-modal fusion network is used to process the multi-parameter matrix to obtain a processed multi-dimensional vector. Therefore, dynamic ECG, PPG, and ESG are fused into multi-dimensional cardiac structural and functional data, which can more comprehensively reflect the characteristics of the heart, avoiding the shortcomings of single dynamic ECG, PPG, and ESG data.

[0023] like Figure 5 As shown, Figure 5 The diagram shows noise-reduced dynamic electrocardiogram, photoplethysmography, and seismogram data obtained by applying the multi-dimensional cardiac structural and functional data processing method provided in the embodiments of this application.

[0024] In some embodiments, step S10 further includes the following steps: S11: obtaining raw signals for forming a dynamic electrocardiogram, photoplethysmography, and seismogram of the heart, respectively; S12: processing the raw signals to obtain data of the dynamic electrocardiogram, photoplethysmography, and seismogram of the heart with consistent amplitude after processing; S13: determining the dynamic electrocardiogram, photoplethysmography, and seismogram of the heart based on the processed data.

[0025] Processing raw signals from different sources to make their amplitudes consistent helps to eliminate the influence of the amplitude dimensions of signals from different sources.

[0026] In some embodiments, amplitude normalization is performed on the original signal to standardize the signal to the [-1,1] interval, which helps to eliminate the influence of amplitude dimensions of signals from different sources and makes the data stabilize faster when the data is subsequently processed using a feedforward neural network.

[0027] In some embodiments, step S20 further includes the following steps: S21: Segmenting the data of dynamic electrocardiogram, photoplethysmography (PPG), and seismogram in time sequence to obtain the dynamic electrocardiogram, PPG, and seismogram after time sequence segmentation; S22: Extracting feature vectors from the dynamic electrocardiogram, PPG, and seismogram of different time sequence segments respectively; S23: Training the dynamic convolution kernel and scaling factor of the feedforward neural network using the feature vectors to obtain the trained feedforward neural network; S24: Obtaining the output signals of the dynamic electrocardiogram, PPG, and seismogram after different time sequence segments through the feedforward neural network.

[0028] During data acquisition, noise originates from interference, baseline drift, and sensor errors. Using dynamic convolution kernels allows for adjustments to convolution parameters based on the characteristics of different temporal segments, thus better adapting to the noise characteristics of different temporal segments and enabling targeted noise reduction for better denoising results.

[0029] In some embodiments, step S22 further includes the following steps: extracting feature values ​​from dynamic electrocardiograms, photoplethysmography, and seismograms of different time segments; and obtaining a feature value vector based on the feature values. The feature values ​​include: root mean square value, peak-to-peak value, skewness, steepness, mean, predefined frequency band energy percentage, spectral kurtosis, and dominant frequency value.

[0030] In some embodiments, step S23 further includes the following steps: determining the noise type based on the feature value vectors of different time segments; determining the scaling factor based on the noise type; determining the static reference kernel; determining the weights of the static reference kernel and the dynamic convolution kernel based on the scaling factor; determining the adaptive convolution kernel based on the weights of the static reference kernel and the dynamic convolution kernel; and obtaining the trained feedforward neural network based on the adaptive convolution kernel.

[0031] During data acquisition, noise originates from interference, baseline drift, and sensor errors. By adjusting the ratio of dynamic to static convolution kernels using a scaling factor to account for different noise types, the noise reduction process can be effectively reduced while also becoming more adaptable.

[0032] In some embodiments, in step S40, based on the dynamic electrocardiogram obtained in step S30, the peaks and troughs of the Q wave, R wave, S wave, and T wave are extracted. Based on the peaks and troughs of the Q wave, R wave, S wave, and T wave, the heart rate oscillations, heart rate decay force, heart rate, time interval between adjacent R waves, duration of the Q wave, R wave, and S wave, and the time from the beginning of the Q wave, R wave, and S wave to the end of the T wave after heart rate correction are determined. Based on the isovolumetric oscillation obtained in step S30, the isovolumetric contraction initiation point, aortic valve opening point, and aortic valve closing point are extracted. Based on the isovolumetric contraction initiation point, aortic valve opening point, and aortic valve closing point, the pre-ejection period and left ventricular ejection event are determined.

