Electrocardiogram monitoring and data analysis system based on artificial intelligence

Abstract

The invention provides an electrocardiogram monitoring and data analysis system based on artificial intelligence, which relates to the technical field of data analysis, and comprises a data acquisition module used for connecting an electrocardiogram sensor and generating an electrocardiogram signal time sequence data set; the de-noising processing module is used for calling the joint sensor network, executing synchronous data acquisition to establish an additional data set, and performing data de-noising to establish a data de-noising result; the feature extraction module is used for calling a feature extraction network, activating time domain, frequency domain, nonlinear and deep learning feature sub-channels to perform feature extraction, and establishing a feature set; and the analysis and identification module is used for performing similarity analysis and comparison on the joint features and a preset feature library and performing data identification. The technical problems that in the prior art, electrocardiogram data analysis mostly stays at a basic feature extraction stage, comprehensive analysis capacity for multi-dimensional features is lacked, complex information in data cannot be fully mined, and consequently the accuracy and comprehensiveness of signal analysis are insufficient are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and particularly to an electrocardiogram monitoring and data analysis system based on artificial intelligence. Background Art

[0002] Modern electrocardiogram monitoring is not only limited to the hospital environment, but also widely applied to fields such as intelligent wearable devices, telemedicine, and home health management. These systems collect users' physiological data through electrocardiogram sensors and, in combination with artificial intelligence and big data analysis technologies, perform real-time processing, abnormal analysis, and risk prediction on electrocardiogram data.

[0003] However, the analysis of electrocardiogram data often involves multi-dimensional features. Existing electrocardiogram analysis technologies mostly stay at the basic feature extraction stage and lack the comprehensive analysis ability of multi-dimensional features. Traditional methods often analyze electrocardiogram signals from a single angle and cannot fully mine the complex information in the signals, resulting in insufficient comprehensiveness of the analysis results. Moreover, during the collection process of electrocardiogram data, it is easily interfered by factors such as the external environment, the sensor itself, and the user's physiological state. Existing electrocardiogram denoising technologies mostly rely on traditional filtering methods. Although they can remove noise to a certain extent, it is difficult to handle complex and variable environmental interference and physiological noise, which leads to ineffective improvement of signal quality and further affects the accuracy of subsequent data analysis. Summary of the Invention

[0004] This application provides an electrocardiogram monitoring and data analysis system based on artificial intelligence, aiming to solve the technical problems that existing electrocardiogram data analysis mostly stays at the basic feature extraction stage, lacks the comprehensive analysis ability of multi-dimensional features, cannot fully mine the complex information in the data, and leads to insufficient accuracy and comprehensiveness of signal analysis.

[0005] The present application discloses an artificial intelligence-based ECG monitoring and data analysis system, the system comprising: a data acquisition module, used to connect an ECG sensor, record the ECG signal acquired by the ECG sensor, and generate an ECG signal time series data set; a denoising processing module, used to call a joint sensor network, perform synchronous data acquisition during the ECG sensor acquisition process, establish an additional data set, the additional data set includes the user's physiological characteristic data and the acquired environmental data, perform data denoising of the ECG signal time series data set based on the additional data set, and establish a data denoising result, wherein the data denoising includes environmental denoising and physiological joint interference denoising; a feature extraction module, used to call a feature extraction network, activate the time domain feature sub-channel, frequency domain feature sub-channel, nonlinear feature sub-channel and deep learning feature sub-channel of the feature extraction network to extract features of the data denoising result, and establish a feature set; an analysis and identification module, used to perform similarity analysis and comparison between the joint features and the preset feature library after feature combination of the feature set, and perform data identification according to the similarity analysis and comparison results.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: Through the joint sensor network, the user's physiological characteristic data and environmental data are collected synchronously while the ECG sensor collects the ECG signal, which can ensure the consistency and integrity of the collected data in time, and help reduce the analysis errors caused by data asynchrony or missing; based on the collected user's physiological characteristic data and environmental data, an additional data set is established, and the ECG signal is denoised based on the additional data set, and the environmental noise and physiological joint interference are effectively removed, thereby reducing the impact of interference sources on the quality of the ECG signal, which can significantly improve the signal-to-noise ratio of the ECG signal, making the ECG signal purer and more reliable, and facilitating subsequent feature extraction and analysis; through the activation of the time domain feature sub-channel, frequency domain feature sub-channel, nonlinear feature sub-channel and deep learning feature sub-channel in the feature extraction network, it is possible to extract from the denoised ECG signal Multi-level features are extracted, including time domain features, frequency domain features, nonlinear features and deep learning features. This multi-dimensional and multi-level feature extraction method can provide more comprehensive and detailed signal features for subsequent analysis, which is helpful to deeply explore potential health information; the extracted features are combined to form a more complete joint feature, which helps to reduce the deviation that may be caused by single feature extraction, and identify the similarity with known patterns by performing similarity analysis and comparison between the joint features and the preset feature library, so as to realize pattern recognition in a big data environment and improve the intelligence and automation level of data processing; data identification is performed based on the similarity analysis and comparison results, and the newly collected ECG signals can be marked as corresponding states. This intelligent data processing method not only improves the efficiency of ECG signal analysis, but also makes the entire monitoring process more automated.

