CN115944302A - Unsupervised multi-mode electrocardiogram abnormity detection method based on attention mechanism - Google Patents
Unsupervised multi-mode electrocardiogram abnormity detection method based on attention mechanism Download PDFInfo
- Publication number
- CN115944302A CN115944302A CN202211589031.6A CN202211589031A CN115944302A CN 115944302 A CN115944302 A CN 115944302A CN 202211589031 A CN202211589031 A CN 202211589031A CN 115944302 A CN115944302 A CN 115944302A
- Authority
- CN
- China
- Prior art keywords
- sequence
- electrocardio
- model
- time domain
- domain sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses an unsupervised multi-mode electrocardiogram abnormity detection method based on an attention mechanism, wherein the abnormity detection of an electrocardiogram is automatically carried out unsupervised according to a body surface electrophysiological signal, the frequency domain signal characteristic is also considered by a model on the basis of a time sequence electric signal, an original signal is reconstructed by encoding and decoding an input signal, and when the difference between the reconstructed time sequence signal and the input time sequence signal is greater than a certain value, the electrocardiogram is judged to be abnormal. According to the method, the time domain and frequency domain signals of the ECG are subjected to feature extraction, so that normal ECG heartbeat data and abnormal ECG heartbeat data can be distinguished under the condition of no supervision and no label use; under the condition that the electrocardiogram data with labels is limited, the method assists doctors to automatically screen abnormal electrocardiograms, and has clinical diagnosis and therapeutic reference significance.
Description
Technical Field
The invention belongs to the technical field of prediction of heart electrophysiological diseases, and particularly relates to an unsupervised multi-modal electrocardiogram abnormity detection method based on an attention mechanism.
Background
Cardiovascular diseases a series of heart or blood vessel related diseases are collectively called as the main diseases threatening human life, and the death rate caused by the diseases is still the first. ECG (electrocardiogram) is one of biological signals which are researched and applied to medical clinic at the earliest time by human beings, and the electrocardiogram reflects the health condition of human heart and is an important basis for clinical diagnosis of cardiovascular diseases. With the rapid increase of the number of electrocardiograms, it is very important to use the AI algorithm to assist the doctor in automatic abnormality diagnosis.
Many existing ECG automated diagnostic algorithms are based on supervised learning of data analyzed and annotated by expert physicians. However, since experts can only process a small amount of electrocardiogram data, and most of the data comes from the same pattern, the supervised learning based automatic diagnosis algorithm has its limitations. In addition, the distribution of each type of label in the label data is usually unbalanced, which can affect the diagnostic analysis effect of the model; unlabeled data is also valuable, which may further reveal information about potential pathologies within the cardiovascular system of a patient. Therefore, it is very important to develop an algorithm capable of performing an unsupervised diagnosis of electrocardiographic data.
At present, most of the existing unsupervised ECG abnormality detection or classification methods do not fully utilize and mine the characteristics of data, and can be roughly classified into two types: one is to use clustering algorithms to make unsupervised decisions, such as the literature [ AbawajH, kelarev AV V, chowdhury M. Multistage approach for clustering and classification of ECG data [ J ]. Computer methods and programs in biomedicine,2013,112 (3): 720-730] to use Gaussian mixture models and K-means clustering algorithms to transform ECG data into numerical features to complete abnormality classification. Another is to detect ECG anomalies end-to-end through deep learning, for example, the document [ Pereira J, silverera M.Unsequently rendering and analysis detection in ECG sequences [ J ]. International Journal of Data Mining and Bioinformatics,2019,22 (4): 389- ] decodes and reconstructs the input by using the recursive network as the basis of the self-encoder. Nevertheless, the above-described methods have the drawback that the information of the ECG is not fully exploited and mined and the model lacks features that incorporate more dimensions.
Disclosure of Invention
In view of the above, the present invention provides an unsupervised multi-modal electrocardiogram anomaly detection method based on an attention mechanism, which exploits hidden features in time domain and frequency domain data by using the attention mechanism to reconstruct time domain input unsupervised, and can well solve the problems that electrocardiogram data containing tags is lacked and a supervised model cannot achieve a wide range of automatic diagnosis effects.
