Cardiac rhythm data processing systems, methods, and storage media

Abstract

The application discloses a heart rhythm data processing system, method and storage medium. The heart rhythm data processing system comprises a wearable device and a cloud server. The wearable device is configured with a data acquisition module and a data uploading module. The cloud server is configured with a data processing module and a result feedback module. The data acquisition module is used to determine a data acquisition mode in response to a user triggering operation, and acquire target user's to-be-processed heart rhythm signal data and motion state data within a preset time length according to the data acquisition mode. The data uploading module is used to upload the to-be-processed heart rhythm signal data and the motion state data to the cloud server. The data processing module is used to process the to-be-processed heart rhythm signal data and the motion state data based on a heart rhythm data classification model, and obtain a target heart rhythm recognition result. The result feedback module is used to send the target heart rhythm recognition result to the wearable device, thereby improving the accuracy, interpretability and scene adaptability of the heart rhythm type determination.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a cardiac rhythm data processing system, method, and storage medium. Background Technology

[0002] Cardiac arrhythmia is a common cardiovascular disease, including various types such as atrial fibrillation, premature ventricular contractions, and ventricular tachycardia. If not treated in time, it may lead to serious complications.

[0003] Currently, the existing methods for determining heart rhythm type mainly involve analyzing electrocardiogram (ECG) data to identify the corresponding heart rhythm type. However, these methods rely on limited data and cannot guarantee the accuracy and reliability of the determined heart rhythm type. Summary of the Invention

[0004] This invention provides a heart rhythm data processing system, method, and storage medium, which improves the accuracy, interpretability, and scenario adaptability of heart rhythm type determination.

[0005] According to one aspect of the present invention, a cardiac rhythm data processing system is provided, comprising: a wearable device and a cloud server, wherein the wearable device is configured with: a data acquisition module and a data upload module, and the cloud server is configured with: a data processing module and a result feedback module, wherein... The data acquisition module is configured to: in response to a user-triggered operation, determine a data acquisition mode corresponding to the user-triggered operation, and acquire, based on the data acquisition mode, the target user's unprocessed heart rhythm signal data and motion state data within a preset time period; wherein, the target user is a user wearing the wearable device; The data upload module is configured to upload the heart rhythm signal data to be processed and the motion state data to the cloud server. The data processing module is configured to: process the heart rhythm signal data to be processed and the motion state data based on a pre-trained heart rhythm data classification model to obtain a target heart rhythm recognition result corresponding to the target user; wherein, the heart rhythm data classification model includes at least: a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module, and the target heart rhythm recognition result includes: the target heart rhythm type corresponding to the target user and explanatory information corresponding to the target heart rhythm type; The result feedback module is configured to send the target heart rhythm recognition result to the wearable device.

[0006] According to another aspect of the present invention, a method for processing cardiac rhythm data is provided, applied to a cardiac rhythm data processing system, the cardiac rhythm data processing system comprising: a wearable device and a cloud server, the wearable device being configured with: a data acquisition module and a data upload module, and the cloud server being configured with: a data processing module and a result feedback module, the method comprising: The data acquisition module responds to a user-triggered operation, determines the data acquisition mode corresponding to the user-triggered operation, and acquires the target user's unprocessed heart rhythm signal data and motion status data within a preset time period based on the data acquisition mode; wherein, the target user is a user wearing the wearable device; The data upload module uploads the heart rhythm signal data to be processed and the motion status data to the cloud server. The data processing module processes the heart rhythm signal data and motion state data based on a pre-trained heart rhythm data classification model to obtain a target heart rhythm recognition result corresponding to the target user. The heart rhythm data classification model includes at least a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module. The target heart rhythm recognition result includes the target heart rhythm type corresponding to the target user and explanatory information corresponding to the target heart rhythm type. The target heart rhythm recognition result is sent to the wearable device through the result feedback module.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the functions of the cardiac rhythm data processing system according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided that stores computer instructions for causing a processor to execute and implement the functions of the cardiac rhythm data processing system according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the functions of a cardiac rhythm data processing system as described in any embodiment of the present invention.

[0010] The technical solution of this invention involves a wearable device's data acquisition module responding to a user-triggered operation within a heart rhythm data processing system. This module determines the data acquisition mode corresponding to the user's operation and, based on the acquisition mode, acquires the target user's heart rhythm signal data and motion state data within a preset time period. The data upload module then uploads this data to a cloud server, achieving low-power, high-precision, and real-time data acquisition and uploading, providing data support for subsequent determination of the target heart rhythm type. Correspondingly, the cloud server's data processing module processes the heart rhythm signal data and motion state data based on a pre-trained heart rhythm data classification model, obtaining the target heart rhythm identification result corresponding to the target user, ensuring the interpretability and accuracy of the target heart rhythm type. Furthermore, deploying the heart rhythm data classification model on a cloud server reduces the load on the wearable device, ensuring low power consumption and device miniaturization. This invention ensures the scenario adaptability of heart rhythm type detection by acquiring heart rhythm signal data and motion state data under different data acquisition modes. Furthermore, by completely migrating the computationally intensive high-precision heart rhythm data classification model to the cloud, the wearable device only performs extremely simple and power-saving operations, thereby ensuring extremely low terminal power consumption. At the same time, it utilizes the powerful computing resources of the cloud to achieve high detection accuracy and low end-to-end latency, improving the efficiency and accuracy of heart rhythm type determination. This fundamentally breaks the contradiction between accuracy and power consumption, and realizes high-precision, interpretable, and personalized real-time heart rhythm type detection.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of the structure of a heart rhythm data processing system provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the code for determining the heart rhythm signal data to be processed according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the process of a data processing module performing model processing according to an embodiment of the present invention; Figure 4This is a flowchart illustrating a cardiac rhythm data processing method provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device that implements the heart rhythm data processing method of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0015] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0017] Example 1 Figure 1 This is a schematic diagram of a heart rhythm data processing system provided in Embodiment 1 of the present invention. This embodiment is applicable to situations where the heart rhythm type of a target user needs to be determined. See also... Figure 1 The heart rhythm data processing system provided in this embodiment of the invention includes a wearable device and a cloud server. The wearable device is configured with a data acquisition module 110 and a data upload module 120. The cloud server is configured with a data processing module 130 and a result feedback module 140. The following describes the specific structure of the heart rhythm data processing system.

[0018] The data acquisition module 110 is configured to: respond to a user-triggered operation, determine the data acquisition mode corresponding to the user-triggered operation, and acquire the target user's unprocessed heart rhythm signal data and motion state data within a preset time period based on the data acquisition mode.

[0019] The target user is the user wearing the wearable device. The user-triggered action can be an action taken by the target user or a person associated with the target user, selecting an action on the wearable device. Optionally, the user can trigger the action by clicking a control associated with the "data acquisition mode" on the wearable device's display interface. It should also be noted that the person associated with the target user can be a doctor or other personnel authorized by the target user.

[0020] The data acquisition mode is related to the target user's corresponding cardiac rhythm signal data acquisition scenario. Optionally, the data acquisition mode can be divided into a first mode and a second mode. The first mode can be understood as the data acquisition mode in scenarios where the target user wears the wearable device daily. The second mode can be understood as the data acquisition mode in scenarios where the target user is in a hospital intensive care unit. It should be noted that the cardiac rhythm signal data to be processed differs in different data acquisition modes. The preset duration can be a pre-set data acquisition duration. Optionally, the preset duration can be different for different data acquisition modes.

[0021] The cardiac rhythm signal data to be processed can be data associated with the target user's cardiac rhythm signal. Optionally, the cardiac rhythm signal data to be processed may include pulse wave signal data and / or electrocardiogram signal data, etc. Motion state data can be used to characterize the motion state of the target user wearing the wearable device. Optionally, the motion state data can be triaxial acceleration data. The triaxial acceleration data is used to characterize the target user's acceleration in three orthogonal directions (X, Y, Z axes).