[0007] The above description is only an overview of the technical solution of the present application. In order to better understand the technical means of the present application, it can be implemented according to the content of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specific embodiments of the present application are given. Brief Description of the Drawings

[0008] Figure 1 It is a schematic structural diagram of an electrocardiogram monitoring and data analysis system based on artificial intelligence provided by an embodiment of the present application.

[0009] Figure 2 It is a schematic implementation flowchart of a denoising processing module in an electrocardiogram monitoring and data analysis system based on artificial intelligence provided by an embodiment of the present application.

[0010] Description of the reference numerals: data acquisition module 10, denoising processing module 20, feature extraction module 30, analysis and identification module 40. Detailed Embodiments

[0011] By providing an electrocardiogram monitoring and data analysis system based on artificial intelligence in an embodiment of the present application, the technical problem that electrocardiogram data analysis in the prior art mostly stays in the basic feature extraction stage, lacks the comprehensive analysis ability of multi-dimensional features, and cannot fully mine the complex information in the data, resulting in insufficient accuracy and comprehensiveness of signal analysis is solved.

[0012] After introducing the basic principle of the present application, the following will specifically introduce various non-limiting embodiments of the present application in conjunction with the drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0013] As Figure 1 shown, an embodiment of the present application provides an electrocardiogram monitoring and data analysis system based on artificial intelligence, and the system includes: A data acquisition module 10, configured to connect to an electrocardiogram sensor, record the electrocardiogram signals collected by the electrocardiogram sensor, and generate an electrocardiogram signal time series data set.

[0014] Connect the electrocardiogram (ECG) sensor. The ECG sensor consists of electrodes that are attached to the human body surface, such as the chest, to collect the electrical activity signals of the heart. The sensor has high precision and high sensitivity and can accurately capture weak ECG signals. Configure the working mode and sampling frequency of the ECG sensor to ensure that the dynamic changes of the ECG signals can be accurately captured. The ECG sensor collects the potential changes at various points on the human body surface through the electrodes. These potential changes are reflections of the electrophysiological activities during the contraction and relaxation of the heart. To obtain accurate ECG signals, multiple electrodes are used, such as a 12-lead electrocardiogram. Each electrode independently collects electrical signals from different parts, and these electrical signals reflect the electrical activities of different regions of the heart. The collected ECG signals are continuous electrical signal data and need to be sampled at certain time intervals. For example, 500 sample data are collected per second. Arrange the collected ECG signals in chronological order to form an ECG signal time series dataset.

[0015] The denoising processing module 20 is used to call the combined sensor network to perform synchronous data collection during the acquisition of the ECG sensor, establish an additional dataset. The additional dataset includes the user's physiological characteristic data and the collected environmental data, and perform data denoising on the ECG signal time series dataset based on the additional dataset to establish a data denoising result. Among them, data denoising includes environmental denoising and physiological combined interference denoising.

[0016] Call the combined sensor network. The combined sensor network is a system composed of multiple sensors that can simultaneously collect various types of data, including environmental data and user physiological data. The data collection of this network is synchronized with the signal acquisition process of the ECG sensor. Data synchronization can be achieved through hardware triggering, timestamp marking, or network protocols to ensure that multiple datasets can be collected within the same time frame.

[0017] The obtained additional dataset includes the user's physiological characteristic data and the collected environmental data. Among them, the physiological characteristic data comes from the data of other physiological sensors and is used to provide additional information about the user's physical state, such as facial expression and muscle activity data. Monitor the user's emotions or muscle activities through an electromyogram sensor or a facial expression recognition device. Especially in a resting electrocardiogram, emotional fluctuations or muscle contractions may interfere with the ECG signals; the environmental data includes the collected information related to the environment, and these data help to identify the noise caused by the external environment, such as electromagnetic environment data. Monitor the electromagnetic interference in the environment through an electromagnetic field intensity sensor, such as the electromagnetic waves generated by surrounding electronic devices. Electromagnetic noise will interfere with the quality of the ECG signals.

