Electrocardiogram anomaly monitoring method and apparatus, device and storage medium

Abstract

An electrocardiogram anomaly monitoring method and apparatus, a computer device and a storage medium. The method comprises: (S210) acquiring PPG data of a user; (S220) inputting the PPG data into a pre-trained ECG anomaly monitoring model, so as to obtain an output value of the ECG anomaly monitoring model; wherein the ECG anomaly monitoring model is pre-trained for, on the basis of the input PPG data, outputting an output value indicating whether ECG data corresponding to the input PPG data is estimated to have a cardiac function anomaly; (S230) and when the output value of the ECG anomaly monitoring model indicates that the ECG data of the user is estimated to have a cardiac function anomaly, sending an alert to the user equipment associated with the user.

Description

Electrocardiogram abnormality monitoring method, device, equipment and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application filed with the China Patent Office on July 18, 2023, with application number 202310881402.6 and invention name “Electrocardiogram Abnormality Monitoring Method, Device, Equipment and Storage Medium”, the entire contents of which are incorporated by reference into this application. Technical Field

[0003] The present application relates to the technical field of electrocardiogram monitoring, and in particular to a method, apparatus, device and storage medium for monitoring abnormal electrocardiograms. Background Art

[0004] Electrocardiogram (ECG) is a medical examination method widely used to monitor / classify cardiac abnormalities (such as myocardial infarction, ventricular hypertrophy, heart failure, etc.).

[0005] The 12-lead ECG is a standard method for obtaining cardiac function information in primary care. It is measured by placing 10 electrodes on the skin surface of the chest and limbs to record the electrical activity of the heart. The ECG waveform consists of a QRS complex representing ventricular depolarization, a P wave representing atrial depolarization, and a T wave representing ventricular repolarization. Such a waveform is an informative and reliable measurement method reflecting cardiac function and is therefore widely used in clinical practice for the diagnosis of heart diseases. 12-lead ECG devices can measure ECG data more comprehensively, but they are usually bulky and equipped with electrodes, a central unit, and accessories such as a display and keyboard. It is difficult for ordinary users to conveniently use 12-lead ECG devices to perform daily continuous monitoring of their own ECG.

[0006] In contrast, portable ECG devices such as smartwatches and fitness trackers are smaller but only measure a single ECG lead, resulting in incomplete ECG data. Furthermore, they require user-initiated action, such as grasping a probe to close a conductive circuit, for measurement to occur. This requires the user to choose the appropriate moment to initiate ECG measurement. The paroxysmal nature of cardiac abnormalities, such as arrhythmias, makes choosing the right timing for ECG measurement challenging even for clinical experts. Consequently, continuous ECG acquisition and monitoring using portable ECG devices is difficult.

[0007] Therefore, the above existing ECG measurement methods still have room for improvement.

[0008] Summary of the Invention

[0009] According to various embodiments of the present application, a method, apparatus, device, and storage medium for monitoring electrocardiogram (ECG) abnormalities are provided.

[0010] In one aspect, a method for monitoring ECG abnormalities is provided, comprising: acquiring photoplethysmography (PPG) data of a user; inputting the PPG data into a pre-trained ECG abnormality monitoring model to obtain an output value of the ECG abnormality monitoring model; wherein the ECG abnormality monitoring model is pre-trained to output an output value indicating whether ECG data corresponding to the input PPG data is estimated to have abnormal cardiac function based on the input PPG data; and when the output value of the ECG abnormality monitoring model indicates that the ECG data is estimated to have abnormal cardiac function, causing a user device associated with the user to issue an alarm.

[0011] On the other hand, an ECG abnormality monitoring device is provided, comprising: a data acquisition module for acquiring PPG data of a user; a model monitoring module for inputting the PPG data into a pre-trained ECG abnormality monitoring model to obtain an output value of the ECG abnormality monitoring model; wherein the ECG abnormality monitoring model is configured to output an output value indicating whether ECG data corresponding to the input PPG data is estimated to have abnormal cardiac function based on the input PPG data; and an abnormality alarm module for causing a user device associated with the user to issue an alarm when the output value of the ECG abnormality monitoring model indicates that the ECG data is estimated to have abnormal cardiac function.

[0012] On the other hand, an ECG abnormality monitoring device is provided, including a PPG data detector, an output device, a memory, and a processor; the PPG data detector is used to monitor the PPG data of a user and transmit the measured PPG data to the processor; the memory stores a computer program; and when the processor executes the computer program, it is used to receive PPG data from the PPG data detector and implement the above method, so that when the output value of the ECG abnormality monitoring model indicates that the user's ECG data is estimated to have abnormal cardiac function, the output device is used to cause a user device associated with the user to issue an alarm.

[0013] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program implements the above method when executed by a processor.

[0014] The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will become apparent from the description, drawings, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to better describe and illustrate the embodiments and / or examples of the inventions disclosed herein, reference may be made to one or more of the accompanying drawings. The additional details or examples used to describe the accompanying drawings should not be considered to limit the scope of the disclosed inventions, the presently described embodiments and / or examples, and any of the best modes currently understood for these inventions.

[0016] FIG1 is a diagram illustrating an application environment of an ECG abnormality monitoring method according to an embodiment;

[0017] FIG2 is a schematic flow chart of an ECG abnormality monitoring method according to an embodiment;

[0018] FIG3 is an example of a PPG data segment retained by preprocessing in one embodiment;

[0019] FIG4 is an example of a data segment that is not filtered out by pre-processing in one embodiment;

[0020] FIG5 is a schematic diagram of the architecture of an ECG abnormality monitoring model according to one embodiment;

[0021] FIG6 is a schematic diagram of the architecture of a residual-dual convolutional attention block in one embodiment;

[0022] FIG7 is a schematic diagram of the architecture of a dual convolutional attention block in one embodiment;

[0023] FIG8 is a schematic diagram of the architecture of a convolutional attention block in one embodiment;

[0024] FIG9 is an example of ECG raw data segments and corresponding PPG raw data segments for leads II, V, and AVR in one embodiment;

[0025] FIG10 is an example of an ECG pre-processed data segment marked as normal and its corresponding PPG pre-processed data segment in one embodiment;

[0026] FIG11 is an example of an ECG pre-processed data segment marked as abnormal and its corresponding PPG pre-processed data segment in one embodiment;

[0027] FIG12 is a schematic structural diagram of an ECG abnormality monitoring device according to one embodiment;

[0028] FIG13 is a schematic diagram of the structure of an ECG abnormality monitoring device in one embodiment. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0030] Due to the close relationship between cardiac function and ECG data, and the rich information contained in ECG data, especially the ECG waveform structure and implied time information, the 12-lead ECG is a standard method for monitoring cardiac dysfunction in clinical practice. However, existing ECG measurement devices are inconvenient to use. For example, 12-lead ECG devices are bulky and inconvenient for daily monitoring. Portable ECG devices can usually only measure one ECG lead and usually require the user to "actively" sample the ECG through the device, making it difficult to determine the appropriate measurement timing.

[0031] PPG is an optical technology that measures the cardiac cycle by detecting changes in blood volume. PPG can be measured "passively" through pulse oximeters, for example, and is often embedded as a function in wearable devices such as fitness bands and smart watches. The PPG waveform consists of a systolic wave and a diastolic wave, which is simpler and smoother than the ECG waveform. PPG can be conveniently used for long-term continuous monitoring of physiological parameters such as pulse and respiratory rate. However, it is very difficult to classify or monitor cardiac abnormalities based on PPG because it is a peripheral measurement signal that indirectly monitors the operation of the heart, and its relatively smooth shape reflects less cardiac information compared to ECG. Therefore, existing PPG is mostly used to measure physiological parameters such as pulse rate, respiratory rate, and blood pressure.