[0033] Based on the photoplethysmogram obtained in step S30, the pulse wave initiation point, main wave peak, trough, and dicrotic wave point are extracted to determine the pulse wave conduction time and the slope of the rising branch.

[0034] Characteristic parameters include rate oscillation, heart rate decay force, heart rate, time interval between adjacent R waves, duration of Q waves, R waves and S waves, time from the beginning of Q waves, R waves and S waves to the end of T waves after heart rate correction, pre-ejection period, left ventricular ejection event, pulse wave conduction time, and ascending limb slope.

[0035] By extracting feature points from different images and determining the feature parameters of different images, it is helpful to use the relationship between different parameters to correct each other and to stitch the feature parameters together.

[0036] In some embodiments, taking left ventricular ejection events as an example, they are correlated with multiple extracted features. Heart rate oscillations and heart rate decay force obtained from Holter monitoring are used to assess cardiac autonomic regulation and vagal nerve activity; neurological dysfunction predicts heart failure and a decrease in left ventricular ejection events. The systolic peak amplitude obtained from seismographs directly characterizes the strength of myocardial contractility and is a strong feature positively correlated with left ventricular ejection events. Abnormal systolic oscillation frequency indicates asynchronous myocardial contraction, affecting ejection efficiency and negatively correlated with left ventricular ejection events. Pulse wave velocity obtained from photoplethysmography can indicate arteriosclerosis, which increases cardiac afterload. Long-term effects lead to a decrease in left ventricular ejection events. A longer pulse wave rise time indicates greater ejection resistance and slower velocity, negatively correlated with left ventricular ejection events. The diastolic perfusion index indirectly reflects cardiac output levels; insufficient perfusion suggests a possible decrease in left ventricular ejection events.

[0037] By understanding the relationship between feature parameters and related features, feature parameters can be corrected. The mutual corroboration of the relationships between multiple features and feature parameters can improve the accuracy of the correction.

[0038] In some embodiments, step S50 further includes the following steps: S51: determining the data quality of each parameter of the dynamic electrocardiogram, photoplethysmography, and seismogram in the multi-parameter matrix; S52: obtaining the factors affecting the parameters based on the determined quality; S53: correcting the parameters using the factors; S54: performing time-series splicing on the corrected parameters to obtain the corrected multi-parameter matrix.

[0039] It is understandable that data acquired by wearable devices may be biased due to various interference factors. By identifying the factors affecting the parameters through parameter quality analysis, and then correcting the parameters using these factors, it is possible to compensate for and correct parameters that are severely affected by noise using data from different sources. Furthermore, compared to the traditional method of simply splicing together time sequences, correcting the parameters using the factors affecting them before splicing can result in higher data reliability.

[0040] In some embodiments, step S51 further includes the following steps: S511: weighting each parameter of the previous time period's dynamic electrocardiogram, photoplethysmography, and seismogram; S512: obtaining the characteristic parameters of the current time period's dynamic electrocardiogram, photoplethysmography, and seismogram; S513: using the characteristic parameters obtained in S512 and the characteristic parameters obtained in the current measurement to determine the quality of the data.

[0041] Data on the dynamic changes of the human heart is continuous over time. By using feature parameters obtained by weighting previous time periods and feature parameters of the current time period to judge data quality, and avoiding the use of other methods for processing data mutations, the reliability of data processing can be improved.

[0042] In some embodiments, step S52 further includes the following steps: S521: determining the temporal information, morphological stability, and noise level of the dynamic electrocardiogram, photoplethysmography, and seismogram based on the quality of the data; S522: obtaining the factors affecting the parameters using the temporal information, morphological stability, and noise level of the dynamic electrocardiogram, photoplethysmography, and seismogram.

[0043] Because data on the dynamic changes of the human heart exhibit temporal continuity and certain patterns, and noise is the primary factor causing data errors, obtaining influencing factors based on temporal information, morphological stability, and noise levels avoids using a single-dimensional influencing factor for judgment. This approach better reflects the impact of temporal information, morphological stability, and noise levels on the parameters. The influencing factors obtained through this step improve the reliability of the data.

[0044] In some embodiments, in step S522, the factor conforms to the following expression: .

