Heart failure monitoring and evaluating method and system

Abstract

The invention provides a heart failure monitoring and evaluating method and system, and relates to the field of medical monitoring, and the method comprises the steps: collecting feedback information from each terminal device, so as to generate a data set containing patient information, I-lead electrocardiogram waveform conditions, heart sound data and time axes corresponding to each examination accurate to milliseconds; identifying and classifying the collected data set, sorting the data, and storing the sorted data in a partitioned manner according to the type and the information of the patient; performing feature extraction from the stored image data to obtain an image data result; and carrying out data processing on the stored audio data, and respectively storing processing results to form data vectors of single check so as to obtain single check vector data. According to the invention, methods of information acquisition, data processing, machine learning and the like are integrated, and powerful support is provided for home monitoring of heart failure and general management of communities.

Description

Technical Field

[0001] The present invention relates to the field of medical monitoring technology, and in particular to a heart failure monitoring and evaluation method and system. Background Art

[0002] In traditional heart failure monitoring and evaluation, some rely mainly on the doctor's clinical experience and regular hospital examinations. The problem with this method is that its monitoring frequency is limited and it is difficult to capture real-time changes in the patient's condition.

[0003] In addition, due to the limitations of data collection and processing methods, some traditional methods are unable to fully and accurately reflect the patient's true condition. For example, the collection of key physiological information such as electrocardiograms and heart sounds may be affected by multiple factors such as the environment, equipment performance, and operator skill level, resulting in uneven data quality. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a heart failure monitoring and evaluation method and system, which integrates information collection, data processing, machine learning and other methods, and provides strong support for home monitoring of heart failure and general practice management in the community.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows: In a first aspect, a heart failure monitoring and evaluation method is provided, the method comprising: Collect feedback information from each terminal device to generate a data set containing patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds; Identify and classify the collected data sets, organize the data, and store the organized data by type based on the patient's information; Extract features from the stored image data to obtain image data results; Performing data processing on the stored audio data, and storing the processing results respectively to form data vectors of a single inspection, so as to obtain single inspection vector data; Analyze and evaluate the single inspection vector data and feed back the evaluation results to the terminal; According to the evaluation results and image data results, critical values ​​and model quality are obtained, and single inspection vector data, critical values ​​and model quality are fed back to the terminal device to realize information feedback; Based on the feedback information, a data curve is obtained to achieve clinical heart failure monitoring.

[0006] Furthermore, feedback information from each terminal device is collected to generate a data set containing patient information, I lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds, including: Collect feedback information from various terminal devices, where the terminal device is a portable examination device held by the patient or community general practitioner, and the terminal device has built-in diagnostic information collection module, user and patient information management module, data feedback and reception module, and early warning module; From the feedback information, the audio sequence, image sequence and I-lead ECG sequence are accurately processed to correspond to the moments on the time scale, and the interval duration between the sequence features of the three sequences or the features in a single sequence are processed and extracted to generate a data set containing patient information, I-lead ECG waveform conditions, heart sound data, and the time axis corresponding to each examination accurate to milliseconds.

[0007] Furthermore, the collected data sets are identified and classified, the data are sorted, and the sorted data are partitioned and stored according to the patient's information, including: According to the collected data set, the electrocardiogram waveform data and the heart sound data are classified respectively to obtain different types of data; Develop corresponding processing strategies based on the characteristics of different types of data and processing equipment; According to the data classification and processing strategy, the data sorting tasks are assigned to the corresponding processing equipment and algorithms to achieve data sorting; The sorted data is stored in partitions according to the patient's information.

[0008] Further, feature extraction is performed from the stored image data to obtain image data results, including: The stored image data is passed Perform feature extraction to obtain image data results, where: is the output feature map after the maximum pooling operation at the kth channel position The value of It's location The corresponding pooling area, In the pooling area The coordinates inside, M and N are the height and width of the convolution kernel, Is the input image or the feature map of the previous layer at position The value of is the kth convolution kernel at position The weight value of is the bias term associated with the kth convolution kernel.

[0009] Further, the stored audio data is processed, and the processing results are stored separately to form a data vector of a single inspection to obtain single inspection vector data, including: Identify all local maximum and minimum values ​​of the audio data, interpolate the local maximum and minimum values ​​to form an upper envelope and a lower envelope, calculate the mean of the upper and lower envelopes, subtract the mean from the original data signal to obtain a new data signal, and check whether the condition of the intrinsic mode function is met. If not, iterate the new data signal as the original data signal until the condition is met to obtain a number of intrinsic mode function components; According to several intrinsic mode function components, the waveform diagram of each component and its three-dimensional characteristic spectrum in time domain, frequency domain and energy domain are formed, and artificial subjective screening is performed to obtain specific stage intrinsic mode function components; The intrinsic mode function components are stored separately to form a data vector for a single inspection.