[0032] This application proposes an ECG abnormality monitoring method that uses the continuously monitored, ubiquitous PPG signal to detect possible ECG abnormalities, including but not limited to common cardiac function abnormalities such as atrioventricular block, sinus rhythm abnormalities, bundle branch block, atrial fibrillation, and premature beats. This method prompts the user to undergo an ECG examination to determine cardiac risks.

[0033] The ECG abnormality monitoring method provided in this application can be applied to the application environment shown in Figure 1. Therein, the terminal 102 communicates with the server 104 through a network. The terminal 102 can be a terminal device capable of measuring PPG data. For example, the terminal 102 can be provided with a PPG data detector, such as a pulse oximeter. For example, the terminal 102 can be a portable wearable terminal device such as a smart watch, a smart bracelet, or a smart headset with a built-in PPG data detector. The terminal 102 can also have other forms, such as a medical monitor equipped with a pulse oximeter, or a system consisting of a smart bracelet with a built-in pulse oximeter and a smart phone that wirelessly communicates with the smart bracelet, etc. Those skilled in the art will be able to think of many other variations, which are not limited in this application. The server 104 can be implemented as an independent server or a server cluster consisting of multiple servers.

[0034] The ECG anomaly monitoring method of the present application may include a training phase and a testing phase. During the training phase, an ECG anomaly monitoring model may be trained using a training data set to obtain a trained ECG anomaly monitoring model. The anomalies include, but are not limited to, common ECG lead anomalies such as atrioventricular block, sinus rhythm abnormalities, bundle branch block, atrial fibrillation, and premature beats. The training process may be executed on the server 104, and after the training is completed to obtain the trained ECG anomaly monitoring model, the trained ECG anomaly monitoring model is deployed from the server 104 to the terminal 102 for use. After the training is completed to obtain the trained ECG anomaly monitoring model, during the testing phase, the terminal 102 may use its PPG detector to monitor the user's PPG data and execute the ECG anomaly monitoring method of the present application embodiment. When the output value of the ECG anomaly monitoring model indicates that the user's ECG data is estimated to be abnormal, the user device associated with the user issues an alarm, for example, causing the terminal 102 itself to issue an alarm and / or causing other associated devices outside the terminal 102 to issue an alarm. It can be understood that where some or all of the steps of the above-mentioned ECG abnormality monitoring method are performed can be appropriately adjusted between the terminal 102 and the server 104. For example, the trained ECG abnormality monitoring model can also be deployed on the server 104. The terminal 102 sends the collected PPG data to the server 104, and the server 104 uses the model to monitor and send the monitoring results back to the terminal 102, etc. This application does not limit this.

[0035] Furthermore, the above-mentioned alarm can be used to instruct to further perform an ECG measurement on the user. The ECG measurement can be performed by the terminal 102. For example, the terminal 102 can be equipped with an ECG measurement device, such as ECG electrodes or ECG sensors, so that the ECG measurement of the user can be performed using the terminal 102. The ECG measurement can also be performed by any other ECG measurement device outside the terminal 102.

[0036] In one embodiment, as shown in FIG2 , a method for monitoring ECG abnormalities is provided. In a testing phase, the method includes the following steps S210 - S230 :

[0037] S210, obtaining the user's PPG data.

[0038] PPG is a noninvasive method that uses photoelectric means to detect changes in blood volume in living tissue. PPG can be detected using a PPG data detector, any device capable of noninvasively detecting a user's PPG. A PPG data detector may include a light source and a photodetector. The light source emits light of a predetermined wavelength toward the tissue being measured, while the photodetector receives light reflected or transmitted by the tissue. A PPG signal is generated based on the intensity of the light reflected or transmitted by the tissue. For example, a PPG data detector may be a pulse oximeter.

[0039] In one example, the pulse oximeter can be built into terminal 102, which can be, for example, a smartwatch. When a user wears the smartwatch, the pulse oximeter located at the bottom of the smartwatch dial (close to the user's skin) detects PPG data from the inside of the user's wrist. In other examples, terminal 102 can have other forms, and the PPG data measurement site is not limited to the inside of the user's wrist, but can be other sites such as the inside of the lower ankle or the chest. It will be understood that the measurement site of the test data collected during the testing phase is the same as the measurement site of the training data used during the training phase.

[0040] PPG data is data that characterizes the change in the user's pulse wave amplitude over time, which can be expressed as one-dimensional data, that is, a PPG one-dimensional data sequence obtained by arranging the pulse wave sampling values ​​in sampling time order.

[0041] In one embodiment, the acquired PPG data of the user is the PPG data of the user monitored in real time by a portable terminal device.

[0042] In one example, terminal 102 continuously samples and monitors the test site of the user under test at a predetermined sampling frequency. Terminal 102 extracts PPG data segments of a predetermined duration from the continuously monitored PPG signal as PPG data to be input into the model. The predetermined sampling frequency and predetermined duration can be set according to actual needs. For example, the predetermined sampling frequency can be 125 Hz, and the predetermined duration can be 10 seconds. Thus, a 10-second PPG data segment can include approximately 1250 sample values. The data represented by the dark gray curve in FIG3 illustrates an example of a 10-second PPG raw data segment. The horizontal axis represents time, represented by the sampling point number, and the vertical axis represents amplitude. The horizontal and vertical axes in the ECG and PPG data segments shown in the remaining figures of this application have similar meanings.

[0043] The raw PPG data collected may contain noise or low quality, which may affect the accuracy of the monitoring results. Therefore, the collected raw data can be preprocessed and then used as input into the model for monitoring.

[0044] In one embodiment, step S210 may include obtaining the user's raw PPG data and preprocessing the raw PPG data to obtain preprocessed PPG data. The preprocessed PPG data is then used as the PPG data to be input into the model. Accordingly, in the subsequent step S220, the preprocessed PPG data may be input into a pre-trained ECG abnormality monitoring model.

[0045] In one embodiment, preprocessing the PPG raw data to obtain PPG preprocessed data may include one or more of the following steps:

[0046] 1) Flatness detection and signal screening:

[0047] In this step, it is detected whether the flatness of the signal meets a predetermined standard. If so, the signal is retained; otherwise, the signal is discarded.

[0048] For example, for a 10-second segment of PPG raw data, if the sample values ​​of any 60 consecutive sampling points (approximately 0.5 seconds) do not exceed the threshold (1e-5), it is considered a flat signal and the entire 10-second data segment is discarded. Otherwise, the data segment is retained.

[0049] 2) Normalization:

[0050] In this step, the signal is normalized to zero mean and unit variance.

[0051] For example, each sampling value in the data segment retained in the above-mentioned flatness detection and signal screening steps can be normalized to zero mean and unit variance to obtain a normalized data segment.

[0052] 3) Filtering:

[0053] In this step, the signal is filtered.

[0054] For example, a third-order bandpass Butterworth filter may be used to filter the normalized data segments in the above step to obtain filtered data segments, wherein the filter has a low-band cutoff of 0.5 Hz and a high-band cutoff of 8 Hz.

[0055] 4) Peak detection and signal screening:

[0056] In this step, it is detected whether the number of effective peaks per unit length of the signal reaches a predetermined standard. If so, the signal is retained; otherwise, the signal is discarded.

[0057] For example, the Python toolbox "HeartpPy" can be used to detect the number of valid peaks in the data segment after the above step. If the number of valid peaks in the 10s data segment is less than 5 (corresponding to less than 30bpm), the data segment is discarded. Otherwise, the data segment is retained.