[0022] Specifically, when a user-triggered operation is detected that a control associated with a data acquisition mode has been triggered, the corresponding data acquisition mode is determined. Based on the data acquisition mode, the target user's heart rhythm data and motion status data within a preset time period are acquired. This enables the acquisition of signal data, providing data support for subsequent heart rhythm type determination.

[0023] Based on the above embodiments, the data acquisition module 110 is configured to acquire the target user's unprocessed heart rhythm signal data and motion state data within a preset time period in the following manner: when the data acquisition mode is determined to be a first mode, at least two unprocessed pulse wave data and motion state data of the target user within a preset time period corresponding to the first mode are acquired, and the at least two unprocessed pulse wave data are filtered based on the motion state data to obtain filtered data corresponding to the first mode; when the data acquisition mode is determined to be a second mode, at least two unprocessed pulse wave data, electrocardiogram data, and motion state data of the target user within a preset time period corresponding to the second mode are acquired, and the at least two unprocessed pulse wave data and electrocardiogram data are filtered based on the motion state data to obtain filtered data corresponding to the second mode; and the unprocessed heart rhythm signal data are determined based on the filtered data corresponding to the data acquisition mode.

[0024] The preset duration corresponding to the first mode can be pre-set, representing the duration for acquiring pulse wave data to be processed under the first mode. The pulse wave data to be processed can be time-series signal data generated by detecting changes in blood volume, used to reflect the periodic fluctuations in blood volume within blood vessels caused by heartbeats.

[0025] At least two pulse wave data points to be processed can be acquired from at least two pulse wave sensor sites configured in the wearable device. Optionally, the at least two pulse wave sensor sites can be a wrist photoplethysmography (PPG) sensor site and a fingertip PPG sensor site. Accordingly, the acquired pulse wave data to be processed are wrist pulse wave data and fingertip pulse wave data, respectively. Filtering can be an adaptive filtering process that uses motion state data as a noise reference for the pulse wave data to be processed. The filtered data can be the data obtained by filtering the at least two pulse wave data points to be processed. Optionally, the filtered data can include at least two filtered pulse wave data points to be processed.

[0026] Correspondingly, the preset duration corresponding to the second mode can be a pre-set duration for acquiring pulse wave data to be processed under the second mode. The electrocardiogram (ECG) data can be the ECG signal data of the target user acquired within the preset duration. Correspondingly, the filtered data corresponding to the second mode includes at least two filtered pulse wave data sets and filtered ECG data.

[0027] It should also be noted that the cardiac rhythm signal data to be processed can differ depending on the data acquisition mode. In the first mode, to reduce data transmission power consumption, the cardiac rhythm signal data to be processed can be time-domain signal data obtained by weighting and fusing at least two filtered pulse wave data according to empirical weights. In the second mode, the cardiac rhythm signal data to be processed can be at least two filtered pulse wave data and filtered electrocardiogram data.

[0028] Specifically, in the first data acquisition mode, at least two pulse wave data points and motion state data points of the target user within a preset time period corresponding to the first mode are acquired. Using the motion state data as a noise reference, adaptive filtering is performed on the at least two pulse wave data points to obtain at least two filtered pulse wave data points. The at least two filtered pulse wave data points are then weighted and fused to obtain fused time-domain signal data. This fused time-domain signal data is used as the heart rhythm signal data to be processed corresponding to the first mode.

[0029] When the data acquisition mode is in the first mode, at least two pulse wave data, electrocardiogram (ECG) data, and motion status data of the target user within a preset time period corresponding to the second mode are acquired. Using the motion status data as a noise reference, adaptive filtering is performed on the at least two pulse wave data and ECG data to obtain at least two filtered pulse wave data and filtered ECG data. The at least two filtered pulse wave data and filtered ECG data are then used as the cardiac rhythm signal data to be processed corresponding to the second mode.

[0030] For example, the following explanation uses at least two pulse wave data points as wrist PPG signal data collected by wrist PPG sensor sites and fingertip PPG signal data collected by fingertip PPG sensor sites, and motion state data as triaxial acceleration data.

[0031] When the wearable device's data acquisition module detects that the target user has triggered the selection of a data acquisition mode, the wrist PPG sensor site configured with the wearable device's data acquisition module acquires data from the target user at a preset sampling rate of 200Hz, obtaining wrist PPG signal data. Correspondingly, the fingertip PPG sensor site, assisted by green light, acquires the target user's fingertip PPG signal data at a preset sampling rate of 200Hz. Correspondingly, the wearable device's data acquisition module's triaxial accelerometer acquires the target user's triaxial acceleration data at a preset sampling rate of 50Hz.

[0032] See Figure 2 , Figure 2This is a schematic diagram illustrating the code for determining the heart rhythm signal data to be processed. Using the triaxial acceleration data (imu) as the noise reference (noise_ref), adaptive filtering is performed on the wrist PPG signal data (ppg_wrist) using the Least Mean Square (LMS) algorithm, resulting in the filtered wrist PPG signal data (filtered_wrist). Similarly, adaptive filtering is performed on the fingertip PPG signal data (ppg_finger), resulting in the filtered fingertip PPG signal data (filtered_finger). This process eliminates motion artifacts and ensures data reliability.

[0033] Optionally, the above data can also be filtered using the Normalized Least Mean Square (NLMS) adaptive filtering algorithm. Specifically, this can be represented by the following function: ; in, Represents the adaptive filter coefficients. Indicates step size, This indicates the noise reduction error. This represents a noise reference determined based on triaxial acceleration data. Motion artifact suppression is achieved by iteratively updating w to minimize the error.

[0034] For the first mode (data acquisition mode in everyday wearable device scenarios), the following function can be used to achieve weighted fusion of filtered wrist PPG signal data and filtered fingertip PPG signal data: ; ; in, The fusion weight (between 0 and 1) indicates that a more reliable channel has a larger weight. The sensitivity coefficient / temperature coefficient (which controls the degree of sensitivity of the weights to the difference in signal-to-noise ratio) is represented by ISNR, which can be understood as "instantaneous signal-to-noise ratio index" (a type of signal quality index). This indicates the quality difference between two channels (e.g., the absolute value or difference between ISNR1 and ISNR2). This represents the filtered wrist PPG signal data. This represents the filtered fingertip PPG signal data. This represents the fused time-domain signal data.

[0035] See Figure 2The fusion weight can be 0.6. Weighted fusion processing is performed based on the fusion weight of 0.6 corresponding to the filtered wrist PPG signal data (filtered_wrist) and the weight coefficient of 0.4 corresponding to the filtered fingertip PPG signal data (filtered_finger), resulting in the fused time-domain signal data. To ensure data reliability, the fused time-domain signal data can also be verified against a signal-to-noise ratio threshold. The verified fused time-domain signal data is then used as the cardiac rhythm signal data to be processed in the first mode.

[0036] Accordingly, the collected electrocardiogram (ECG) data is filtered to obtain filtered ECG data. For the second mode (data acquisition mode in a hospital intensive care unit scenario), the filtered wrist PPG signal data (filtered_wrist), the filtered fingertip PPG signal data (filtered_finger), and the filtered ECG data are used as the heart rhythm signal data to be processed corresponding to the second mode.

[0037] It should be noted that the heart rhythm signal data to be processed is heart rhythm signal data with corresponding timestamps.

[0038] Optionally, to save network bandwidth, the data acquisition module 110 is also configured to: detect abnormal data in the filtered data according to a preset threshold corresponding to the data acquisition mode, determine abnormal heart rhythm signal data, and upload the abnormal heart rhythm signal data as heart rhythm signal data to be processed to the cloud server.

