CN118490184A

CN118490184A - Atrial fibrillation measurement method and device based on PPG waveform signals and pressure pulse signals

Info

Publication number: CN118490184A
Application number: CN202410952766.3A
Authority: CN
Inventors: 张少林; 段光杰; 赵海涛; 王利明
Original assignee: Shenzhen Jiemeirui Technology Co ltd
Current assignee: Shenzhen Jiemeirui Technology Co ltd
Priority date: 2024-07-16
Filing date: 2024-07-16
Publication date: 2024-08-16
Anticipated expiration: 2044-07-16
Also published as: CN118490184B

Abstract

The application relates to the technical field of deep learning, and discloses an atrial fibrillation measurement method and device based on a PPG waveform signal and a pressure pulse signal. The method comprises the following steps: respectively acquiring an initial PPG waveform signal and an initial pressure pulse signal of a tested object; baseline drift removal and high-frequency noise filtering are carried out, and a target PPG waveform signal and a target pressure pulse signal are obtained; performing feature extraction to obtain a first feature parameter set of the PPG waveform signal and a second feature parameter set of the pressure pulse signal; performing feature mapping and vector conversion to obtain a corresponding first feature mapping vector and a corresponding second feature mapping vector; performing feature combination and main component dimension reduction to obtain a fusion feature mapping vector; the atrial fibrillation state is judged through the double-layer stacked support vector machine model, and the atrial fibrillation state judgment result is obtained.

Description

Atrial fibrillation measurement method and device based on PPG waveform signals and pressure pulse signals

Technical Field

The application relates to the technical field of deep learning, in particular to an atrial fibrillation measurement method and device based on a PPG waveform signal and a pressure pulse signal.

Background

Atrial fibrillation is becoming a common arrhythmia condition and is becoming more and more interesting. Early detection and accurate monitoring of atrial fibrillation is of great importance to prevent further complications such as stroke and heart failure.

Traditional atrial fibrillation detection methods mainly rely on Electrocardiogram (ECG) monitoring, and although the accuracy is high, the traditional atrial fibrillation detection methods are difficult to widely apply in daily life due to high equipment cost, complex operation and large interference to life of patients. Therefore, it is important to find a non-invasive, portable and accurate atrial fibrillation detection method. PPG (photoplethysmography) and pressure pulse signals are technologies of great interest in recent years in non-invasive physiological signal monitoring. These two techniques can reflect the state of the cardiovascular system by detecting changes in blood volume and pulse waveforms. However, there are limitations to using PPG or pressure pulse signals alone for atrial fibrillation detection. For example, PPG signals are susceptible to motion artifacts and ambient light interference, resulting in reduced detection accuracy; the pressure pulse signal may be affected by the sensor position and contact pressure, resulting in signal instability.

Disclosure of Invention

The application provides an atrial fibrillation measurement method and device based on PPG waveform signals and pressure pulse signals, according to the method, fusion analysis is carried out on the PPG waveform signal and the pressure pulse signal, so that the accuracy of atrial fibrillation measurement is improved.

In a first aspect, the present application provides an atrial fibrillation measurement method based on a PPG waveform signal and a pressure pulse signal, the atrial fibrillation measurement method based on the PPG waveform signal and the pressure pulse signal comprising:

Synchronous data acquisition is carried out on a tested object by using a PPG sensor and a pressure sensor respectively to obtain an initial PPG waveform signal and an initial pressure pulse signal;

baseline drift removal and high-frequency noise filtering are respectively carried out on the initial PPG waveform signal and the initial pressure pulse signal, so as to obtain a target PPG waveform signal and a target pressure pulse signal;

Performing time domain feature extraction and frequency domain feature extraction on the target PPG waveform signal to obtain a first feature parameter set of the PPG waveform signal, and performing time domain feature extraction on the target pressure pulse signal to obtain a second feature parameter set of the pressure pulse signal;

Performing feature mapping and vector conversion on the first feature parameter set and the second feature parameter set respectively to obtain a corresponding first feature mapping vector and a corresponding second feature mapping vector;

performing feature combination and principal component dimension reduction on the first feature mapping vector and the second feature mapping vector to obtain a fusion feature mapping vector;

And inputting the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector into a preset double-layer stacked support vector machine model to judge the atrial fibrillation state, so as to obtain an atrial fibrillation state judging result.