[0058] 5) Skewness SQI detection and signal screening:

[0059] In this step, it is detected whether the skewness of the signal meets the predetermined standard. If so, the signal is retained; otherwise, the signal is discarded.

[0060] For example, the skewness is calculated using a sliding window with a window width of 250 samples (two seconds) and a step size of 125 samples (one second). For the 10-second data segment retained in the aforementioned peak detection and signal screening steps, the skewness of the sample values ​​in each of the nine time windows (0-2s, 1-3s, 2-4s, 3-5s, 4-6s, 5-7s, 6-8s, 7-9s, and 8-10s) can be calculated. If the majority of the nine calculated skewnesses (over 50%) are negative, the data segment is considered to be a low-quality signal and is discarded. Otherwise, the data segment is retained.

[0061] 6) Replacement of outlier samples:

[0062] In this step, outlier sampling values ​​in the signal are detected, and the detected outlier sampling values ​​are replaced with the median of the sampling values ​​to obtain a signal after eliminating the outlier sampling values.

[0063] For example, a Hampel filter can be applied to detect outliers. For every 10 consecutive sample values ​​in the data segment retained from the aforementioned skewness SQI detection and signal screening steps, the median absolute deviation (MAD) is calculated. The standard deviation (std) of the data segment is then estimated based on the MAD value assuming a normal distribution. If a sample value differs from the median of the data segment by 3 std, the sample value is detected as an outlier. The detected outlier is replaced with the median of the data segment. This allows for the removal of a large number of outliers at the beginning and end of the data segment.

[0064] Step 1) can be considered a quality screening step. Steps 2) and 3) are cleaning and denoising, and steps 4) to 6) are quality control steps. Referring to FIG3 , FIG3 shows a PPG data segment retained after the above-mentioned preprocessing steps 1) to 6), wherein the dark gray curve represents the original data segment, and the light gray curve represents the preprocessed data segment obtained after preprocessing. Referring to FIG4 , FIG4 shows examples of data segments that were screened out due to failure to pass the different quality control criteria in the above-mentioned preprocessing steps. In FIG4 , (a) represents a data segment that was screened out due to substandard flatness, (b) represents a data segment that was screened out due to substandard number of valid peaks, and (c) represents a data segment that was screened out due to substandard skewness.

[0065] In this example, for any raw data segment, the preprocessed data segment obtained after the above preprocessing steps can be used to input into the model for monitoring in the subsequent steps. However, if any raw data segment is discarded in any of the above steps, it will no longer be used for monitoring. In other words, only preprocessed data segments that meet the requirements after being filtered out by the preprocessing steps are used for monitoring, thereby ensuring monitoring accuracy. It will be understood that the preprocessing operations performed on the raw data during the testing phase can be the same as those performed on the raw data during the training phase to ensure the applicability of the model.

[0066] S220 , inputting the PPG data into a pre-trained ECG abnormality monitoring model to obtain an output value of the ECG abnormality monitoring model.

[0067] Among them, the ECG anomaly monitoring model of the present application is trained using a training data set including a data pair including PPG data and a label indicating whether the ECG data corresponding to the PPG data is abnormal. The architecture and training process of the ECG anomaly monitoring model will be described later.

[0068] The trained ECG anomaly monitoring model can be used to output an output value indicating whether the ECG data corresponding to the input PPG data is estimated to have an anomaly based on the input PPG data.

[0069] For example, when the 10s data segment obtained in the above step is input into the ECG abnormality monitoring model, the ECG abnormality monitoring model can output the monitoring category probability As the output value, the probability of the monitoring category By comparing it with the set threshold T, it can be determined whether the corresponding ECG data is estimated to be abnormal. indicates that the ECG data estimate is abnormal when indicates that there is no abnormality in the ECG data.

[0070] S230 : When the output value of the ECG abnormality monitoring model indicates that an abnormality exists in the estimated ECG data, a user device associated with the user sends an alarm.

[0071] In one embodiment, the user device associated with the user in S230 may include the aforementioned terminal 102, such as a portable terminal device worn by the user, to inform the user that he or she has an ECG abnormality risk and should take ECG measurements in a timely manner.

[0072] In another embodiment, the user device associated with the user in S230 may also be a monitoring terminal device of a guardian associated with the user, such as a terminal device used by the user's family member or caregiver, or a terminal device used by the user's family doctor or attending physician, or a terminal device or server of the monitoring agency to which the user belongs, etc., to inform the guardian that the user is at risk of ECG abnormality and that ECG measurement should be performed in a timely manner.

[0073] In one embodiment, causing a user device associated with a user to issue an alarm includes one or more combinations of the following: displaying visual information on a display screen of the user device indicating that an abnormality is estimated in the user's ECG data, causing an indicator light of the user device to light up indicating that an abnormality is estimated in the user's ECG data, causing a vibrator of the user device to vibrate indicating that an abnormality is estimated in the user's ECG data, and causing a sound output device of the user device to sound a sound prompt indicating that an abnormality is estimated in the user's ECG data.

[0074] For example, the content that causes the user device associated with the user to issue an alarm includes not only the monitoring result informing that an abnormality is estimated in the ECG data, but also suggestion information suggesting that the user actively take ECG measurements.

[0075] Furthermore, in one embodiment, the above method may further include: when the output value of the ECG abnormality monitoring model indicates that the ECG data is estimated to be abnormal, making an alarm call to a predetermined alarm receiving device.

[0076] The alarm call may be initiated by an alarm initiating device, which may be any suitable terminal or server. For example, the alarm call may be initiated by a portable terminal device worn by the user. In this case, the portable terminal device needs to be a terminal device with a phone call function, such as a smartwatch or a smart bracelet and a mobile phone paired with the smart bracelet. Alternatively, the alarm call may be initiated by, for example, the server 104 or other servers or terminal devices.

[0077] The alarm call can be automatically initiated by the terminal or server or decided based on the user's instructions. For example, the terminal or server can provide the user with the option of initiating an alarm call through a portable terminal device worn by the user, and then decide whether to make an alarm call to the predetermined alarm receiving device based on the user's selection input of the option. Alternatively, after providing the above options to the user, if the terminal or server does not receive the user's selection input within a predetermined period of time, it may be considered that the user may be in a dangerous situation such as losing consciousness, and the alarm call will be automatically made to the predetermined alarm receiving device.

[0078] The predetermined alarm receiving device for receiving the alarm call can be pre-set as needed. For example, the predetermined alarm receiving device may include the monitoring terminal device of the guardian associated with the user, and / or the alarm terminal of the designated medical institution (for example, the hospital emergency telephone terminal that can be dialed by dialing 112), etc.

[0079] Depending on the pre-set settings or the user's selected input, the alarm call can be a real-time call that allows the user to talk to the recipient at the predetermined alarm receiving device. The alarm call can also include pre-set voice broadcast content, for example, it can include voice broadcast content that informs the user that the ECG data is estimated to have abnormal monitoring results. In addition, it can also include voice broadcast content such as the user's name, gender, age, residence and other personal information and the user's real-time location, so that the alarm call recipient can quickly know the user's current condition and real-time location.

[0080] In one embodiment, the method may further include: when the output value of the ECG abnormality monitoring model indicates that the ECG data is estimated to be abnormal, performing ECG measurement on the user to obtain a measurement result indicating whether the user's ECG is abnormal.