[0039] The preset threshold can be a pre-set standard value for the filtered data corresponding to the data acquisition mode. That is, for the first mode, the preset threshold may include the pulse wave threshold corresponding to each filtered pulse wave data point. For the second mode, the preset threshold may include the pulse wave threshold corresponding to each filtered pulse wave data point and the signal threshold of the filtered electrocardiogram data.

[0040] Abnormal heart rhythm signal data can be heart rhythm signal data that does not meet a preset threshold in the filtered data. For example, if the RR interval variation in PPG signal data exceeds the threshold of 0.2 seconds, then the corresponding PPG signal data can be regarded as abnormal heart rhythm signal data.

[0041] Specifically, for the first mode, based on the pulse wave threshold corresponding to each filtered pulse wave data point, abnormal data detection is performed on the corresponding filtered pulse wave data to identify abnormal pulse wave data. The resolution of the abnormal pulse wave data is then adjusted to obtain high-resolution abnormal pulse wave data. This high-resolution abnormal pulse wave data is then uploaded to the cloud server as abnormal heart rhythm signal data.

[0042] In the second mode, based on the pulse wave threshold corresponding to each filtered pulse wave data point, abnormal data detection is performed on the corresponding filtered pulse wave data to identify abnormal pulse wave data. Similarly, based on the signal threshold corresponding to the filtered electrocardiogram (ECG) data, abnormal data detection is performed on the filtered ECG data to obtain abnormal ECG data. The resolution of the abnormal pulse wave data and abnormal ECG data is adjusted to obtain high-resolution abnormal pulse wave data and high-resolution abnormal ECG data. These high-resolution abnormal pulse wave data and high-resolution abnormal ECG data are then uploaded to the cloud server as abnormal heart rhythm signal data. Based on the above, the end-to-end warning delay caused by the transmission from the wearable device to the cloud server can be reduced.

[0043] The data upload module 120 is configured to upload the heart rhythm signal data and motion status data to be processed to the cloud server.

[0044] Specifically, to save power consumption of wearable devices, the heart rate signal data and motion status data to be processed can be mixed and compressed. The compressed heart rate signal data and motion status data can then be uploaded to the cloud server according to a preset upload cycle. The preset upload cycle can be a pre-set data upload period for the heart rate signal data and motion status data, such as 30 seconds.

[0045] Optionally, the hybrid compression process can be: based on the LZW (Lempel-Ziv-Welch) lossless data compression algorithm and the Discrete Cosine Transform (DCT) lossy compression algorithm, performing hybrid compression processing on the cardiac rhythm signal data and motion state data to be processed, with a compression ratio greater than 10:1 and a reconstruction error less than 5%. The DCT lossy compression algorithm can be: DCT(PPG) = first k retained coefficients (k = signal length / 10). Here, PPG represents the pulse wave data in the cardiac rhythm signal data to be processed, and k represents the number of retained coefficients (compression ratio control parameter).

[0046] Optionally, when uploading the compressed heart rhythm signal data and motion status data to the cloud server according to a preset upload cycle, it can be uploaded in the form of one data packet every 10 seconds (data size less than 500 bytes), and an alarm packet can be sent immediately after detecting a transmission abnormality to ensure that the packet loss rate is less than 1%. At the same time, adaptive network switching (power consumption less than 8mW) can be performed to adaptively switch between 5G network, 4G network and Wi-Fi.

[0047] It should also be noted that wearable devices can be equipped with adaptive calibration circuits for periodic verification (e.g., 30 minutes per cycle) of the acquired heart rate signal data and motion status data, so as to effectively control the accuracy decay within 2% and ensure the reliability and accuracy of the data.

[0048] The data processing module 130 is configured to process the heart rhythm signal data and motion state data to be processed based on a pre-trained heart rhythm data classification model to obtain the target heart rhythm recognition result corresponding to the target user.

[0049] The heart rhythm data classification model includes at least: a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module. The feature extraction module based on the dual-domain attention mechanism is used to extract the time-domain and frequency-domain features corresponding to the heart rhythm signal data to be processed. The number of multiple heart rhythm type recognition modules is consistent with the pre-set number of heart rhythm types. Heart rhythm types may include: normal heart rhythm, atrial fibrillation, premature ventricular contractions, ventricular tachycardia, etc.

[0050] The input features of the heart rhythm type identification module can be fused features obtained by fusing the time-domain and frequency-domain features corresponding to the heart rhythm signal data to be processed. The heart rhythm type identification module is used to assess the probability and risk level of the corresponding heart rhythm type. The output module is used to comprehensively analyze the outputs of all heart rhythm type identification modules to output the target heart rhythm identification result. The target heart rhythm identification result includes: the target heart rhythm type corresponding to the target user and explanatory information corresponding to the target heart rhythm type. The target heart rhythm type can be determined comprehensively based on the outputs of all heart rhythm type identification modules, representing the heart rhythm type to which the target user belongs. The explanatory information can be used to explain the reasons for determining the target user to have the target heart rhythm type, and the reasons for determining the risk information of the target heart rhythm type.

[0051] Specifically, based on the pre-trained heart rhythm data classification model, the feature extraction module using a dual-domain attention mechanism processes the heart rhythm signal data and motion state data to obtain frequency domain features and time domain features. These frequency domain features and time domain features are then fused to obtain fused features. Multiple heart rhythm type recognition modules process the fused features separately, obtaining the output results corresponding to each module. Based on the output results of multiple heart rhythm type recognition modules, the target heart rhythm recognition result for the target user is determined.

[0052] The cloud server is also equipped with a model training module, which is configured to: acquire sample heart rhythm data corresponding to different data acquisition modes and theoretical heart rhythm recognition results corresponding to the sample heart rhythm data; input the sample heart rhythm data into the heart rhythm data classification model to be trained to obtain the output heart rhythm recognition result; determine the contrastive learning loss value, weighted cross-entropy classification loss value, and mean squared error loss value based on the output heart rhythm recognition result and the theoretical heart rhythm recognition result; perform weighted fusion processing on the contrastive learning loss value, weighted cross-entropy classification loss value, and mean squared error loss value to obtain the total loss value; and correct the model parameters of the heart rhythm data classification model to be trained based on the total loss value to obtain the trained heart rhythm data classification model.

[0053] The sample heart rhythm data differs depending on the data acquisition mode. Sample heart rhythm data can be sample heart rhythm signal data (including at least two sample pulse wave data) corresponding to the first mode and sample motion state data corresponding to the sample heart rhythm signal data. Sample heart rhythm data can also be sample heart rhythm signal data (including at least two sample pulse wave data and sample electrocardiogram data) corresponding to the second mode and sample motion state data corresponding to the sample heart rhythm signal data. The theoretical heart rhythm recognition result can be a pre-determined standard heart rhythm type corresponding to the sample heart rhythm data. The output heart rhythm recognition result can be the heart rhythm type determined by the heart rhythm data classification model to be trained.

[0054] Contrastive learning loss measures the ability of a trained heart rhythm classification model to distinguish between similar and dissimilar heart rhythm data in the feature space. Weighted cross-entropy classification loss measures the performance of the trained model in heart rhythm type classification. Mean squared error loss measures the difference between the theoretical heart rhythm recognition result and the output heart rhythm recognition result, thus measuring the prediction accuracy of the trained heart rhythm classification model.

[0055] The total loss value can be obtained by weighted fusion of the contrastive learning loss value, the weighted cross-entropy classification loss value, and the mean squared error loss value. Alternatively, the total loss value can be determined using the following function: ; in, This represents the contrastive learning loss value. This represents the weight coefficients corresponding to the contrastive learning loss values. This represents the cross-entropy classification loss value. This represents the weight coefficient corresponding to the cross-entropy classification loss value. This represents the mean squared error loss value. This represents the weighting coefficient corresponding to the mean squared error loss value. This represents the total loss value.