An unsupervised multi-mode electrocardiogram abnormity detection method based on an attention mechanism comprises the following steps:
(1) Collecting multi-lead electrocardiosignals from the body surface of a patient, and taking each heartbeat cycle as a group of electrocardio time domain sequences;
(2) Carrying out normalization processing and frequency domain conversion on each group of electrocardio time domain sequences to obtain corresponding electrocardio frequency domain sequences;
(3) Constructing a reconstruction model based on an attention mechanism, wherein the reconstruction model comprises two coding modules and a decoding module, the two coding modules are respectively used for coding an electrocardio time domain sequence and an electrocardio frequency domain sequence, and the obtained coding features are spliced and then decoded into an ECG waveform sequence consistent with an input dimension through the decoding module;
(4) The electrocardio time domain sequence and the electrocardio frequency domain sequence are input into the model one by one in pairs, and the average error between the ECG waveform sequence output by the model and the input electrocardio time domain sequence is the minimum to be a loss function, so that the model is trained;
(5) Inputting the time domain sequence and the frequency domain sequence of the electrocardiosignal to be detected into a trained reconstruction model, and judging whether the electrocardiosignal to be detected is abnormal or not according to the reconstruction error between the model output ECG waveform sequence and the input electrocardio time domain sequence.
Further, the normalization processing in the step (2) adopts a maximum-minimum value normalization strategy, and the frequency domain conversion adopts wavelet transformation.
Furthermore, two coding modules in the reconstruction model have the same structure, but do not share the weight in the training process, and independently perform feature learning.
Further, the time domain sequence and the frequency domain sequence of the electrocardiogram input into the model need to be position-coded first, that is, time position information is added to the amplitude corresponding to each moment in the sequence, which is specifically as follows:
PE (pos,2i) =sin(pos/10000 2i/d )
PE (pos,2i+1) =cos(pos/10000 2i/d )
wherein: PE (polyethylene) (pos,2i) Representing time-position information added at even positions in the sequence, PE (pos,2i+1) The time position information added to the odd positions in the sequence is represented, pos represents the time point corresponding to each amplitude value in the sequence, d represents the dimension of the code, and i is a natural number.
Furthermore, the coding module adopts a multi-head attention mechanism and residual connection, wherein the multi-head attention mechanism is formed by overlapping a plurality of self-attention mechanisms; meanwhile, in the process of forward propagation of the learning parameters by the coding module, layer Normalization is carried out on the parameters of each Layer, and the activation value of each Layer is normalized.
Further, the calculation process of the self-attention mechanism is as follows:
Q=X embedding *W Q
K=X embedding *W K
V=X embedding *W V
wherein: attention (Q, K, V) is the output of the Attention mechanism, X embedding Is an electrocardio time domain sequence or an electrocardio frequency domain sequence after position coding, Q, K, V is a query vector, a key vector, a value vector and W Q 、W K 、W V A weight matrix corresponding to the query vector, the key vector and the value vector, d k For the output dimension of the self-attention mechanism, T denotes transpose.
Further, the encoding characteristics output by the two encoding modules are spliced along the time axis direction, namely the dimension after splicing is doubled, and then the characteristics are decoded into an ECG waveform sequence consistent with the input dimension through a linear mapping layer of the decoding module.
Further, the process of training the model in step (4) is as follows:
4.1 initializing model parameters, including a bias vector and a weight matrix of each layer, a learning rate and an optimizer;
4.2, inputting the electrocardio time domain sequence and the electrocardio frequency domain sequence into the model, transmitting and outputting the model in the forward direction to obtain a corresponding reconstruction result, namely an ECG waveform sequence, and calculating a loss function L between the output ECG waveform sequence and the input electrocardio time domain sequence;
and 4.3, continuously and iteratively updating the model parameters by using an optimizer through a gradient descent method according to the loss function L until the loss function L is converged, and finishing training.