[0081] The ECG measurement may be performed by the terminal 102 equipped with an ECG measurement device, or by another ECG measurement device different from the terminal 102, such as a 12-lead ECG device installed in a medical institution, or a home ECG measurement device. For example, after the terminal 102 issues an alarm to remind the user to perform an ECG measurement in step S230, the terminal 102 may determine whether to trigger the ECG measurement of the user by the terminal 102 based on the user input. For another example, the terminal 102 may automatically trigger the ECG measurement of the user when the output value indicates that the ECG data is estimated to be abnormal. For another example, after the terminal 102 issues an alarm to remind the user / guardian to perform an ECG measurement in step S230, the user / guardian may choose to perform an ECG measurement of the user using another ECG measurement device other than the terminal 102. For example, the user may go to a medical institution to perform a more professional and comprehensive ECG measurement using a 12-lead ECG and obtain a measurement result indicating whether the ECG is abnormal. When the terminal 102 performs the above-mentioned ECG measurement, the terminal 102 can analyze the measured ECG data by itself or via an external device such as a server, and push the measured ECG data and the measurement result of whether the ECG is abnormal to the user through the terminal 102.

[0082] The above-mentioned ECG abnormality monitoring method, since PPG data can be obtained by measuring the user in a passive manner through a portable wearable device such as a smart watch, can conveniently and continuously monitor the user by obtaining the user's PPG data. By cleverly using a label indicating whether the ECG data corresponding to the PPG data is abnormal to train the data pairs formed by marking the PPG data, an ECG abnormality monitoring model can be obtained that can output an output value indicating whether the ECG data corresponding to the input PPG data is estimated to be abnormal based on the input PPG data. The model can be used to monitor whether the corresponding ECG data is likely to be abnormal based on the measured PPG data, and when an abnormality is detected, the user device associated with the user will issue an alarm to suggest the user to actively perform ECG measurement to obtain an accurate result of whether the ECG is abnormal.

[0083] The above describes the method steps that the terminal 102 or the server 104 may perform during the testing phase, wherein in step S220 of the testing phase, a pre-trained ECG abnormality monitoring model is used. This requires that in the training phase before the testing phase, the model architecture of the ECG abnormality monitoring model is pre-constructed, and the constructed ECG abnormality monitoring model is trained to optimize and determine the model parameters of the ECG abnormality monitoring model to obtain a trained signal quality assessment model.

[0084] This application proposes a dual-convolutional attention network (DCA-Net) as the above-mentioned ECG abnormality monitoring model. The dual-convolutional attention network is characterized in that it includes one or more residual-dual convolutional attention (Res-DAC) blocks. Each residual-dual convolutional attention block is formed by adding a dual-convolutional-attention (DCA) block to the residual block. Among them, the dual-convolutional attention block includes a first convolutional attention block for convolving data in the channel dimension and a second convolutional attention block for convolving data in the time dimension.

[0085] In one embodiment, as shown in FIG5 , the ECG abnormality monitoring model may include at least a plurality of residual-dual convolution attention blocks 1 to N arranged in series.

[0086] In one embodiment, as shown in Figure 5, the ECG abnormality monitoring model may also include an input convolution layer and a pooling layer located before the multiple residual-double convolution attention blocks and arranged in series with the multiple residual-double convolution attention blocks, and an average pooling layer and a fully connected layer located after the multiple residual-double convolution attention blocks and arranged in series with the multiple residual-double convolution attention blocks.

[0087] Among them, the input convolution layer receives the PPG data input to the ECG anomaly monitoring model, the output of the input convolution layer is fed into the pooling layer, the output of the pooling layer is fed into the first residual-double convolution attention block in multiple residual-double convolution attention blocks, the output of the previous one in multiple residual-double convolution attention blocks is fed into the next one, and the output of the last residual-double convolution attention block in multiple residual-double convolution attention blocks is fed into the average pooling layer, the output of the average pooling layer is fed into the fully connected layer, and the fully connected layer outputs the output value of the ECG anomaly monitoring model.

[0088] The last layer of the fully connected layer can have a single neuron, which outputs a 1*1 dimensional output value, that is, the probability The probability By comparing it with the set threshold T, the monitoring result of whether the ECG is estimated to be abnormal can be obtained.

[0089] For example, each convolution layer in the model, such as the input convolution layer, can be a one-dimensional convolution layer to adapt to the processing requirements of the PPG one-dimensional data sequence.

[0090] It can be understood that the model architecture in Figure 5 is only an example. Under the basic premise that the model includes one or more residual-dual convolution attention blocks as proposed in this application, those skilled in the art can also make many variations to the model architecture. For example, the number and serial / parallel mode of the residual-dual convolution attention blocks can be changed, and the number and position of the convolution layers and pooling layers in the model can also be changed.

[0091] In one embodiment, referring to Figures 6 and 7, each of the multiple residual-dual convolution attention blocks may include a first main path and a first short-circuit branch path. One or more convolution layers and a dual convolution attention block are serially arranged on the first main path. The dual convolution attention block includes a first convolution attention block and a second convolution attention block arranged in series or in parallel, the first convolution attention block is used to perform convolution in the channel dimension, and the second convolution attention block is used to perform convolution in the time dimension. The first short-circuit branch path is connected in parallel to one or more convolution layers and both ends of the dual convolution attention block arranged in series on the first main path.

[0092] Among them, the former of one or more convolutional layers and dual convolutional attention blocks and the first short-circuit branch path receive the output of the previous path, the latter of one or more convolutional layers and dual convolutional attention blocks receive the output of the former, and the output of the last of one or more convolutional layers and dual convolutional attention blocks is added to the output matrix of the first short-circuit branch path and fed into the subsequent path.

[0093] In one embodiment, as shown in FIG6 , a batch normalization layer and an activation function layer are respectively provided on the output side of each convolution layer in each residual-dual convolution attention block. In each residual-dual convolution attention block, the batch normalization layer and the activation function layer on the output side of each convolution layer except the last convolution layer are arranged before the next convolution layer, and the batch normalization layer on the output side of the last convolution layer is arranged between the convolution layer and the dual convolution attention block, the dual convolution attention block is arranged before the addition operation with the output of the first short-circuit branch path, and the batch normalization layer arranged on the output side of the last convolution layer is arranged after the addition operation with the output of the first short-circuit branch path.

[0094] In one specific example, as shown in Figure 6, each residual-dual convolutional attention block is configured with two one-dimensional convolutional layers. The activation function layer is a ReLU activation function layer. In other examples, the number and dimensions of the convolutional layers in the residual-dual convolutional attention block can be changed as needed, and the activation function layer can also be a different type of activation function layer.

[0095] In one embodiment, as shown in FIG7 , the first convolutional attention block and the second convolutional attention block are set in series, and the first convolutional attention block is located before the second convolutional attention block. Placing the first convolutional attention block in terms of channel before the second convolutional attention block in terms of time allows the dual convolutional attention blocks to first learn the channel-to-channel interaction and then learn the temporal interaction based on the channel participation output, which performs better than the two in parallel or the first convolutional attention block in terms of channel being placed in the latter order. In other examples, the second convolutional attention block may be located before the first convolutional attention block or the first convolutional attention block and the second convolutional attention block may be set in parallel.

[0096] In one embodiment, as shown in FIG8 , each of the first convolutional attention block and the second convolutional attention block includes: a second main path and a second short-circuit branch path. The second main path is provided with a maximum pooling layer and an average pooling layer, a convolution layer, and an activation function layer. The maximum pooling layer and the average pooling layer are provided in parallel, and the parallel maximum pooling layer and the average pooling layer are sequentially provided in series with the convolution layer and the activation function layer. The second short-circuit branch path is connected in parallel to both ends of the maximum pooling layer and the average pooling layer, the convolution layer, and the activation function layer provided on the second main path.