[0018] Based on the collected additional data set, a denoising operation is performed to clean the noise and interference in the electrocardiogram (ECG) signal time series data. Specifically, using the environmental data collected in the additional data set, such as electromagnetic field intensity, power supply fluctuations, etc., the interference caused by external environmental noise is identified and removed; by using the physiological characteristic data, the interference caused by physiological factors such as the user's emotional changes, movements, or muscle activities is identified and removed. After completing the environmental denoising and physiological combined interference denoising, a data denoising result is established, and the denoised data will be provided as input to subsequent feature extraction to improve the accuracy and reliability of electrocardiogram analysis.

[0019] The feature extraction module 30 is used to call the feature extraction network and activate the time-domain feature sub-channel, frequency-domain feature sub-channel, non-linear feature sub-channel, and deep learning feature sub-channel of the feature extraction network to extract the features of the data denoising result and establish a feature set.

[0020] Call the feature extraction network, which includes multiple different feature sub-channels. Each channel is responsible for extracting different features of the signal. The network extracts the multi-dimensional features of the signal by activating different feature sub-channels. Among them, the time-domain feature sub-channel is responsible for extracting features from the time domain (i.e., time series) of the ECG signal. The time-domain features include time-domain statistical features, instantaneous features, and periodic features. The time-domain features reflect the changes of the ECG signal on the time axis and can reveal information such as the periodicity of the heartbeat and amplitude changes; the frequency-domain feature sub-channel is used to convert the ECG signal from the time domain to the frequency domain and analyze the frequency components of the signal. The frequency-domain analysis of the ECG signal helps to capture some changes with strong periodicity or regularity, such as heart rate, the fundamental frequency of the electrocardiogram, etc.; the non-linear feature sub-channel uses non-linear analysis methods such as Poincare plots and Lyapunov exponents to extract complex dynamic features from the ECG signal. These features can reveal the complex laws of the signal, such as the non-linear response of the heart and abnormal signal patterns; the deep learning feature sub-channel uses deep neural networks, such as convolutional neural networks and long short-term memory networks, to learn the high-level features in the signal. These features can reflect the complex patterns and long-term dependence relationships of the ECG signal.

[0021] Integrate the features extracted from each feature extraction channel to establish a feature set, which is a multi-dimensional data set that comprehensively describes the denoised signal.

[0022] The analysis and identification module 40 is used to perform feature combination on the feature set, then execute the similarity analysis and comparison between the combined features and the preset feature library, and perform data identification according to the similarity analysis and comparison results.

[0023] Feature combination is to combine time-domain, frequency-domain, non-linear, and deep learning features to form a comprehensive feature vector. These features can be combined through weighted averaging or concatenation to fuse the feature information extracted from different feature channels, thereby obtaining a more comprehensive description of the signal.

[0024] Perform a similarity analysis and comparison between the combined features and a preset feature library. The preset feature library contains known and labeled electrocardiogram signal features as a reference standard. Similarity measurement methods include Euclidean distance, cosine similarity, Manhattan distance, etc. The purpose of the comparison is to determine the type of the signal by measuring the similarity between the feature set of the current signal and the known signals. According to the results of the similarity analysis, assign a label to the current electrocardiogram signal to automatically identify the data of the signal. For example, if the current signal is similar to the feature set of a healthy electrocardiogram, it is labeled as normal; if the signal is similar to the features of certain abnormal electrocardiograms, such as arrhythmia, atrial fibrillation, etc., it is labeled as the corresponding abnormal type.

[0025] Furthermore, as Figure 2 shown, the combined sensor network is called to perform synchronous data acquisition during the acquisition of electrocardiogram sensors, and an additional data set is established, including: Call the electromagnetic field intensity sensor in the combined sensor network to perform the acquisition of electromagnetic interference data in the acquisition environment and establish electromagnetic environment data; call the current and voltage sensors in the combined sensor network to perform the acquisition of device power data in the acquisition environment and establish power fluctuation data, and use the electromagnetic environment data and the power fluctuation data as environmental data; call the acceleration sensor in the combined sensor network to perform the acquisition of user's motion data and establish a motion data set; call the image acquisition sensor and electromyogram sensor in the combined sensor network to perform the acquisition of user's facial expression and muscle activity data and establish a physiological data set; use the motion data set and the physiological data set as the user's physiological feature data, and establish the additional data set with the physiological feature data and the environmental data.