Further, the expression of the loss function L is as follows:
wherein: y is i For the ith amplitude, x, in the output ECG waveform sequence i Is the ith amplitude in the input electrocardio time domain sequence, and n is the dimension of the electrocardio time domain sequence.
Further, the optimizer adopts an Adam optimizer.
Further, in the step (5), a reconstruction error between the ECG waveform sequence output by the model and the input ECG time domain sequence is calculated, and if the reconstruction error is greater than a set threshold, it is determined that the ECG signal to be detected is abnormal; the threshold is set according to the statistical probability distribution, namely the threshold is the sum of the mean and the variance of the training reconstruction errors.
According to the method, the time domain and frequency domain signals of the ECG are subjected to feature extraction, so that normal ECG heartbeat data and abnormal ECG heartbeat data can be distinguished under the condition of no supervision and no label. Therefore, the present invention can help a doctor or a patient to perform efficient and accurate ECG abnormality detection.
Drawings
FIG. 1 is a schematic flow chart of the unsupervised multi-modal abnormal electrocardiogram detection method of the present invention.
FIG. 2 is a schematic structural diagram of a sequence feature encoding module in the model of the present invention.
FIG. 3 is a diagram showing the predicted results of the normal ECG and abnormal ECG data according to the model of the present invention.
Detailed Description
In order to more specifically describe the present invention, the following detailed description is provided for the technical solution of the present invention with reference to the accompanying drawings and the specific embodiments.
As shown in FIG. 1, the unsupervised multi-modal electrocardiogram anomaly detection method based on the attention mechanism of the invention comprises the following steps:
(1) Electrophysiological signal data are acquired from the patient's body surface and processed into heartbeat signals, i.e. data for each heartbeat cycle.
(2) And carrying out normalization processing on all heartbeat data, and calculating by adopting a maximum-minimum normalization strategy.
Wherein: x represents the signal amplitude at a certain time on each lead of the electrocardiogram data, x min Representing the minimum value of amplitude, x, of the lead heartbeat signal max Representing the maximum amplitude, x, of the lead heartbeat signal norm The results after normalization are shown.
(3) And performing frequency domain conversion on the time domain electrocardiogram data of each lead.
Wherein: a is a scale parameter, b is a translation parameter, and f (t) is a certain electrocardiogram time sequence signal.
(4) The electrocardiographic data of each lead is recorded as X = [ X ] 1 ,x 2 ,…,x n ]The corresponding continuous wavelet transform frequency domain is denoted as X f =[f 1 ,f 2 ,…,f n ]The time domain and frequency domain signals of the electrocardiogram are used as model input parts.
(5) Two input sequence feature coding modules are constructed, the two modules have the same structure, but do not share the weight in the training process, feature learning is independently carried out, and the input of the two coding modules is the electrocardiogram time domain and the corresponding frequency domain conversion result.
(6) And simultaneously, carrying out position coding on the input time domain sequence and the input frequency domain sequence, namely adding time position information to the amplitude corresponding to each moment of the original sequence.
PE (pos,2i) =sin(pos/10000 2i/d )
PE (pos,2i+1) =cos(pos/10000 2i/d )
Wherein: pos is the time point of each amplitude value in the heartbeat signal, and d represents the dimension of coding; the above two calculation methods represent that the odd and even positions of the electrocardiogram time domain sequence or the frequency domain sequence are respectively encoded.
(7) The sequence after position coding is input to a sequence feature coding module, which mainly comprises a Multi-Head Attention mechanism (Multi-Head Attention) and residual connection, as shown in fig. 2, wherein the Multi-Head Attention mechanism is composed of a self-Attention mechanism (self-Attention), and in the process of forward propagation of learning parameters in the feature coding module, the model performs layer normalization on each layer of parameters, and normalizes the activation value of each layer.
(8) The self-attention mechanism (self-attention) is the focus of sequence feature learning, and the part is to model the input sequence by defining three learnable matrices.