[0097] Among them, the second short-circuit branch path, the maximum pooling layer and the average pooling layer respectively receive the output of the previous path, the output of the maximum pooling layer and the output of the average pooling layer are combined (concatenation) to obtain the output fed into the convolution layer, the output of the convolution layer is fed into the activation function layer, and the output of the activation function layer is multiplied by the output of the second short-circuit branch path and fed into the subsequent path.

[0098] In one specific example, as shown in FIG8 , the first / second convolutional attention block includes a one-dimensional convolutional layer. The activation function layer is a Sigmoid activation function layer. In other examples, the number and dimensions of the convolutional layers in the first / second convolutional attention block can be changed as needed, and the activation function layer can also be another type of activation function layer.

[0099] The convolutional attention block relies on applying a convolution operation to a certain dimension of the data to learn the interactions between data points along that dimension. Compared to the MLP attention module, convolutional attention is more computationally efficient because no parameters are required in the convolution layer. As can be seen in Figure 8, the first and second convolutional attention blocks have similar architectures, but they are applied in different directions: the first convolutional attention block is applied in the convolution channel direction, and the second convolutional attention block is applied in the time domain direction.

[0100] In the above embodiments, the specific configurations of the parameters of each layer / block, such as activation function, kernel size, step size, padding, etc., can be set according to actual needs.

[0101] Specifically, for the specific examples shown in Figures 7 and 8, taking the convolution layer in Figure 8 as a one-dimensional convolution layer with a convolution kernel size of 7 as an example, let the input of the first convolution attention block in the channel (also the input of the DCA module) be Where N is the sampling point / batch size, C is the number of convolution channels output from the original residual block, and D is the signal length (denoted as the time dimension). First Convolutional Attention Block The output is:

[0102] where g(X) = [p1(X) || p2(X)] D (2)

[0103] Where p1(X)=max_pool D (X) (3)

[0104] And, p2(X)=mean_pool D (X) (4)

[0105] In equation (1), represents the Hadamard product (element-wise matrix multiplication); σ(·) represents the Sigmoid activation function, represents a 1D convolution layer with a kernel size of 7 and convolution in the channel direction. [·||·] in Equation (2) D The subscript D in equations (3) and (4) indicates that the pooling layer is also applied in the time dimension. Output channel-wise attention weights, which are further extended to X of the same shape to scale X. Then the scaled X (denoted as ) is used as input to the temporal attention.

[0106] Similarly, let the output of the second convolutional attention block in time be

[0107] in,

[0108] in,

[0109] and,

[0110] Equations (7)-(8) indicate that the maximum pooling and average pooling are first performed on the channel dimension, so and Then, the pooled matrix is ​​combined along the channel dimension, (Equation (6)). The one-dimensional convolutional layer convolutions in the time dimension and outputs a single channel such that Finally, this convolved matrix is ​​activated by the Sigmoid activation function and used to scale the input time domain.

[0111] In a specific example, a DCA-Net model of a specific example can be obtained by replacing all two-dimensional convolutional layers in the known ResNet-34 model with one-dimensional convolutional layers, replacing the last fully connected layer in the ResNet-34 model with a fully connected layer with a 512-dimensional input and a 1-dimensional output, and adding a double convolutional attention block before adding the residual to each residual block of the ResNet-34 model. It can be understood that the DCA-Net model of this specific example will include a serial input convolution layer, a pooling layer, 16 residual-double convolutional attention blocks, an average pooling layer, and a fully connected layer. The architecture of the DCA-Net model of this specific example can be shown in Figure 5, where N is N=16, and the architecture of each residual-double convolutional attention block can be shown in Figures 6 to 8. Among them, for example, in each residual-double convolution attention block, the convolution kernel size of the first convolution attention block and the second convolution attention block can be 7, the step size is 1, the average pooling layer can be a one-dimensional adaptive average pooling layer (AdaptiveAvgPool1d), and the parameters of other layers and blocks in the model can refer to the parameter settings of the known ResNet-34 model for example.

[0112] For example, taking batch training as an example, the forward propagation process of the ResNet-34 model of this specific example is explained. When the batch size is 64, when a one-dimensional PPG data of 64x1x1250 is input to the DCA-Net model of this specific example, after passing through the input convolution layer (with a stride of 2), 64×64×625 data is output, and after passing through the (maximum) pooling layer (with a stride of 2), 64×64×313 data is output. The data is then input into the first residual-double convolutional attention block. After passing through the convolution layer, batch normalization layer, ReLU activation function layer, convolution layer, and batch normalization layer shown in Figure 6, it is input into the double convolutional attention layer. The input data first enters the first convolutional attention block in terms of the channel. In the first convolutional attention block, see Figure 7, the average pooling layer performs average pooling on the input data in the time dimension and outputs a 64×64×1 matrix. At the same time, the maximum pooling layer performs maximum pooling on the input data in the time dimension and outputs another 64×64×1 matrix. These two matrices are combined along the time dimension (similar to a splicing operation) to obtain a 64×64×2 matrix. This 64×64×2 matrix is ​​convolved along the channel dimension through a convolutional layer to learn 64×64×1 attention weights. It is then replicated and expanded along the temporal dimension to form a 64×64×313 attention score. This attention score is Hadamard-multiplied with the original input data output via the second short-circuit branch path, outputting a 64×64×313 matrix. Next, the 64×64×313 matrix output by the first convolutional attention block is input to the second convolutional attention block in time. Within this second convolutional attention block, as shown in Figure 7, the average pooling layer performs average pooling on the 64×64×313 matrix along the channel dimension, outputting a 64×1×313 matrix. Simultaneously, the max pooling layer performs max pooling on the input data along the channel dimension, outputting another 64×1×313 matrix. These two matrices are combined to form a 64×2×313 matrix. The 64×2×313 matrix is ​​convolved along the time dimension through the convolution layer to obtain a 64×1×313 attention weight, which is then replicated and expanded along the channel dimension into a 64×64×313 attention score. This attention score is Hadamard multiplied with the original input 64×64×313 matrix output through the second short-circuit branch path, and a 64×64×313 matrix is ​​output. This 64×64×313 matrix is ​​then input into the next residual-double convolution attention block as the output of the second convolution attention block. After 16 residual-double convolution attention blocks, the output data is then passed through the average pooling layer and the fully connected layer to output the output value of the model.

[0113] In the training phase, the ECG abnormality monitoring model constructed in any of the above embodiments is trained to determine a trained ECG abnormality monitoring model. The ECG abnormality monitoring model with the above structure can be trained using various conventional or novel training methods.

[0114] To perform model training, a training dataset needs to be constructed. In this application, the training dataset can include multiple pairs of PPG data and labels indicating whether the corresponding ECG data is abnormal. The PPG data included in the training dataset, as well as the corresponding ECG data required for labeling, can be collected from multiple cases corresponding to multiple subjects. The number of subjects from which the data is sourced can be set as needed, for example, but not limited to, hundreds, thousands, or tens of thousands, to obtain data that statistically reflects the characteristics of the majority of users. For each of the multiple subjects, one or more PPG data and corresponding ECG data can be collected. Each ECG data item is then labeled, for example, manually or with machine assistance by a physician, to indicate whether it indicates cardiac function abnormality. The label is then assigned to the PPG data corresponding to the ECG data, thereby obtaining one or more pairs of PPG data and labels indicating whether the corresponding ECG data is abnormal for the subject. By combining the multiple pairs of PPG data and labels indicating whether the corresponding ECG data is abnormal collected from the multiple subjects, a training dataset comprising data from multiple subjects can be obtained. The aforementioned multiple subjects may include at least some patients with abnormal cardiac function, so that at least some PPG data and corresponding ECG data with characteristics of abnormal cardiac function can be collected. Furthermore, the aforementioned multiple subjects may also include some subjects with normal cardiac function, or, although only patients with abnormal cardiac function are included, at least some PPG data and corresponding ECG data with characteristics of normal cardiac function (e.g., normal data collected during a period when the patient is not ill) can be collected. For example, in practice, the PPG data and corresponding ECG data used to construct the training dataset can be collected from clinical medical data collected from a large number of different subjects during hospitalization at one or more hospitals. It is worth noting that it is understood that the subjects from which the training data used in the training phase originates do not need to be or be included in the subjects tested in the testing phase. For example, a large amount of clinical data from a medical system can be used to construct a training dataset to train the model, and the trained model can then be deployed on the terminal 102. Thereafter, any user can use the terminal 102 to perform the aforementioned testing phase process to monitor their own electrocardiogram abnormalities.

