Heart rate monitoring method based on multi-sensor fusion

Abstract

The invention discloses a multi-sensor fusion heart rate monitoring method, and relates to the field of medical health monitoring, and the method comprises the steps: 1, collecting an original mixed body motion signal, and separating a normal BCG signal and an abnormal BCG signal from the original mixed body motion signal; 2, training a heart rate sign extraction model, and extracting heart rate signs; 3, the heart rate features are used for training a classification network model, and a heart rate judgment result is output; and 4, inputting the new heart rate data into the heart rate sign extraction model, extracting feature representation, and inputting the feature representation into the trained classification network model to obtain a judgment result of the new heart rate data. According to the technical scheme provided by the invention, the accuracy and efficiency of detection analysis are improved by utilizing an artificial intelligence technology and a deep learning algorithm and applying a self-supervision method of contrast learning, and the feature representation of positive and abnormal BCG signals is learned and distinguished by comparing the positive and abnormal BCG signals, so that the dependence on manual marking data is reduced, and the labor cost and errors are reduced.

Description

Technical Field

[0001] The present invention relates to the field of medical health monitoring, and in particular to a heart rate monitoring method using multi-sensor fusion. Background Art

[0002] By using artificial intelligence technology and deep learning algorithms, heart rate monitoring can be achieved for infants, the elderly and patients, potential heart problems or other health abnormalities can be detected, suggestions can be given on heart and physical health conditions, and personalized health advice and intervention measures can be provided to users. For example, abnormal heart rates can be detected in time, and users can be reminded to seek medical treatment in time. Therefore, technicians in this field are committed to developing a heart rate monitoring method based on multi-sensor fusion, building a model based on the collected body motion signals, and obtaining a heart rate judgment result. Summary of the invention

[0003] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is how to perform modeling and analysis based on the collected original body motion signals to obtain the heart rate judgment result.

[0004] To achieve the above object, the present invention provides a multi-sensor fusion heart rate monitoring method, comprising the following steps: Step 1, collecting the original mixed body motion signal, and separating the normal BCG signal and the abnormal BCG signal; Step 2: training a heart rate sign extraction model using the signal collected in step 1 to extract the heart rate sign; Step 3: Using the heart rate feature to train a classification network model and outputting a heart rate judgment result; Step 4: input the new heart rate data into the heart rate sign extraction model, extract feature representation, and then input the feature representation into the trained classification network model to obtain the judgment result of the new heart rate data.

[0005] Furthermore, the step 2 includes the following sub-steps: Step 2.1, randomly selecting N signals from the original mixed body motion signals of step 1, and separating the normal BCG signals therefrom as positive sample data sets, and the abnormal BCG signals as negative sample data sets; Step 2.2: Send the original mixed body motion signal, the positive sample data set and the negative sample data set to the corresponding encoder to capture the time-frequency features of the BCG signal, and record the feature representations as , +、 , wherein parameters are shared between the encoders; Step 2.3: Through a Project network, the feature representation , +、 Mapped into another space, and then normalized and calculated the contrast loss function to obtain the heart rate sign.

[0006] Furthermore, in step 2.2, the feature representation + is obtained by taking the time-frequency features obtained from the original mixed body motion signal and the positive sample data set as the input of Transformer and fusing the features through the multi-head attention mechanism.

[0007] Furthermore, the encoder in step 2.2 adopts a convolutional neural network, a recurrent neural network, LSTM, Resnet or Unet.

[0008] Furthermore, the Project network in step 2.3 is CNN.

[0009] Furthermore, the normalization in step 2.3 is specifically performed by mapping to a hypersphere through standardization.

[0010] Furthermore, the contrast loss function in step 2.3 is InfoNCE Loss, and the specific calculation formula is: in, Represents the contrast loss function, and the numerator represents the similarity of the positive sample The exponential value after temperature parameter adjustment, the denominator is the similarity between the current sample and all samples The sum of the exponential values ​​adjusted by the temperature parameter, is the total number of samples of all samples, including positive samples and abnormal samples; is a temperature parameter used to adjust the distribution of similarity.

[0011] Furthermore, the similarity of the positive samples The calculation function is cosine similarity, specifically: .