[0056] Specifically, sample heart rhythm data corresponding to different data acquisition modes are acquired, and the theoretical heart rhythm recognition result corresponding to the sample heart rhythm data is determined. The sample heart rhythm data is input into the heart rhythm data classification model to be trained, and the output heart rhythm recognition result is obtained. Based on the output heart rhythm recognition result and the theoretical heart rhythm recognition result, the contrastive learning loss value, the weighted cross-entropy classification loss value, and the mean squared error loss value are determined. The contrastive learning loss value, the weighted cross-entropy classification loss value, and the mean squared error loss value are weighted and fused using the above total loss value determination function to obtain the total loss value. The model parameters of the heart rhythm data classification model to be trained are then corrected based on the total loss value.

[0057] When using the total loss value to correct the model parameters in the heart rhythm data classification model to be trained, the convergence of the loss function can be used as a training objective. This includes checking if the training error is less than a preset error, if the error change tends to stabilize, or if the current number of iterations equals a preset number. If convergence is detected, such as the training error of the loss function being less than the preset error, or the error change trend tending to stabilize, it indicates that the heart rhythm data classification model to be trained has been successfully trained, and iterative training can be stopped. If convergence has not been detected, other sample heart rhythm data can be obtained to continue training the heart rhythm data classification model until the training error of the loss function is within a preset range. When the training error of the loss function converges, the trained heart rhythm data classification model is obtained. That is, when the heart rhythm signal data and motion state data to be processed are input into this trained heart rhythm data classification model, the target heart rhythm recognition result corresponding to the target user can be accurately obtained.

[0058] By training a heart rhythm data classification model based on sample heart rhythm data from different data acquisition modes, the heart rhythm data classification model can process heart rhythm signal data and motion state data under different data acquisition modes.

[0059] For example, in the second data acquisition mode, more than 1000 training samples are acquired, each of which includes sample heart rhythm data and the corresponding theoretical heart rhythm recognition result. The sample heart rhythm data includes: acquired wrist PPG signal data, fingertip PPG signal data, and sample electrocardiogram signal data. It should be noted that the event sparsity of the theoretical heart rhythm recognition results corresponding to all training samples is less than 5%.

[0060] The training samples are subjected to mixed supervised training according to the training parameters. Contrastive learning loss is applied using an extended SimCLR, and weighted cross-entropy loss and mean squared error loss based on event frequency and individual variation are determined to adjust the model parameters of the heart rhythm data classification model to be trained, resulting in a well-trained heart rhythm data classification model. The training parameters may include: training batch size (e.g., 32), number of iterations (e.g., 200), learning rate (e.g., 1e-3), and weight decay (e.g., 1e-4). It should also be noted that the theoretical heart rhythm recognition results (event labels) in the training samples are accurate to the second to support efficient training of the heart rhythm data classification model on sparse events, improving the model's ability to identify complex heart rhythm arrhythmias.

[0061] The result feedback module 140 is configured to send the target heart rhythm recognition result to the wearable device.

[0062] Specifically, the cloud server feeds back the identified target heart rhythm recognition result to the wearable device, so that the target heart rhythm recognition result can be displayed on the corresponding interface of the wearable device.

[0063] In conjunction with the above embodiments, the data acquisition module 110 is further configured to: when the received target heart rhythm recognition result meets the preset conditions, adjust the data acquisition mode to the second mode, so as to acquire the heart rhythm signal data to be processed based on the second mode.

[0064] The preset conditions can be pre-set conditions that require adjustment of the data acquisition mode based on the target heart rhythm recognition results. Optionally, if the target heart rhythm type in the target heart rhythm recognition results is a preset heart rhythm type, or if the target heart rhythm type in multiple consecutive target heart rhythm recognition results is a preset arrhythmia type, then the target heart rhythm recognition results are determined to meet the preset conditions. The preset arrhythmia type can be ventricular tachycardia, etc.

[0065] Specifically, when the received target heart rhythm recognition result meets the preset conditions, the data acquisition mode is adjusted to the second mode to acquire the heart rhythm signal data to be processed again according to the second mode, and to realize the identification of the target user's heart rhythm type.

[0066] It should also be noted that after receiving the target heart rhythm recognition result, the target user or personnel associated with the target user can adjust the data collection mode of the wearable device according to the target heart rhythm recognition result, so as to adjust it to the second mode.

[0067] Optionally, the target heart rhythm recognition result can be determined to meet the preset conditions based on the explanation information of the target heart rhythm type in the target heart rhythm recognition result.

[0068] Optionally, the wearable device is also equipped with a data feedback module, which is configured to: after receiving the target heart rhythm recognition result, obtain user feedback data corresponding to the target heart rhythm recognition result; and send the user feedback data back to the cloud server so that the cloud server can adjust the model parameters of the heart rhythm data classification model based on the user feedback data.

[0069] The user feedback data may include: assessment information determined by the target user or personnel associated with the target user in evaluating the target heart rhythm identification results, and the corresponding feedback heart rhythm signal data. For example, the target user or doctor may confirm the target heart rhythm identification results to determine their accuracy (e.g., the assessment information may be that the target heart rhythm type is correct, incorrect, or uncertain), and based on the determined assessment information, determine the corresponding feedback heart rhythm signal data from the heart rhythm signal data to be processed.

[0070] Specifically, to achieve dynamic optimization of the heart rhythm data classification model, after receiving the target heart rhythm identification result, evaluation information corresponding to the target heart rhythm identification result can be obtained. Based on the evaluation information, feedback heart rhythm signal data that matches the evaluation information in the heart rhythm data to be processed can be determined. Based on the feedback heart rhythm signal data and the evaluation information, user feedback data can be determined.

[0071] Optionally, a DreamFusion-like generative model can be used to process the feedback heart rhythm signal data to obtain variable heart rhythm signal data that matches the physiological characteristics of the target user. This variable heart rhythm signal data is then used as the feedback heart rhythm signal data and combined with evaluation information to determine the user feedback data.

[0072] Optionally, the feedback heart rhythm signal data can be processed by slight morphological scaling, time axis jitter, superposition of motion noise related to triaxial acceleration, and frequency band perturbation to obtain adjusted feedback heart rhythm signal data. This adjusted feedback heart rhythm signal data is then used as the feedback heart rhythm signal data and combined with evaluation information to obtain user feedback data.

[0073] Correspondingly, it should also be noted that after the cloud server receives user feedback data, it can adjust and process the heart rhythm data classification model based on the user feedback data.

[0074] Optionally, the parameters of multiple heart rhythm type recognition modules in the heart rhythm data classification model can be adjusted based on user feedback data. That is, multiple heart rhythm type recognition modules can be adjusted using the functions corresponding to the FedProx algorithm below: ; in, This represents the cross-entropy loss value. This is a hyperparameter and can be set to 0.01. This indicates the parameters of the current heart rhythm type recognition module. This represents the global model parameters.

[0075] For example, the target user or doctor can confirm the target heart rhythm recognition result (true positive / false positive / uncertain) on the wearable device or a medical terminal associated with the wearable device to determine the assessment information. The assessment information is then bound to the corresponding waveform segment in the heart rhythm signal data to be processed to obtain an individual small sample (corresponding to the user feedback data). The individual small sample is sent back to the cloud server to update the individual threshold / calibration parameters of the heart rhythm data classification model on the cloud server; or to trigger small-step fine-tuning in the cloud (e.g., updating only the adapter / warning head parameters) to improve the subsequent recognition accuracy of the target user.