[0115] For example, the process of constructing a training dataset may include the following steps:

[0116] 1) Obtain ECG signals of multiple leads and their corresponding PPG signals.

[0117] The correspondence between the ECG signal and the PPG signal means that the ECG signal and the PPG signal are measured simultaneously on the same subject to ensure that the ECG signal and the PPG signal are time-aligned and correlated with each other.

[0118] For example, ECG signals from multiple leads and their corresponding PPG signals can be obtained from a matching subset of the MIMIC-III Waveform Database (WDB). The MIMIC-III WDB is linked to clinical records—a matching subset can contain a large number of patients, and all patients in the matching subset are identified and matched to the medical record system. The MIMIC-III WDB includes ECG signals from multiple leads and their paired PPG signals. For each patient, waveform data of varying lengths is collected at different time points during their hospitalization. These signals are sampled at a frequency of 125 Hz. Because different ECG recordings have varying numbers and types of leads, in this example, data from the three most common and representative ECG leads—leads II, V, and AVR—are selected. Therefore, ECG signal recordings using these three leads and their corresponding PPG signal recordings are included in this analysis. However, in other examples, more or fewer ECG leads may be selected, and the types of selected leads may vary. Furthermore, to reduce noise levels, only recordings with a duration exceeding a predetermined value, for example, two minutes, are selected. Signals shorter than two minutes will be filtered out.

[0119] 2) Extracting multiple leads of ECG raw data segments and their corresponding PPG raw data segments of a predetermined length from the acquired multiple leads of ECG signals and their corresponding PPG signals.

[0120] During the training process, the predetermined duration of the data segments to be captured for training data is the same as that of the data segments captured during the test process, for example, 10 seconds. This allows a second 10-second data segment to be captured from each ECG / PPG signal to avoid instability in the initial sampling period. An example of the resulting 10-second ECG raw data segments for Leads II, V, and AVR, along with the corresponding 10-second PPG raw data segments, is shown in Figure 9.

[0121] 3) Preprocessing the intercepted ECG raw data segments of the plurality of leads of the predetermined time length and their corresponding PPG raw data segments respectively, so as to preprocess and screen out the ECG preprocessed data segments of the plurality of leads of the predetermined time length and their corresponding PPG preprocessed data segments that meet the requirements.

[0122] For the 10 s PPG raw data segment, the same preprocessing process including steps 1) to 6) as in the test process described above can be performed to screen out the PPG preprocessed data segments that meet the requirements.

[0123] For 10s of ECG raw data segments of leads II, V, and AVR, a 5th-order high-pass Butterworth filter of 0.5Hz can be used, followed by power line filtering with a power line frequency of 50Hz to obtain 10s of ECG pre-processed data segments of leads II, V, and AVR. Then, a predetermined signal quality assessment method is used to assess the signal quality of each ECG pre-processed data segment. The assessment outputs one of three quality categories: unacceptable, almost unacceptable, or excellent. When the ECG pre-processed data segment of a certain lead is assessed as excellent, the ECG pre-processed data segment of the lead is retained. Conversely, when the ECG pre-processed data segment of a certain lead is assessed as unacceptable or almost unacceptable, the ECG pre-processed data segment of the lead is discarded.

[0124] Thus, for any set of 10s of multiple-lead ECG raw data segments and their corresponding PPG raw data segments, if any one of the lead ECG raw data segments or the PPG raw data segments cannot be retained through the above preprocessing process, the entire set of data segments is discarded. Only when all of the multiple-lead ECG raw data segments and their corresponding PPG raw data segments in a set are retained through the above preprocessing process can the set of multiple-lead ECG raw data segments and their corresponding PPG raw data segments of the predetermined duration be obtained.

[0125] 4) For each group of ECG preprocessed data segments of a predetermined duration and their corresponding PPG preprocessed data segments, determining a label that indicates whether the ECG preprocessed data segments of the multiple leads are abnormal, and forming a data pair with the label and the PPG preprocessed data segment, thereby obtaining a training dataset including multiple groups of PPG data and a label indicating whether the ECG data corresponding to the PPG data is abnormal.

[0126] Since ECG labels are not provided in the MIMIC-III WDB, for example, a known ECG classification model with high accuracy, such as AutoNet, can be used to efficiently label whether ECG pre-processed data segments are abnormal. Referring to Figures 10 and 11, Figure 10 shows an example of an ECG pre-processed data segment labeled as normal and its corresponding PPG pre-processed data segment, and Figure 11 shows an example of an ECG pre-processed data segment labeled as abnormal and its corresponding PPG pre-processed data segment. In other alternative examples, professional technicians, such as clinicians, can also manually label whether ECG pre-processed data segments are abnormal to ensure the accuracy of the labels. In this way, multiple sets of data pairs such as PPG pre-processed data segments and labels indicating whether the corresponding pre-processed data segments are abnormal can be obtained to form the required training dataset.

[0127] After the training data set is constructed, the ECG abnormality monitoring model can be trained using the training data set to determine a trained ECG abnormality monitoring model.

[0128] In one embodiment, the training process of the ECG abnormality monitoring model includes:

[0129] Obtain a training data set; the training data set includes multiple sets of PPG data and data pairs with labels indicating whether the corresponding ECG data is abnormal;

[0130] Divide the training dataset into training set, validation set and test set;

[0131] The ECG abnormality monitoring model is trained using the training set to obtain the fitted model parameters. The ECG abnormality monitoring model determined by the fitted model parameters is verified using the validation set and the model hyperparameters are adjusted. The performance of the trained ECG abnormality monitoring model determined by the adjusted hyperparameters and the trained model parameters is evaluated using the test set.

[0132] For example, the DCA-Net model of the above-mentioned specific example can be applied to the PPG data segments processed based on the MIMIC-III WDB, and binary classification is performed using labels indicating whether the corresponding ECG data segments are abnormal. The 38,320 sets of PPG data segments and labels indicating whether the corresponding ECG data segments are abnormal are divided into training, validation, and test sets in a ratio of 64%:16%:20%. The model parameters are learned using the training set and verified in the validation set to avoid overfitting. The test set is only used to report the results. The model performance was evaluated using specificity, sensitivity / recall, precision, accuracy, AUROC (area under the receiver operating characteristic curve), and AUPRC (area under the precision-recall curve). The model was further randomly repeated 10 times, and the average and standard deviation of the above evaluation indicators were used.

[0133] The DCA-Net model can be implemented using PyTorch version 1.11.0 and CUDA version 11.7. The loss function is binary cross entropy. Further experiments adjust for class imbalance by increasing the loss weight. The model is optimized with the Adam optimizer using default hyperparameter settings and trained with a batch size of 64. A stride-wise learning rate scheduler is applied to reduce the learning rate by a factor of 0.1 every 10 epochs. Early stopping is also applied—training is terminated if the validation loss does not decrease compared to the current lowest loss for five consecutive epochs. The model with the lowest loss is always saved and applied to the test set for evaluation.