[0012] Furthermore, the similarity between the current sample and all samples The calculation function is cosine similarity, specifically: .

[0013] Furthermore, the classification network model in step 3 first removes the Project network in the heart rate sign extraction model, then fixes the Encoder network, and trains it using a cross entropy loss function.

[0014] The multi-sensor fusion heart rate monitoring method provided by the present invention has at least the following technical effects: The technical solution provided by the present invention improves the accuracy and efficiency of detection and analysis by utilizing artificial intelligence technology and deep learning algorithms and applying a self-supervised method of contrastive learning. It also learns to distinguish the feature representations of positive and abnormal BCG signals by comparing them, thereby reducing dependence on manually labeled data and reducing labor costs and errors.

[0015] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a heart rate monitoring method according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0018] Example 1 like Figure 1 FIG. 1 is a flowchart of a multi-sensor fusion heart rate monitoring method provided by an embodiment of the present invention. The method adopts a two-stage training method, which specifically includes the following steps: Step 1, collecting the original mixed body motion signal, and separating the normal BCG signal and the abnormal BCG signal; Step 2: Use the signal collected in step 1 to train a heart rate sign extraction model to extract heart rate signs; Step 3: Use the heart rate features to train the classification network model and output the heart rate judgment result; Step 4: Input the new heart rate data into the heart rate sign extraction model, extract the feature representation, and then input the feature representation into the trained classification network model to obtain the judgment result of the new heart rate data.

[0019] In particular, the ballistocardiograph (BCG) is a non-invasive, contactless and convenient means of monitoring cardiovascular function. In step 1, the BCG component usually appears as a slightly higher frequency vibration than the respiratory component, with a frequency range of 0.5 to 10 Hz. In medicine, it is generally believed that the heart rate is normal between 40 and 150 beats per minute, that is, the frequency is 0.67 to 2.5 Hz. The acquisition method of the original mixed body motion signal is as follows: First, a bandpass filter of 0.5Hz-3Hz bandpass filter is designed to preliminarily extract the BCG signal. Then, the first-order derivative or second-order derivative of the preliminarily extracted BCG signal is calculated to help identify the slope change points, which are usually candidates for peaks. A threshold is set to identify possible peaks. Check the connected rising and falling parts above the set threshold, mark the local maximum as the peak, compare the current sample point with the previous and next sample points to determine whether it is a peak, and set the minimum distance between peaks to avoid detecting repeated peaks, thereby removing pseudo peaks. The largest point among the peak values ​​is taken as the J point in the BCG signal.

[0020] The beating of the heart is caused by the depolarization and repolarization of myocardial cell tissue. In the normal pacemaker of the heart, the sinoatrial node, the impulse generated is transmitted to the atrium through the atrial conduction system, stimulating the atrial myocytes, and then transmitting the impulse to the atrioventricular node. Then, the atrioventricular node stimulates the entire ventricular muscle through the left and right bundle branches, completing a heart beat. This process can be observed and recorded through electrocardiographic signals in electrophysiology. A standard BCG heartbeat mainly includes H, I, J, K, L, M, and N waves. Among them, the signal of the HL interval constitutes a wave group shaped like the letter "w", representing the process of heart contraction, and L, M, and N represent the process of heart relaxation. The time interval between adjacent J peaks is called the "J-J interval". Calculating the inverse of the JJ interval is the instantaneous heart rate based on the BCG signal. By locating the J peak in the BCG signal peak group, not only can the heart rate information be calculated, but also the foundation for the calculation of other sleep indicators and subsequent analysis can be laid.

[0021] Example 2 Based on Example 1, step 2 includes the following sub-steps: Step 2.1, randomly select N signals from the original mixed body motion signals in step 1 to form a batch. For any signal in the batch, separate the normal BCG signal as the positive sample data set and the abnormal BCG signal as the negative sample data set; Step 2.2: Send the original mixed body motion signal, positive sample data set and negative sample data set to the corresponding encoder to capture the time-frequency characteristics of the BCG signal, and record the feature representation as , +、 , where parameters are shared between encoders; Step 2.3: Through a Project network, the feature representation , +、 Mapped to another space (usually with a smaller dimension), normalized (mapped to a hypersphere) and calculated the contrast loss function to obtain heart rate signs.