[0076] It should also be noted that, to better accommodate individual physiological differences, the target user's individual baseline data can be included as part of the user feedback data. This allows for adjustments to the heart rate data classification model by incorporating user feedback data based on the individual baseline data. The individual baseline data may include: the target user's user attribute information, resting heart rate range, RR variability, and signal data quality indicators, etc.

[0077] For example, when a target user first wears the wearable device, user attribute information such as age, gender, and medical history can be obtained with the user's authorization. Furthermore, with authorization, exercise and sleep information can be continuously acquired, serving as individual baseline data. This individual baseline data is used to perform temperature calibration or threshold adaptation on the heart rhythm data classification model to update multiple heart rhythm data recognition modules. This enables stratified initialization or stratified threshold determination based on user attribute information, and adds uncertainty or confidence level gating to characterize the accuracy of the determined target heart rhythm type. This allows for triggering the uploading of more heart rhythm signal data or higher sampling rate heart rhythm signal data, or prompting for retesting, when confidence levels are low.

[0078] The technical solution of this embodiment utilizes the wearable device's data acquisition module within the heart rhythm data processing system to respond to user-triggered operations. It determines the data acquisition mode corresponding to the user's operation and, based on this mode, acquires the target user's heart rhythm signal data and motion state data within a preset time period. This data is then uploaded to a cloud server via a data upload module, achieving low-power, high-precision, and real-time data acquisition and uploading, providing data support for subsequent determination of the target heart rhythm type. Correspondingly, the cloud server's data processing module processes the heart rhythm signal data and motion state data based on a pre-trained heart rhythm data classification model, obtaining the target heart rhythm identification result corresponding to the target user, ensuring the interpretability and accuracy of the target heart rhythm type. Furthermore, deploying the heart rhythm data classification model on a cloud server reduces the load on the wearable device, ensuring low power consumption and device miniaturization. This invention ensures the scenario adaptability of heart rhythm type detection by acquiring heart rhythm signal data and motion state data under different data acquisition modes. Furthermore, by completely migrating the computationally intensive high-precision heart rhythm data classification model to the cloud, the wearable device only performs extremely simple and power-saving operations, thereby ensuring extremely low terminal power consumption. At the same time, it utilizes the powerful computing resources of the cloud to achieve high detection accuracy and low end-to-end latency, improving the efficiency and accuracy of heart rhythm type determination. This fundamentally breaks the contradiction between accuracy and power consumption, and realizes high-precision, interpretable, and personalized real-time heart rhythm type detection.

[0079] Example 2 Figure 3 This is a schematic flowchart illustrating the data processing module performing model processing according to Embodiment 2 of the present invention. This embodiment is a refinement of the data processing module performing model processing based on the above embodiments. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here. Figure 3 As shown, the data processing module is configured to process the heart rhythm signal data and motion state data to be processed based on a pre-trained heart rhythm data classification model in the following manner to obtain the target heart rhythm recognition result corresponding to the target user: S210. Based on the feature extraction module, feature extraction is performed on the cardiac rhythm signal data to be processed to obtain the input features.

[0080] The feature extraction module can be used to extract features from the cardiac rhythm signal data to be processed. Optionally, the feature extraction module can be a Transformer encoder based on a multi-layer attention mechanism. The input features can be features obtained by performing time-domain and frequency-domain feature extraction on the cardiac rhythm signal data to be processed.

[0081] Specifically, the heart rhythm data classification model is used to extract features from the heart rhythm signal data to be processed, thereby obtaining the input features.

[0082] Optionally, the feature extraction module based on the dual-domain attention mechanism includes: a feature extraction unit based on the temporal domain attention mechanism, a feature extraction unit based on the frequency domain attention mechanism, and a feature fusion unit. The feature extraction unit based on the temporal domain attention mechanism is used to extract temporal features from the cardiac rhythm signal data to be processed. The feature extraction unit based on the frequency domain attention mechanism is used to extract frequency domain features from the cardiac rhythm signal data to be processed. The feature fusion unit is used to perform feature fusion processing on the features extracted by the feature extraction unit based on the temporal domain attention mechanism and the features extracted by the feature extraction unit based on the frequency domain attention mechanism.

[0083] The data processing module 130 is configured to obtain the input features in the following manner: a feature extraction unit based on a time-domain attention mechanism extracts time-domain features from the heart rhythm signal data to be processed, obtaining the time-domain features to be processed; a feature extraction unit based on a frequency-domain attention mechanism extracts frequency-domain features from the heart rhythm signal data to be processed, obtaining the frequency-domain features to be processed; the heart rhythm signal data and motion state data are processed based on a preset activation function to obtain adaptive weight coefficients; and the time-domain features and frequency-domain features to be processed are fused according to the adaptive weight coefficients to obtain fused features, which are then used as the input features.

[0084] The time-domain features to be processed can be features obtained by extracting time-domain features from the cardiac rhythm signal data to be processed. Optionally, the time-domain features to be processed can be obtained through the following time-domain determination function: ; in, Represents the time-domain features to be processed. It is obtained by mapping the time-domain input representation corresponding to the heart rhythm signal data to be processed, and is used to represent the query, key, and value matrix corresponding to the time-domain features to be processed; wherein, the time-domain input representation corresponding to the heart rhythm signal data to be processed can be obtained by feature extraction of the heart rhythm signal data to be processed; Indicates the feature dimension.

[0085] Correspondingly, the frequency domain features to be processed can be the features obtained by frequency domain feature extraction from the cardiac rhythm signal data to be processed. Optionally, the frequency domain features to be processed can be determined by the following process: frequency domain transformation of the cardiac rhythm signal data to be processed is performed based on Fast Fourier Transform (FFT) or Short-Time Fourier Transform (STFT) to obtain a frequency domain input representation. The frequency domain input representation is then processed based on a frequency domain attention mechanism to obtain the frequency domain features to be processed. This can be obtained using the following frequency domain determination function: ; in, Represents the frequency domain features to be processed. It is obtained by mapping the frequency domain input representation corresponding to the heart rhythm signal data to be processed, and is used to characterize the query, key, and value matrix corresponding to the frequency domain features to be processed.

[0086] The preset activation function can be a pre-set activation function used to determine the adaptive weight coefficients. Optionally, the preset activation function can be a sigmoid activation function. The adaptive weight coefficients can be the weight coefficients applied when fusing the time-domain features and frequency-domain features to be processed.

[0087] Optionally, the adaptive weighting coefficients can be determined using the following function: ; in, This represents the feature representation corresponding to the cardiac rhythm signal data to be processed. This represents the feature representation corresponding to the motion state data. This represents a trainable weight matrix. This represents the bias coefficient. It can be the sigmoid activation function.

[0088] Fusion features can be obtained by fusing the time-domain features and frequency-domain features to be processed. Fusion features can be used to account for both morphological anomalies and frequency band energy anomalies. Optionally, fusion features can be obtained using the following fusion feature determination function: ; in, Indicates the adaptive weighting coefficient. This represents the fusion feature, which is the feature to be input, i.e., the feature used to input the heart rhythm type recognition module. Represents the time-domain features to be processed. This represents the frequency domain features to be processed.

[0089] Specifically, a feature extraction unit based on a temporal attention mechanism extracts temporal features from the cardiac rhythm signal data to be processed, obtaining the temporal features to be processed. Correspondingly, a feature extraction unit based on a frequency domain attention mechanism extracts frequency domain features from the cardiac rhythm signal data to be processed, obtaining the frequency domain features to be processed. Feature representations corresponding to the cardiac rhythm signal data to be processed and feature representations corresponding to the motion state data are determined. Based on the aforementioned adaptive weight coefficients, a function is used to process the feature representations of the cardiac rhythm signal data and the motion state data to obtain adaptive weight coefficients. Based on the adaptive weight coefficients, the temporal and frequency domain features to be processed are fused using a weighted feature calculation, obtaining a fused feature. This fused feature is used as the input feature for the cardiac rhythm type recognition module.