[0134] It should be understood that, although the various steps in the flowchart of FIG2 are shown in sequence according to the instructions of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in FIG2 may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0135] Performance Testing

[0136] To verify the superior architectural performance of the DCA-Net model proposed in this application, the DCA-Net model described in the specific example above was used to monitor ECG abnormalities based on PPG data, along with existing neural network models ResNet-18, ResNet-34, ResNet-50, and ECA-Net. The results are shown in Table 1 below. In Table 1, the numbers shown are the average score percentages of 10 random repetitions, with the standard deviations shown in parentheses. The best average accuracy, AUROC, and AUPRC are highlighted in bold.

[0137] Table 1 Performance comparison results of the DCA-Net model of a specific example of this application and the existing neural network model.

[0138] As shown in the table above, the proposed DCA-Net achieves the best average precision, AUROC, and AUPRC scores. In a one-sided t-test, DCA-Net significantly outperforms ECA-Net in AUPRC, with a p-value of 0.01. Furthermore, DCA-Net's performance is more stable than ECA-Net due to the smaller standard deviations of all evaluation measures.

[0139] In one embodiment, as shown in FIG12 , an ECG abnormality monitoring device 1200 is provided, comprising:

[0140] The data acquisition module 1210 is used to obtain the user's PPG data;

[0141] Model monitoring module 1220, configured to input PPG data into a pre-trained ECG anomaly monitoring model to obtain an output value of the ECG anomaly monitoring model; wherein the ECG anomaly monitoring model is trained using a training dataset comprising data pairs of PPG data and labels indicating whether ECG data corresponding to the PPG data is abnormal; the ECG anomaly monitoring model is configured to output an output value indicating whether the ECG data corresponding to the input PPG data is estimated to be abnormal based on the input PPG data;

[0142] The abnormality alarm module 1230 is configured to cause a user device associated with the user to issue an alarm when the output value of the ECG abnormality monitoring model indicates that an abnormality exists in the ECG data estimate.

[0143] In one embodiment, the ECG abnormality monitoring device 1200 further includes a model training module 1240, and the model monitoring module 1220 is further used to obtain a trained ECG abnormality monitoring model from the model training module 1240. The model training module 1240 is used to train the ECG abnormality monitoring model to obtain a trained ECG abnormality monitoring model.

[0144] For example, the data acquisition module 1210 , the model monitoring module 1220 , and the abnormality alarm module 1230 may be arranged in the terminal 102 , while the model training module 1300 may be arranged in the server 104 .

[0145] The specific definition of the ECG abnormality monitoring device 1200 can be found in the definition of the ECG abnormality monitoring method above and will not be repeated here. Each module in the ECG abnormality monitoring device 1200 can be implemented in whole or in part via software, hardware, or a combination thereof. Each of these modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0146] In one embodiment, as shown in FIG13 , there is provided an ECG abnormality monitoring device 1300 , comprising a PPG data detector 1310 , a memory 1320 , and a processor 1330 ;

[0147] The PPG data detector 1310 is used to monitor the user's PPG data and transmit the measured PPG data to the processor;

[0148] The memory 1320 stores computer programs; and

[0149] When the processor 1330 executes the computer program, it is configured to receive PPG data from the PPG data detector 1310 and implement the method of any of the above embodiments.

[0150] In one embodiment, the PPG data detector may be a pulse oximeter.

[0151] In one embodiment, the ECG abnormality monitoring device 1300 further includes an output device 1340, which may include a display screen, an indicator light, a vibrator, and / or a sound output device. The display screen may be a liquid crystal display or an electronic ink display screen, for example. The sound output device may be a device with a sound output function, such as a speaker or a buzzer. Thus, when the ECG abnormality monitoring device 1300 itself is used to issue an alarm, the alarm may be issued via the output device 1340.

[0152] In one embodiment, the ECG abnormality monitoring device 1300 also has a phone-making function. For example, the ECG abnormality monitoring device 1300 may also include a dialing device with a phone-making function. The dialing device can be implemented, for example, by a smart watch, mobile phone or server associated with the user, so that when an alarm call needs to be made to a predetermined alarm receiving device, the alarm call can be made through the dialing device.

[0153] In one embodiment, the ECG abnormality monitoring device 1300 also has an ECG measurement function. For example, the ECG abnormality monitoring device 1300 may further include an ECG measurement device to further perform ECG measurement on the user using the ECG measurement device when the ECG abnormality is estimated using PPG data.

[0154] It can be understood that the ECG abnormality monitoring device 1300 can be any device that can execute the method of the above-mentioned embodiment of the present application, and it may have various different forms. For example, it may include the aforementioned terminal 102, or it may also include the aforementioned terminal 102 and server 104.

[0155] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, the following steps are implemented:

[0156] Get the user's PPG data;

[0157] Inputting the PPG data into a pre-trained ECG anomaly monitoring model to obtain an output value of the ECG anomaly monitoring model; wherein the ECG anomaly monitoring model is trained using a training data set including data pairs of PPG data and labels indicating whether ECG data corresponding to the PPG data is abnormal, and the ECG anomaly monitoring model is configured to output an output value indicating whether the ECG data corresponding to the input PPG data is estimated to be abnormal based on the input PPG data;

[0158] When the output value of the ECG anomaly monitoring model indicates that an anomaly is estimated to exist in the ECG data, a user device associated with the user is caused to issue an alarm.

[0159] In other embodiments, when the computer program is executed by a processor, the steps of the method of any of the above embodiments are further implemented.

[0160] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0161] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0162] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for monitoring abnormal electrocardiogram, characterized in that: The method comprises: Get the user's PPG data; Inputting the PPG data into a pre-trained ECG abnormality monitoring model to obtain an output value of the ECG abnormality monitoring model; wherein the ECG abnormality monitoring model is pre-trained to output an output value indicating whether the ECG data corresponding to the input PPG data is estimated to have abnormal cardiac function based on the input PPG data; When the output value of the ECG abnormality monitoring model indicates that the ECG data estimates that there is abnormal heart function, a user device associated with the user is caused to issue an alarm.

2. The abnormality monitoring method according to claim 1, characterized in that: The ECG abnormality monitoring model is trained using a training data set, where the training data set includes PPG data and a label indicating whether there is abnormal cardiac function in the ECG data corresponding to the PPG data.

3. The ECG abnormality monitoring method according to claim 2, characterized in that: The process of constructing the training data set includes: Obtain ECG signals of multiple leads and their corresponding PPG signals; Extracting, from the acquired ECG signals of the multiple leads and their corresponding PPG signals, multiple ECG raw data segments of the leads and their corresponding PPG raw data segments of the leads of a predetermined time length; Preprocessing the intercepted ECG raw data segments of the plurality of leads of the predetermined time length and the corresponding PPG raw data segments respectively to obtain a plurality of groups of ECG preprocessed data segments of the plurality of leads of the predetermined time length and the corresponding PPG preprocessed data segments; For each group of ECG preprocessed data segments of a plurality of leads of the predetermined time period and their corresponding PPG preprocessed data segments, determining a label indicating whether the ECG preprocessed data segments of the plurality of leads have abnormal cardiac function; The labels and corresponding PPG pre-processed data segments are combined into data pairs, thereby obtaining the training data set including a plurality of sets of the data pairs.

4. The method for monitoring abnormal electrocardiogram according to claim 1, characterized in that: The obtaining of the user's PPG data and inputting the PPG data into a pre-trained ECG abnormality monitoring model includes: Get the user's PPG raw data; Preprocessing the PPG raw data to obtain PPG preprocessed data; The PPG preprocessed data is input into a pre-trained ECG abnormality monitoring model.