[0010] Furthermore, the single inspection vector data is analyzed and evaluated, and the evaluation results are fed back to the terminal, including: For a single check vector data pass Perform outlier detection to obtain the test results, where: is the position of a data point relative to the overall distribution of the data set; is the value of a particular data point, n is the total number of data points, k and is the index variable; The detection result is evaluated, an evaluation result is obtained, and the evaluation result is fed back to the terminal.

[0011] Furthermore, according to the evaluation results and the image data results, the critical value and the model quality are obtained, and the single inspection vector data, the critical value and the model quality are fed back to the terminal device to realize the feedback of information, including: Analyze the evaluation results and image data, identify critical values, and mark the identified critical values ​​to obtain critical values; Build a model that generates evaluation results and image data results, and perform quality assessment to obtain model quality; The single inspection vector data, critical value and model quality are fed back to the terminal device to realize information feedback.

[0012] In a second aspect, a heart failure monitoring and evaluation system includes: A data collection module is used to collect feedback information from various terminal devices to generate a data set containing patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds; The data sorting module is used to identify and classify the collected data sets, sort the data, and store the sorted data by type according to the patient's information; The data analysis module is used to extract features from the stored image data to obtain image data results; perform data processing on the stored audio data, and store the processing results separately to form a data vector of a single inspection to obtain single inspection vector data; Big data intelligent module, used to analyze and evaluate single inspection vector data and feed back the evaluation results to the terminal; The early warning module is used to obtain critical values ​​and model quality based on the evaluation results and image data results, and to feed back single-inspection vector data, critical values, and model quality to the terminal device to realize information feedback; based on the feedback information, the data curve is obtained to realize clinical heart failure monitoring.

[0013] According to a third aspect, a computing device includes: one or more processors; The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described.

[0014] In a fourth aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method described is implemented.

[0015] The above solution of the present invention includes at least the following beneficial effects: The terminal device of the present invention not only integrates the diagnostic information acquisition module of I-lead electrocardiogram, heart sound collection, patient information and other data, but also integrates the user and patient information management module, data feedback and receiving module, and early warning module, which can integrate heart sound, I-lead electrocardiogram and patient information to generate comprehensive data set feedback. The terminal has the characteristics of simple operation and no special requirements for the site. It can not only meet the daily monitoring of heart failure patients themselves, but also meet the daily chronic disease management of community general practitioners, which helps to improve the clinical information collection density of patients and add additional evidence chains for the clinical evaluation of patients. The central data processing processes the data through Hilbert-Huang transform to perform intrinsic mode function segmentation, which can identify characteristic sequences and waveforms, calculate and process heart failure-related indicators, and the method is scientific and effective, and can identify heart failure related. The system can also provide automatic or manual feedback and critical value warning according to clinical characteristics, which helps to identify emergencies and deal with them in time, and improve the prognosis of heart failure. The establishment of a patient heart sound database can form long-term data management and help to formulate individualized clinical treatment plans for patients.

[0016] The artificial intelligence-based heart failure monitoring system of the present invention integrates multiple methods and parameter indicators to achieve emergency identification and individualized chronic disease management for heart failure patients. It integrates information collection, data processing, machine learning and other methods, provides strong support for home monitoring of heart failure and general practice management in the community, brings health benefits to patients, and contributes new wisdom to the rational use of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a heart failure monitoring and evaluation method provided by an embodiment of the present invention.

[0018] Figure 2 It is a schematic diagram of a heart failure monitoring and evaluation system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in a form and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0020] like Figure 1 As shown, an embodiment of the present invention provides a heart failure monitoring and evaluation method, the method comprising the following steps: Step 11, collecting feedback information from each terminal device to generate a data set including patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds; Step 12, identifying and classifying the collected data set, organizing the data, and storing the organized data by type partition according to the patient's information; Step 13, extracting features from the stored image data to obtain image data results; Step 14, performing data processing on the stored audio data, and storing the processing results respectively to form a data vector of a single inspection, so as to obtain single inspection vector data; Step 15, analyzing and evaluating the single inspection vector data, and feeding back the evaluation result to the terminal; Step 16, according to the evaluation results and the image data results, the critical value and the model quality are obtained, and the single inspection vector data, the critical value and the model quality are fed back to the terminal device to realize information feedback; Step 17, obtaining a data curve based on the feedback information to achieve clinical heart failure monitoring.

