Atrial fibrillation detection method and device, electronic equipment, storage medium and program product

Abstract

The invention discloses an atrial fibrillation detection method and device, electronic equipment, a storage medium and a program product. The method comprises the steps that a ballistocardiogram signal uploaded by a pressure sensor arranged in bedding is acquired; determining effective signal segments in the ballistocardiogram signal according to the periodic characteristics of the ballistocardiogram signal; and obtaining time-frequency characteristics of the effective signal segments, and generating an atrial fibrillation detection result according to the time-frequency characteristics. According to the invention, a ballistocardiogram signal can be captured for a long time in a non-contact manner through the pressure sensor arranged in the bedding, and comprehensive data support is provided for early discovery of atrial fibrillation; by screening high-quality effective signal fragments, misjudgment or missed judgment caused by invalid data can be reduced, and the accuracy of atrial fibrillation detection is improved; rhythm laws in the signals can be reflected more accurately by using the time-frequency characteristics, and the reliability of atrial fibrillation detection is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to an atrial fibrillation detection method and device, electronic equipment, storage medium and program product. BACKGROUND

[0002] Atrial fibrillation is a common arrhythmia, and the patient population is large. Its serious complications such as stroke and heart failure caused by atrial fibrillation are significant. At present, the traditional monitoring method mainly relies on electrocardiogram and 24-hour dynamic electrocardiogram. However, the traditional monitoring method has obvious limitations: the symptomless and paroxysmal characteristics of early atrial fibrillation make it difficult for the traditional monitoring method to effectively capture short-term or paroxysmal atrial fibrillation; dynamic electrocardiogram monitoring needs to wear chest leads or electrode patches, and the monitoring time is limited, which cannot cover a longer time, especially the night heart rhythm. SUMMARY

[0003] The present application provides an atrial fibrillation detection method, device, electronic equipment, storage medium and program product to provide comprehensive data support for early detection of atrial fibrillation and improve the accuracy and reliability of atrial fibrillation detection.

[0004] In one aspect of the present application, an atrial fibrillation detection method is provided, comprising:

[0005] obtaining a ballistocardiogram signal uploaded by a pressure sensor built in a bedding;

[0006] determining an effective signal segment in the ballistocardiogram signal according to the periodicity characteristics of the ballistocardiogram signal;

[0007] obtaining the time-frequency characteristics of the effective signal segment, and generating an atrial fibrillation detection result according to the time-frequency characteristics.

[0008] In one aspect of the present application, an atrial fibrillation detection device is provided, comprising:

[0009] a first signal acquisition module configured to obtain a ballistocardiogram signal uploaded by a pressure sensor built in a bedding;

[0010] a second signal acquisition module configured to determine an effective signal segment in the ballistocardiogram signal according to the periodicity characteristics of the ballistocardiogram signal;

[0011] a result acquisition module configured to obtain the time-frequency characteristics of the effective signal segment, and generate an atrial fibrillation detection result according to the time-frequency characteristics.

[0012] In another aspect of the present application, an electronic equipment is provided, comprising:

[0013] at least one processor; and

[0014] a memory in communication with the at least one processor;

[0015] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the atrial fibrillation detection method of any of the embodiments of the present application.

[0016] According to another aspect of the embodiments of the present application, a computer readable storage medium is provided, comprising: computer instructions for enabling a processor to perform the atrial fibrillation detection method of any of the embodiments of the present application when executed by the processor.

[0017] According to another aspect of the embodiments of the present application, a computer program product is provided, comprising: a computer program for enabling a processor to perform the atrial fibrillation detection method of any of the embodiments of the present application when executed by the processor.

[0018] The present application obtains the ballistocardiogram signal uploaded by the pressure sensor built in the bedding, determines the effective signal segment in the ballistocardiogram signal according to the periodicity characteristics of the ballistocardiogram signal, obtains the time-frequency characteristics of the effective signal segment, and generates the atrial fibrillation detection result according to the time-frequency characteristics. The present application can realize contactless and long-time capture of the ballistocardiogram signal through the pressure sensor built in the bedding, provides comprehensive data support for early discovery of atrial fibrillation; through screening of high-quality effective signal segments, the present application can reduce misjudgment or omission caused by invalid data, and improves the accuracy of atrial fibrillation detection; the time-frequency characteristics can more accurately reflect the rhythm law in the signal, and improve the reliability of atrial fibrillation detection.

[0019] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0021] Figure 1 is a flow chart of an atrial fibrillation detection method according to the first embodiment of the present application;

[0022] Figure 2 is another flow chart of an atrial fibrillation detection method according to the second embodiment of the present application;

[0023] Figure 3 is another flow chart of an atrial fibrillation detection method according to the third embodiment of the present application;

[0024] Figure 4 Another flow chart of the atrial fibrillation detection method provided for the fourth embodiment of the present application;

[0025] Figure 5 A schematic diagram of a band-pass filtered Holter signal provided for the fourth embodiment of the present application;

[0026] Figure 6 A schematic diagram of an atrial fibrillation detection device provided for the fourth embodiment of the present application;

[0027] Figure 7 Another structural diagram of an atrial fibrillation detection device provided for the fifth embodiment of the present application;

[0028] Figure 8 A block diagram of an electronic device for executing an atrial fibrillation detection method provided for the sixth embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the personnel in the technical field better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by the personnel in the field without creative labor should belong to the protection scope of the present application.

[0030] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product or equipment including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or equipment.

[0031] Figure 1 A flow chart of an atrial fibrillation detection method provided for the embodiments of the present application, the embodiments of the present application can be applicable to the scene of detecting whether the atrial fibrillation event occurs to the to-be-detected object at night or during rest. The method can be executed by an atrial fibrillation detection device, which can be realized in the form of hardware and / or software, and can be configured in a semiconductor test dedicated server or a distributed computing cluster composed of multiple servers and the like. As shown in the figure, the method comprises: Figure 1 ​

[0032] S110, acquire the ballistocardiogram signal uploaded by the pressure sensor built in the bedding.

[0033] The pressure sensor refers to a device capable of sensing pressure changes and converting the pressure changes into an electrical signal. In the embodiments of the present application, the pressure sensor can sense the micro-vibration of the body surface caused by the heart beat and blood flow of the human body and convert the micro-vibration into an electrical signal. It can be understood that the electrical signal is the ballistocardiogram signal. The pressure sensor can be installed inside the bedding or on the upper surface of the bedding. For example, the bedding can be a mattress or a sofa. The deployment position of the pressure sensor can include a longitudinal deployment position in the projection area of the thoracic spine heart or a transverse deployment position in the projection area covering the fourth intercostal level of the midaxillary line.

[0034] The ballistocardiogram signal can be understood as an electrical signal formed by the micro-vibration of the body surface caused by the heart beat and blood flow of the human body. It can reflect the rhythm of the heart activity and can be used to judge the atrial fibrillation condition of the to-be-detected object. For example, the rhythm of the heart activity and the atrial fibrillation condition of the to-be-detected object can be judged by observing the waveform of the ballistocardiogram signal. If the waveform of the ballistocardiogram signal is observed to be periodically repeated with small peak-to-peak interval variation, it can be explained that the rhythm of the heart activity of the to-be-detected object is fixed and the atrial fibrillation condition of the to-be-detected object is non-atrial fibrillation. If the waveform of the ballistocardiogram signal is observed to be fuzzy or even disappear with random peak-to-peak interval, it can be explained that the rhythm of the heart activity of the to-be-detected object is non-fixed rhythm and the atrial fibrillation condition of the to-be-detected object is atrial fibrillation.

[0035] Specifically, the ballistocardiogram signal actively uploaded or passively uploaded by the pressure sensor is acquired through wired connection transmission or wireless transmission. The ballistocardiogram signal is an electrical signal converted by the pressure sensor after sensing the micro-vibration of the body surface caused by the heart beat and blood flow of the human body.

[0036] S120, determine the effective signal segment in the ballistocardiogram signal according to the periodicity characteristic of the ballistocardiogram signal.

[0037] The periodicity characteristic can be understood as a regularity characteristic that repeatedly appears in the ballistocardiogram signal, which is used to screen the effective signal segment. For example, the regularity characteristic can be manifested as that the waveform of the ballistocardiogram signal is periodically repeated and / or the peak value interval also presents a stable periodicity.

[0038] The effective signal segment can be understood as a signal segment with clear periodicity characteristic screened from the original ballistocardiogram signal. The effective signal segment is a signal segment with strong periodicity characteristic, which can more accurately reflect the heart activity state. For example, the periodicity characteristic of the signal can be reflected by using the periodicity strength or the periodicity segment similarity, and the effective signal segment can be determined by calculating the periodicity strength or the periodicity segment similarity of the ballistocardiogram signal.

[0039] Specifically, the ballistocardiogram signal can be divided into a plurality of signal segments, and it is detected whether a regularity feature repeatedly appears in each signal segment, and if the regularity feature repeatedly appears, the corresponding signal segment is taken as an effective signal segment.

[0040] Further, in the embodiment of the present application, before determining the effective signal segment in the ballistocardiogram signal, a signal preprocessing operation can be performed on the ballistocardiogram signal, for example, the signal preprocessing operation can include: filtering out high-frequency interference and ultra-low frequency baseline drift of the ballistocardiogram signal by using a band-pass filter, or performing signal normalization processing on the ballistocardiogram signal, etc.

[0041] S130, acquiring a time-frequency feature of the effective signal segment, and generating an atrial fibrillation detection result according to the time-frequency feature.

[0042] The time-frequency feature refers to a feature obtained by converting the effective signal segment into a time-frequency domain and performing feature extraction, for example, the effective signal segment can be converted into a time-frequency domain by using a short-time Fourier transform or a wavelet transform, and the time-frequency feature can be extracted in the time-frequency domain by using a preset feature model such as a machine learning model or a neural network model. The time-frequency feature of the effective signal segment can reflect the frequency distribution rule of the heart activity.

[0043] The atrial fibrillation detection result refers to a result of whether the to-be-detected object has atrial fibrillation, which is obtained according to the time-frequency feature of the ballistocardiogram signal, for example, the atrial fibrillation detection result at least includes two types of results of atrial fibrillation and non-atrial fibrillation.

[0044] Specifically, the effective signal segment can be converted into a time-frequency domain by using a short-time Fourier transform or a wavelet transform, an effective time-frequency signal of the effective signal segment is obtained, the effective time-frequency signal is input into a preset feature extraction model such as a machine learning model or a neural network model for extracting the time-frequency feature, the time-frequency feature of the effective signal segment is obtained, and then the time-frequency feature is input into another model such as a machine model or a neural network for generating the atrial fibrillation detection result, and the atrial fibrillation detection result is obtained.

[0045] The embodiment of the present application acquires the ballistocardiogram signal uploaded by the pressure sensor built in the bedding, determines the effective signal segment in the ballistocardiogram signal according to the periodicity characteristics of the ballistocardiogram signal, acquires the time-frequency characteristics of the effective signal segment, and generates an atrial fibrillation detection result according to the time-frequency characteristics. The embodiment of the present application can realize contactless and long-time capture of the ballistocardiogram signal through the pressure sensor built in the bedding, and provides comprehensive data support for early discovery of atrial fibrillation; through screening of high-quality effective signal segments, the embodiment of the present application can reduce misjudgment or omission caused by invalid data, and improve the accuracy of atrial fibrillation detection; the embodiment of the present application can more accurately reflect the rhythm law in the signal by using the time-frequency characteristics, and improve the reliability of atrial fibrillation detection.

[0046] Embodiment two

[0047] Figure 2 The embodiment two of the present application provides a program diagram of another atrial fibrillation detection method, and the embodiment of the present application is a refinement of the above-mentioned embodiment. Specifically, the embodiment of the present application refines how to determine the effective signal segment and how to acquire the specific steps of the time-frequency characteristics.

[0048] As shown in Figure 2 , another atrial fibrillation detection method can include the following specific steps:

[0049] S210, acquiring a ballistocardiogram signal uploaded by a pressure sensor built in bedding.

[0050] S220, dividing the ballistocardiogram signal into at least two signal segments according to a preset fixed interval.

[0051] The preset fixed interval can be understood as a kind of preset fixed time length, which is used for dividing the ballistocardiogram signal. The continuous ballistocardiogram signal can be divided into multiple equal-length signal segments according to the preset fixed interval. The preset fixed interval determines the time length and the number of the signal segments. Specifically, the time length of each signal segment is equal to the preset fixed interval; the ratio of the total time length of the continuous ballistocardiogram signal to the preset fixed interval is the number of the signal segments. Dividing the ballistocardiogram signal according to the preset fixed interval can ensure that the signal segments have a uniform analysis scale in the time dimension. Further, in the embodiment of the present application, the preset fixed interval can be flexibly adjusted according to different task scenarios. For example, when it is necessary to capture the short-term fluctuation of the signal more finely, the preset fixed interval can be shortened; when it is necessary to observe the medium and long-term trend of the signal, the preset fixed interval can be lengthened.

[0052] The signal segment refers to a part of the signal with a fixed time length, which is divided from the continuous ballistocardiogram signal according to the preset fixed interval, and is used as a basic unit for subsequent calculation of the autocorrelation sequence and analysis of the periodic intensity.

[0053] Specifically, a suitable preset fixed interval can be set according to different task scenarios, a time interval parameter of the preset fixed interval can be set through programming, a ballistocardiogram signal is automatically segmented into equal-length signal segments by using a signal processing tool, or the ballistocardiogram signal is manually segmented and processed according to the preset fixed interval in data analysis software, and the signal segments are obtained.

[0054] S230, acquiring a self-correlation sequence corresponding to each signal segment by using a self-correlation analysis method.

[0055] The self-correlation sequence can be understood as a group of numerical sequences obtained by processing the signal segment by using the self-correlation analysis method, and is used to quantify the strength of the periodicity feature of the signal segment. The more obvious the secondary peak in the self-correlation sequence is, the stronger the periodicity feature of the signal segment is, that is, the strength of the periodicity feature of the signal segment can be determined by extracting the secondary peak value in the self-correlation sequence.

[0056] Specifically, each signal segment is processed by using the self-correlation analysis method to obtain a self-correlation sequence corresponding to each signal segment. The self-correlation analysis method is a general technology for signal processing, and will not be described in detail in the embodiments of the present application.

[0057] S240, acquiring a highest secondary peak value and a main peak value in each self-correlation sequence, and taking a ratio of the highest secondary peak value to the main peak value as a periodicity strength corresponding to each signal segment.

[0058] The highest secondary peak value refers to a numerical value corresponding to the highest peak except the main peak in the self-correlation sequence, and the main peak value refers to a numerical value corresponding to the highest peak in the self-correlation sequence. The ratio of the highest secondary peak value to the main peak value can be used as an index for measuring the strength of the periodicity of the signal, and can be used to determine whether the signal segment has a strong periodicity. It should be noted that the highest secondary peak value should be a peak value having a fixed lag relationship with the main peak value in time, rather than an arbitrary secondary peak.

[0059] The periodicity strength can be understood as an index for measuring the strength of the periodicity feature of the signal segment, and is calculated by the ratio of the highest secondary peak value to the main peak value, and can reflect the strength of the periodicity of the signal segment. The higher the periodicity strength is, the more stable the periodicity of the signal is, and the higher the accuracy and stability of the atrial fibrillation detection are.

[0060] Specifically, the highest secondary peak and the main peak in the self-correlation sequence are extracted, the highest secondary peak value of the highest secondary peak and the main peak value of the main peak are acquired, the highest secondary peak is taken as the numerator, and the main peak value is taken as the denominator to obtain the ratio of the highest secondary peak value to the main peak value, and the ratio is taken as the periodicity strength of the signal segment.

[0061] S250, taking a signal segment corresponding to a periodicity strength greater than a preset strength threshold as an effective signal segment.

[0062] The preset intensity threshold can be understood as a preset critical value for determining whether the signal segment is sufficient as an effective signal segment. For example, the preset intensity threshold can be determined based on average statistical results of the ballistocardiogram signal cycle intensity of a large number of healthy people, or based on cycle intensity difference data of the atrial fibrillation patients and the healthy people, and the difference critical point of the cycle intensity difference of the atrial fibrillation patients and the healthy people is taken as the preset intensity threshold.

[0063] The effective signal segment can be understood as a signal segment with a cycle intensity greater than the preset intensity threshold, and the effective signal segment contains stable heart beat cycle information.

[0064] Specifically, the preset intensity threshold can be determined based on average statistical results of the ballistocardiogram signal cycle intensity of a large number of healthy people, or the difference critical point of the cycle intensity difference of the atrial fibrillation patients and the healthy people is taken as the preset intensity threshold, and the cycle intensity of each signal segment is compared with the preset intensity threshold, and the signal segment corresponding to the cycle intensity greater than the preset intensity threshold is taken as the effective signal segment.

[0065] For example, the signal segments can also be arranged in descending order according to the cycle intensity, and a preset proportion, for example, the first 20% or 30% of the number of cycle intensities corresponding to the signal segments, can be selected, and the signal segments corresponding to the number of cycle intensities corresponding to the preset proportion can be determined as the effective signal segments.

[0066] S260, performing short-time Fourier transform on the effective signal segment to obtain an effective time-frequency signal of the effective signal segment.

[0067] The effective time-frequency signal can be understood as a signal obtained by performing short-time Fourier transform on the effective signal segment, which contains information of the effective signal segment in time and frequency, and can intuitively reflect the frequency distribution change of the effective signal segment at different time points.

[0068] Specifically, the short-time Fourier transform is performed on the effective signal segment to obtain the effective time-frequency signal of the effective signal segment.

[0069] For example, the short-time Fourier transform calculation formula at least includes:

[0070]

[0071] wherein, is the effective signal segment, is a window function, is a time shift parameter, is a frequency parameter.

[0072] S270, input the effective time-frequency signal into a preset feature extraction model for feature extraction to obtain a time-frequency feature, and generate an atrial fibrillation detection result according to the time-frequency feature.

[0073] The preset feature extraction model can be understood as a pre-trained machine learning model or neural network model, which is used to extract the time-frequency feature from the effective time-frequency signal. The preset feature extraction model can automatically extract the time-frequency feature reflecting the frequency distribution rule of the heart activity according to the input effective time-frequency signal.

[0074] Specifically, the effective time-frequency signal is input into the preset feature extraction model which has been pre-trained, to extract the time-frequency feature reflecting the frequency distribution rule of the heart activity. Then, the time-frequency feature is input into a preset atrial fibrillation detection model. The preset atrial fibrillation detection model can be used to process the time-frequency feature to obtain the result of whether the to-be-detected object has atrial fibrillation, i.e., the atrial fibrillation detection result. The preset feature extraction model or the preset atrial fibrillation detection model can be a machine learning model or a neural network model.

[0075] In the embodiment of the present application, the ballistocardiogram signal uploaded by the pressure sensor built in the bedding is acquired, the ballistocardiogram signal is divided into at least two signal segments according to a preset fixed interval, the autocorrelation analysis method is used to acquire the autocorrelation sequence corresponding to each signal segment, the highest order peak value and the main peak value in each autocorrelation sequence are acquired, the ratio of the highest order peak value to the main peak value is taken as the cycle intensity corresponding to each signal segment, the signal segment corresponding to the cycle intensity greater than a preset intensity threshold is taken as an effective signal segment, the short-time Fourier transform is performed on the effective signal segment to obtain the effective time-frequency signal of the effective signal segment, the effective time-frequency signal is input into a preset feature extraction model for feature extraction to obtain a time-frequency feature, and an atrial fibrillation detection result is generated according to the time-frequency feature. In the embodiment of the present application, the ballistocardiogram signal is segmented, the complexity of subsequent calculation is reduced, and the efficiency of atrial fibrillation detection is improved. Through piece-by-piece analysis, the local features can be avoided to be covered due to the too long original ballistocardiogram signal, and the accuracy of atrial fibrillation detection is further improved. The ratio of the highest order peak value to the main peak value is defined as the cycle intensity, the periodicity of the signal segment can be accurately quantified, and a reliable basis is provided for subsequent screening of effective signal segments. Through the analysis of atrial fibrillation based on the effective signal segment, the interference of invalid data can be reduced, and the accuracy of the atrial fibrillation detection result is improved. Through the preset feature extraction model, the time-frequency feature can be efficiently and accurately extracted, and the efficiency and accuracy of the atrial fibrillation detection result are further improved.

[0076] Further, the embodiment of the present application also supplements step S270, specifically, supplements the preset feature extraction model at least includes one of the following: local time-frequency feature extraction module, stage time-frequency feature extraction module and global time-frequency feature extraction module; also refines step S270, specifically, refines how to extract time-frequency features in the effective time-frequency signal, including:

[0077] Calling the local time-frequency feature extraction module to extract the local time-frequency features of the effective time-frequency signal; calling the stage time-frequency feature extraction module to extract the stage time-frequency features of the effective time-frequency signal; calling the global time-frequency feature extraction module to extract the global time-frequency features of the effective time-frequency signal.

[0078] The local time-frequency feature extraction module can be understood as a functional module for capturing micro-scale local detail information from the effective time-frequency signal, for example, the micro-scale local detail information is the local time-frequency feature in the extremely narrow frequency segment. The local time-frequency feature refers to the feature extracted from the micro-scale region of the effective time-frequency signal, which can reflect the local spectral characteristics of the effective time-frequency signal, for example, the local time-frequency feature can include: the instantaneous amplitude value, energy distribution or adjacent time-frequency point micro-texture fluctuation of a single frequency point in the time-frequency diagram, etc.

[0079] The stage time-frequency feature extraction module can be understood as a functional module for capturing medium-scale region information from the effective time-frequency signal, for example, the medium-scale region information can be the stage time-frequency feature on a specific frequency segment. The stage time-frequency feature refers to the feature extracted from the medium-scale region of the effective time-frequency signal, which can reflect the spectral distribution law of the effective time-frequency signal in the medium range, for example, the stage time-frequency feature can include: average amplitude in a certain continuous time period, amplitude variation trend in a specific frequency segment, or texture fluctuation characteristics in a medium-sized region, etc.

[0080] The global time-frequency feature extraction module can be understood as a functional module for capturing global information of the whole scale from the effective time-frequency signal, for example, the global information of the whole scale is the global time-frequency feature of the effective time-frequency signal in the full frequency range. The global time-frequency feature refers to the feature extracted from the global scale region of the effective time-frequency signal, which can reflect the spectral characteristics of the effective time-frequency signal in the whole range, for example, the global time-frequency feature can include: main frequency distribution, spectral entropy or overall smoothness in the full frequency range, etc.

[0081] Specifically, the effective time-frequency signal is taken as the input of the local time-frequency feature extraction module, and is input to the local time-frequency feature extraction module. The local time-frequency feature extraction module processes and analyzes the effective time-frequency signal, and outputs the local time-frequency feature of the effective time-frequency signal. The effective time-frequency signal is taken as the input of the stage time-frequency feature extraction module, and is input to the stage time-frequency feature extraction module. The stage time-frequency feature extraction module processes and analyzes the effective time-frequency signal, and outputs the stage time-frequency feature of the effective time-frequency signal. The effective time-frequency signal is taken as the input of the global time-frequency feature extraction module, and is input to the global time-frequency feature extraction module. The global time-frequency feature extraction module processes and analyzes the effective time-frequency signal, and outputs the global time-frequency feature of the effective time-frequency signal.

[0082] Further, in the embodiment of the present application, different sizes of convolution kernels can be used to realize the extraction of time-frequency features of different granularities. For example, 6x6 convolution kernels, 12x12 convolution kernels, and 22x22 convolution kernels can be used for the local time-frequency feature extraction module, the stage time-frequency feature extraction module, and the global time-frequency feature extraction module, respectively, or 5x5 convolution kernels, 10x10 convolution kernels, and 20x20 convolution kernels can be used for the local time-frequency feature extraction module, the stage time-frequency feature extraction module, and the global time-frequency feature extraction module, respectively.

[0083] Further, the embodiment of the present application further comprises that the local time-frequency feature extraction module, the stage time-frequency feature extraction module, and the global time-frequency feature extraction module in the preset feature extraction model each comprise at least two convolution layers, and each convolution layer contains different numbers of convolution kernels.

[0084] Specifically, the local time-frequency feature extraction module, the stage time-frequency feature extraction module, and the global time-frequency feature extraction module in the preset feature extraction model each comprise at least two convolution layers, and each convolution layer can be provided with different numbers of convolution kernels. By setting a double-layer convolution layer structure, subtle changes in the effective signal segment can be effectively captured, and the sensitivity of detection can be improved.

[0085] Embodiment Three

[0086] Figure 3 A flowchart of another atrial fibrillation detection method is provided for the third embodiment of the present application. The third embodiment of the present application is a refinement of the above-mentioned embodiments. Specifically, the specific steps of how to generate the atrial fibrillation detection result are refined.

[0087] As shown in Figure 3 Another atrial fibrillation detection method can include the following specific steps:

[0088] S310, acquiring a ballistocardiogram signal uploaded by a pressure sensor built in a bedding.

[0089] S320, determining an effective signal segment in the ballistocardiogram signal according to the periodicity feature of the ballistocardiogram signal.

[0090] S330, acquire the time-frequency feature of the effective signal segment, call a preset atrial fibrillation detection model to detect the time-frequency feature, and obtain an atrial fibrillation detection result; wherein the preset atrial fibrillation detection model comprises a cross-scale attention module, a feature fusion module, and a classification module.

[0091] The preset atrial fibrillation detection model can be understood as a model for obtaining an atrial fibrillation detection result according to input features, and examples of the preset atrial fibrillation detection model include a machine learning model or a neural network model. The preset atrial fibrillation detection model at least includes a cross-scale attention module, a feature fusion module, and a classification module. The cross-scale attention module can be understood as a module for assigning different weights to time-frequency features of different scales (local, stage, and global) in the preset atrial fibrillation detection model. The feature fusion module can be understood as a module for performing splicing processing on the weighted time-frequency features of different scales in the preset atrial fibrillation detection model, and can fuse features of different scales into a fusion feature vector. The classification module can be understood as a module for processing the fusion feature vector to determine whether it is atrial fibrillation in the preset atrial fibrillation detection model, and can output an atrial fibrillation detection result including two categories of atrial fibrillation and non-atrial fibrillation according to the features of the fusion feature vector.

[0092] Specifically, the effective signal segment is input into a feature extraction model to obtain the time-frequency feature of the effective signal segment, and the time-frequency feature of the effective signal segment is input into the preset atrial fibrillation detection model. The preset atrial fibrillation detection model can process and analyze the time-frequency feature to output an atrial fibrillation detection result. It should be noted that the preset atrial fibrillation detection model includes a cross-scale attention module, a feature fusion module, and a classification module. The time-frequency feature of the effective signal segment is first input into the cross-scale attention module, which is used to assign different weights to time-frequency features of different scales (local, stage, and global). The feature fusion module is used to perform splicing processing on the weighted local time-frequency feature, the weighted stage time-frequency feature, and the weighted global time-frequency feature to obtain a fusion feature vector. The classification module is used to process the fusion feature vector to obtain an atrial fibrillation detection result including two categories of atrial fibrillation and non-atrial fibrillation.

[0093] In the embodiment of the present application, the ballistocardiogram signal uploaded by the pressure sensor built in the bedding is acquired, the effective signal segment in the ballistocardiogram signal is determined according to the periodic characteristics of the ballistocardiogram signal, the time-frequency feature of the effective signal segment is acquired, and the time-frequency feature is input into the preset atrial fibrillation detection model for detection to obtain an atrial fibrillation detection result. In the embodiment of the present application, the atrial fibrillation detection result is obtained by using the preset atrial fibrillation detection model, which can realize automatic atrial fibrillation detection and improve the efficiency of atrial fibrillation detection.

[0094] Further, the embodiment of the present application also refines step S330, specifically, the working process of the cross-scale attention module, the feature fusion module and the classification module is refined, including:

[0095] S3301, calling the cross-scale attention module to determine the first allocation weight of the local time-frequency feature, the second allocation weight of the stage time-frequency feature and the third allocation weight of the global time-frequency feature based on the interaction feature matrix, and obtaining the weighted local time-frequency feature, the weighted stage time-frequency feature and the weighted global time-frequency feature; wherein the interaction feature matrix is generated by tensor product operation on the local time-frequency feature, the stage time-frequency feature and the global time-frequency feature.

[0096] Wherein, the interaction feature matrix can be understood as a kind of matrix generated by tensor product operation on the local time-frequency feature, the stage time-frequency feature and the global time-frequency feature, which is used to determine the allocation weight of each time-frequency feature, and can reflect the interaction relationship between different scale time-frequency features.

[0097] The first allocation weight can be understood as a kind of weight of the local time-frequency feature determined by the cross-scale attention module based on the interaction feature matrix, which is used to determine the importance of the local time-frequency feature in subsequent processing.

[0098] The second allocation weight can be understood as a kind of weight of the stage time-frequency feature determined by the cross-scale attention module based on the interaction feature matrix, which is used to determine the importance of the stage time-frequency feature in subsequent processing.

[0099] The third allocation weight can be understood as a kind of weight of the global time-frequency feature determined by the cross-scale attention module based on the interaction feature matrix, which is used to determine the importance of the global time-frequency feature in subsequent processing.

[0100] Specifically, the local time-frequency feature, the stage time-frequency feature and the global time-frequency feature output by the preset feature extraction module are input into the cross-scale attention module, the cross-scale attention module uses the interaction feature matrix to assign the first allocation weight to the local time-frequency feature, the second allocation weight to the stage time-frequency feature and the third allocation weight to the global time-frequency feature, and outputs the weighted local time-frequency feature, the weighted stage time-frequency feature and the weighted global time-frequency feature, wherein the interaction feature matrix is generated by tensor product operation on the local time-frequency feature, the stage time-frequency feature and the global time-frequency feature.

[0101] S3302, calling the feature fusion module to perform splicing processing on the weighted local time-frequency feature, the weighted stage time-frequency feature and the weighted global time-frequency feature, and obtaining the fusion feature vector.

[0102] The fusion feature vector can be understood as a vector obtained by splicing processing of the weighted local time-frequency feature, the weighted stage time-frequency feature and the weighted global time-frequency feature by the feature fusion module, and the fusion feature vector comprehensively includes information of different scale time-frequency features, and is used as an input vector of the classification module and is used for atrial fibrillation detection.

[0103] Specifically, the weighted local time-frequency feature, the weighted stage time-frequency feature and the weighted global time-frequency feature are input into the feature fusion module, and the feature fusion module is used for splicing processing of the weighted local time-frequency feature, the weighted stage time-frequency feature and the weighted global time-frequency feature, to obtain the fusion feature vector.

[0104] S3303, calling the classification module to process the fusion feature vector to obtain an atrial fibrillation detection result including two types of atrial fibrillation and non-atrial fibrillation.

[0105] Specifically, the fusion feature vector is input into the classification module as an input of the classification module, the classification module processes and analyzes the fusion feature vector, and outputs an atrial fibrillation detection result, which can include two types of results of atrial fibrillation and non-atrial fibrillation.

[0106] The embodiment of the application can determine the distribution weight of different scale time-frequency features by the cross-scale attention module, highlight the influence of important features on the atrial fibrillation detection result, and improve the accuracy of atrial fibrillation detection; the features of different scales are fused by the feature fusion module, so that the classification module can more accurately determine whether it is atrial fibrillation, and the accuracy of atrial fibrillation detection is further improved.

[0107] Embodiment four

[0108] Figure 4 Another atrial fibrillation detection method flowchart provided by the third embodiment of the application; Figure 5 A schematic diagram of a heart alluvial chart signal filtered by a band-pass filter provided by the third embodiment of the application; Figure 6 Another atrial fibrillation detection device schematic diagram provided by the third embodiment of the application. The embodiment of the application is an optimization of the above-mentioned embodiments, and specifically, the specific steps of the pre-processing operation of the heart alluvial chart signal and another atrial fibrillation detection device are supplemented.

[0109] As Figure 4The other atrial fibrillation detection method comprises the following steps: first, when a human subject, i.e. a to-be-detected object, lies on a mattress, a heart impulse cardiogram (BCG) signal generated by heart pumping can be captured in real time by a pressure sensor built in the mattress, the pressure sensor covers the thoracic spine heart projection area longitudinally to collect pressure data generated by the to-be-detected object on the mattress, the pressure sensor converts the pressure data into an electrical signal, and the electrical signal is the BCG signal; the BCG signal is uploaded to a local gateway cloud server through a wireless module transmission. Figure 5 As shown in the formula, it is a BCG signal after band-pass filtering. The quality of the respiratory signal is evaluated, and after the evaluation, the effective signal segment with better quality will be automatically selected to obtain an effective time-frequency signal after short-time Fourier transform. The time-frequency diagram is input into a neural network model, and inference is started immediately to obtain an atrial fibrillation (AF) probability. When the probability is greater than or equal to a probability threshold, the atrial fibrillation detection result is determined to be atrial fibrillation, and when the probability is less than the probability threshold, the atrial fibrillation detection result is determined to be non-atrial fibrillation. The atrial fibrillation detection result can be pushed to a mobile client in real time. The mobile terminal interface can dynamically display the BCG waveform and the like. The method can realize AF non-invasive screening with high accuracy, break through the bottleneck of traditional electrocardiogram monitoring, and reduce the risk of stroke in the aging population.

[0110] The neural network model at least comprises a preset feature extraction model, a cross-scale attention module, a feature fusion module and a classification module; the preset feature extraction model comprises a local time-frequency feature extraction module, a stage time-frequency feature extraction module and a global time-frequency feature extraction module, the local time-frequency feature extraction module, the stage time-frequency feature extraction module and the global time-frequency feature extraction module each comprise at least two convolution layers, and each convolution layer comprises different numbers of convolution kernels, for example, 36 convolution kernels and 64 convolution kernels.

[0111] It is worth mentioning that in the data preprocessing stage, the design of the band-pass filter aims to effectively suppress high-frequency interference (such as power frequency noise and electromyographic interference) and ultra-low frequency baseline drift, thereby highlighting the periodic BCG signal changes caused by respiration. According to the periodic characteristics of the respiratory signal, further select the periodicity significant and waveform stable segments from the filtered signal as the basis for subsequent analysis. The core of this selection process is to objectively evaluate the signal quality, that is, to judge the degree of fit between the patient and the mattress. Poor fit can introduce motion artifacts or cause signal attenuation, thereby affecting the recognizability of the respiratory signal. The autocorrelation analysis method is used to quantify the periodicity of the respiratory signal, by calculating the autocorrelation function of the signal and evaluating its periodicity strength by finding the height of the secondary peak. The obvious periodic peak presented in the autocorrelation sequence is a direct manifestation of the stability of the respiratory rhythm; on the contrary, if the autocorrelation function decays rapidly or the peak is chaotic, it indicates that the signal quality of this segment is poor. After completing the autocorrelation evaluation of all signal segments, each signal segment is scored according to the periodicity strength, and sorted from high to low according to the score. Finally, the top five signal segments with the best quality are selected for further filtering and feature extraction. These high-quality signal segments will help improve the accuracy and reliability of subsequent processing.

[0112] Previous BCG-based atrial fibrillation (AF) detection research mainly focuses on time domain features, but the time series representation of BCG signals faces the problems of low information density and waveform diversity. In order to overcome these problems, instead of directly using time series as input for feature extraction and classification model, short-time Fourier transform (STFT) is used to convert the effective signal segment into time-frequency signal, which retains the frequency information and provides the time index of the corresponding component, providing high-density time and spectral information for model analysis.

[0113] The effective time-frequency diagram obtained by short-time Fourier transform will be used as the input of the neural network model to obtain the time-frequency features of the ballistocardiogram signal.

[0114] In the preset feature extraction model, three different scale convolution branches are used to extract local time-frequency features, stage time-frequency features and global time-frequency features by using 5*5, 10*10 and 20*20 convolution kernels respectively. This design aims to capture multi-granularity features in the BCG signal time-frequency graph. The 5*5 convolution kernel is good at extracting local detail features, such as the tiny fluctuations and edge information in the AF signal; the 10*10 convolution kernel can capture medium-range patterns and identify the periodic and stage features of the AF attack; and the 20*20 convolution kernel is responsible for extracting the difference information between different global periods and grasping the macro structure of the entire time-frequency graph. The multi-scale parallel processing method gradually understands the signal features from local to global, which is suitable for the detection task of AF, a complex arrhythmia signal. Studies have shown that multi-scale feature extraction can effectively improve the recognition accuracy of AF features, reduce missed and false detections. In addition, the design of convolution branches of different scales also enhances the robustness of the model to noise and interference in the BCG signal, making the AF detection more stable and reliable.

[0115] Each convolution branch contains two convolution layers with 36 and 64 convolution kernels respectively. This hierarchical structure design follows the progressive principle of feature extraction. The first layer of 36 convolution kernels is responsible for extracting basic features such as signal edges, textures and other simple features. These features correspond to the basic components of the AF signal. As the signal passes through the second layer of 64 convolution kernels, the model can combine these basic features into more complex patterns and identify complex features such as the unique rhythm features of AF. This feature extraction process from simple to complex is similar to the way the human cognitive system understands signals, first recognizing basic elements and then understanding their combined meaning. In the AF detection task, this structure can effectively capture the subtle changes in the BCG signal before and after the AF attack, improving the sensitivity of the detection. Increasing the depth and width of the convolution layer can improve the feature expression ability of the preset feature extraction model, but it will also increase the computational complexity and the risk of overfitting. Therefore, the embodiment of the present application adopts a design of two convolution layers with gradually increasing channel numbers, which maintains the lightweight of the model while ensuring sufficient feature extraction capability, suitable for the deployment requirements of real-time AF detection.

[0116] Further, in each convolution layer, the Rectified Linear Unit (ReLU) activation function is used to set negative features to zero and retain valid positive values, enhancing the sensitivity of the model to AF signal mutation features; batch normalization optimization technology is used to standardize the feature distribution of each layer input, accelerate training convergence and reduce noise interference on features, providing clean features for the subsequent attention module.

[0117] Further, the neural network model further comprises a cross-scale attention module, a feature fusion module, and a classification module.

[0118] In the cross-scale attention module, high-order tensor operations are used to replace traditional matrix operations to model the nonlinear correlation between multi-scale features. Specifically, the cross-scale attention module regards the feature tensors output by the three branches in the preset feature extraction model as high-order tensors, and generates a multi-scale interaction feature matrix through tensor product operation. This design breaks through the limitation of traditional attention mechanisms that can only process two-dimensional features, and is more consistent with the complex dependence relationship between local details and global patterns in AF signals. In the multi-scale attention calculation, the cross-scale attention module introduces a dynamic weight distribution mechanism to adjust the contribution of different scale features based on the multi-scale interaction feature matrix. For example, when detecting the high-frequency oscillation characteristics specific to AF, the local detail feature weight of the 5x5 branch is automatically increased; when identifying the sustained rhythm abnormalities of atrial fibrillation, the global context feature weight of the 20x20 branch is enhanced. This dynamic adjustment simulates the selective attention mechanism in the human visual system, enabling the neural network model to focus on the key time-frequency patterns in AF signals, significantly improving the robustness to noise interference.

[0119] The feature fusion module is used to integrate the weighted feature tensors output by the three branches of the cross-scale attention module, solving the information redundancy problem caused by traditional concatenation operations. The specific process includes: first, performing global average pooling on the output features of each branch output by the cross-scale attention module to generate a one-dimensional channel description vector; second, learning the inter-channel dependency through two fully connected layers; and finally, generating a one-dimensional channel weight vector. In the feature fusion module, the one-dimensional channel weight vector after fusion can be added to the feature tensors output by the three branches in the feature extraction model, alleviating gradient vanishing and ensuring that the unique features of each branch are not lost.

[0120] The classification module adopts a lightweight fully connected structure, including two hidden layers, to maintain detection efficiency while enhancing discrimination ability. The input is the one-dimensional channel weight vector output by the feature fusion module, the first layer (128 neurons) hidden layer uses the ReLU activation function to extract high-order abstract features, and the second layer (64 neurons) hidden layer randomly discards 50% of the neurons, forcing the network to learn redundant features to prevent overfitting. The classification module also includes an output layer: using the Sigmoid function to generate an atrial fibrillation probability value (0-1), which has smooth gradient characteristics suitable for binary classification tasks, and the probability output can intuitively reflect the confidence of AF attack. The random inactivation of this module makes the model more robust to noise in BCG signals. In the classification module, the output layer can be combined with the cross-entropy loss function to accelerate convergence and improve small sample classification ability. In clinical applications, the probability threshold can be dynamically adjusted to balance the risk of missed diagnosis and misdiagnosis.

[0121] As shown in Figure 6 , a kind of atrial fibrillation detection device is shown, it includes: mattress, power supply, monitoring APP and cloud server.The object to be detected lies on mattress, BCG signal is collected by pressure sensor built-in mattress, BCG signal is uploaded to cloud server, cloud server includes neural network model, responsible for output atrial fibrillation detection search.Atrial fibrillation detection result can be sent to monitoring APP, and atrial fibrillation detection result is shown using monitoring APP.

[0122] Example five

[0123] Figure 7 Another structure schematic view of atrial fibrillation detection device provided for the embodiment five of the present application is shown.As shown in Figure 7 , the device includes: first signal acquisition module 410, second signal acquisition module 420 and result acquisition module 430.

[0124] First signal acquisition module 410 is used to acquire ballistocardiogram signal uploaded by pressure sensor built-in bedding;

[0125] Second signal acquisition module 420 is used to determine effective signal segment in ballistocardiogram signal according to periodicity feature of ballistocardiogram signal;

[0126] Result acquisition module 430 is used to acquire time-frequency feature of effective signal segment, and generate atrial fibrillation detection result according to time-frequency feature.

[0127] Optionally, second signal acquisition module 420 includes: signal division unit, used to divide ballistocardiogram signal into at least two signal segments according to preset fixed interval;Sequence acquisition unit is used to obtain the autocorrelation sequence corresponding to each signal segment by autocorrelation analysis method;Periodicity intensity acquisition unit is used to obtain the highest peak value in each autocorrelation sequence and main peak value, and the ratio of highest peak value and main peak value is used as the periodicity intensity corresponding to each signal segment;Signal extraction unit is used to extract the signal segment corresponding to periodicity intensity greater than preset intensity threshold as effective signal segment.

[0128] Optionally, result acquisition module 430 includes: time-frequency signal acquisition unit, used to perform short-time Fourier transform on effective signal segment to obtain effective time-frequency signal of effective signal segment;Characteristic extraction unit is used to input effective time-frequency signal into preset characteristic extraction model to extract feature, and obtain time-frequency feature.

[0129] Optionally, the feature extraction unit comprises: a local feature extraction subunit configured to call a local time-frequency feature extraction module to perform local time-frequency feature extraction on the effective time-frequency signal to obtain local time-frequency features; a stage feature extraction subunit configured to call a stage time-frequency feature extraction module to perform stage time-frequency feature extraction on the effective time-frequency signal to obtain stage time-frequency features; and a global feature extraction subunit configured to call a global time-frequency feature extraction module to perform global time-frequency feature extraction on the effective time-frequency signal to obtain global time-frequency features.

[0130] Further, the local time-frequency feature extraction module, the stage time-frequency feature extraction module and the global time-frequency feature extraction module in the preset feature extraction model in the embodiment of the application each comprise at least two convolution layers, and each convolution layer comprises different numbers of convolution kernels.

[0131] Optionally, the result acquisition module 430 comprises: a result acquisition unit configured to input the time-frequency features into a preset atrial fibrillation detection model for detection to obtain an atrial fibrillation detection result; wherein the preset atrial fibrillation detection model comprises a cross-scale attention module, a feature fusion module and a classification module.

[0132] The result acquisition unit comprises: a weight distribution subunit configured to call the cross-scale attention module to determine a first distribution weight of the local time-frequency features, a second distribution weight of the stage time-frequency features and a third distribution weight of the global time-frequency features based on an interaction feature matrix, and obtain weighted local time-frequency features, weighted stage time-frequency features and weighted global time-frequency features; wherein the interaction feature matrix is generated by performing tensor product operation on the local time-frequency features, the stage time-frequency features and the global time-frequency features; a feature splicing subunit configured to call the feature fusion module to perform splicing processing on the weighted local time-frequency features, the weighted stage time-frequency features and the weighted global time-frequency features to obtain a fusion feature vector; and a result classification subunit configured to call the classification module to process the fusion feature vector to obtain an atrial fibrillation detection result comprising two categories of atrial fibrillation and non-atrial fibrillation.

[0133] The atrial fibrillation detection device provided in the embodiment of the application can execute the atrial fibrillation detection method provided in any embodiment of the application, and has the beneficial effects corresponding to the execution method.

[0134] Embodiment six

[0135] The embodiment five of the application provides an electronic device for executing an atrial fibrillation detection method, a computer readable storage medium and a computer program product.

[0136] Figure 8A structural diagram of an electronic device that can be used to implement the atrial fibrillation detection method of any of the present embodiments is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices such as personal digital processors, cellular telephones, smart phones, wearable devices (e.g., headsets, glasses, watches, etc.), and other similar computing devices. The components shown in the present embodiments, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present embodiments described and / or claimed in this document.

[0137] As shown in Figure 8 The electronic device includes at least one processor 11, and a memory, such as a Read-Only Memory (ROM) 12, a Random Access Memory (RAM) 13, etc., connected to the at least one processor 11 in communication, where the memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer programs stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. In the RAM 13, various programs and data required for device operation can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An Input / Output (I / O) interface 15 is also connected to the bus 14.

[0138] Various components in the electronic device are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc., an output unit 17, such as various types of displays, a speaker, etc., a storage unit 18, such as a magnetic disk, an optical disk, etc., and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0139] The processor 11 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit, a graphics processing unit, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, a digital signal processor, and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as the atrial fibrillation detection method.

[0140] In some embodiments, the atrial fibrillation detection method can be implemented as a computer program tangibly embodied in a computer readable storage medium, e.g., storage unit 18. In some embodiments, parts or all of the computer program can be loaded and / or installed onto the electronic device via, e.g., ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the atrial fibrillation detection method can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the atrial fibrillation detection method by any other suitable means, e.g., with the aid of firmware.

[0141] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, specially designed application specific integrated circuits, application specific standard products, chips, microprocessors, computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input system, and at least one output system.

[0142] Computer programs implementing methods of embodiments of the application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing system to produce a machine, such that the computer program, when executed, implements the functions / acts specified in the flowcharts and / or block diagrams. The computer program can be executed entirely on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0143] In the context of embodiments of the present application, a computer- readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, system, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a RAM, a ROM, an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0144] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display system (e.g., a cathode ray tube or a liquid crystal display monitor) for displaying information to the user and a keyboard and a pointing system (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of systems can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0145] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network, a wide area network, a blockchain network, and the Internet.

[0146] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and virtual private server service.

[0147] It should be understood that the steps shown above can be reordered, added, or deleted using various forms of programming. For example, the steps described in the present application can be performed in parallel, in series, or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, and the present application is not limited herein.

[0148] The above detailed description does not constitute a limitation on the protection scope of the embodiments of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting atrial fibrillation, characterized in that, The method includes: Acquire cardiac impaction signals uploaded by the pressure sensor built into the bedding; The effective signal segments within the cardiac impactogram signal are determined based on the periodic characteristics of the cardiac impactogram signal; The time-frequency characteristics of the effective signal segment are obtained, and the atrial fibrillation detection result is generated based on the time-frequency characteristics.

2. The method according to claim 1, characterized in that, The step of determining the effective signal segment within the cardiac impactogram signal based on the periodic characteristics of the cardiac impactogram signal includes: The cardiac impaction signal is divided into at least two signal segments according to a preset fixed interval; Autocorrelation analysis was used to obtain the autocorrelation sequences corresponding to each of the signal segments. Obtain the highest peak value and the main peak value in each autocorrelation sequence, and use the ratio of the highest peak value to the main peak value as the periodic intensity corresponding to each signal segment; Signal segments with periodic intensities greater than a preset intensity threshold are considered as valid signal segments.

3. The method according to claim 1, characterized in that, The acquisition of the time-frequency characteristics of the effective signal segment includes: The effective time-frequency signal of the effective signal segment is obtained by performing a short-time Fourier transform on the effective signal segment. The effective time-frequency signal is input into a preset feature extraction model for feature extraction to obtain the time-frequency features.

4. The method according to claim 3, characterized in that, The preset feature extraction model includes at least one of the following: a local time-frequency feature extraction module, a stage time-frequency feature extraction module, and a global time-frequency feature extraction module. The step of inputting the effective time-frequency signal into the preset feature extraction model for feature extraction to obtain the time-frequency features includes: The local time-frequency feature extraction module is invoked to extract local time-frequency features from the effective time-frequency signal, thereby obtaining local time-frequency features; The stage time-frequency feature extraction module is invoked to extract stage time-frequency features from the effective time-frequency signal to obtain stage time-frequency features; The global time-frequency feature extraction module is invoked to extract global time-frequency features from the effective time-frequency signal, thereby obtaining global time-frequency features.

5. The method according to claim 4, characterized in that, The local time-frequency feature extraction module, the stage time-frequency feature extraction module, and the global time-frequency feature extraction module in the preset feature extraction model each include at least two convolutional layers, and each convolutional layer contains a different number of convolutional kernels.

6. The method according to claim 1, characterized in that, The time-frequency features include: local time-frequency features, stage time-frequency features, and global time-frequency features. Generating atrial fibrillation detection results based on the time-frequency features includes: A preset atrial fibrillation detection model is invoked to detect the time-frequency features, thereby obtaining the atrial fibrillation detection result. The preset atrial fibrillation detection model includes a cross-scale attention module, a feature fusion module, and a classification module, comprising: The cross-scale attention module is invoked to determine the first allocation weight of the local time-frequency feature, the second allocation weight of the stage time-frequency feature, and the third allocation weight of the global time-frequency feature based on the interaction feature matrix, and to obtain the weighted local time-frequency feature, the weighted stage time-frequency feature, and the weighted global time-frequency feature; wherein, the interaction feature matrix is ​​generated by performing tensor product operation on the local time-frequency feature, the stage time-frequency feature, and the global time-frequency feature; The feature fusion module is invoked to concatenate the weighted local time-frequency features, the weighted stage time-frequency features, and the weighted global time-frequency features to obtain a fused feature vector; The classification module is invoked to process the fused feature vector to obtain atrial fibrillation detection results including both atrial fibrillation and non-atrial fibrillation categories.

7. An atrial fibrillation detection device, characterized in that, The device includes: The first signal acquisition module is used to acquire the cardiac impact signal uploaded by the pressure sensor built into the bedding. The second signal acquisition module is used to determine the effective signal segments within the cardiac impact signal based on the periodic characteristics of the cardiac impact signal; The result acquisition module is used to acquire the time-frequency characteristics of the effective signal segment and generate atrial fibrillation detection results based on the time-frequency characteristics.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the atrial fibrillation detection method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the atrial fibrillation detection method according to any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the atrial fibrillation detection method according to any one of claims 1-6.

Images