[0090] For example, the heart rhythm data classification model can be a multimodal heart rhythm large model (PPG-to-ArrhythmiaLLM), which is a Transformer-based model with hundreds of billions of parameters. The backbone network (feature extraction module) of this model can be a Transformer encoder based on a multi-layer self-attention mechanism. In the first data acquisition mode, the input heart rhythm signal data to be processed is the time-domain signal data after weighted fusion of at least two filtered pulse wave data according to empirical weights. In the second data acquisition mode, the input heart rhythm signal data to be processed can be at least two filtered pulse wave data and filtered electrocardiogram data. Input dimension: [batch, sequence length, feature dimension], for example, can be set to [32, 2000, 128]. The heart rhythm signal data to be processed is processed by the Transformer encoder based on a multi-layer self-attention mechanism to obtain the input features for the heart rhythm type recognition module. Based on the above processing, the signal-to-noise ratio can be improved by more than 20%.

[0091] It should be noted that during the training phase of a heart rhythm data classification model, the theoretical heart rhythm identification result can be determined by synchronizing ECG (electrocardiogram data) and clinical event labels (second-level timestamps, such as "atrial fibrillation onset: t=10s"). In other words, during the training phase of a heart rhythm data classification model, synchronized ECG can be used for event alignment and high-confidence label generation.

[0092] S220. For multiple heart rhythm type recognition modules, the input features are processed based on the heart rhythm type recognition modules to obtain recognition attribute information corresponding to the heart rhythm type recognition modules; wherein, the recognition attribute information includes: probability information and risk information of the heart rhythm type.

[0093] The heart rhythm type identification module outputs identification attribute information for the corresponding heart rhythm type. Optionally, the heart rhythm type identification module can be a small CNN / LSTM distillation sub-model. Multiple heart rhythm type identification modules can include at least: a heart rhythm type identification module for atrial fibrillation, a heart rhythm type identification module for premature ventricular contractions, a heart rhythm type identification module for ventricular tachycardia, and a heart rhythm type identification module for normal heart rhythm. The probability information in the identification attribute information is used to characterize the probability that the target user belongs to the corresponding heart rhythm type. The risk information is used to characterize the risk level or urgency level of the target user within the corresponding heart rhythm type. For example, the risk information can be represented by a risk score between 0 and 100. Optionally, the risk information can be determined by combining individual baseline data, the heart rhythm signal data to be processed, and motion state data.

[0094] Specifically, for multiple heart rhythm type recognition modules, the input features are processed separately for each heart rhythm type recognition module to obtain the probability information and risk information of the heart rhythm type to which the heart rhythm type recognition module belongs.

[0095] For example, referring to the above example, let's take the heart rhythm type recognition module as a distillation sub-model, the probability information as the probability of the corresponding heart rhythm type, and the risk information as the risk score of the corresponding heart rhythm type as an example. The input features are processed by the distillation sub-model corresponding to the heart rhythm type to determine the probability and risk score corresponding to the corresponding heart rhythm type. It should be noted that a corresponding confidence level can also be output to adjust the parameters of the heart rhythm data classification model based on the confidence level.

[0096] S230. Based on the output module, process the probability information of multiple heart rhythm types to determine the target heart rhythm type.

[0097] The target heart rhythm type can be the heart rhythm type corresponding to the highest probability information among multiple probability information.

[0098] Specifically, from the probability information of multiple heart rhythm types, the heart rhythm type corresponding to the highest probability information is determined, and this heart rhythm type is identified as the target heart rhythm type.

[0099] S240. Process the target heart rhythm type and the risk information corresponding to the target heart rhythm type according to the preset rule engine, and determine the explanatory information corresponding to the target heart rhythm type.

[0100] The preset rule engine can be a pre-configured, hard-coded clinical guideline rule engine. The explanatory information may include the risk level corresponding to the risk information, as well as explanations for determining the target user's target heart rhythm type and the reasons for determining the risk level of the target heart rhythm type.

[0101] Specifically, based on a preset rule engine, the target heart rhythm type and risk information are processed to determine the risk level corresponding to the risk information. Based on the risk level, it is determined whether the current conditions for an arrhythmia alarm are met, and corresponding treatment suggestions are provided. Correspondingly, the reasons for the target heart rhythm type and the reasons for determining the risk level of the target heart rhythm type can also be determined. The information determined above serves as the explanatory information corresponding to the target heart rhythm type.

[0102] It should also be noted that different interpretations can be provided for different data acquisition modes. If the current data acquisition mode is the second mode, the heart rhythm signal data to be processed can be further processed based on the target heart rhythm type and risk information to obtain an annotated signal waveform and corresponding textual suggestions. The interpretation information is then updated based on the annotated signal waveform and the corresponding textual suggestions to obtain interpretation information suitable for the second mode.

[0103] For example, in conjunction with the above example, the output of the heart rhythm data classification model (target heart rhythm type, risk information, confidence level, and duration and heart rate interval corresponding to the target heart rhythm type in the heart rhythm signal data to be processed) is substituted into the clinical guideline rule engine. (For example, the input information is analyzed by the atrial fibrillation duration threshold, tachycardia heart rate threshold, etc. determined by the rule engine based on the AHA (American Heart Association) and ESC (European Society of Cardiology) guidelines, so as to output "whether the alarm conditions / risk stratification / recommended treatment level are met".)

[0104] Optionally, the target heart rhythm type, the risk information corresponding to the target heart rhythm type, and the heart rhythm signal data to be processed can be input into the LLM model to obtain Grad-CAM hot-labeled abnormal waveforms (red-marked irregular RR segments) and natural language explanations (e.g., "P waves disappear, RR intervals are absolutely irregular, which meets the AHA definition of atrial fibrillation").

[0105] For example, based on the above embodiments, the heart rhythm data processing method can be described through the following example. In the first mode, which is a data acquisition mode for everyday wear, the target user wears a watch-type terminal device (wearable device). Sensors in the wearable device collect PPG signal data from the wrist and fingertips. Simultaneously, motion artifact calibration is performed on the wrist and fingertips PPG signal data based on the collected triaxial acceleration data (corresponding to motion state data) to obtain the heart rhythm signal data and motion state data to be uploaded to the cloud server.

[0106] The data packets corresponding to the heart rhythm signal data and motion status data to be processed are compressed according to a preset compression ratio and uploaded to the cloud server in a short time. The heart rhythm data classification model deployed on the cloud server processes the uploaded data, detects persistent atrial fibrillation, and issues an early warning in a short time. Accordingly, the wearable device vibrates and displays an abnormal waveform heatmap, natural language explanation (such as "P wave disappearance, RR interval absolutely irregular, consistent with atrial fibrillation characteristics") and suggestions (such as "It is recommended to rest immediately and consult a doctor"). In this mode, the wearable device's power consumption is kept at a low level, resulting in a long battery life.

[0107] In the second mode, which is the data collection mode in a hospital intensive care unit scenario, the heart rhythm data classification model deployed on the cloud server is trained using an expanded modal dataset (wrist PPG signal data, fingertip PPG signal data, synchronous ECG data, and theoretical heart rhythm recognition results). After wearing the device for a period of time, the heart rhythm data classification model can be individually fine-tuned based on user feedback data (FedProx algorithm) to adapt the model to the user's specific physiological characteristics. In actual testing, it successfully provided an early warning of a malignant ventricular tachycardia, allowing the user to seek medical attention promptly. The heart rhythm data classification model deployed on the cloud server achieves high specificity and strictly follows the AHA guidelines to output graded risk (high risk: call for help immediately).