[0026] The role of the electromagnetic field intensity sensor is to monitor the electromagnetic interference in the environment. The electromagnetic interference can come from device power supplies, wireless communication devices, etc., and may affect the quality of electrocardiogram signals, resulting in data noise. By measuring the intensity of the electromagnetic field per unit time with the sensor, the real-time intensity value of the electromagnetic interference is obtained as the electromagnetic environment data, which is used to describe the electromagnetic interference characteristics in the environment that may affect electrocardiogram signals.

[0027] The current-voltage sensor is used to monitor the power supply stability of the device, especially voltage fluctuations and current changes. These fluctuations may be caused by the instability of the power system, changes in device load, etc., and may affect the signals generated by the electrocardiogram sensor. The current-voltage sensor monitors the voltage and current fluctuations of the power supply in real time, collects data and generates power fluctuation data, specifically including the amplitude and frequency of voltage fluctuations, which characterize the fluctuation characteristics in the power supply; the amplitude and frequency of current fluctuations, which reflect the changes in the device's power consumption status. These fluctuations may affect the signal acquisition accuracy. The collected electromagnetic environment data and power fluctuation data are merged to form a comprehensive environmental data.

[0028] The main function of the acceleration sensor is to monitor the user's motion state. It can capture the acceleration changes of various parts of the body and obtain the user's three-dimensional motion data through a three-axis accelerometer (X, Y, Z axes). These data can reflect the user's motion direction, intensity and frequency. The motion data set composed of the data collected by the acceleration sensor in real time is used to analyze the user's motion state, and then for subsequent physiological feature recognition and noise removal, especially the pseudo-signal interference caused by motion.

[0029] An image acquisition sensor, such as a camera, is used to collect the user's facial expressions. Facial expressions are important reflections of the user's emotions, psychological states and physiological reactions. Changes in facial expressions can affect the electrocardiogram signal, especially in the case of large emotional fluctuations. By analyzing the facial images collected by the image sensor, facial expression changes, such as smiling, frowning, tension, etc., are extracted, and then the user's emotional state is identified; the electromyogram sensor is used to monitor the activity state of the user's muscles. The electromyogram sensor can identify the contraction and relaxation of muscles by sensing the electrical activity of muscles (electromyogram signals), which can reflect the motion and tension states of the user's body. The collected facial expression data and electromyogram activity data will form a physiological data set.

[0030] The motion data set and the physiological data set are merged as the user's physiological feature data. Further, the physiological feature data and the environmental data are merged to establish an additional data set. By merging these data, a comprehensive multi-source data set can be formed, which can provide rich background information for subsequent signal processing and denoising algorithms, and improve the accuracy and precision of data denoising.

[0031] Furthermore, data denoising of the electrocardiogram signal time series data set is performed based on the additional data set to establish a data denoising result, including: Perform temporal alignment on the said additional dataset and the electrocardiogram signal temporal dataset; extract environmental noise features using the environmental data in the additional dataset to establish the environmental noise feature extraction result; perform temporal backtracking analysis based on the environmental noise feature extraction result and the temporal alignment result, and establish a backtracking influence window; complete environmental denoising using the backtracking influence window, and establish a data denoising result based on the environmental denoising.

[0032] Perform temporal alignment on the additional dataset and the electrocardiogram signal temporal dataset. The goal is to align the time axes of these datasets so that all data points correspond to the same timestamp, thereby enabling synchronous processing of the data.

[0033] Environmental noise, including electromagnetic interference, power supply fluctuations, etc., will interfere with the electrocardiogram signal and affect the accuracy of the signal. Extract environmental noise features from the environmental data in the additional dataset, including extracting the spectral features, fluctuation amplitude, etc. of electromagnetic noise using the electromagnetic field intensity sensor data to obtain electromagnetic noise features, and the frequency components of electromagnetic interference can be extracted through frequency domain analysis (such as FFT transformation); use the current and voltage sensor data to extract the features of power supply fluctuations, such as the amplitude and frequency of voltage fluctuations, to help identify the impact of power supply noise on the signal. After feature extraction, establish the environmental noise feature extraction result to provide a reference for the subsequent denoising process.

[0034] Based on the extracted environmental noise features and the temporal alignment result, analyze the impact of environmental noise on the electrocardiogram signal through temporal backtracking analysis. The backtracking influence window refers to determining the time range in which the noise affects the electrocardiogram signal by analyzing the temporal relationship between the noise features and the signal. Specifically, through backtracking analysis, find out the specific impact of environmental noise features on the electrocardiogram signal within a specific time period. For example, when the environmental noise intensity exceeds a certain threshold, the electrocardiogram signal may be interfered. According to the backtracking analysis result, determine the time window that affects the signal quality, that is, the time period affected by the noise. This window reflects the duration of the impact of environmental noise on the electrocardiogram signal.