In a second aspect, the present application provides an atrial fibrillation measurement device based on a PPG waveform signal and a pressure pulse signal, the atrial fibrillation measurement device based on a PPG waveform signal and a pressure pulse signal comprising:

The acquisition module is used for synchronously acquiring data of a tested object by using the PPG sensor and the pressure sensor respectively to obtain an initial PPG waveform signal and an initial pressure pulse signal;

The removing module is used for respectively carrying out baseline drift removal and high-frequency noise filtering on the initial PPG waveform signal and the initial pressure pulse signal to obtain a target PPG waveform signal and a target pressure pulse signal;

The extraction module is used for carrying out time domain feature extraction and frequency domain feature extraction on the target PPG waveform signal to obtain a first feature parameter set of the PPG waveform signal, and carrying out time domain feature extraction on the target pressure pulse signal to obtain a second feature parameter set of the pressure pulse signal;

the mapping module is used for carrying out feature mapping and vector conversion on the first feature parameter set and the second feature parameter set respectively to obtain a corresponding first feature mapping vector and a corresponding second feature mapping vector;

the combination module is used for carrying out feature combination and main component dimension reduction on the first feature mapping vector and the second feature mapping vector to obtain a fusion feature mapping vector;

And the judging module is used for inputting the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector into a preset double-layer stacked support vector machine model to judge the atrial fibrillation state, so as to obtain an atrial fibrillation state judging result.

A third aspect of the present application provides a computer apparatus comprising: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the computer device to perform the atrial fibrillation measurement method described above based on the PPG waveform signal and the pressure pulse signal.

A fourth aspect of the application provides a computer readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the above-described atrial fibrillation measurement method based on a PPG waveform signal and a pressure pulse signal.

According to the technical scheme provided by the application, the PPG sensor and the pressure sensor are used for synchronous data acquisition, so that the time alignment of two signals is ensured, and the data deviation caused by time asynchronization is avoided. The preprocessing steps of baseline drift removal and high-frequency noise filtering are adopted, so that interference factors in signals are effectively removed, the purity of the signals is improved, time domain features and frequency domain features of the PPG waveform signals are extracted, and detailed time domain feature extraction is performed on pressure pulse signals, so that the change of the cardiovascular system is ensured to be captured from different dimensions. The extracted characteristic parameters cover a plurality of aspects of signals, enrich characteristic sets and improve the identification capability of atrial fibrillation states. The feature parameters of the PPG waveform signal and the pressure pulse signal are combined to form a high-dimensional feature vector, and then the feature vector is subjected to dimension reduction processing by methods such as principal component analysis and the like, so that redundant information is effectively reduced, and the most distinguished feature is reserved. The construction of the fusion characteristic fully utilizes the complementarity of the two signals, enhances the detection capability of the atrial fibrillation state and improves the discrimination performance of the model. The atrial fibrillation state is judged by adopting a double-layer stacked support vector machine model, and the robustness and generalization capability of the model are improved through multi-level classification and weighted fusion. In the classifying process, the accuracy and the reliability of atrial fibrillation detection are further improved through confidence calculation and feature enhancement processing, the atrial fibrillation state of a tested object is monitored in real time, and the data acquisition and the processing are carried out through portable equipment, so that the real-time judgment and the monitoring of the atrial fibrillation state are realized.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained based on these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of an embodiment of an atrial fibrillation measurement method based on a PPG waveform signal and a pressure pulse signal according to an embodiment of the present application;

fig. 2 is a schematic diagram of an embodiment of an atrial fibrillation measurement device according to an embodiment of the present application based on PPG waveform signals and pressure pulse signals.

Detailed Description

The embodiment of the application provides an atrial fibrillation measurement method and device based on a PPG waveform signal and a pressure pulse signal. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For ease of understanding, a specific flow of an embodiment of the present application is described below, referring to fig. 1, and an embodiment of an atrial fibrillation measurement method based on a PPG waveform signal and a pressure pulse signal in the embodiment of the present application includes:

Step 101, respectively using a PPG sensor and a pressure sensor to acquire synchronous data of a tested object to obtain an initial PPG waveform signal and an initial pressure pulse signal;

It is to be understood that the implementation subject of the present application may be an atrial fibrillation measurement device based on PPG waveform signals and pressure pulse signals, and may also be a terminal or a server, which is not limited herein. The embodiment of the application is described by taking a server as an execution main body as an example.