[0021] In the embodiment of the present invention, by collecting feedback information from various terminal devices, multiple physiological data of the patient, including electrocardiogram waveforms, heart sounds, etc., can be obtained comprehensively and in real time, and a data set containing an accurate time axis can be generated, which is helpful to track and analyze the dynamic changes of the patient's condition and improve the timeliness and accuracy of diagnosis. Identifying and classifying data can ensure the accuracy and consistency of data, reduce errors in the analysis process, and store data by type partitions, which is convenient for subsequent rapid retrieval and call, and improves data processing efficiency. Extracting features from image data can deeply mine hidden information in the image, providing strong support for the early detection and diagnosis of heart failure. In-depth analysis of audio data helps to capture abnormal sound patterns related to heart failure and improve the sensitivity of diagnosis. Forming a data vector for a single examination facilitates quantitative comparison and trend analysis of data, providing a scientific basis for the adjustment of treatment plans. Analyzing and evaluating the vector data of a single examination can timely discover potential changes in the condition. Feeding back the evaluation results to the terminal helps to realize the remote and intelligent medical services and improve the convenience of patients' medical treatment. Acquiring critical values ​​and model quality can timely alert medical staff to the patient's critical condition, reduce the occurrence of medical accidents, and feed back single-check vector data, critical values, and model quality to the terminal device, realizing comprehensive and real-time information sharing and improving the collaboration efficiency of the medical team. The curve of the data obtained based on the feedback information can intuitively display the changing trend of the patient's condition, provide an intuitive and effective tool for clinical heart failure monitoring, realize clinical heart failure monitoring, help to timely detect and deal with heart failure events, and improve the patient's quality of life and prognosis.

[0022] In a preferred embodiment of the present invention, the above step 11 may include: Step 111, collecting feedback information from various terminal devices, wherein the terminal device is a portable examination device held by a patient or a community general practitioner, and the terminal device has a built-in diagnostic information collection module, a user and patient information management module, a data feedback and receiving module, and an early warning module; Step 112, from the feedback information, the audio sequence, image sequence and I-lead ECG sequence are accurately processed on the time scale, and the interval duration between the sequence features of the three sequences or the features in a single sequence are processed and extracted to generate a data set containing patient information, I-lead ECG waveform conditions, heart sound data and the time axis corresponding to each examination accurate to milliseconds.

[0023] In an embodiment of the present invention, feedback information is collected through a portable inspection device held by a patient or a community general practitioner, which greatly improves the portability and popularity of data collection. It can be easily used in different environments and scenarios, making heart failure monitoring closer to the patient's daily life and helping to detect the disease in time. The terminal device has multiple built-in modules, including diagnostic information collection, user and patient information management, data feedback and reception, and early warning modules, which realize the full process management from data collection to preliminary processing and then to timely feedback. This highly integrated design improves work efficiency and simplifies the operation process. By accurately processing the audio sequence, image sequence, and I-lead electrocardiogram sequence on the time scale, the time synchronization between different data types is ensured. This precise correspondence is crucial for subsequent data analysis and disease judgment, and improves the accuracy of diagnosis. Processing and extracting the sequence features of the three sequences by interval duration or features in a single sequence helps to deeply explore the potential information in the data. This feature extraction method can capture the subtle changes in the disease and provide strong support for the formulation of personalized treatment plans. Generates a data set containing patient information, I-lead ECG waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds, providing medical staff with a comprehensive and detailed picture of the patient's condition.

[0024] The specific implementation steps of the present invention are: The terminal device is a portable inspection device held by the patient or community general practitioner. The terminal device has built-in diagnostic information collection module, user and patient information management module, data feedback and receiving module, and early warning module; the user and patient information management module can upload and view patient information, and can adjust the early warning critical value according to the patient's condition. The data vector of each data monitoring of the patient can be used as the result feedback of a single analysis or long-term change observation, and a single patient model can also be built and trained. The diagnostic information collection module, the I-lead ECG information collector connected to the wrist electrodes of both hands, the handheld heart sound collector, the information infusion module, and the precision time recording module are uploaded in the form of data vectors after the collection is completed to ensure that valid information for analysis can be uploaded after each collection. After the audio sequence, image sequence and I-lead electrocardiogram sequence are accurately processed on the time scale, the interval length between the sequence features of the three sequences or the features in a single sequence are processed and extracted, and the pre-set operation is performed to obtain a data set containing patient information, I-lead electrocardiogram waveform, heart sound data and the time axis corresponding to each examination accurate to milliseconds. The heart sound data includes the amplitude of the first heart sound (S1) / the amplitude of the second heart sound (S2): the amplitude of the first and second heart sounds, that is, the energy, mainly depends on the contractility of the myocardium. The greater the contractility of the myocardium, the greater its amplitude. Therefore, by analyzing and comparing its amplitude, the myocardial contractility can be reflected, thereby evaluating cardiac function. In order to avoid the influence of individual differences and other related factors, the amplitude parameters of the heart sound are extracted by the relative comparison method. After measuring the amplitudes of S1 and S2 respectively, the ratio is taken as a characteristic index; Second, the diastolic time (D) / systolic time (S): the length of the diastolic period reflects the degree of ventricular filling on the one hand, and the blood supply of the myocardium on the other hand. If the diastolic period is long enough, the myocardium will have adequate blood supply and strong contractile force; otherwise, it will not be the case. The present invention uses the ratio to measure the strength of cardiac function. 3. The electromechanical excitation time of the heart is the time from the start of QRS in the I lead electrocardiogram sequence to the time of S1 in the heart sound sequence. This interval represents the time interval from the start of ventricular electrical excitation to the mechanical activity of the left ventricle.