[0035] According to the time period marked by the backtracking influence window, use denoising algorithms such as filtering, weighted average, etc. to remove the environmental noise components in the electrocardiogram signal. Through the environmental denoising process, generate a data denoising result, including the electrocardiogram signal dataset after denoising, and this dataset removes the noise interference caused by environmental factors.

[0036] Furthermore, the establishing of the data denoising result based on the environmental denoising includes: Jointly perceive the user's emotions using facial expression and muscle activity data in the additional dataset, and establish the result of joint emotion perception; perform physiological joint interference denoising on the electrocardiogram signal time series dataset based on the joint emotion perception result and the time series alignment result; establish the data denoising result based on physiological joint interference denoising and environmental denoising.

[0037] The facial expression data is the change in the user's facial expression obtained through an image acquisition sensor, which is used to analyze the facial expression state, such as smiling, frowning, etc. These expression changes can reflect the user's emotional state; the muscle activity data is the user's muscle activity data obtained using an electromyogram sensor, especially the facial muscle activity related to emotional changes, such as tension, pleasure, anxiety, etc. Combine the emotion perception results of facial expressions and muscle activities to generate a comprehensive joint emotion perception result, including identifying the user's emotional state, such as happy, tense, anxious, calm, etc. This result reflects the user's emotional state at a certain moment and can specifically remove the electrocardiogram signal interference caused by emotional fluctuations.

[0038] Physiological state changes such as the user's emotional fluctuations and stress may interfere with the electrocardiogram signal. If these interferences are not removed, they will affect the accuracy of the electrocardiogram signal and the quality of subsequent analysis. Combine the obtained joint emotion perception result with the time series aligned electrocardiogram signal data to locate the specific impact of emotional changes on the electrocardiogram signal. According to the change in the emotional state, identify abnormal fluctuations in the electrocardiogram signal. For example, when the emotion changes from calm to anxious, the electrocardiogram may show short-term fluctuations or abnormal signals. Use denoising methods such as emotion-based weighted filtering and dynamically adjusting denoising parameters to remove the physiological interference caused by emotional fluctuations. Through this process, the fluctuations or abnormalities in the electrocardiogram signal caused by the user's emotions will be effectively removed, making the signal more stable and suitable for subsequent analysis.

[0039] Combine the physiological joint interference denoising result and the environmental denoising result, and further optimize them through techniques such as weighting and fusion to establish the data denoising result. This process precisely removes interference factors by jointly processing physiological and environmental data, thereby providing high-quality electrocardiogram signal data.

[0040] Furthermore, the establishment of the feature set includes: Extract time domain statistical features using the time domain feature sub-channel. The time domain statistical features include mean, standard deviation, peak value, skewness, and kurtosis; extract instantaneous features using the time domain feature sub-channel. The instantaneous features include instantaneous frequency and instantaneous amplitude; perform waveform analysis of the denoising result using the time domain feature sub-channel to establish periodic features; establish a time domain feature set with the time domain statistical features, the instantaneous features, and the periodic features, and establish the feature set according to the time domain feature set.

[0041] Through the time-domain feature sub-channel, time-domain statistical features are extracted from the original time-series data of the electrocardiogram (ECG) signal using mathematical formulas or computational methods. Time-domain statistical features refer to the statistics directly extracted from the time-series data of the signal and are used to describe the global properties of the signal, including the mean, standard deviation, peak value, skewness, and kurtosis.

[0042] Among them, the mean is used to measure the baseline value of the signal and reflects the overall offset of the signal. The mean in the ECG signal can help identify whether there is baseline drift; the standard deviation reflects the magnitude of the signal fluctuation, that is, the degree of dispersion of the signal. A signal with a larger standard deviation usually indicates that the signal changes violently, while a smaller standard deviation means that the signal is relatively stable; the peak value measures the maximum and minimum values of the signal. This feature is used to identify extreme changes in the signal. For example, the peak waves in the ECG signal, such as the P wave, R wave, etc., are used to detect abnormal events; the skewness measures the symmetry of the signal distribution and reflects whether the signal waveform tends to a certain direction. Changes in skewness can help identify whether there is asymmetry in the signal, such as whether there are abnormal waveforms; the kurtosis measures the sharpness of the signal, that is, the steepness of the signal wave peak. High kurtosis indicates large signal fluctuations, while low kurtosis indicates that the signal is relatively smooth or lacks violent changes. Abnormal fluctuations in the electrocardiogram or intense electrocardiogram activities may be manifested as higher kurtosis.