Q=X embedding *W Q
K=X embedding *W K
V=X embedding *W V
Wherein: x embedding Representing the input sequence after position coding, W Q 、W K And W V Representing three learnable matrices, Q, K and V, the vectors that attention mechanism requires comparative calculations.
(9) The self-attention mechanism then calculates the relationships between the various time instants of the sequence from Q, K and V, and reassigns the weights to modify V.
X embedding =X embedding +Attention(Q,K,V)
Wherein:in order to change the attention moment array into a standard normal distribution, softmax is normalized, and then new X is updated and obtained on the basis of the normalization embedding 。
(10) And then, on the basis, the sequence is subjected to linear mapping, the layer normalization operation is carried out on the parameters of the activation layer, and a multi-head attention mechanism is formed by superposing a plurality of self-attention mechanisms.
(11) The time domain data and the frequency domain data are respectively processed by the corresponding coding modules to obtain corresponding coding characteristics, and then the coding characteristics are spliced to form new coding characteristics, wherein the splicing mode is splicing along the time axis direction, namely the dimension after splicing is converted into twice of the original dimension.
(12) Finally, the features are decoded into the original input time-domain consistent dimension via the linear mapping layer.
(13) And (3) calculating a reconstruction error of a decoding output and a time domain input adopted by a target function of the model, wherein the reconstruction error is calculated as an absolute value average error function.
Wherein: x, y are time domain input and decoded output, y i And x i Respectively, the reconstructed sequence and its corresponding original sequence amplitude.
(14) And updating model parameters of the whole model by adopting an Adam optimizer.
m t =β 1 m t-1 +(1-β 1 )g t
Wherein: m is t And v t Are estimates of the first moment (mean) and the second moment (with a bias) of the gradient, respectively, initialized to a 0 vector, g t Forward calculation of function f for a certain time instant A The gradient of (a) of (b) is,correction of gradient estimate for Adam, beta 1 Has a default value of 0.9, beta 2 0.999, e is 10 -8 。
(15) Finally, an abnormal threshold value is set according to the reconstruction error of the model to the normal electrocardiogram, and then the abnormal detection task of the electrocardiogram is completed on the basis of the threshold value. The threshold value set here is set according to statistical probability distribution, namely the threshold value is the sum of the mean and variance of the training reconstruction error; when the reconstruction error of the model is larger than the threshold value, the input sample is judged to be an abnormal sample, otherwise, the input sample is a normal sample.
To verify the effectiveness of the present invention, we trained and tested the model of the present invention using the public electrocardiogram data set ECG5000, wherein ECG5000 is a single lead electrocardiogram data of 20 hours for a patient, which contains 5000 complete time series heartbeat data. We used 80% of the normal electrocardiogram data as training data of the unsupervised model and the remaining 20% of the normal electrocardiogram data and abnormal electrocardiogram data as test data, and fig. 3 shows the results predicted by the model of the present invention on the normal electrocardiogram and abnormal electrocardiogram. It can be seen that compared to many related anomaly detection algorithms, the model of the present invention can effectively distinguish between normal and abnormal ecg samples without supervision, and can accurately make an automatic diagnosis.
The previous description of the specific embodiments is provided to enable any person skilled in the art to make or use the present invention. It will be readily apparent to those skilled in the art that various modifications to the specific embodiments described above may be made, and the generic principles described herein may be applied to other embodiments without the use of the inventive faculty. Therefore, the present invention is not limited to the above embodiments, and those skilled in the art should make improvements and modifications to the present invention based on the disclosure of the present invention within the protection scope of the present invention.