[0045] in , , The coefficient at the current time is determined by the parameters of the previous N periods and the mass at the current time. As a factor, This is a time-series information factor for dynamic electrocardiography, photoplethysmography, and seismograph. It serves as a stability factor for the morphology of dynamic electrocardiograms, photoplethysmography, and seismographs. This is the noise magnitude factor for dynamic electrocardiogram, photoplethysmography, and seismograph.

[0046] By using coefficients determined by the parameters of the previous N periods and the quality at the current moment, the time series information factor, the morphological stability factor, and the noise magnitude factor are fitted as factors affecting the parameters. A single factor can reflect the weights of reliable and suspicious parameters, making the factors more adaptable. Furthermore, the subsequent correction and splicing strategies for the parameters are determined based on the magnitude of the factors, resulting in higher reliability of the final data.

[0047] In some embodiments, step S53 further includes the following step: S531: determining the appropriate parameters for Holter monitoring, photoplethysmography, and cardiac oscillation. Figure 3 S532: Determine the corrected parameters based on the factor and the offset correction term.

[0048] The magnitude of the factor reflects the weights of the reliable and doubtful parameters. Correction based on offset correction terms from three different signal sources makes the parameter correction more stable and allows the data to stabilize more quickly during subsequent processing.

[0049] In some embodiments, the corrected parameters determined in step S532 conform to the following expression: .

[0050] For the offset correction terms of the three signals, These are the parameters before correction. These are the corrected parameters.

[0051] Because the data obtained from Holter monitoring, photoplethysmography, and seismography during the cardiac cycle have time-dependent, morphologically complementary, and physiologically coupled characteristics, parameters are corrected by offset correction terms of the three signals and factors of influence parameters fitted by time-series information factors, morphological stability factors, and noise magnitude factors. When a parameter is significantly affected by noise, the erroneous parameter can be corrected by parameters from other sources. The feature point localization error of parameters under the same time series is co-optimized, making the correction more stable, accurate, and reliable.

[0052] Embodiments of this application also provide a processing system for multi-dimensional cardiac structural and functional data, comprising the following modules: a data acquisition module configured to acquire dynamic electrocardiograms, photoplethysmography (PPG), and seismograms of the heart; a feedforward neural network configured to denoise the data from the dynamic electrocardiograms, PPG, and seismograms, respectively, and obtain output signals of the time-series segmented dynamic electrocardiograms, PPG, and seismograms after denoising; and a processor configured to receive the output signal from the feedforward neural network and to: fuse the output signals from different time-series segmented data to obtain the data from the dynamic electrocardiograms, PPG, and seismograms processed by the feedforward neural network; determine the feature parameters of the dynamic electrocardiograms, PPG, and seismograms based on the obtained data; combine and concatenate the determined feature parameters in chronological order to obtain a multi-parameter matrix; process the multi-parameter matrix using a multimodal fusion network to obtain a processed multi-dimensional vector; and output the multi-dimensional vector.

[0053] The system provided in this application acquires dynamic electrocardiograms (ECGs), photoplethysmography (PPGs), and seismograms through a data acquisition module. A feedforward neural network denoises the data from these three datasets, resulting in denoised ECGs, PPGs, and seismograms, thus increasing data reliability. A processor fuses ECGs, PPGs, and seismograms from different time segments and determines feature parameters. These determined feature parameters are then combined and concatenated in chronological order to obtain a multi-parameter matrix. A multi-modal fusion network processes this multi-parameter matrix to obtain a processed multi-dimensional vector, which is then output by the processor. This system integrates dynamic ECGs, PPGs, and seismograms into multi-dimensional cardiac structural and functional data. This multi-dimensional data provides a more comprehensive reflection of cardiac characteristics, avoiding the limitations of relying solely on individual ECG, PPG, or seismogram data.

[0054] In some embodiments, the processing system further includes a data storage module for storing the raw data output after denoising by the feedforward neural network and the result data output after processor processing. This configuration allows for in-system analysis and backup of the raw and result data, reducing data latency when transmitting data to the cloud for analysis and enabling faster access to analysis results.