[0108] The technical solution of this embodiment extracts features from the heart rhythm signal data to be processed through the feature extraction module of the heart rhythm data classification model, obtaining the input features. This achieves dual-domain feature extraction of the heart rhythm signal data, providing reliable data support for subsequent determination of the target heart rhythm type. For multiple heart rhythm type recognition modules, the input features are processed based on the heart rhythm type recognition modules to obtain recognition attribute information corresponding to the heart rhythm type recognition modules. This enables the determination of each heart rhythm type to which the target user belongs, thereby determining the target user's target heart rhythm type and the corresponding explanatory information. A preset rule engine processes the risk information corresponding to the target heart rhythm type to obtain explanatory information, improving the interpretability of the heart rhythm type determination and increasing the target user's trust.

[0109] Example 3 Figure 4 This is a flowchart illustrating a heart rhythm data processing method according to Embodiment 3 of the present invention. This embodiment is applicable to situations where the heart rhythm type of a target user needs to be determined. The method can be executed by a heart rhythm data processing system, which includes a wearable device and a cloud server. The wearable device is configured with a data acquisition module and a data upload module, and the cloud server is configured with a data processing module and a result feedback module. Figure 4 As shown, the method includes: S310. Through the data acquisition module, in response to the user's triggered operation, determine the data acquisition mode corresponding to the user's triggered operation, and according to the data acquisition mode, acquire the target user's unprocessed heart rhythm signal data and motion status data within a preset time period.

[0110] The target users are those who wear wearable devices.

[0111] Optionally, the following method is used to acquire the target user's heart rhythm signal data and motion state data within a preset time period: When the data acquisition mode is determined to be the first mode, at least two pulse wave data and motion state data of the target user within a preset time period corresponding to the first mode are acquired, and the at least two pulse wave data to be processed are filtered based on the motion state data to obtain filtered data corresponding to the first mode; When the data acquisition mode is determined to be the second mode, at least two pulse wave data, electrocardiogram data, and motion state data of the target user within a preset time period corresponding to the second mode are acquired, and the at least two pulse wave data and electrocardiogram data to be processed are filtered based on the motion state data to obtain filtered data corresponding to the second mode; Based on the filtered data corresponding to the data acquisition mode, the heart rhythm signal data to be processed is determined.

[0112] Optionally, embodiments of the present invention further include: performing abnormal data detection on the filtered data according to a preset threshold corresponding to the data acquisition mode to determine abnormal heart rhythm signal data; and uploading the abnormal heart rhythm signal data as heart rhythm signal data to be processed to a cloud server.

[0113] S320: Upload the heart rhythm signal data and motion status data to be processed to the cloud server through the data upload module.

[0114] S330. Through the data processing module, the heart rhythm signal data and motion state data to be processed are processed based on the pre-trained heart rhythm data classification model to obtain the target heart rhythm recognition result corresponding to the target user.

[0115] The heart rhythm data classification model includes at least: a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module. The target heart rhythm recognition result includes: the target heart rhythm type corresponding to the target user and the explanatory information corresponding to the target heart rhythm type.

[0116] Optionally, the target heart rhythm recognition result is determined as follows: The feature extraction module extracts features from the heart rhythm signal data to be processed, obtaining the input features; for multiple heart rhythm type recognition modules, the input features are processed based on the heart rhythm type recognition modules to obtain recognition attribute information corresponding to the heart rhythm type recognition modules; wherein, the recognition attribute information includes: probability information and risk information of the corresponding heart rhythm type; the output module processes the probability information of multiple corresponding heart rhythm types to determine the target heart rhythm type; and the preset rule engine processes the target heart rhythm type and the risk information corresponding to the target heart rhythm type to determine the interpretation information corresponding to the target heart rhythm type.

[0117] Optionally, the feature extraction module based on the dual-domain attention mechanism includes: a feature extraction unit based on the temporal domain attention mechanism, a feature extraction unit based on the frequency domain attention mechanism, and a feature fusion unit. In this embodiment of the invention, the method for determining the input features can be as follows: the feature extraction unit based on the temporal domain attention mechanism performs temporal feature extraction on the cardiac rhythm signal data to be processed, obtaining the temporal features to be processed; the feature extraction unit based on the frequency domain attention mechanism performs frequency feature extraction on the cardiac rhythm signal data to be processed, obtaining the frequency features to be processed; the cardiac rhythm signal data to be processed and the motion state data are processed based on a preset activation function to obtain adaptive weight coefficients; and the temporal features to be processed and the frequency features to be processed are fused according to the adaptive weight coefficients to obtain fused features, which are then used as the input features.

[0118] Before processing the heart rhythm signal data based on the trained heart rhythm data classification model, the model can be trained as follows: Obtain sample heart rhythm data corresponding to different data acquisition modes and theoretical heart rhythm recognition results corresponding to the sample heart rhythm data; input the sample heart rhythm data into the heart rhythm data classification model to be trained to obtain the output heart rhythm recognition result; determine the contrastive learning loss value, weighted cross-entropy classification loss value, and mean squared error loss value based on the output heart rhythm recognition result and the theoretical heart rhythm recognition result; perform weighted fusion processing on the contrastive learning loss value, weighted cross-entropy classification loss value, and mean squared error loss value to obtain the total loss value; and correct the model parameters of the heart rhythm data classification model to be trained based on the total loss value to obtain the trained heart rhythm data classification model.

[0119] S340: The target heart rhythm recognition result is sent to the wearable device through the result feedback module.

[0120] Optionally, when the received target heart rhythm recognition result meets the preset conditions, the data acquisition mode is adjusted to the second mode to acquire the heart rhythm signal data to be processed based on the second mode.

[0121] Optionally, after receiving the target heart rhythm recognition result, user feedback data corresponding to the target heart rhythm recognition result is obtained; the user feedback data is then sent back to the cloud server so that the cloud server can adjust the model parameters of the heart rhythm data classification model based on the user feedback data.

[0122] The technical solution of this embodiment utilizes the wearable device's data acquisition module within the heart rhythm data processing system to respond to user-triggered operations. It determines the data acquisition mode corresponding to the user's operation and, based on this mode, acquires the target user's heart rhythm signal data and motion state data within a preset time period. This data is then uploaded to a cloud server via a data upload module, achieving low-power, high-precision, and real-time data acquisition and uploading, providing data support for subsequent determination of the target heart rhythm type. Correspondingly, the cloud server's data processing module processes the heart rhythm signal data and motion state data based on a pre-trained heart rhythm data classification model, obtaining the target heart rhythm identification result corresponding to the target user, ensuring the interpretability and accuracy of the target heart rhythm type. Furthermore, deploying the heart rhythm data classification model on a cloud server reduces the load on the wearable device, ensuring low power consumption and device miniaturization. This invention ensures the scenario adaptability of heart rhythm type detection by acquiring heart rhythm signal data and motion state data under different data acquisition modes. Furthermore, by completely migrating the computationally intensive high-precision heart rhythm data classification model to the cloud, the wearable device only performs extremely simple and power-saving operations, thereby ensuring extremely low terminal power consumption. At the same time, it utilizes the powerful computing resources of the cloud to achieve high detection accuracy and low end-to-end latency, improving the efficiency and accuracy of heart rhythm type determination. This fundamentally breaks the contradiction between accuracy and power consumption, and realizes high-precision, interpretable, and personalized real-time heart rhythm type detection.

[0123] Example 4 Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0124] like Figure 5As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0125] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0126] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the functions configured in the cardiac rhythm data processing system.

[0127] In some embodiments, the functions configured for the heart rhythm data processing system may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the functions configured for the heart rhythm data processing system described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the functions configured for the heart rhythm data processing system by any other suitable means (e.g., by means of firmware).