Claims (10)
1. An unsupervised multi-mode electrocardiogram abnormity detection method based on an attention mechanism comprises the following steps:
(1) Collecting multi-lead electrocardiosignals from the body surface of a patient, and taking each heartbeat cycle as a group of electrocardio time domain sequences;
(2) Carrying out normalization processing and frequency domain conversion on each group of electrocardio time domain sequences to obtain corresponding electrocardio frequency domain sequences;
(3) Constructing a reconstruction model based on an attention mechanism, wherein the reconstruction model comprises two coding modules and a decoding module, the two coding modules are respectively used for coding an electrocardio time domain sequence and an electrocardio frequency domain sequence, and the obtained coding characteristics are spliced and decoded into an ECG waveform sequence consistent with the input dimension by the decoding module;
(4) The electrocardio time domain sequence and the electrocardio frequency domain sequence are input into the model one by one in pairs, and the average error between the ECG waveform sequence output by the model and the input electrocardio time domain sequence is the minimum to be a loss function, so that the model is trained;
(5) Inputting the time domain sequence and the frequency domain sequence of the electrocardiosignal to be detected into a trained reconstruction model, and judging whether the electrocardiosignal to be detected is abnormal or not according to the reconstruction error between the model output ECG waveform sequence and the input electrocardio time domain sequence.
2. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: the normalization processing in the step (2) adopts a maximum-minimum value normalization strategy, and the frequency domain conversion adopts wavelet transformation.
3. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: the two coding modules in the reconstruction model have the same structure, but do not share the weight in the training process, and independently perform feature learning.
4. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: the electrocardio time domain sequence and the electrocardio frequency domain sequence input into the model need to be subjected to position coding, namely time position information is added to the amplitude corresponding to each moment in the sequence, and the method specifically comprises the following steps:
PE (pos,2i) =sin(pos/10000 2i/d )
PE (pos,2i+1) =cos(pos/10000 2i/d )
wherein: PE (polyethylene) (pos,2i) Representing time position information added at even positions in the sequence, PE (pos,2i+1) The time position information added to the odd positions in the sequence is represented, pos represents a time point corresponding to each amplitude value in the sequence, d represents the dimension of the code, and i is a natural number.
5. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: the coding module adopts a multi-head attention mechanism and residual connection, wherein the multi-head attention mechanism is formed by overlapping a plurality of self-attention mechanisms; meanwhile, in the process of forward propagation of the learning parameters by the coding module, layer Normalization is carried out on the parameters of each Layer, and the activation value of each Layer is normalized.
6. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 5, characterized in that: the calculation process of the self-attention mechanism is as follows:
Q=X embedding *W Q
K=X embedding *W K
V=X embedding *W V
wherein: attention (Q, K, V) is the output of the self-Attention mechanism, X embedding Is an electrocardio time domain sequence or an electrocardio frequency domain sequence after position coding, Q, K, P is a query vector, a key vector, a value vector and W Q 、W K 、W V A weight matrix corresponding to the query vector, the key vector and the value vector, d k For the output dimension of the self-attention mechanism, T represents transpose.
7. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: the encoding characteristics output by the two encoding modules are spliced along the time axis direction, namely the dimension after splicing is doubled, and then the characteristics are decoded into an ECG waveform sequence consistent with the input dimension through a linear mapping layer of the decoding module.
8. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: the process of training the model in the step (4) is as follows:
4.1 initializing model parameters, including a bias vector and a weight matrix of each layer, a learning rate and an optimizer;
4.2, inputting the electrocardio time domain sequence and the electrocardio frequency domain sequence into the model, transmitting and outputting the model in the forward direction to obtain a corresponding reconstruction result, namely an ECG waveform sequence, and calculating a loss function L between the output ECG waveform sequence and the input electrocardio time domain sequence;
and 4.3, continuously and iteratively updating model parameters by using an optimizer through a gradient descent method according to the loss function L until the loss function L is converged, and finishing training.
9. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 8, characterized in that: the expression of the loss function L is as follows:
wherein: y is i For the ith amplitude, x, in the output ECG waveform sequence i Is the ith amplitude in the input electrocardio time domain sequence, and n is the dimension of the electrocardio time domain sequence.