[0043] Through the time-domain feature sub-channel, instantaneous features are extracted from the original time-series data of the ECG signal using computational methods for instantaneous frequency and instantaneous amplitude, such as Hilbert transform, short-time Fourier transform, etc. Instantaneous features refer to the local change characteristics of the signal at a certain moment. These features reflect the dynamic changes of the signal, including instantaneous frequency and instantaneous amplitude. Among them, the instantaneous frequency reflects the change of the signal frequency. The change of the instantaneous frequency in the ECG signal can reveal the rapid changes in heart activities, such as sudden arrhythmias or fluctuations in the heart state; the instantaneous amplitude reflects the instantaneous change of the signal amplitude. The change of the instantaneous amplitude helps to reveal the law of amplitude fluctuations in the electrocardiogram, especially the mutation of abnormal waveforms.

[0044] The time-domain feature sub-channel is used to perform waveform analysis on the ECG signal to identify the periodic fluctuations of the signal. Based on the periodic waveforms, such as the heartbeat cycle, period features are extracted from the denoised signal. Period features describe the law of periodic changes of the signal and are particularly suitable for reflecting the repeated waveforms in the ECG signal, such as the periodic fluctuations of the heart beating. Period features include the period of the signal, the repeated pattern, and the fluctuation pattern within the period.

[0045] Integrate the extracted time-domain statistical features, instantaneous features, and period features to establish a time-domain feature set. Based on the time-domain feature set, a complete feature set is constructed. Through the analysis of the feature set, rich input information can be provided for the subsequent processing, pattern recognition, and anomaly detection of the ECG signal.

[0046] Furthermore, establishing the feature set according to the time-domain feature set includes: Performing frequency spectrum analysis on the denoising result based on the frequency-domain feature sub-channel to extract frequency-domain features; establishing a feature set according to the frequency-domain features and the time-domain features.

[0047] Frequency-domain features refer to the features extracted from the time-domain data of a signal through frequency analysis, which describe the distribution of the signal in different frequency components. Specifically, based on the frequency-domain feature sub-channel, applying the fast Fourier transform or other frequency analysis methods to the denoised electrocardiogram signal to convert the time-domain signal to the frequency domain, obtaining the frequency spectrum of the signal, and calculating frequency components through the frequency spectrum, such as power spectral density, band energy, frequency peak, etc. These frequency-domain features help to identify the contributions of different frequency components to the electrocardiogram signal and can reveal low-frequency or high-frequency abnormalities in the signal. For example, for electrocardiogram signals, frequency-domain analysis can help identify low-frequency noise, heart rate changes, and frequency components of heart diseases.

[0048] Time-domain features mainly reflect the basic shape, periodicity, and instantaneous changes of the signal, while frequency-domain features focus on the frequency distribution and energy components of the signal. Combining the two to establish a feature set can effectively complement each other's information and avoid the analysis blind spots caused by relying only on a certain type of feature.

[0049] Furthermore, establishing the feature set according to the frequency-domain features and the time-domain features includes: Configuring multiple feature extraction scales for the deep learning feature sub-channel, performing multi-scale feature extraction on the denoising result to establish a multi-scale extraction result; performing cross-channel information exchange on the multi-scale extraction result, and performing deep feature fusion according to the cross-channel information exchange result to establish a deep feature fusion result; performing secondary attention feature extraction on the denoising result based on the deep feature fusion result to establish an attention feature extraction result; outputting the attention feature extraction result and the deep feature fusion result as deep learning features; establishing a feature set according to the deep learning features, the frequency-domain features, and the time-domain features.

[0050] Multi-scale feature extraction refers to extracting features at different scales, that is, at different resolutions or different time / frequency ranges, to capture different levels and detailed information of the signal. In deep learning, multi-scale feature extraction is usually achieved through multi-scale convolutional kernels in a convolutional neural network. These convolutional kernels have different sizes or different strides and can scan the input signal at different scales to extract information at different levels.

[0051] Configure multiple feature extraction scales for the deep learning feature sub-channels, that is, configure convolutional layers of multiple scales. Each convolutional layer uses convolutional kernels of different sizes to process the input signal. For example, small convolutional kernels (such as 3x3 or 5x5) can be used to capture detailed information, and large convolutional kernels (such as 7x7 or 9x9) can be used to capture the global trend of the signal. Perform convolutional operations on the denoised signal, apply convolutional kernels of different scales respectively, extract features at different levels, and establish multi-scale extraction results.