[0025] l The present invention finally selects S1 / S2, D / S and cardiac electromechanical excitation time to measure the degree of heart failure. These parameters can be obtained by calculation in the above-mentioned parameter-time axis information. The features in the parameter column are identified and marked through the results of machine learning, and then the precise time values ​​between the corresponding features are measured to accurately calculate the parameters. Select and correct the cardiac electromechanical excitation time. Perform heart failure prognosis and risk prediction. The cardiac electromechanical excitation time is the ratio of the cardiac electromechanical excitation time to the RR interval. The normal value is cardiac electromechanical excitation time / (RR) ≥ 0.15. This indicator is related to the decreased left ventricular systolic function. An increase in the cardiac electromechanical excitation time value indicates the possibility of re-admission for heart failure, which can also be obtained through the aforementioned machine learning and time measurement method.

[0026] In a preferred embodiment of the present invention, the above step 12 may include: Step 121, classifying the electrocardiogram waveform data and the heart sound data respectively according to the collected data set to obtain different types of data; Step 122, formulate corresponding processing strategies according to the characteristics of different types of data and processing equipment; Step 123, according to the data classification and processing strategy, the data sorting task is assigned to the corresponding processing equipment and algorithm to realize data sorting; Step 124, the sorted data is stored in partitions according to type based on the patient's information.

[0027] In an embodiment of the present invention, the electrocardiogram waveform data and the heart sound data are classified separately, which helps to form a clear and orderly data structure, formulate processing strategies according to the characteristics of different types of data and processing equipment, and ensure the accuracy and effectiveness of data processing. According to the classification and processing strategy of the data, the data sorting tasks are assigned to the corresponding processing equipment and algorithms, so that the tasks are reasonably allocated and can be processed in a timely manner, thereby improving the efficiency and response speed of data processing. The sorted data is stored in partitions according to the patient's information, ensuring the organization and searchability of the data. Through partitioned storage, the privacy and security of patients can be better protected. Different types of data are stored separately, which reduces the risk of data leakage and meets the requirements of medical information security.

[0028] The specific implementation steps of the present invention are: Step 121, extract all ECG waveform data from the data set, use waveform recognition algorithm to automatically classify the data according to the characteristics of the ECG waveform, such as P wave, QRS complex, T wave, etc., and mark the classified ECG data into different types, such as normal ECG, arrhythmia ECG, myocardial ischemia ECG; extract all heart sound data from the data set, and use audio processing technology to extract features of heart sounds, such as time domain and frequency domain features.

[0029] Step 122, analyze the characteristics of the processing equipment and evaluate the available processing equipment, including its computing power, memory size, supported algorithms and tools; formulate processing strategies for different types of data (such as normal electrocardiogram, abnormal electrocardiogram, normal heart sounds, abnormal heart sounds, etc.) in combination with the characteristics of the processing equipment. For example, for devices with stronger computing power, more complex algorithms can be assigned for data processing; for devices with limited memory, it is necessary to optimize data storage and transmission methods.

[0030] Step 123, according to the formulated processing strategy, select appropriate processing equipment and algorithms for different types of data, assign the classified data to corresponding processing equipment and algorithms according to type, compare the collected parameters with the normal values ​​and output the results, check the data of a single collection unit with the data set to detect the data quality of a single collection unit, and predict the subsequent clinical indicators through the large data set and output the results.

[0031] Step 124, create different data partitions in the storage system according to the patient's information and data type, each partition is used to store a certain type of data for a specific patient, such as the "patient A-normal electrocardiogram" partition and the "patient B-abnormal heart sound" partition, and store the sorted data into the corresponding partition according to the type.

[0032] In a preferred embodiment of the present invention, the above step 13 may include: The stored image data is passed Perform feature extraction to obtain image data results, where: is the output feature map after the maximum pooling operation at the kth channel position The value of It's location The corresponding pooling area, In the pooling area The coordinates inside, M and N are the height and width of the convolution kernel, Is the input image or the feature map of the previous layer at position The value of is the kth convolution kernel at position The weight value of is the bias term associated with the kth convolution kernel.