Specifically, the sampling frequencies of the PPG sensor and the pressure sensor are set, the sampling frequency of the PPG sensor is set to be f1, the sampling frequency of the pressure sensor is set to be f2, and both f1 and f2 are preset sampling frequencies which are larger than 500 Hz. By means of the high sampling frequency setting, it is ensured that the acquired signals have sufficient time resolution, so that rapidly changing physiological signal characteristics are accurately captured. And acquiring a first PPG waveform signal of the finger of the tested object according to the set PPG sensor, wherein the acquisition duration is t1, and obtaining the first acquired PPG waveform signal. Meanwhile, according to the set pressure sensor, the wrist artery of the tested object is subjected to first pressure pulse signal acquisition, the acquisition duration is t1, and the pressure pulse signal acquired for the first time is obtained. The synchronous data acquisition ensures the spatio-temporal correspondence of the two signals. And aligning sampling starting time points of the PPG waveform signal acquired for the first time and the pressure pulse signal acquired for the first time to obtain the PPG waveform signal and the pressure pulse signal after the first synchronization. By time alignment, time errors caused by different starting time of different sensors are eliminated, and the synchronism of signals is ensured. And acquiring a second PPG waveform signal of the finger of the tested object according to the set PPG sensor, wherein the acquisition duration is t2, and obtaining a second acquired PPG waveform signal. Meanwhile, according to the set pressure sensor, the wrist artery of the tested object is subjected to second pressure pulse signal acquisition, the acquisition duration is t2, and the pressure pulse signal acquired for the second time is obtained. The data volume can be increased through multiple acquisitions, so that the accuracy and the robustness of signal analysis are improved. And aligning sampling starting time points of the PPG waveform signal acquired for the second time and the pressure pulse signal acquired for the second time to obtain a PPG waveform signal and a pressure pulse signal after the second synchronization. And repeating the acquisition and alignment steps until the acquisition of the PPG waveform signals and the pressure pulse signals for N times of preset acquisition times is completed, and obtaining the PPG waveform signals and the pressure pulse signals after N times of synchronization. And more stable and reliable synchronous signal data are obtained through repeated acquisition and alignment for a plurality of times. And performing time sequence splicing on the PPG waveform signals after N times of synchronization to obtain spliced PPG waveform signals, and performing time sequence splicing on the pressure pulse signals after N times of synchronization to obtain spliced pressure pulse signals. The time series splice can combine the signals acquired for multiple times, thereby forming continuous long-time series signals, and facilitating subsequent feature extraction and analysis. And calculating the signal-to-noise ratio SNR1 of the spliced PPG waveform signals, and calculating the signal-to-noise ratio SNR2 of the spliced pressure pulse signals. The quality of the signal is evaluated by calculation of the signal-to-noise ratio, thereby determining the proportion of effective information in the signal. If SNR1 is smaller than the preset signal-to-noise threshold η1 or SNR2 is smaller than the preset signal-to-noise threshold η2, the positions of the PPG sensor and the pressure sensor need to be adjusted, and the above-mentioned acquisition and processing steps are re-executed. Adjusting the position of the sensor can optimize the acquisition condition of the signal, thereby improving the signal-to-noise ratio. repeating the above process until SNR1 is greater than or equal to η1 and SNR2 is greater than or equal to η2, and taking the spliced PPG waveform signal and the spliced pressure pulse signal obtained at this time as a final initial PPG waveform signal and an initial pressure pulse signal.