[0052] Cross-channel information exchange refers to sharing information, interacting and collaborating among multiple feature channels in the network, so as to enhance the effect of feature learning. Specifically, for each feature channel, such as convolutional layers of different scales, the information exchange operation can be to interact the information of different channels through connection, weighted average or other methods. The purpose is to enhance the connection between channels, so that the features of each channel can provide valuable information for other channels. After cross-channel information exchange, the information of multiple channels will be fused to some extent to form a deep feature fusion result. This process helps to improve the representation ability of the model by combining deep features of multiple scales and capture more complex signal features.

[0053] Secondary attention feature extraction is to extract more important features by applying a further attention mechanism to the deep feature fusion result. The key is to emphasize the extraction of the most relevant features in the denoised signal through the attention mechanism.

[0054] Specifically, in the deep learning model, the attention mechanism usually includes two forms: spatial attention and channel attention. The spatial attention mechanism weights the features at specific spatial positions by evaluating the importance of the feature maps at different positions in space, highlighting the key parts of the signal; the channel attention mechanism evaluates the importance of the features of each channel and focuses on the channel information that is most valuable for the analysis of the electrocardiogram signal by weighting each channel. After obtaining the deep feature fusion result, apply the attention mechanism to this result, calculate the importance of each feature position or channel, and weight according to its contribution, so as to highlight the important signal features. Through secondary attention feature extraction, the attention feature extraction result is finally obtained, that is, the weighted features, which are important parts in signal analysis and can provide a more accurate signal description.

[0055] Integrate the attention feature extraction result and the deep feature fusion result to obtain a deep learning feature containing multi-dimensional and rich information. This feature combines deep features from different scales and different channels, as well as important signal features extracted through the attention mechanism.

[0056] Combining frequency-domain features, time-domain features, and deep learning features, a comprehensive feature set is finally established, which will serve as the input data for the model and provide multi-dimensional information of the electrocardiogram signal.

[0057] Furthermore, the system further includes: An early warning reporting module, which is used to perform early warning matching based on the similarity analysis comparison result, establish an early warning matching result, and perform an early warning response according to the early warning matching result.

[0058] The similarity analysis comparison result includes the similarity between the electrocardiogram signal and the known healthy state or abnormal pattern in the database. For example, if the similarity between the current signal and a certain healthy abnormal state is high, it indicates an abnormality, then an early warning is triggered. On the contrary, if the similarity is low, it indicates that the current signal is within the normal range, and no early warning needs to be triggered. According to the similarity analysis comparison result, an early warning matching result is generated, including the abnormal type, similarity score, health risk level, etc.

[0059] According to the similarity score and health risk level in the early warning matching result, different early warning response levels are set. For example, in the case of low risk, health advice is provided; in the case of high risk, emergency response measures are immediately initiated to notify doctors and patients to ensure timely handling of potential health problems.

[0060] In summary, the artificial intelligence-based electrocardiogram monitoring and data analysis system provided by the embodiments of the present application has the following technical effects: Through the joint sensor network, the user's physiological characteristic data and environmental data are collected synchronously while the ECG sensor collects the ECG signal, which can ensure the consistency and integrity of the collected data in time, and help reduce the analysis errors caused by data asynchrony or missing; based on the collected user's physiological characteristic data and environmental data, an additional data set is established, and the ECG signal is denoised based on the additional data set, and the environmental noise and physiological joint interference are effectively removed, thereby reducing the impact of interference sources on the quality of the ECG signal, which can significantly improve the signal-to-noise ratio of the ECG signal, making the ECG signal purer and more reliable, and facilitating subsequent feature extraction and analysis; through the activation of the time domain feature sub-channel, frequency domain feature sub-channel, nonlinear feature sub-channel and deep learning feature sub-channel in the feature extraction network, it is possible to extract from the denoised ECG signal Multi-level features are extracted, including time domain features, frequency domain features, nonlinear features and deep learning features. This multi-dimensional and multi-level feature extraction method can provide more comprehensive and detailed signal features for subsequent analysis, which is helpful to deeply explore potential health information; the extracted features are combined to form a more complete joint feature, which helps to reduce the deviation that may be caused by single feature extraction, and identify the similarity with known patterns by performing similarity analysis and comparison between the joint features and the preset feature library, so as to realize pattern recognition in a big data environment and improve the intelligence and automation level of data processing; data identification is performed based on the similarity analysis and comparison results, and the newly collected ECG signals can be marked as corresponding states. This intelligent data processing method not only improves the efficiency of ECG signal analysis, but also makes the entire monitoring process more automated.