[0033] In an embodiment of the present invention, through convolution operation and maximum pooling, key features can be extracted from the original image, and the features are reduced in dimension at the same time. The maximum pooling operation can reduce the size of the feature map, thereby reducing the computational complexity of the subsequent layers. The convolution and pooling operations have a certain translation invariance, that is, a slight movement of the target in the image will not have a significant effect on the feature extraction results, which enhances the robustness of the model and enables the model to better handle various changes that may occur in actual scenes. Reducing the number of parameters by maximum pooling can prevent model overfitting to a certain extent. Overfitting is a problem in which the machine learning model performs well on training data but has reduced generalization ability on new data. Reducing the number of parameters helps the model maintain good generalization performance on new data. The weights and bias items of the convolution kernel are automatically adjusted during the training process to select the most useful features for the task, so that the model can more accurately identify the key information in the image. Through multiple layers of convolution and pooling operations, it is possible to gradually extract high-level features from low-level features, thereby forming a hierarchical feature representation.

[0034] The specific implementation steps of the present invention are: When training the model, image recognition feature extraction uses convolutional neural networks after image modification. Through multi-layer convolution and pooling operations, it can extract local features of the image, gradually form a high-level abstract representation, and automatically learn image features; and use pre-trained deep learning models as feature extractors. These models are pre-trained on large-scale data sets and can capture the complex features of the image. When training the CNN model, the network weights are optimized through the back-propagation algorithm and the activation function is applied to introduce nonlinear operations to minimize the function loss and enhance the model's expressiveness. After the trained model, the image information is subjected to feature extraction technology through the formula to extract the local features of the image. The intrinsic mode function components that have been processed by Hilbert-Huang transformation and manual processing capture the local features and long-distance dependencies in the time series. Activation functions (such as ReLU) and pooling layers (such as maximum pooling) are applied in learning to enhance the nonlinear expression ability of the model and reduce the feature dimension. And the model is regularly evaluated on an independent test set, and the quality of the model is measured using indicators such as accuracy, recall, and F1 score. In a preferred embodiment of the present invention, the above step 14 may include: Step 141, identifying all local maximum values ​​and minimum values ​​of the audio data, interpolating the local maximum values ​​and minimum values ​​respectively to form an upper envelope and a lower envelope, calculating the mean of the upper and lower envelopes, subtracting the mean from the original data signal to obtain a new data signal, and checking whether the condition of the intrinsic mode function is met. If not, iterating the new data signal as the original data signal until the condition is met to obtain a plurality of intrinsic mode function components; Step 142, based on a number of intrinsic mode function components, a waveform diagram of each component and a three-dimensional characteristic spectrum of the time domain, frequency domain and energy domain are formed, and artificial subjective screening is performed to obtain specific stage intrinsic mode function components; Step 143, storing the intrinsic mode function components separately to form a data vector for a single inspection.

[0035] In an embodiment of the present invention, by identifying all local maxima and minima of audio data and interpolating to form upper and lower envelopes, the original data signal can be more accurately decomposed, which helps to extract the intrinsic modal function components in the signal. Calculating the mean of the upper and lower envelopes and subtracting them from the original data signal can effectively remove the noise and interference components in the signal and improve the purity and quality of the signal. Through the iterative process, the new data signal is continuously processed as the original data signal until the conditions of the intrinsic modal function components are met, which can ensure that the extracted intrinsic modal function components are more accurate and reliable. Forming the waveform diagram of each component and its three-dimensional characteristic spectrum in the time domain-frequency domain-energy domain can fully display the characteristic information of the audio data. This multi-dimensional display method helps to understand the data characteristics more deeply. Through artificial subjective screening, specific stage intrinsic modal function components can be obtained. After the intrinsic modal function components are stored separately, a data vector of a single inspection is formed. By forming a unified data vector format, the data between different inspections can be ensured to be consistent and comparable.

[0036] The specific implementation steps of the present invention are: Step 141, traverse the audio data to find all local maxima and minima. A local maximum is a point where a data point is larger than its adjacent data point, while a local minimum is the opposite. Interpolate all local maxima (such as cubic spline interpolation) to form an upper envelope. This line will pass through all local maximum points, and perform the same interpolation on all local minima to form a lower envelope. This line will pass through all local minimum points. At each data point, calculate the average of the upper envelope and the lower envelope to obtain the mean line. At each data point, subtract the value of the mean line from the original audio data to obtain a new data signal, and check whether it meets the conditions of the intrinsic mode function component: in the entire data set, the number of maxima and the number of zero crossings must be equal or differ by at most one; at any point, the mean of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is zero. If the new data signal does not meet the conditions of the intrinsic mode function, it is used as a new original data signal and the above steps are repeated until the conditions are met. By repeatedly performing the above steps, several intrinsic mode function components can be obtained, each of which represents the characteristics of different frequencies in the original signal.