[0128] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0129] The computer program configured to implement the functions of the cardiac rhythm data processing system of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer program causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer program can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0130] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0131] Example 5 Embodiment 5 of the present invention also provides a computer-readable storage medium storing computer instructions for causing a processor to perform the functions configured in the aforementioned cardiac rhythm data processing system, namely, The data acquisition module is configured to: in response to a user-triggered operation, determine a data acquisition mode corresponding to the user-triggered operation, and acquire, based on the data acquisition mode, the target user's unprocessed heart rhythm signal data and motion state data within a preset time period; wherein, the target user is a user wearing the wearable device; The data upload module is configured to upload the heart rhythm signal data to be processed and the motion state data to the cloud server. The data processing module is configured to: process the heart rhythm signal data to be processed and the motion state data based on a pre-trained heart rhythm data classification model to obtain a target heart rhythm recognition result corresponding to the target user; wherein, the heart rhythm data classification model includes at least: a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module, and the target heart rhythm recognition result includes: the target heart rhythm type corresponding to the target user and explanatory information corresponding to the target heart rhythm type; The result feedback module is configured to send the target heart rhythm recognition result to the wearable device.

[0132] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0133] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0134] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0135] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0136] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A cardiac rhythm data processing system, characterized in that, The cardiac rhythm data processing system includes a wearable device and a cloud server. The wearable device is configured with a data acquisition module and a data upload module. The cloud server is configured with a data processing module and a result feedback module. The data acquisition module is configured to: in response to a user-triggered operation, determine a data acquisition mode corresponding to the user-triggered operation, and acquire, based on the data acquisition mode, the target user's unprocessed heart rhythm signal data and motion state data within a preset time period; wherein, the target user is a user wearing the wearable device; The data upload module is configured to upload the heart rhythm signal data to be processed and the motion state data to the cloud server. The data processing module is configured to: process the heart rhythm signal data to be processed and the motion state data based on a pre-trained heart rhythm data classification model to obtain a target heart rhythm recognition result corresponding to the target user; wherein, the heart rhythm data classification model includes at least: a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module, and the target heart rhythm recognition result includes: the target heart rhythm type corresponding to the target user and explanatory information corresponding to the target heart rhythm type; The result feedback module is configured to send the target heart rhythm recognition result to the wearable device.

2. The system according to claim 1, characterized in that, The data acquisition module is configured to acquire the target user's unprocessed heart rhythm signal data and motion status data within a preset time period using the following method: When the data acquisition mode is determined to be the first mode, at least two pulse wave data to be processed and the motion state data of the target user within a preset time period corresponding to the first mode are acquired, and the at least two pulse wave data to be processed are filtered based on the motion state data to obtain filtered data corresponding to the first mode. When the data acquisition mode is determined to be the second mode, at least two pulse wave data, electrocardiogram data and motion state data of the target user within a preset time period corresponding to the second mode are acquired, and the at least two pulse wave data and electrocardiogram data to be processed are filtered based on the motion state data to obtain filtered data corresponding to the second mode. The heart rhythm signal data to be processed is determined based on the filtered data corresponding to the data acquisition mode.

3. The system according to claim 2, characterized in that, The data acquisition module is also configured to: Abnormal data detection is performed on the filtered data based on a preset threshold corresponding to the data acquisition mode to determine abnormal heart rhythm signal data; The abnormal heart rhythm signal data is uploaded to the cloud server as the heart rhythm signal data to be processed.

4. The system according to claim 1, characterized in that, The data processing module is configured to process the heart rhythm signal data to be processed and the motion state data based on a pre-trained heart rhythm data classification model in the following manner to obtain the target heart rhythm recognition result corresponding to the target user: Based on the feature extraction module, feature extraction is performed on the heart rhythm signal data to be processed to obtain the input features; For multiple heart rhythm type recognition modules, the input features are processed based on the heart rhythm type recognition modules to obtain recognition attribute information corresponding to the heart rhythm type recognition modules; wherein, the recognition attribute information includes: probability information and risk information of the heart rhythm type to which the module belongs; Based on the output module, the probability information of multiple corresponding heart rhythm types is processed to determine the target heart rhythm type; The target heart rhythm type and the risk information corresponding to the target heart rhythm type are processed by a preset rule engine to determine the explanatory information corresponding to the target heart rhythm type.

5. The system according to claim 4, characterized in that, The feature extraction module based on the dual-domain attention mechanism includes: a feature extraction unit based on the time-domain attention mechanism, a feature extraction unit based on the frequency-domain attention mechanism, and a feature fusion unit. The data processing module is configured to extract features from the heart rhythm signal data to be processed based on the feature extraction module in the following manner to obtain the input features: The time-domain feature extraction unit based on the time-domain attention mechanism is used to perform time-domain feature extraction on the heart rhythm signal data to be processed, so as to obtain the time-domain features to be processed. The frequency domain feature extraction unit based on the frequency domain attention mechanism is used to perform frequency domain feature extraction on the heart rhythm signal data to be processed, so as to obtain the frequency domain features to be processed. The heart rhythm signal data and motion state data to be processed are processed based on a preset activation function to obtain adaptive weighting coefficients; The time-domain features and frequency-domain features to be processed are fused according to the adaptive weighting coefficients to obtain fused features, which are then used as input features.

6. The system according to claim 1, characterized in that, The cloud server is also equipped with a model training module, which is configured as follows: Acquire sample heart rhythm data corresponding to different data acquisition modes and theoretical heart rhythm recognition results corresponding to the sample heart rhythm data; The sample heart rhythm data is input into the heart rhythm data classification model to be trained to obtain the output heart rhythm recognition result; Based on the output heart rhythm recognition result and the theoretical heart rhythm recognition result, determine the contrastive learning loss value, the weighted cross-entropy classification loss value, and the mean squared error loss value; The contrastive learning loss value, the weighted cross-entropy classification loss value, and the mean squared error loss value are weighted and fused to obtain the total loss value. The model parameters of the heart rhythm data classification model to be trained are corrected based on the total loss value to obtain the trained heart rhythm data classification model.

7. The system according to claim 1, characterized in that, The data acquisition module is also configured to: When the received target heart rhythm recognition result meets the preset conditions, the data acquisition mode is adjusted to the second mode to acquire the heart rhythm signal data to be processed based on the second mode.

8. The system according to claim 1, characterized in that, The wearable device is also equipped with a data feedback module, which is configured as follows: After receiving the target heart rhythm recognition result, obtain the user feedback data corresponding to the target heart rhythm recognition result; The user feedback data is sent back to the cloud server so that the cloud server can adjust the model parameters of the heart rhythm data classification model based on the user feedback data.

9. A method for processing cardiac rhythm data, characterized in that, An application is made in a cardiac rhythm data processing system, the system comprising: a wearable device and a cloud server; the wearable device is configured with: a data acquisition module and a data upload module; the cloud server is configured with: a data processing module and a result feedback module; the method includes: The data acquisition module responds to a user-triggered operation by determining a data acquisition mode corresponding to the user-triggered operation, and acquires the target user's unprocessed heart rhythm signal data and motion status data within a preset time period based on the data acquisition mode; wherein, the target user is a user wearing the wearable device. The data upload module uploads the heart rhythm signal data to be processed and the motion status data to the cloud server. The data processing module processes the heart rhythm signal data and motion state data based on a pre-trained heart rhythm data classification model to obtain a target heart rhythm recognition result corresponding to the target user. The heart rhythm data classification model includes at least a feature extraction module based on a dual-domain attention mechanism, multiple heart rhythm type recognition modules, and an output module. The target heart rhythm recognition result includes the target heart rhythm type corresponding to the target user and explanatory information corresponding to the target heart rhythm type. The target heart rhythm recognition result is sent to the wearable device through the result feedback module.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, enable the heart rhythm data processing system of any one of claims 1-8 to perform its functions.

Images