[0022] In particular, in step 2.2, the feature representation + is to use the time-frequency features obtained from the original mixed body motion signal and the positive sample data set as the input of the Transformer, and fuse the features through the multi-head attention mechanism. The multi-head attention of the Transformer can help the model learn the association and interaction between different features, thereby helping the model focus on the key features in the signal and extract feature representations that are more beneficial to the classification task.

[0023] In particular, the encoder in step 2.2 adopts a convolutional neural network, a recurrent neural network, LSTM, Resnet, or Unet.

[0024] Example 3 Based on Example 2, the Project network in step 2.3 is CNN.

[0025] In particular, the contrast loss function in step 2.3 is InfoNCE Loss, and the specific calculation formula is: in, Represents the contrast loss function, and the numerator represents the similarity of the positive sample The exponential value after temperature parameter adjustment is to make the positive sample have a higher contribution in the loss function; the denominator is the similarity between the current sample and all samples. The sum of the exponential values ​​adjusted by the temperature parameter is equivalent to considering the comparison between the current sample and all samples; is the total number of samples of all samples, including positive samples and abnormal samples; It is a temperature parameter used to adjust the distribution of similarity, usually a small positive number. The similarity calculation function takes the dot product of the representation vector after L2 regularization or the cosine similarity between the representation vectors.

[0026] In particular, the similarity of the positive samples The calculation function is cosine similarity, specifically: .

[0027] In particular, the similarity between the current sample and all samples The calculation function is cosine similarity, specifically: .

[0028] Example 4 Based on Example 2 or 3, the classification network model in step 3 first removes the Project network in the heart rate sign extraction model, then fixes the Encoder network, and trains it with the cross entropy loss function for subsequent heart rate classification tasks.

[0029] The use process of the multi-sensor fusion heart rate monitoring method provided by the embodiment of the present invention is as follows: 1. Input the new heart rate data samples to be tested into the trained model and extract the feature representation; 2. Input the feature representation of this new sample into a trained classification network model for classification to obtain normal or abnormal heart rate judgment results.

[0030] This abnormal heart rate detection method based on contrastive learning can effectively utilize the characteristics of normal heart rate signals and enhance the model's ability to learn abnormal patterns. Through parameter sharing, the generalization of the model can be improved and the risk of overfitting can be reduced.

[0031] While performing heart rate detection, traditional heart rate variability (HRV) analysis is performed to further analyze heart rate health status, including: calculating the time interval (JJ interval) between consecutive heart rate events, and performing HRV analysis to evaluate the activity of the autonomic nervous system. Heart rate variability refers to the changes in the speed of the heart rate rhythm over time, and is a method for analyzing the difference in successive sinus heart rates. HRV reflects the activity of the autonomic nervous system and quantitatively evaluates the tension and balance of the cardiac sympathetic nerves and vagus nerves, thereby judging its effect on the condition and prevention of cardiovascular diseases. The larger the HRV value, the better the recovery and rest, and the healthier. Studies have shown that heart rate variability can also be applied to sleep staging, coronary heart disease, hypertension and heart failure detection, especially the continuous monitoring of the prognosis of cardiovascular diseases, and preventing further deterioration by understanding the changes in the heart.

[0032] HRV parameters were obtained through time domain analysis and nonlinear analysis.

[0033] Among them, time domain analysis is a simple statistical measure for calculating the JJ interval, including: Mean HR: The average of all JJ intervals.

[0034] Standard Deviation (SDNN): The overall standard deviation of the JJ interval, reflecting the overall HRV.

[0035] Mean NN50: The average number of consecutive JJ interval differences greater than 50 milliseconds.

[0036] Standard Deviation NN50 (SDNN Index): The standard deviation of the number of NN50, reflecting short-term HRV.

[0037] Triangular index (TINN): triangular distribution index of JJ interval differences.

[0038] Frequency domain analysis: Calculate the power spectral density (PSD) through fast Fourier transform (FFT) or other frequency domain analysis methods. Determine the following frequency domain indicators: Total Power: The power of the entire spectrum.