[0037] Step 142, for each intrinsic modal function component, draw its waveform diagram to show the change of the component over time, perform spectrum analysis (such as Fourier transform) on each intrinsic modal function component to obtain its representation in the frequency domain, calculate the energy distribution of each intrinsic modal function component to form an energy spectrum, and combine the time domain waveform, spectrum and energy spectrum to form a three-dimensional characteristic spectrum. According to the characteristic spectrum, artificially screen out the stage intrinsic modal function components with specificity. These components may represent key features or abnormal events in the audio data.

[0038] Step 143, the selected specific intrinsic mode function components are stored separately, which can be saved as a numerical file. For a single inspection, all stored intrinsic mode function components are combined into a data vector. This vector can represent the feature set of the audio data in the inspection.

[0039] In a preferred embodiment of the present invention, the above step 15 may include: Step 151, check the vector data for a single time Perform outlier detection to obtain the test results, where: is the position of a data point relative to the overall distribution of the data set; is the value of a particular data point, n is the total number of data points, k and is the index variable; The detection result is evaluated, an evaluation result is obtained, and the evaluation result is fed back to the terminal.

[0040] In an embodiment of the present invention, by performing outlier detection on single-check vector data, data points that do not conform to the overall distribution of the data set can be accurately identified, which helps to discover potential errors, anomalies or data that does not meet expectations, and improve the reliability of data quality. Outlier detection is one of the important steps in data cleaning. By identifying and processing outliers, noise and interference in the data can be eliminated, making subsequent data analysis more accurate and effective. The outlier detection method used in this step can be adjusted and optimized according to the characteristics of the data, and has good adaptability. Evaluating the test results and feeding back the evaluation results to the terminal in a timely manner can ensure that medical staff are aware of the quality of the data in a timely manner. Through the feedback of the evaluation results, medical staff can more clearly understand which data is reliable and which data may have problems. This helps them make diagnosis or treatment decisions more quickly and accurately, and improves the quality and efficiency of medical services.

[0041] The specific implementation steps of the present invention are: Step 151, obtain a data vector for a single inspection, which is composed of the inherent intrinsic mode function components processed in the previous steps. According to the characteristics and requirements of the data, use outlier detection for each data point in the data vector. For statistical-based methods, calculate the statistics of each data point, and determine whether it is an outlier based on a preset threshold. For machine learning-based methods, train a model to identify abnormal patterns in the data, and apply the model to detect outliers. The detected outliers and their related information (such as index, value, degree of abnormality, etc.) are recorded.

[0042] Step 152, further analyze the detected outliers, and determine whether the outliers have practical significance or may be caused by errors in data collection or processing based on domain knowledge and data background; if the outliers do represent important information in the data, they need to be retained and relevant personnel informed. If the outliers are caused by errors, they may need to be excluded or corrected.

[0043] In a preferred embodiment of the present invention, the above step 16 may include: Step 161, analyzing the evaluation results and the image data, identifying the critical value, and marking the identified critical value to obtain the critical value; Step 162, constructing a model for generating evaluation results and image data results, and performing quality evaluation to obtain a model quality evaluation result; Step 163, feeding back the single inspection vector data, critical value, and model quality to the terminal device to realize information feedback.

[0044] In an embodiment of the present invention, by analyzing the evaluation results and image data, critical values ​​can be quickly identified, and the identified critical values ​​can be marked, which can significantly improve their visibility in complex data and help medical staff quickly locate and pay attention to these key information. Timely identification and marking of critical values ​​helps to ensure the safety of patients and reduce medical risks caused by missing information or untimely processing. By building a model to generate evaluation results and image data results, a large amount of data can be automatically processed and analyzed, significantly improving work efficiency. Quality evaluation of the model can ensure the accuracy and reliability of the model, thereby enhancing the trust in the evaluation results and image data results. High-quality models provide medical staff with powerful decision support tools, helping them to make more scientific and reasonable diagnosis and treatment plans. Feedback of single-check vector data, critical values, and model quality to the terminal device in a timely manner ensures that medical staff can obtain key information in the first place and respond quickly. By receiving feedback information through the terminal device, medical staff can remotely monitor the patient's condition.