10. The unsupervised multi-modal electrocardiographic abnormality detection method according to claim 1, characterized in that: calculating a reconstruction error between the ECG waveform sequence output by the model and the input ECG time domain sequence in the step (5), and if the reconstruction error is greater than a set threshold value, judging that the electrocardiosignal to be detected is abnormal; the threshold is set according to the statistical probability distribution, namely the threshold is the sum of the mean and the variance of the training reconstruction errors.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211589031.6A CN115944302A (en) | 2022-12-09 | 2022-12-09 | Unsupervised multi-mode electrocardiogram abnormity detection method based on attention mechanism |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211589031.6A CN115944302A (en) | 2022-12-09 | 2022-12-09 | Unsupervised multi-mode electrocardiogram abnormity detection method based on attention mechanism |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115944302A true CN115944302A (en) | 2023-04-11 |
Family
ID=87285285
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211589031.6A Pending CN115944302A (en) | 2022-12-09 | 2022-12-09 | Unsupervised multi-mode electrocardiogram abnormity detection method based on attention mechanism |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115944302A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117076936A (en) * | 2023-10-16 | 2023-11-17 | 北京理工大学 | Time sequence data anomaly detection method based on multi-head attention model |
CN117796817A (en) * | 2024-02-29 | 2024-04-02 | 山东大学齐鲁医院 | Method and system for rapid detection and early warning of acute myocardial infarction |
-
2022
- 2022-12-09 CN CN202211589031.6A patent/CN115944302A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117076936A (en) * | 2023-10-16 | 2023-11-17 | 北京理工大学 | Time sequence data anomaly detection method based on multi-head attention model |
CN117796817A (en) * | 2024-02-29 | 2024-04-02 | 山东大学齐鲁医院 | Method and system for rapid detection and early warning of acute myocardial infarction |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Liu et al. | Deep learning in ECG diagnosis: A review | |
Houssein et al. | An automatic arrhythmia classification model based on improved marine predators algorithm and convolutions neural networks | |
Liu et al. | Classification of heart diseases based on ECG signals using long short-term memory | |
CN115944302A (en) | Unsupervised multi-mode electrocardiogram abnormity detection method based on attention mechanism | |
Ullah et al. | Classification of arrhythmia in heartbeat detection using deep learning | |
CA3069891A1 (en) | Discovering novel features to use in machine learning techniques, such as machine learning techniques for diagnosing medical conditions | |
Rajan et al. | A generative modeling approach to limited channel ECG classification | |
Luo et al. | Multi-classification of arrhythmias using a HCRNet on imbalanced ECG datasets | |
CN115470828A (en) | Multi-lead electrocardiogram classification and identification method based on convolution and self-attention mechanism | |
CN115281688A (en) | Cardiac hypertrophy multi-label detection system based on multi-mode deep learning | |
Xie et al. | Intelligent analysis of premature ventricular contraction based on features and random forest | |
Singh et al. | Short and noisy electrocardiogram classification based on deep learning | |
Zhang et al. | Automatic diagnosis of arrhythmia with electrocardiogram using multiple instance learning: From rhythm annotation to heartbeat prediction | |
CN115177262A (en) | Heart sound and electrocardiogram combined diagnosis device and system based on deep learning | |
Allam et al. | A deformable CNN architecture for predicting clinical acceptability of ECG signal | |
Lakhdari et al. | A new vision of a simple 1D Convolutional Neural Networks (1D-CNN) with Leaky-ReLU function for ECG abnormalities classification | |
Chen et al. | Detecting atrial fibrillation in ICU telemetry data with weak labels | |
Dambal et al. | Premature Ventricular Contraction Classification Based on Spiral Search-Manta Ray Foraging and Bi-LSTM | |
Gordon et al. | PVC detection using a convolutional autoencoder and random forest classifier | |
Zaorálek et al. | Patient-adapted and inter-patient ECG classification using neural network and gradient boosting | |
Ansari et al. | Estimating age and gender from electrocardiogram signals: A comprehensive review of the past decade | |
Liao et al. | Automatic classification of heartbeats using ECG signals via higher order hidden Markov model | |
Neifar et al. | Deep Generative Models for Physiological Signals: A Systematic Literature Review | |
JP2023104885A (en) | Electrocardiographic heart rate multi-type prediction method based on graph convolution | |
Pal et al. | Two-Stage Classifier for Resource Constrained On-Board Cardiac Arrhythmia Detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |