A method for dividing SCG signal heartbeat period

Abstract

The application discloses a heartbeat period division method of SCG signals, comprising the following steps: S1, acquiring a plurality of human body SCG signals through a sensor and storing the SCG signals in a memory to form an SCG signal database; S2, marking each SCG signal in the SCG signal database with a heartbeat period to divide the SCG signal into a plurality of complete SCG signal heartbeat periods. The heartbeat period division method of the SCG signals disclosed by the application firstly performs noise reduction preprocessing on the SCG signals, then marks the processed signal data, and trains the signal data through a bidirectional LSTM network model with attention enhancement, the trained model can judge whether a single heartbeat period is accurately divided, and the self-adaptive sliding window algorithm is used to adjust the heartbeat period, so that the complete SCG signal heartbeat period is obtained.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of heartbeat cycle division, and particularly relates to a SCG signal heartbeat cycle division method. BACKGROUND

[0002] Cardiovascular disease has become one of the factors that seriously affect people's life and health, so it is very important to detect the state of the cardiovascular system and evaluate the health. At present, electrocardiogram (ECG), ultrasound, and angiography are mainly used to detect the state of the heart and evaluate the health. Among them, the ultrasound and angiography technology has high professional requirements for the operator, while the ECG is often used for early diagnosis of heart disease due to its non-invasive, simple operation, low cost and other characteristics. However, only when the heart is abnormal, the abnormal condition can be determined by detecting the ECG signal, and it is difficult to monitor the ECG of a person for a long time. In addition, the ECG signal has low sensitivity and specificity for structural heart disease. With the development of sensor technology, signal processing and artificial intelligence technology, some scholars propose to use seismocardiography (SCG) composed of heart mechanical vibration signals to detect heart disease. Compared with the ECG signal, the SCG signal is more convenient to collect and easier to monitor for a long time, and has higher sensitivity and accuracy for some heart diseases (such as myocardial ischemia, myocardial infarction, etc.). The SCG signal can reflect the mechanical movement of the heart and has important value for diagnosing heart disease, which can provide comprehensive and accurate diagnostic information.

[0003] SCG signals are primarily acquired from chest cavity vibrations caused by heartbeats, and there are currently two main methods: contact and non-contact. Contact acquisition methods mainly utilize accelerometers and gyroscopes strapped to the chest cavity; however, this method is not suitable for long-term wear by users, nor is it appropriate for burn patients, patients with infectious diseases, or infants. Non-contact methods primarily use radar technology to capture the displacement of the chest cavity during heartbeats, and then analyze and process this data. In recent years, there has been considerable research on extracting human heart rate and respiratory signals using microwave and millimeter-wave (0-300 GHz) radar. However, lower frequency and low bandwidth radar has limited sensing accuracy and resolution, resulting in poor detection capabilities for weak SCG signals and an inability to further acquire abnormal cardiac information. Higher-frequency terahertz bands are needed for more precise SCG signal acquisition and analysis. The terahertz band (0.1-10 THz) has advantages such as large usable bandwidth, good polarity, and harmlessness to the human body, making it a popular choice for researchers. First, the terahertz band offers a large usable bandwidth, which can better improve the range resolution of radar. Second, its short wavelength makes it more sensitive to weak signals; for example, a terahertz wave with a wavelength of 1 mm can detect a phase amplitude of approximately 1.26 rad for a signal with a vibration amplitude of only 100 μm. Furthermore, the shorter wavelength also gives terahertz signals better directionality, reducing the influence of irrelevant information.

[0004] Currently, there is a growing body of research on the analysis and processing of terahertz radar SCG signals. Among these studies, the segmentation of individual heartbeat cycles within the SCG signal is fundamental to understanding different cardiac abnormalities and is frequently used to calculate heart rate variability (HRV), a crucial indicator for predicting sudden cardiac death and arrhythmic events. Therefore, accurate segmentation of heartbeat cycles within the SCG signal is of paramount importance. However, due to the indistinct characteristics and strong temporal correlation of terahertz radar SCG signals, and the high event correlation and coupling between heartbeat cycles, segmenting individual heartbeat cycles over long time series presents significant challenges.