[0055] In some embodiments, the data acquisition module is configured to: acquire raw signals for forming a dynamic electrocardiogram, photoplethysmography (PPG), and seismogram of the heart, respectively; process the raw signals to obtain processed data of the dynamic electrocardiogram, PPG, and seismogram of the heart with consistent amplitude; and determine the dynamic electrocardiogram, PPG, and seismogram of the heart based on the processed data.

[0056] Processing raw signals from different sources to make their amplitudes consistent helps to eliminate the influence of the amplitude dimensions of raw signals from different sources.

[0057] In some embodiments, the data acquisition module performs amplitude normalization processing on the original signal, standardizing the signal to the [-1,1] interval, which helps to eliminate the influence of amplitude dimensions of signals from different sources and makes the data stabilize faster when the data is subsequently processed using a feedforward neural network.

[0058] In some embodiments, the multi-dimensional cardiac structural and functional data processing system provided in this application can also collect user information and posture and motion information data. This configuration, by training the feedforward neural network with individualized user data, enables the processing system to perform individualized data denoising, resulting in better denoising performance. Simultaneously, since different users have different underlying cardiac conditions, collecting user information also helps improve the accuracy of subsequent data analysis results.

[0059] In some embodiments, the feedforward neural network is configured to: segment the data of dynamic electrocardiogram, photoplethysmography (PPG), and seismogram in time sequence to obtain the dynamic electrocardiogram, PPG, and seismogram after time segmentation; extract feature vectors from the dynamic electrocardiogram, PPG, and seismogram of different time segments respectively; train the dynamic convolution kernel and scaling factor of the feedforward neural network using the feature vectors to obtain the trained feedforward neural network; and acquire the output signals of the dynamic electrocardiogram, PPG, and seismogram after different time segments of the feedforward neural network.

[0060] During data acquisition, noise originates from interference, baseline drift, and sensor errors. Using dynamic convolution kernels allows for adjustments to convolution parameters based on the characteristics of different temporal segments, thus better adapting to the noise characteristics of different temporal segments and enabling targeted noise reduction for better denoising results.

[0061] In some embodiments, the feedforward neural network is further configured to: extract feature values ​​from dynamic electrocardiograms, photoplethysmography, and seismograms of different time segments; and obtain a feature value vector based on the feature values. The feature values ​​include: root mean square value, peak-to-peak value, skewness, steepness, mean, predefined frequency band energy percentage, spectral kurtosis, and dominant frequency value.

[0062] In some embodiments, the feedforward neural network is trained by the following steps: determining the noise type based on the feature vectors of different time segments; determining the scaling factor based on the noise type; determining the static reference kernel; determining the weights of the static reference kernel and the dynamic convolution kernel based on the scaling factor; determining the adaptive convolution kernel based on the weights of the static reference kernel and the dynamic convolution kernel; and obtaining the trained feedforward neural network based on the adaptive convolution kernel.

[0063] During data acquisition, noise originates from interference, baseline drift, and sensor errors. By adjusting the ratio of dynamic to static convolution kernels using a scaling factor to account for different noise types, the noise reduction process can be effectively reduced while also becoming more adaptable.

[0064] In some embodiments, the processor is further configured to: determine the quality of data for each parameter of the dynamic electrocardiogram, photoplethysmography, and seismogram in the multi-parameter matrix; obtain factors affecting the parameters based on the determined quality; correct the parameters using the factors; and perform time-series splicing on the corrected parameters to obtain a corrected multi-parameter matrix.

[0065] It is understandable that data acquired by wearable devices may be biased due to interference from various factors. The processor identifies the factors affecting the parameters by assessing parameter quality, and then corrects the parameters based on these factors. This allows for the compensation and correction of parameters severely affected by noise using different data sets. Furthermore, compared to the simple temporal concatenation of traditional methods, correcting the parameters using the influencing factors before concatenation results in higher data reliability.

[0066] In some embodiments, the processor is further configured to: weight each parameter of the previous time period's Holter ECG, photoplethysmography, and seismogram to obtain characteristic parameters of the current time period's Holter ECG, photoplethysmography, and seismogram, and determine the quality of the data using the obtained characteristic parameters and the currently measured characteristic parameters.

[0067] The data on dynamic changes in the heart are continuous over time. The processor is set to use the feature parameters obtained by weighting the previous time period and the feature parameters of the current time period to judge the data quality, avoiding the use of other methods used to process data mutations, which can improve the reliability of data processing.