[0061] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An artificial intelligence-based ECG monitoring and data analysis system, characterized in that: The system comprises: A data acquisition module, used for connecting to an electrocardiogram sensor, recording electrocardiogram signals acquired by the electrocardiogram sensor, and generating an electrocardiogram signal time series data set; A denoising processing module is used to call the joint sensor network, perform synchronous data collection during the electrocardiogram sensor collection process, establish an additional data set, the additional data set includes the user's physiological characteristic data and the collected environmental data, perform data denoising of the electrocardiogram signal time series data set based on the additional data set, and establish a data denoising result, wherein the data denoising includes environmental denoising and physiological joint interference denoising; A feature extraction module is used to call the feature extraction network, activate the time domain feature sub-channel, frequency domain feature sub-channel, nonlinear feature sub-channel and deep learning feature sub-channel of the feature extraction network to extract the features of the data denoising results, and establish a feature set; The analysis and identification module is used to perform similarity analysis and comparison between the combined features and the preset feature library after combining the features of the feature set, and to identify the data according to the similarity analysis and comparison results.

2. The artificial intelligence-based ECG monitoring and data analysis system according to claim 1, characterized in that: The calling of the joint sensor network to perform synchronous data collection during the electrocardiogram sensor collection process to establish an additional data set includes: Calling the electromagnetic field strength sensor in the joint sensor network to collect electromagnetic interference data of the collection environment and establish electromagnetic environment data; Calling the current and voltage sensors in the joint sensor network to collect power data of the equipment in the collection environment, establishing power fluctuation data, and using the electromagnetic environment data and the power fluctuation data as environment data; Calling the acceleration sensor in the joint sensor network to collect the user's motion data and establish a motion data set; Calling the image acquisition sensor and the electromyography sensor in the joint sensor network to collect the user's facial expression and muscle activity data and establish a physiological data set; The motion data set and the physiological data set are used as the user's physiological characteristic data, and the additional data set is established with the physiological characteristic data and environmental data.

3. The artificial intelligence-based ECG monitoring and data analysis system according to claim 2, characterized in that: The step of performing data denoising on the ECG signal time series data set based on the additional data set to establish a data denoising result includes: Performing time series alignment on the additional data set and the ECG signal time series data set; Using the environmental data in the additional data set to extract environmental noise features, and establishing environmental noise feature extraction results; Performing a timing backtracking analysis according to the environmental noise feature extraction result and the timing alignment result, and establishing a backtracking impact window; The environment denoising is completed using the retrospective impact window, and the data denoising result is established based on the environment denoising.

4. The artificial intelligence-based ECG monitoring and data analysis system according to claim 3, characterized in that: The step of establishing a data denoising result according to the environment denoising includes: Use facial expression and muscle activity data in the additional dataset to perform joint perception of user emotions and establish joint emotion perception results; Performing physiological joint interference denoising of the electrocardiogram signal time series data set based on the emotion joint perception result and the time series alignment result; Data denoising results are established based on physiological joint interference denoising and environmental denoising.

5. The artificial intelligence-based ECG monitoring and data analysis system according to claim 1, characterized in that: The establishing of the feature set comprises: Extracting time domain statistical features using the time domain feature subchannel, wherein the time domain statistical features include mean, standard deviation, peak, skewness, and kurtosis; Extracting instantaneous features using the time domain feature subchannel, wherein the instantaneous features include instantaneous frequency and instantaneous amplitude; Using the time domain feature sub-channel to perform waveform analysis on the denoising result and establish periodic features; A time domain feature set is established based on the time domain statistical feature, the instantaneous feature, and the periodic feature, and a feature set is established based on the time domain feature set.

6. The artificial intelligence-based ECG monitoring and data analysis system according to claim 5, characterized in that: The establishing of a feature set according to the time domain feature set comprises: Perform frequency spectrum analysis on denoising results based on frequency domain feature sub-channels to extract frequency domain features; A feature set is established according to the frequency domain features and the time domain features.

7. The artificial intelligence-based ECG monitoring and data analysis system according to claim 6, characterized in that: The establishing of a feature set according to the frequency domain features and the time domain features comprises: Configure multiple feature extraction scales of deep learning feature sub-channels, perform multi-scale feature extraction of denoising results, and establish multi-scale extraction results; Perform cross-channel information exchange on the multi-scale extraction results, perform deep feature fusion based on the cross-channel information exchange results, and establish deep feature fusion results; Perform secondary attention feature extraction on the denoising result based on the deep feature fusion result, and establish the attention feature extraction result; Output the attention feature extraction result and the deep feature fusion result as deep learning features; A feature set is established according to the deep learning features, the frequency domain features and the time domain features.

8. The artificial intelligence-based ECG monitoring and data analysis system according to claim 1, characterized in that: The system further comprises: The early warning output module is used to perform early warning matching based on similarity analysis comparison results, establish early warning matching results, and perform early warning response according to the early warning matching results.

Images