[0039] Low frequency power (LF Power): The power in the range of 0.04-0.15 Hz, reflecting the combined effects of the sympathetic and parasympathetic nerves.

[0040] HF Power: The power in the range of 0.15-0.4 Hz, which mainly reflects the parasympathetic nerve activity.

[0041] LF / HF ratio: assesses the balance between sympathetic and parasympathetic nervous activity.

[0042] Nonlinear analysis includes: Compute nonlinear statistical measures, such as the Poisson distribution index of heart rate (Poincare Plot).

[0043] The approximate entropy (ApEn) or sample entropy (SampEn) of heart rate evaluates the complexity and predictability of heart rate.

[0044] Based on the calculated HRV parameters, evaluate the activity and balance of the autonomic nervous system, draw graphs of HRV parameters, such as time series graphs, histograms, power spectra, etc., to intuitively display the HRV characteristics, and establish a user feedback mechanism, including: designing a user feedback mechanism. When an abnormal heart rate is detected, the user is prompted visually or auditorily, and further medical examinations are recommended.

[0045] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A multi-sensor fusion heart rate monitoring method, characterized in that: The method comprises the following steps: Step 1, collecting the original mixed body motion signal, and separating the normal BCG signal and the abnormal BCG signal; Step 2: training a heart rate sign extraction model using the signal collected in step 1 to extract the heart rate sign; Step 3: Using the heart rate feature to train a classification network model and outputting a heart rate judgment result; Step 4: input the new heart rate data into the heart rate sign extraction model, extract feature representation, and then input the feature representation into the trained classification network model to obtain the judgment result of the new heart rate data.

2. The multi-sensor fusion heart rate monitoring method according to claim 1, characterized in that: The step 2 includes the following sub-steps: Step 2.1, randomly selecting N signals from the original mixed body motion signals of step 1, and separating the normal BCG signals therefrom as positive sample data sets, and the abnormal BCG signals as negative sample data sets; Step 2.2: Send the original mixed body motion signal, the positive sample data set and the negative sample data set to the corresponding encoder to capture the time-frequency features of the BCG signal, and record the feature representations as , +、 , wherein parameters are shared between the encoders; Step 2.3: Through a Project network, the feature representation , +、 Mapped into another space, and then normalized and calculated the contrast loss function to obtain the heart rate sign.

3. The multi-sensor fusion heart rate monitoring method according to claim 2, characterized in that: In step 2.2, the feature representation + is obtained by taking the time-frequency features obtained from the original mixed body motion signal and the positive sample data set as the input of Transformer and fusing the features through the multi-head attention mechanism.

4. The multi-sensor fusion heart rate monitoring method according to claim 2, characterized in that: The encoder in step 2.2 adopts convolutional neural network, recurrent neural network, LSTM, Resnet or Unet.

5. The multi-sensor fusion heart rate monitoring method according to claim 2, characterized in that: The Project network in step 2.3 is CNN.

6. The multi-sensor fusion heart rate monitoring method according to claim 2, characterized in that: The normalization in step 2.3 is specifically to map to a hypersphere through standardization.

7. The multi-sensor fusion heart rate monitoring method according to claim 2, characterized in that: The contrast loss function in step 2.3 is InfoNCE Loss, and the specific calculation formula is: in, Represents the contrast loss function, and the numerator represents the similarity of the positive sample The exponential value after temperature parameter adjustment, the denominator is the similarity between the current sample and all samples The sum of the exponential values ​​adjusted by the temperature parameter, is the total number of samples of all samples, including positive samples and abnormal samples; is a temperature parameter used to adjust the distribution of similarity.

8. The multi-sensor fusion heart rate monitoring method according to claim 7, characterized in that: The similarity of the positive sample The calculation function is cosine similarity, specifically: 。 9. The multi-sensor fusion heart rate monitoring method according to claim 7, characterized in that: The similarity between the current sample and all samples The calculation function is cosine similarity, specifically: 。 10. The multi-sensor fusion heart rate monitoring method according to claim 2, characterized in that: The classification network model in step 3 first removes the Project network in the heart rate sign extraction model, then fixes the Encoder network, and trains it using the cross entropy loss function.

Images