[0068] In some embodiments, the processor is further configured to: determine the timing information, morphological stability, and noise level of the dynamic electrocardiogram, photoplethysmography, and seismogram based on the quality of the data; and obtain the factors affecting the parameters using the timing information, morphological stability, and noise level of the dynamic electrocardiogram, photoplethysmography, and seismogram.

[0069] Because cardiac dynamics data exhibit temporal continuity and certain patterns, and noise is the primary factor causing data errors, using temporal information, morphological stability, and noise levels to identify influencing factors better reflects the impact on these parameters. This aligns with the nature of medical diagnostic parameters and enhances data reliability.

[0070] The methods and systems provided in this application can be applied to devices used for cardiac monitoring. Here, we will further explain how to apply this system using an example of a device employing a multi-dimensional cardiac structural and functional data processing system based on embodiments of this application.

[0071] See Figure 2 and Figure 3 , Figure 2 This is a schematic diagram of the structure of the device 10 for processing multi-dimensional cardiac structural and functional data provided by the embodiments of this application; Figure 3 This is a schematic diagram of the principle of the device 10 for processing multi-dimensional cardiac structural and functional data provided in the embodiments of this application.

[0072] like Figure 2 As shown, the device 10 includes a body 11 and a patch 12. The body 11 is configured to acquire dynamic electrocardiogram, photoplethysmography, and seismogram data of the heart, process the data, and store the raw data and the processed data. The patch 12 is fixedly connected to the body and is used to fix the body 11 to the user's skin surface and assist the body 11 in acquiring data.

[0073] By setting the patch 12, the main body 11 can be easily fixed to the user's skin surface, and the patch 12 can make the main body 11 fit the curvature of the human skin more closely, assisting the main body 11 in collecting the user's heart data.

[0074] In some embodiments, such as Figure 3As shown, the main body 11 includes a data acquisition module 111, a data processing module 112, a data storage module 113, and a power supply module 114. The data acquisition module 111 is used to acquire dynamic electrocardiogram, photoplethysmography, and seismogram data of the heart, and transmit the data to the data processing module 112. The data processing module 112 uses a feedforward neural network and a processor in the multi-dimensional cardiac structural and functional data processing system provided in the second aspect embodiment of this application to transmit and process the data acquired by the data acquisition module, and transmits the original data after noise reduction by the feedforward neural network and the result data after processor processing to the data storage module 113. The data storage module 113 is used to store the original data and the result data. The power supply module 114 is used to supply power to the data acquisition module 111, the data processing module 112, and the data storage module 113.

[0075] By integrating the data acquisition module 111, data processing module 112, data storage module 113, and power supply module 114 into the main body 11, the device 10 can be easily used by the user without requiring additional setup. As long as the user attaches the device 10 to the designated position on the body and keeps the device 10 in the activated state, the device 10 will automatically acquire, process, and store the data.

[0076] In some embodiments, the data acquisition module 111 includes a dynamic electrocardiogram sensor 1111, a photoplethysmography (PPG) sensor 1112, and a seismograph sensor 1113; the dynamic electrocardiogram sensor 1111 is used to acquire dynamic electrocardiogram data of the heart, the PPG sensor 1112 is used to acquire PPG data, and the seismograph sensor 1113 is used to acquire seismograph data.

[0077] In some embodiments, the data processing module 112 includes an edge computing module and a data transmission module; the edge computing module is used to perform noise reduction on the data using a feedforward neural network and to fuse the data using a processor locally on the device 10 to obtain a multi-parameter matrix and a multi-dimensional vector after processing by a multi-modal fusion network, and to analyze the processed multi-dimensional vector; the data transmission module is used to transmit the processed result data and analysis results to the data storage module 113.

[0078] In some embodiments, such as Figure 3 As shown, the power supply module 114 supplies power to the data acquisition module 111, the data processing module 112, and the data storage module 113 via the battery 1141. By setting the battery 1141 to supply power to the power supply module 114, the battery 1141 can be directly replaced when the device 10's power is insufficient, avoiding the need for the device 10 to be charged due to insufficient power, which would affect data acquisition.

[0079] See Figure 4 , Figure 4 This is a schematic diagram illustrating the usage flow of the apparatus 10, which applies the multi-dimensional cardiac structural and functional data processing system provided in the embodiments of this application. Figure 4 As shown, in some embodiments, the data transmission module further includes a Bluetooth transmission module and a USB transmission module. The Bluetooth transmission module is used to connect with the mobile phone via Bluetooth and transmit the result data to the mobile APP via Bluetooth connection; the USB transmission module is used to connect with the mobile phone or computer via USB connection and transmit the raw data to the mobile APP or computer via USB connection.

[0080] Device 10 transmits the results data to a mobile app via Bluetooth, allowing users to directly view the processed data and analysis, such as real-time waveforms, parameters, and analysis reports. Users can also query historical data, enabling them to easily and intuitively obtain personal cardiac information and track their heart condition over the long term. Raw data can also be transferred to a mobile phone or computer via USB, allowing users to back up their original data.

[0081] In some embodiments, such as Figure 4 As shown, device 10 also has a cloud server. When a user uses device 10, device 10 collects user data, processes the data, and saves it locally. After the user transfers the data to a mobile app or computer, the user can also upload the data to the cloud server via the mobile app or computer. This setup, using user-uploaded data to train the edge computing module, makes data processing more reliable.

[0082] Regarding the embodiments of this application, it should also be noted that, without conflict, the embodiments of this application and the features in the embodiments can be combined with each other to obtain new embodiments.

[0083] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method of processing multi-dimensional cardiac structure function data, characterized by, It comprises the following steps: S10: respectively obtain the dynamic electrocardiogram, photoplethysmogram and seismocardiogram of the heart; S20: utilize the feedforward neural network to carry out noise reduction processing on the data of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram, and respectively obtain the output signals of the time series segmented dynamic electrocardiogram, photoplethysmogram and seismocardiogram after noise reduction processing; S30: fuse the output signals of different time series, and obtain the data of the dynamic electrocardiogram, photoplethysmogram and seismocardiogram after the feedforward neural network processing; S40: respectively determine the characteristic parameters of the dynamic electrocardiogram, photoplethysmogram and seismocardiogram according to the dynamic electrocardiogram, photoplethysmogram and seismocardiogram obtained in the S30 step; S50: combine and splice the characteristic parameters respectively determined in the S40 step in time sequence, and obtain a multi-parameter matrix; S60: utilize a multi-modal fusion network to process the multi-parameter matrix, and obtain a processed multi-dimensional vector.

2. The method of claim 1, wherein the S20 step further comprises the following steps: S21: segment the data of the dynamic electrocardiogram, photoplethysmogram and seismocardiogram in time sequence, and obtain the dynamic electrocardiogram, photoplethysmogram and seismocardiogram after time sequence segmentation; S22: respectively extract feature value vectors from the dynamic electrocardiogram, photoplethysmogram and seismocardiogram of different time series; S23: utilize the feature value vectors to train the dynamic convolution kernel and the scaling factor of the feedforward neural network, and obtain the trained feedforward neural network; S24: obtain the output signals of different time series of the dynamic electrocardiogram, photoplethysmogram and seismocardiogram after the feedforward neural network.

3. The method of claim 1, wherein in the S40 step, according to the dynamic electrocardiogram obtained in the S30 step, extract the vertex and terminal point of Q wave, the vertex and terminal point of R wave, the vertex and terminal point of S wave, and the vertex and terminal point of T wave, and determine the heart rate oscillation, heart rate damping force, heart rate, time interval between adjacent R waves, time limit of Q wave R wave S wave, and time from the start point of Q wave R wave S wave to the end point of T wave after heart rate correction according to the vertex and terminal point of Q wave, the vertex and terminal point of R wave, the vertex and terminal point of S wave, and the vertex and terminal point of T wave; according to the seismocardiogram obtained in the S30 step, extract the isovolumic contraction start point, aortic valve opening point and aortic valve closing point, and determine the pre-ejection period and left ventricular ejection event according to the isovolumic contraction start point, aortic valve opening point and aortic valve closing point; according to the photoplethysmogram obtained in the S30 step, extract the pulse wave start point, main wave peak, wave trough and double pulse wave point, and determine the pulse wave conduction time and rising slope. ​ ​ The characteristic parameters include rate oscillation, heart rate damping force, heart rate, time interval between adjacent R waves, QRS time, time from the start of QRS to the end of T wave after heart rate correction, pre-ejection period, left ventricular ejection event, pulse wave transmission time, ascending branch slope.

4. The method of claim 1, wherein, The step S50 further comprises the following steps: S51: determining the quality of data of each parameter of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram in the multi-parameter matrix; S52: obtaining a factor affecting the parameter according to the determined quality; S53: correcting the parameter by using the factor; S54: time-series splicing the corrected parameter to obtain a corrected multi-parameter matrix.

5. The method of claim 4, wherein, The step S51 further comprises the following steps: S511: weighting each parameter of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram of the previous period, S512: respectively obtaining characteristic parameters of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram of the current period, S513: determining the quality of data by using the characteristic parameters respectively obtained in S512 and the characteristic parameters obtained by current measurement.

6. The method of claim 4, wherein, The step S52 further comprises the following steps: S521: determining the time-series information, the stability of morphology and the noise size of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram according to the quality of data; S522: obtaining a factor affecting the parameter by using the time-series information, the stability of morphology and the noise size of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram.

7. The method of claim 6, wherein, In the step S522, the factor conforms to the following expression: , wherein , , is a coefficient, the current time coefficient being determined jointly by the parameters of the previous N periods and the current time quality, is a factor, a time sequence information factor of the dynamic electrocardiogram, the photoplethysmogram, and the seismocardiogram, a form stability factor of the dynamic electrocardiogram, the photoplethysmogram, and the seismocardiogram, a noise size factor of the dynamic electrocardiogram, the photoplethysmogram, and the seismocardiogram.

8. The method of claim 4, wherein, The step S53 further comprises the following steps: S531: determining a bias correction term for the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram; S532: determining the corrected parameter according to the factor and the bias correction term.

9. The method of claim 8, wherein, In the step S532, the determined corrected parameter conforms to the following expression: , is an offset correction term for the three signals, is a parameter before correction, is a parameter after correction.

10. A system for processing multi-dimensional cardiac structure function data, the system comprising: It comprises the following modules: A data acquisition module is configured to respectively obtain a dynamic electrocardiogram, a photoplethysmogram and a seismocardiogram of a heart; A feedforward neural network is configured to perform noise reduction processing on data of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram to respectively obtain output signals of time-series segmented dynamic electrocardiogram, photoplethysmogram and seismocardiogram after noise reduction processing; A processor is configured to receive the output signals of the feedforward neural network and is configured to: fuse the output signals of different time segments to obtain data of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram processed by the feedforward neural network; determine feature parameters of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram respectively according to the obtained data of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram; combine and splice the determined feature parameters in time sequence to obtain a multi-parameter matrix; process the multi-parameter matrix by using a multi-modal fusion network to obtain a processed multi-dimensional vector; the processor is further configured to output the multi-dimensional vector.

11. The system of claim 10, wherein the feedforward neural network is configured to: segment data of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram in time sequence to obtain the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram segmented in time sequence; extract feature value vectors from the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram segmented in different time sequences respectively; train dynamic convolution kernels and scaling factors of the feedforward neural network by using the feature value vectors to obtain the trained feedforward neural network; obtain output signals of different time segments of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram processed by the feedforward neural network.

12. The system of claim 10, wherein the processor is further configured to: determine quality of data of each parameter of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram in the multi-parameter matrix, obtain factors affecting the parameters according to the determined quality, correct the parameters by using the factors, perform time sequence splicing on the corrected parameters to obtain a corrected multi-parameter matrix.

13. The system of claim 12, wherein the processor is further configured to: weight each parameter of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram in a previous time period, obtain feature parameters of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram in a current time period respectively, determine the quality of the data by using the obtained feature parameters and feature parameters obtained by current measurement.

14. The system of claim 12, wherein the processor is further configured to: determine time sequence information, stability of morphology and noise size of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram according to the quality of the data, obtain factors affecting the parameters by using the time sequence information, the stability of morphology and the noise size of the dynamic electrocardiogram, the photoplethysmogram and the seismocardiogram.

Images