Step 102, baseline drift removal and high-frequency noise filtering are respectively carried out on an initial PPG waveform signal and an initial pressure pulse signal, so as to obtain a target PPG waveform signal and a target pressure pulse signal;

Specifically, the initial PPG waveform signal is subjected to a first median filtering, and the filtering window size is W1, so as to obtain a PPG waveform signal after the first filtering. And meanwhile, carrying out first median filtering on the initial pressure pulse signals, wherein the size of a filtering window is also W1, and obtaining the pressure pulse signals after the first filtering. The median filter can effectively remove impulse noise in the signal. And extracting the baseline drift component of the PPG waveform signal by a self-adaptive threshold method according to the PPG waveform signal after the first filtering to obtain the baseline drift component of the PPG waveform signal. Similarly, the baseline wander component of the pressure pulse signal is extracted by an adaptive threshold method according to the pressure pulse signal after the first filtering, so as to obtain the baseline wander component of the pressure pulse signal. Baseline wander is extracted to separate out low frequency components of the signal, which are often slow-changing signals due to respiration, body movement, etc. And performing primary spline interpolation on the baseline wander component of the PPG waveform signal to obtain a PPG baseline wander component after primary interpolation. And meanwhile, performing primary spline interpolation on the baseline wander component of the pressure pulse signal to obtain the pressure pulse baseline wander component after the primary interpolation. Spline interpolation helps to smooth the baseline drift component, making it more nearly realistic to the actual low frequency drift trend. Subtracting the first filtered PPG waveform signal from the first interpolated PPG baseline drift component to obtain a first baseline drift removed PPG waveform signal. Correspondingly, subtracting the pressure pulse signal after the first filtering from the pressure pulse baseline wander component after the first interpolation to obtain the pressure pulse signal after the first baseline wander removal. By the method, baseline wander components in the signal can be effectively eliminated, so that the signal returns to a state taking zero as a baseline. And performing wavelet transformation on the PPG waveform signal after the first baseline drift is removed, and performing wavelet transformation on the pressure pulse signal after the first baseline drift is removed at the same time to respectively obtain wavelet coefficients of the PPG waveform signal and the pressure pulse signal. Wavelet transformation is a multi-resolution analysis tool that can decompose signals onto different frequency scales, thereby facilitating the separation of high frequency noise components from the signals. and performing soft threshold processing on the wavelet coefficient of the PPG waveform signal to obtain a denoised PPG wavelet coefficient. Similarly, the wavelet coefficient of the pressure pulse signal is subjected to soft threshold processing to obtain the denoised pressure pulse wavelet coefficient. Soft thresholding is a commonly used denoising method that can effectively suppress high frequency noise while retaining as much as possible the active components of the original signal. And carrying out wavelet inverse transformation on the denoised PPG wavelet coefficient to obtain a PPG waveform signal after the first high-frequency noise removal. And meanwhile, carrying out wavelet inverse transformation on the denoised pressure pulse wavelet coefficient to obtain a pressure pulse signal after the first high-frequency noise removal. And restoring the denoised wavelet coefficient into a time domain signal through wavelet inverse transformation, thereby obtaining a signal with noise filtered. And performing a second median filtering on the PPG waveform signal after the first high-frequency noise removal, wherein the size of a filtering window is W2, and obtaining a PPG waveform signal after the second filtering. Correspondingly, the pressure pulse signal after the first high-frequency noise removal is subjected to a second median filtering, and the size of a filtering window is also W2, so that the pressure pulse signal after the second filtering is obtained. And removing residual impulse noise in the signal through second median filtering, so as to ensure the smoothness and cleanliness of the signal. And repeatedly executing the steps of baseline drift removal and high-frequency noise filtering until the baseline drift component and the high-frequency noise component of the PPG waveform signal are removed, and obtaining a target PPG waveform signal. accordingly, the baseline drift component and the high-frequency noise component of the pressure pulse signal are removed until the target pressure pulse signal is obtained. Through repeated iteration of the steps, the method ensures that all interference components in the signal are sufficiently removed, and a high-quality target signal is obtained.

Step 103, performing time domain feature extraction and frequency domain feature extraction on the target PPG waveform signal to obtain a first feature parameter set of the PPG waveform signal, and performing time domain feature extraction on the target pressure pulse signal to obtain a second feature parameter set of the pressure pulse signal;

Specifically, peak detection is performed on the target PPG waveform signal to obtain a peak point sequence of the PPG waveform signal, and a time interval between adjacent peak points is calculated to obtain a peak interval of the PPG waveform signal. And simultaneously, carrying out peak detection on the target pressure pulse signals to obtain a peak point sequence of the pressure pulse signals, and calculating the time interval between adjacent peak points to obtain the peak interval of the pressure pulse signals. Through peak detection, key points in the signal are identified. And calculating the average peak interval, the standard deviation of the peak interval and the variation coefficient of the peak interval of the PPG waveform signal according to the peak interval of the PPG waveform signal to obtain the time domain characteristic parameters of the PPG waveform signal. And calculating the average peak interval, the standard deviation of the peak interval and the variation coefficient of the peak interval of the pressure pulse signal according to the peak interval of the pressure pulse signal, so as to obtain the time domain characteristic parameters of the pressure pulse signal. The time domain characteristic parameters can reflect the periodicity and stability of the signals and are important indexes for measuring the functions of the cardiovascular system. Performing fast Fourier transform on the peak interval of the PPG waveform signal to obtain a frequency domain signal of the PPG waveform signal, and calculating fundamental wave frequency, fundamental wave energy, first harmonic frequency, first harmonic energy, second harmonic frequency and second harmonic energy of the PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal. The frequency domain characteristic parameters can reveal the energy distribution of the signal over different frequency components. And carrying out spectral entropy analysis on the target PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal. And meanwhile, performing spectral entropy analysis on the target pressure pulse signal to obtain time domain characteristic parameters of the pressure pulse signal. Spectral entropy is an important indicator of signal complexity and randomness, which can help identify unordered states and abnormal fluctuations in the signal. And carrying out Poincare mapping on the target PPG waveform signal to obtain a Poincare mapping diagram of the PPG waveform signal, and calculating the ratio of the major axis to the minor axis of the PPG waveform signal to obtain the time domain characteristic parameter of the PPG waveform signal. similarly, poincare mapping is carried out on the target pressure pulse signal, a Poincare mapping diagram of the pressure pulse signal is obtained, the ratio of the major axis to the minor axis of the pressure pulse signal is calculated, and the time domain characteristic parameter of the pressure pulse signal is obtained. The poincare map can provide a geometric distribution feature of the signal, helping to evaluate heart rate variability and rhythmic variation. And carrying out correlation dimension and Lyapunov exponent analysis on the PPG waveform signal according to the peak point sequence of the PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal. And similarly, carrying out association dimension and Lyapunov index analysis on the pressure pulse signals according to the peak point sequence of the pressure pulse signals to obtain time domain characteristic parameters of the pressure pulse signals. The correlation dimension and Lyapunov index are key indexes for measuring the dynamic complexity and the chaotic characteristic of the signal, and the nonlinear dynamics behavior of the system can be revealed. And combining the time domain characteristic parameters and the frequency domain characteristic parameters of the PPG waveform signals to obtain a first characteristic parameter set of the PPG waveform signals. And meanwhile, combining the time domain characteristic parameters of the pressure pulse signals obtained by extraction to obtain a second characteristic parameter set of the pressure pulse signals.

104, Performing feature mapping and vector conversion on the first feature parameter set and the second feature parameter set respectively to obtain a corresponding first feature mapping vector and a corresponding second feature mapping vector;