5. The method for monitoring abnormal electrocardiogram according to claim 4, characterized in that: The pre-processing includes one or more of the following: Flatness detection and signal screening; Normalization; Filtering; Peak detection and signal screening; Skewness SQI detection and signal filtering outlier sample value replacement.

6. The method for monitoring abnormal electrocardiogram according to claim 1, characterized in that: The ECG abnormality monitoring model includes one of the following: DCA-Net model; ResNet-18 model; ResNet-34 model; ResNet-50 model; ECA-Net model.

7. The method for monitoring abnormal electrocardiogram according to claim 1, characterized in that: The electrocardiogram abnormality monitoring model includes one or more residual-double convolution attention blocks, which are formed by adding a double convolution attention block to a residual block, and the double convolution attention block includes a first convolution attention block for convolution in the channel dimension and a second convolution attention block for convolution in the time dimension.

8. The method for monitoring abnormal electrocardiogram according to claim 7, characterized in that: The electrocardiogram abnormality monitoring model includes a plurality of the residual-double convolution attention blocks arranged in series, and each of the plurality of the residual-double convolution attention blocks includes: A first main path, wherein one or more convolutional layers and the A dual convolutional attention block; wherein the dual convolutional attention block includes the first convolutional attention block and the second convolutional attention block arranged in series or in parallel; and A first short-circuit branch path, wherein the first short-circuit branch path is connected in parallel to two ends of the one or more convolutional layers and the dual convolutional attention block arranged in series on the first main path; Among them, the former of the one or more convolutional layers and the dual convolutional attention blocks and the first short-circuit branch path receive the output of the previous path, the latter of the one or more convolutional layers and the dual convolutional attention blocks receives the output of the former, and the output of the last of the one or more convolutional layers and the dual convolutional attention blocks is added to the output of the first short-circuit branch path and fed into the subsequent path.

9. The method for monitoring abnormal electrocardiogram according to claim 8, characterized in that: The first convolutional attention block and the second convolutional attention block are arranged in series, and the first convolutional attention block is located before the second convolutional attention block.

10. The method for monitoring abnormal electrocardiogram according to claim 8, characterized in that: Each of the first convolutional attention block and the second convolutional attention block comprises: a second main path, wherein a maximum pooling layer and an average pooling layer, a convolution layer and an activation function layer are arranged on the second main path, wherein the maximum pooling layer and the average pooling layer are arranged in parallel, and the maximum pooling layer and the average pooling layer arranged in parallel are arranged in series with the convolution layer and the activation function layer in sequence; and A second short-circuit branch path, wherein the second short-circuit branch path is connected in parallel to both ends of the maximum pooling layer and the average pooling layer, the convolution layer, and the activation function layer provided on the second main path; Among them, the second short-circuit branch path, the maximum pooling layer and the average pooling layer respectively receive the output of the previous path, the output of the maximum pooling layer and the output of the average pooling layer are combined to feed the convolution layer, the output of the convolution layer is fed into the activation function layer, and the output of the activation function layer is multiplied by the output of the second short-circuit branch path and fed into the subsequent path.

11. The method for monitoring abnormal electrocardiogram according to claim 8, characterized in that: The output side of each convolutional layer in each residual-double convolutional attention block is also provided with a batch normalization layer and an activation function layer.

12. The method for monitoring abnormal electrocardiogram according to claim 8, characterized in that: The ECG abnormality monitoring model also includes: An input convolution layer and a pooling layer are located before the plurality of residual-double convolution attention blocks and are arranged in series with the plurality of residual-double convolution attention blocks, and An average pooling layer and a fully connected layer, located after the plurality of residual-double convolutional attention blocks and arranged in series with the plurality of residual-double convolutional attention blocks; Among them, the input convolution layer receives the PPG data input to the ECG abnormality monitoring model, the output of the input convolution layer is fed into the pooling layer, the output of the pooling layer is fed into the first residual-double convolution attention block among the multiple residual-double convolution attention blocks, the output of the last residual-double convolution attention block among the multiple residual-double convolution attention blocks is fed into the average pooling layer, the output of the average pooling layer is fed into the fully connected layer, and the fully connected layer outputs the output value of the ECG abnormality monitoring model.

13. The method for monitoring abnormal electrocardiogram according to any one of claims 1 to 12, characterized in that: The training process of the ECG abnormality monitoring model includes: Acquire a training data set; the training data set includes a plurality of data pairs of PPG data and corresponding ECG data with labels indicating whether there is abnormal cardiac function; Dividing the training data set into a training set, a validation set and a test set; The ECG abnormality monitoring model is trained using the training set to obtain fitted model parameters, the ECG abnormality monitoring model determined by the fitted model parameters is verified using the validation set and the model hyperparameters are adjusted, and the performance of the trained ECG abnormality monitoring model determined by the adjusted hyperparameters and the trained model parameters is evaluated using the test set.

14. The method for monitoring abnormal electrocardiogram according to any one of claims 1 to 12, characterized in that: The user device associated with the user includes a portable terminal device worn by the user and / or a guardian terminal device of a guardian associated with the user, and the acquiring of the PPG data of the user includes monitoring the PPG data of the user in real time by the portable terminal device; The step of causing the user device associated with the user to issue an alarm includes one or more of the following: displaying visual information indicating that the user's ECG data estimates that there is abnormal heart function on the display screen of the user device, causing an indicator light of the user device to emit an alarm indicating that the user's ECG data estimates that there is abnormal heart function, The data is estimated to have abnormal heart function, a light prompt is turned on, a vibrator of the user device is made to vibrate to indicate that the user's ECG data is estimated to have abnormal heart function, and a sound output device of the user device is made to emit a sound prompt to indicate that the user's ECG data is estimated to have abnormal heart function.

15. The method for monitoring abnormal electrocardiogram according to any one of claims 1 to 12, characterized in that: The method further comprises: When the output value of the ECG abnormality monitoring model indicates that the ECG data estimates the presence of abnormal cardiac function, an alarm call is made to a predetermined alarm receiving device.

16. The method for monitoring abnormal electrocardiogram according to any one of claims 1 to 12, characterized in that: The method further comprises: When the output value of the ECG abnormality monitoring model indicates that the ECG data estimates the presence of abnormal heart function, an ECG measurement is performed on the user to obtain a measurement result indicating whether the ECG of the user indicates the presence of abnormal heart function.

17. An electrocardiogram abnormality monitoring device, characterized in that: The device comprises: A data acquisition module is used to obtain the user's PPG data; A model monitoring module, used for inputting the PPG data into a pre-trained ECG abnormality monitoring model to obtain an output value of the ECG abnormality monitoring model; wherein the ECG abnormality monitoring model is pre-trained to output an output value indicating whether the ECG data corresponding to the input PPG data is estimated to have abnormal cardiac function based on the input PPG data; The abnormality alarm module is used to cause a user device associated with the user to issue an alarm when the output value of the ECG abnormality monitoring model indicates that the ECG data estimates that there is abnormal heart function.

18. An electrocardiogram abnormality monitoring device, comprising a PPG data detector, a memory and a processor; The PPG data detector is used to monitor the PPG data of the user and transmit the monitored PPG data to the processor; The memory stores a computer program; and When the processor executes the computer program, it is used to receive the PPG data from the PPG data detector and implement the ECG abnormality monitoring method according to any one of claims 1 to 16.

19. The electrocardiogram abnormality monitoring device according to claim 18, characterized in that: It also includes an output device, which includes a display screen, an indicator light, a vibrator and / or a sound output device.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the electrocardiogram abnormality monitoring method according to any one of claims 1 to 16 is implemented.

Images