[0045] The specific implementation steps of the present invention are: Step 161, collect the evaluation results and related image data, perform data mining and analysis on the evaluation results, use statistical methods, machine learning algorithms or rule engines to identify data points that are abnormal or exceed a predetermined threshold, and analyze the image data, involving image recognition, segmentation and feature extraction techniques to detect abnormalities in the image, such as changes in the size, shape or location of the lesion. Once a critical value is identified, it is marked on the corresponding data point or image.

[0046] Step 162, select a deep learning algorithm to build a model. Select the appropriate model type based on the nature of the problem and the characteristics of the data. For example, build a one-dimensional data convolutional neural network to process the aforementioned intrinsic mode function components that have been transformed by Hilbert-Huang and manually processed to capture local features and long-distance dependencies in the time series. Apply activation functions (such as ReLU) and pooling layers (such as maximum pooling) in learning to enhance the nonlinear expression ability of the model and reduce the feature dimension. And regularly evaluate the model on an independent test set, using indicators such as accuracy, recall, and F1 score to measure the quality of the model. Analyze the model evaluation results to determine whether the model meets the performance requirements. If the model performance is poor, it may be necessary to adjust the model parameters, try different algorithms, or collect more data to retrain the model.

[0047] Step 163, integrate the single inspection vector data, the marked critical value and the model quality assessment result into a unified data structure, transmit it through the wired network according to the system architecture and the access method of the terminal device, and push the prepared feedback data to the terminal device through the established connection.

[0048] In a preferred embodiment of the present invention, the above step 17 may include: Step 171, collating the feedback information, the evaluation results include image data, critical value and model quality; Step 172, analyze the evaluation results in detail, obtain a curve of the data, and classify the evaluation results according to the curve of the data to achieve clinical heart failure monitoring.

[0049] In an embodiment of the present invention, by collating the feedback information, including image data, critical values, and model quality, medical staff can more clearly understand the overall condition of the patient, and conduct an evaluation based on comprehensive feedback information from various aspects, thereby providing a comprehensive evaluation basis for medical staff, helping them to more accurately judge the patient's condition and develop a more appropriate treatment plan. By obtaining the curve of the data, medical staff can more intuitively understand the trend of changes in the patient's condition. Grading the evaluation results according to the curve of the data can achieve more accurate heart failure monitoring, which helps medical staff to understand the patient's condition stage more carefully, thereby developing a more targeted treatment plan. Through detailed analysis and grading of the evaluation results, medical staff can promptly detect abnormal changes in the patient's condition, thereby taking preventive measures or conducting early intervention.

[0050] The specific implementation steps of the present invention are: Step 171 , receiving and analyzing the feedback information sent in step 163 from each terminal device, which includes image data, critical value marks, and model quality assessment results.

[0051] Step 172, a detailed analysis is performed on the sorted data, including statistical analysis, trend analysis, and image feature extraction. For example, for image data, professional image processing software or algorithms are used to identify and analyze key features, generate data curves, and display the changing trends of key indicators (such as heart failure marker concentrations, cardiac function parameters, etc.) over time. Combined with the data curves, critical values, and model quality assessment results, an assessment grading standard is formulated, and the patient's assessment results are graded according to this standard, such as "low risk", "medium risk", and "high risk". The comprehensive influence of multiple factors should be considered during the grading, including but not limited to the trend of the data curve, the frequency and severity of critical values, and the confidence of the model prediction. A detailed clinical heart failure monitoring report is generated based on the grading results.

[0052] like Figure 2 As shown, an embodiment of the present invention further provides a heart failure monitoring and evaluation system 20, comprising: The data collection module 21 is used to collect feedback information from various terminal devices to generate a data set including patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds; The data sorting module 22 is used to identify and classify the collected data set, sort the data, and store the sorted data by type partition according to the patient's information; The data analysis module 23 is used to extract features from the stored image data to obtain image data results; perform data processing on the stored audio data, and store the processing results to form a data vector of a single inspection to obtain single inspection vector data; The big data intelligent module 24 is used to analyze and evaluate the single inspection vector data and feed back the evaluation results to the terminal; The early warning module 25 is used to obtain critical values ​​and model quality based on the evaluation results and image data results, and feed back single-check vector data, critical values, and model quality to the terminal device to achieve information feedback; based on the feedback information, the data curve is obtained to achieve clinical heart failure monitoring.

[0053] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A heart failure monitoring and evaluation method, characterized in that: The method comprises: Collect feedback information from each terminal device to generate a data set containing patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds; Identify and classify the collected data sets, organize the data, and store the organized data by type based on the patient's information; Extract features from the stored image data to obtain image data results; Performing data processing on the stored audio data, and storing the processing results respectively to form data vectors of a single inspection, so as to obtain single inspection vector data; Analyze and evaluate the single inspection vector data and feed back the evaluation results to the terminal; According to the evaluation results and image data results, critical values ​​and model quality are obtained, and single inspection vector data, critical values ​​and model quality are fed back to the terminal device to realize information feedback; Based on the feedback information, a data curve is obtained to achieve clinical heart failure monitoring.