[0005] Existing technologies primarily utilize heart rate acceleration information, segmenting heartbeat cycles based on the similarity between cycles using template matching algorithms and dynamic programming. However, significant challenges exist in template selection and matching. Firstly, fixed templates or multiple templates cannot fully accommodate all heartbeat cycles; secondly, the probability of errors during matching is relatively high. Some technologies divide heartbeat cycles using simple sliding windows after bandpass filtering, but their decoupling ability between different heartbeat cycles is weak, also leading to substantial errors. Other technologies employ cutting-edge artificial intelligence methods, such as recurrent neural networks, to segment SCG signals, achieving some success, but further utilization of the feature information within the heartbeat cycle remains.

[0006] Therefore, the above problems are further improved. SUMMARY

[0007] The main purpose of the present application is to provide a heartbeat period division method for SCG signals, which is characterized by unclear SCG signal features, strong time sequence correlation, and strong heartbeat period coupling. First, the SCG signal is preprocessed by noise reduction, then the processed signal data is labeled, and a bidirectional LSTM network model with attention enhancement is introduced for training. The trained model can judge whether the single heartbeat period is divided accurately. Then, the self-adaptive sliding window algorithm is used to adjust the heartbeat period, and finally the complete SCG signal heartbeat period is obtained.

[0008] To achieve the above purpose, the present application provides a heartbeat period division method for SCG signals, which is used to obtain the heartbeat period of SCG signals, comprising the following steps:

[0009] Step S1: Obtain several human SCG signals through a sensor (preferably a terahertz radar) and store them in a memory to form an SCG signal database;

[0010] Step S2: Label each SCG signal in the SCG signal database with a heartbeat period to divide it into multiple complete SCG signal heartbeat periods;

[0011] Step S3: Input the multiple SCG signal heartbeat periods divided into the pre-built bidirectional LSTM network with attention enhancement for training to obtain a heartbeat period discrimination model including network structure, connection weight, activation value, and characteristic value corresponding heartbeat period type information;

[0012] Step S4: Obtain the SCG signal S t of a human body for a period of time through the sensor again, and store it in the memory after determining the starting position of the time period;

[0013] Step S5: Take (divide) a segment of SCG signal S w (initial heartbeat period) with a time length of w from the SCG signal S t according to the average heart rate of the human body in the time period as a time window w;

[0014] Step S6: Determine whether the number of discrimination update times for the current heartbeat period exceeds the upper limit value. If not, execute step S7, otherwise, output the complete current heartbeat period and delete the segment of signal from the collected SCG signal S t ;

[0015] Step S7: input the intercepted or updated SCG signal into the heartbeat cycle discrimination model to output a judgment on whether the current heartbeat cycle is offset, and perform corresponding matching processing according to the offset condition, so that the heartbeat cycle discrimination model outputs the complete current heartbeat cycle and deletes the segment of the collected SCG signal S t ;

[0016] Step S8: judge whether the current SCG signal S t has been segmented, if yes, end, otherwise, enter step S5 to divide the next SCG signal heartbeat cycle.

[0017] As a further preferred technical solution of the above technical solution, the heartbeat cycle discrimination model of step S3 is obtained by the following steps:

[0018] Step S3.1: divide the selected input single SCG signal heartbeat cycle into multiple time intervals according to the minimum sampling time t s , so as to be used for data operation, and the output of the connection of the positive and negative two directions of the attention enhanced bidirectional LSTM network can be represented as:

[0019]

[0020] , wherein, is the state component of the positive and negative two layers of LSTM model, and is represented as:

[0021]

[0022]

[0023] , wherein x t is the t-th interval of the input SCG signal heartbeat cycle, C t-1 is the value of the memory unit at different t-1 time, after obtaining each output component, each component is summed according to the setting of the attention mechanism and used as the output of the bidirectional LSTM network, and the calculation method is:

[0024]

[0025] , wherein a t is the importance of the current component to the current heartbeat cycle, and is represented as:

[0026]

[0027] , wherein s(K t , Q t ) is a scoring function, and K is the selected standard SCG signal heartbeat cycle.

[0028] Step S3.2: (after completing the processing of a single SCG signal heartbeat cycle) completing the training of multiple SCG signal heartbeat cycles by constructing the relationship between the label information and the Attention value of each SCG signal heartbeat cycle, thereby obtaining a heartbeat cycle discrimination model.

[0029] As a further preferred technical solution of the above technical solution, the step S7 of performing corresponding matching processing according to the offset condition is specifically implemented as the following steps:

[0030] Step S7.1: if the output is judged as a late offset, the end time of the current heartbeat cycle is moved forward by a minimum sampling time t s , to obtain a new heartbeat cycle S' w (w-t s ), and return to step S6;

[0031] Step S7.2: if the output is judged as an early offset, the end time of the current heartbeat cycle is moved backward by a minimum sampling time t t , to obtain a new heartbeat cycle S' t (w+t w ), and return to step S6;

[0032] Step S7.3: if the output is judged as no offset, the complete current heartbeat cycle is directly output.

[0033] As a further preferred technical solution of the above technical solution, in step S5, the average heart rate of the human body is obtained by the electrocardio monitoring device and is stored in the memory together with the SCG signal.

[0034] As a further preferred technical solution of the above technical solution, the late offset in step S7.1 refers to that the end time of the segmented heartbeat cycle is greater than the actual end time, and the early offset in step S7.2 refers to that the end time of the segmented heartbeat cycle is less than the actual end time.

[0035] To achieve the above purpose, the present application further provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the heartbeat cycle division method of the SCG signal when executing the program.

[0036] To achieve the above purpose, the present application further provides a non-transitory computer readable storage medium, which stores a computer program, wherein the computer program is executable on a processor to implement the steps of the heartbeat cycle division method of the SCG signal.

[0037] The present application has the following beneficial effects:

[0038] 1、The SCG signal heartbeat period division method of the present application introduces an attention enhancement mechanism on the basis of self-learning of feature selection, adds matching operation on key timing features in the standard SCG signal heartbeat period, improves the algorithm's capture ability of key feature information, and can improve the accuracy of division of individual heartbeat periods in the SCG signal;

[0039] 2、The present application strengthens the bidirectional learning ability of the algorithm in timing, which can reduce the coupling between each heartbeat period in the SCG signal;

[0040] 3、The present application designs a corresponding adaptive sliding window method for the SCG signal heartbeat period discrimination model obtained by training, which has good robustness. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is the heartbeat period discrimination model training flowchart of the SCG signal heartbeat period division method of the present application.

[0042] Figure 2 is the heartbeat period division flowchart of the SCG signal heartbeat period division method of the present application.

[0043] Figure 3 is the structure diagram of the double-LSTM network model with attention enhancement introduced in the SCG signal heartbeat period division method of the present application.

[0044] Figure 4 is the result schematic diagram of the SCG signal heartbeat period obtained after division in the SCG signal heartbeat period division method of the present application. DETAILED DESCRIPTION

[0045] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought by those skilled in the art. The basic principles of the present application defined in the following description can be applied to other embodiments, modifications, improvements, equivalents and other technical solutions without departing from the spirit and scope of the present application.

[0046] In the preferred embodiments of the present application, those skilled in the art should note that the sensor for obtaining SCG signal, electronic equipment and the like involved in the present application can be regarded as prior art.

[0047] Preferred embodiments.

[0048] As Figures 1-4 shown, the present application discloses a SCG signal heartbeat period division method for accurately obtaining SCG signal heartbeat period, which comprises the following steps:

[0049] Step S1: Obtain several human SCG signals through a sensor (preferably a terahertz radar) and store them in a memory to form an SCG signal database;

[0050] Step S2: Mark each SCG signal in the SCG signal database with a heartbeat cycle to segment into multiple complete SCG signal heartbeat cycles;

[0051] Step S3: Input the segmented multiple SCG signal heartbeat cycles into a pre-built bidirectional LSTM network enhanced with attention to train to obtain a heartbeat cycle discrimination model including network structure, connection weight, activation value, and feature value corresponding heartbeat cycle type information;

[0052] Step S4: Obtain the SCG signal S of a human body for a period of time again through a sensor t , and store the starting position of the time period in the memory;

[0053] Step S5: According to the average heart rate of the human body in the time period as a time window w, cut (segment) a SCG signal S t with a time length of w from the SCG signal S w (initial heartbeat cycle);

[0054] Step S6: Determine whether the number of discrimination updates for the current heartbeat cycle exceeds the upper limit value, if not, execute step S7, otherwise, output the complete current heartbeat cycle, and delete the signal in the collected SCG signal S t ;

[0055] Step S7: Input the cut or updated SCG signal into the heartbeat cycle discrimination model to output whether the current heartbeat cycle is offset, and perform corresponding matching processing according to the offset condition, so that the heartbeat cycle discrimination model outputs the complete current heartbeat cycle and deletes the signal in the collected SCG signal S t ;

[0056] Step S8: Determine whether the current SCG signal S t has been segmented, if yes, end, otherwise, go to step S5 to divide the next SCG signal heartbeat cycle.

[0057] Specifically, the heartbeat cycle discrimination model of step S3 is obtained by the following steps:

[0058] Step S3.1: Select an input single SCG signal heartbeat cycle according to the minimum sampling time t sThe segmentation into multiple time intervals is used for data operation, and the connection of the attention-enhanced bidirectional LSTM network can be represented as:

[0059]

[0060] wherein, is the state component of the forward and reverse two-layer LSTM model, and is represented as:

[0061]

[0062]

[0063] wherein, x t is the t-th interval of the input SCG signal heartbeat period, C t-1 is the value of the memory unit at different t-1 time points, after obtaining each output component, each component is summed according to the setting of the attention mechanism and is used as the output of the bidirectional LSTM network, and the calculation method is:

[0064]

[0065] wherein, α t is the importance of the current component to the current heartbeat period, and is represented as:

[0066]

[0067] wherein, s(K t , Q t ) is a scoring function, and K is a standard SCG signal heartbeat period;

[0068] Step S3.2: (after completing the processing of a single SCG signal heartbeat period) complete the training of multiple SCG signal heartbeat periods by constructing the relationship between the label information of each SCG signal heartbeat period and the Attention value, thereby obtaining a heartbeat period discrimination model.

[0069] More specifically, the corresponding matching processing according to the offset condition in step S7 is implemented as the following steps:

[0070] Step S7.1: if the output is judged as a post-offset, the end time of the current heartbeat period is moved forward by one minimum sampling time t s , to obtain a new heartbeat period S′ w (w-t s ), and return to step S6;

[0071] Step S7.2: if the output is judged as a pre-offset, the end time of the current heartbeat period is moved backward by one minimum sampling time ts , obtaining a new heartbeat cycle S' w (w+t s ), and returning to step S6;

[0072] Step S7.3: If the output is judged as no offset, output the complete current heartbeat cycle directly.

[0073] Further, in step S5, the average heart rate of the human body is acquired by the electrocardio monitoring device and is stored in the memory together with the SCG signal.

[0074] Still further, the post-offset in step S7.1 refers to that the end time of the segmented heartbeat cycle is greater than the actual end time, and the pre-offset in step S7.2 refers to that the end time of the segmented heartbeat cycle is less than the actual end time.

[0075] The application further discloses an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the heartbeat cycle segmentation method of the SCG signal when executing the program.

[0076] The application further discloses a non-transient computer readable storage medium, which stores a computer program, wherein the computer program is executable on the processor to implement the steps of the heartbeat cycle segmentation method of the SCG signal.

[0077] Preferably, for the heartbeat cycle judgment model, other algorithms can be used, such as deep learning or machine learning algorithms, such as convolutional neural network, SVM, etc.

[0078] It is worth mentioning that the technical features of the sensor acquiring the SCG signal, the electronic device and the like involved in the present application patent application should be regarded as prior art, and the specific structure, working principle and possible control mode, spatial arrangement mode of these technical features can be selected by using the conventional selection in the art, and should not be regarded as the invention point of the present application patent, and the present application patent will not be further specifically expanded and described.

[0079] For those skilled in the art, the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents, and any modification, equivalent replacement, improvement and the like within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for dividing a heartbeat cycle of an SCG signal, for obtaining a heartbeat cycle of an SCG signal, the method comprising: determining a first heartbeat cycle of the SCG signal; determining a second heartbeat cycle of the SCG signal; and determining a third heartbeat cycle of the SCG signal. The method comprises the following steps: Step S1: obtaining a plurality of human SCG signals through a sensor and storing in a memory to form an SCG signal database; Step S2: marking each SCG signal in the SCG signal database with a heartbeat period to segment into a plurality of complete SCG signal heartbeat periods; Step S3: inputting the plurality of segmented SCG signal heartbeat periods into a pre-built bidirectional LSTM network enhanced with attention to train, to obtain a heartbeat period discrimination model comprising network structure, connection weight, activation value and characteristic value corresponding heartbeat period type information; Step S4: again acquire the SCG signal S of the human body for a period of time through the sensor t and the starting position of the period of time is determined and stored in the memory; Step S5: According to the average heart rate of the human body in this time period as a time window w, a segment of the SCG signal S t with a time length of w is intercepted from the SCG signal S w ; Step S6: judging whether the number of discrimination update times for the current heart cycle exceeds the upper limit value, if not, executing step S7, otherwise, outputting the complete current heart cycle, and deleting the segment signal in the collected SCG signal S t ​ Step S7: input the intercepted or updated SCG signal into the heartbeat cycle discrimination model to output a judgment on whether the current heartbeat cycle is deviated, and perform corresponding matching processing according to the deviation condition, so that the heartbeat cycle discrimination model outputs the complete current heartbeat cycle and deletes the segment signal in the collected SCG signal S t ​ Step S8: judging whether the current SCG signal S t is already divided, if yes, ending, otherwise, entering step S5 to divide the next SCG signal heartbeat period.

2. The method of claim 1, wherein the heartbeat period of the SCG signal is divided into a plurality of heartbeat periods, and each of the heartbeat periods is divided into a plurality of heartbeat sub-periods. The obtaining of the heartbeat period discrimination model of step S3 is implemented by the following steps: Step S3.1: select the input single SCG signal heartbeat cycle according to the minimum sampling time t s The data is divided into multiple time intervals for operation, and the connection of the attention-enhanced bidirectional LSTM network and the output of the LSTM model in the positive and negative directions can be represented as: wherein, is the state component of the bidirectional LSTM model, denoted as: Wherein, x t is the tth interval of the heartbeat cycle of the input SCG signal, C t-1 is the value of the memory unit at different t-1 time, after obtaining each output component, according to the setting of the attention mechanism, each component is summed according to the weight and taken as the output of the bidirectional LSTM network, and the calculation method is as follows: wherein a t is the importance of the current component for the current heartbeat period, expressed as: where s(K t , Q t ) is a scoring function, and K is the selected standard SCG signal heart period. Step S3.2: training the plurality of SCG signal heartbeat periods by building the relationship between the marking information of each SCG signal heartbeat period and the Attention value, to obtain the heartbeat period discrimination model.

3. The method of claim 2, wherein the heartbeat period is divided into a plurality of sub-periods, and each of the sub-periods is divided into a plurality of time slots. The specific implementation of the corresponding matching processing according to the offset in step S7 is as follows: Step S7.1: If the output is judged to be late, move the end time of the current heartbeat period forward by one minimum sampling time t s , get a new heartbeat period S' w , and return to step S6; Step S7.2: If the output is determined to be leading, move the end time of the current heartbeat period back by one minimum sampling time t s , obtaining a new heartbeat period S' w , and returning to step S6; Step S7.3: if the output is judged as no offset, directly output the complete current heartbeat period.

4. The method of claim 3, wherein the heartbeat period is divided into a plurality of sub-periods, and each of the sub-periods is divided into a plurality of time slots. In step S5, the average heart rate of the human body is obtained by the electrocardio monitoring device and stored in the memory together with the SCG signal.

5. The method of claim 4, wherein the heartbeat period is divided into a plurality of sub-periods, and each of the sub-periods is divided into a plurality of time slots. The post-offset in step S7.1 refers to the end time of the segmented heartbeat period being greater than the actual end time, and the pre-offset in step S7.2 refers to the end time of the segmented heartbeat period being less than the actual end time.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the heartbeat period segmentation method of the SCG signal according to any one of claims 1 to 5.

7. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the heartbeat period segmentation method of the SCG signal according to any one of claims 1 to 5.

Images