2. The heart failure monitoring and evaluation method according to claim 1, characterized in that: Collect feedback information from various terminal devices to generate a data set containing patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds, including: Collect feedback information from various terminal devices, where the terminal device is a portable examination device held by the patient or community general practitioner, and the terminal device has built-in diagnostic information collection module, user and patient information management module, data feedback and reception module, and early warning module; From the feedback information, the audio sequence, image sequence and I-lead ECG sequence are accurately processed to correspond to the moments on the time scale, and the interval duration between the sequence features of the three sequences or the features in a single sequence are processed and extracted to generate a data set containing patient information, I-lead ECG waveform conditions, heart sound data, and the time axis corresponding to each examination accurate to milliseconds.

3. The heart failure monitoring and evaluation method according to claim 1, characterized in that: Identify and classify the collected data sets, organize the data, and store the organized data by type based on the patient's information, including: According to the collected data set, the electrocardiogram waveform data and the heart sound data are classified respectively to obtain different types of data; Develop corresponding processing strategies based on the characteristics of different types of data and processing equipment; According to the data classification and processing strategy, the data sorting tasks are assigned to the corresponding processing equipment and algorithms to achieve data sorting; The sorted data is stored in partitions according to the patient's information.

4. The heart failure monitoring and evaluation method according to claim 3, characterized in that: Perform feature extraction from stored image data to obtain image data results, including: The stored image data is passed Perform feature extraction to obtain image data results, where: is the output feature map after the maximum pooling operation at the kth channel position The value of It's location The corresponding pooling area, In the pooling area The coordinates inside, M and N are the height and width of the convolution kernel, Is the input image or the feature map of the previous layer at position The value of is the kth convolution kernel at position The weight value of is the bias term associated with the kth convolution kernel.

5. The heart failure monitoring and evaluation method according to claim 1, characterized in that: The stored audio data is processed, and the processing results are stored separately to form a data vector of a single inspection to obtain single inspection vector data, including: Identify all local maximum and minimum values ​​of the audio data, interpolate the local maximum and minimum values ​​to form an upper envelope and a lower envelope, calculate the mean of the upper and lower envelopes, subtract the mean from the original data signal to obtain a new data signal, and check whether the condition of the intrinsic mode function is met. If not, iterate the new data signal as the original data signal until the condition is met to obtain a number of intrinsic mode function components; According to several intrinsic mode function components, the waveform diagram of each component and its three-dimensional characteristic spectrum in time domain, frequency domain and energy domain are formed, and artificial subjective screening is performed to obtain specific stage intrinsic mode function components; The intrinsic mode function components are stored separately to form a data vector for a single inspection.

6. The heart failure monitoring and evaluation method according to claim 5, characterized in that: Analyze and evaluate the single inspection vector data and feed back the evaluation results to the terminal, including: For a single check vector data pass Perform outlier detection to obtain the test results, where: is the position of a data point relative to the overall distribution of the data set; is the value of a particular data point, n is the total number of data points, k and is the index variable; The detection result is evaluated, an evaluation result is obtained, and the evaluation result is fed back to the terminal.

7. The heart failure monitoring and evaluation method according to claim 1, characterized in that: According to the evaluation results and image data results, critical values ​​and model quality are obtained, and the single inspection vector data, critical values, and model quality are fed back to the terminal device to realize information feedback, including: Analyze the evaluation results and image data, identify critical values, and mark the identified critical values ​​to obtain critical values; Build a model that generates evaluation results and image data results, and perform quality assessment to obtain model quality; The single inspection vector data, critical value and model quality are fed back to the terminal device to realize information feedback.

8. A heart failure monitoring and evaluation system, characterized in that: include: A data collection module is used to collect feedback information from various terminal devices to generate a data set containing patient information, I-lead electrocardiogram waveform, heart sound data, and the time axis corresponding to each examination accurate to milliseconds; The data sorting module is used to identify and classify the collected data sets, sort the data, and store the sorted data by type according to the patient's information; A data analysis module is used to extract features from the stored image data to obtain image data results; Performing data processing on the stored audio data, and storing the processing results respectively to form data vectors of a single inspection, so as to obtain single inspection vector data; Big data intelligent module, used to analyze and evaluate single inspection vector data and feed back the evaluation results to the terminal; The early warning module is used to obtain critical values ​​and model quality based on the evaluation results and image data results, and to feed back single-inspection vector data, critical values, and model quality to the terminal device to realize information feedback; based on the feedback information, the data curve is obtained to realize clinical heart failure monitoring.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images