Specifically, constructing a Gaussian kernel matrix of the PPG waveform signal according to the first characteristic parameter set; meanwhile, a Gaussian kernel matrix of the pressure pulse signal is constructed according to the second characteristic parameter set. The Gaussian kernel matrix is a matrix based on a Gaussian kernel function, and the data of the original feature space is mapped to a new feature space, so that the structure and the internal relation of the data are better revealed. And carrying out feature decomposition on the Gaussian kernel matrix of the PPG waveform signal to obtain a first feature value and a first feature vector of the PPG waveform signal, and sequencing the first feature value according to the first feature value from large to small to obtain a sequenced first feature value sequence and a corresponding first feature vector sequence. And similarly, carrying out feature decomposition on the Gaussian kernel matrix of the pressure pulse signal to obtain a second feature value and a second feature vector of the pressure pulse signal, and sequencing from large to small according to the second feature value to obtain a sequenced second feature value sequence and a corresponding second feature vector sequence. Feature decomposition is a process of decomposing a matrix into a set of feature vectors and feature values that can be used to represent the main features and trends of the data. And cutting off the first eigenvalue sequence according to a preset first eigenvalue threshold value to obtain a cut-off PPG eigenvalue subsequence, and extracting eigenvectors corresponding to the cut-off PPG eigenvalue subsequence to obtain a PPG eigenvector subsequence. And similarly, according to a preset second eigenvalue threshold value, cutting off the second eigenvalue sequence to obtain a cut-off pressure pulse eigenvalue sub-sequence, and extracting eigenvectors corresponding to the cut-off pressure pulse eigenvalue sub-sequence to obtain a pressure pulse eigenvector sub-sequence. The truncation process is to remove features with smaller feature values and lower contribution to data, thereby simplifying the model and improving the calculation efficiency. Carrying out normalization processing on the PPG feature vector subsequence to obtain a normalized PPG feature vector subsequence; and simultaneously, carrying out normalization processing on the pressure pulse characteristic vector subsequence to obtain a normalized pressure pulse characteristic vector subsequence. The normalization processing can eliminate dimension differences among different features, so that the features are compared on the same scale, and the accuracy and consistency of feature mapping are improved. According to the normalized PPG feature vector subsequence, mapping the first feature parameter set to a high-dimensional space through a kernel function to obtain a kernel matrix of the first feature parameter set; and similarly, according to the normalized pressure pulse feature vector subsequence, mapping the second feature parameter set to a high-dimensional space through a kernel function to obtain a kernel matrix of the second feature parameter set. Nonlinear data is converted into a linearly separable form through kernel function mapping, so that subsequent feature processing and analysis are facilitated. Performing singular value decomposition on the nuclear matrix of the first characteristic parameter set to obtain a left singular matrix, a singular value matrix and a right singular matrix of the first characteristic parameter set; and similarly, performing singular value decomposition on the nuclear matrix of the second characteristic parameter set to obtain a left singular matrix, a singular value matrix and a right singular matrix of the second characteristic parameter set. Singular value decomposition is a process of decomposing a matrix into three sub-matrices, which represent the row space, column space, and singular value space of the original matrix, respectively, and can be used to extract the main feature components of the data. Performing dimension reduction processing on the left singular matrix of the first characteristic parameter set according to the singular value matrix of the first characteristic parameter set to obtain a left singular matrix of the first characteristic parameter set after dimension reduction; And similarly, performing dimension reduction processing on the left singular matrix of the second characteristic parameter set according to the singular value matrix of the second characteristic parameter set to obtain the left singular matrix of the second characteristic parameter set after dimension reduction. The dimension reduction processing is to reduce the dimension of the data, thereby reducing the computational complexity and the storage requirement, and simultaneously preserving the main characteristic information of the data. Performing linear transformation on the first characteristic parameter set through the left singular matrix of the first characteristic parameter set after dimension reduction to obtain a characteristic mapping matrix of the first characteristic parameter set; and similarly, linearly transforming the second characteristic parameter set through the left singular matrix of the second characteristic parameter set after the dimension reduction to obtain a characteristic mapping matrix of the second characteristic parameter set. Linear transformation is an operation of transforming data from one space to another, by which the data can be more easily analyzed and processed in the new space. Vectorizing the feature mapping matrix of the first feature parameter set to obtain a first feature mapping vector; and similarly, carrying out vectorization processing on the feature mapping matrix of the second feature parameter set to obtain a second feature mapping vector. The vectorization processing is to convert the data in the matrix form into the vector form, so that the subsequent machine learning algorithm is convenient to process and analyze.

Step 105, feature combination and principal component dimension reduction are carried out on the first feature mapping vector and the second feature mapping vector, and a fusion feature mapping vector is obtained;

Specifically, the first feature mapping vector and the second feature mapping vector are spliced to obtain a spliced feature mapping vector, and a covariance matrix of the spliced feature mapping vector is calculated. Covariance matrices are used to describe the linear relationship between the components in the feature map vector and their degree of variation. And carrying out feature decomposition on the covariance matrix to obtain a third feature value and a third feature vector of the covariance matrix. The eigenvalue decomposition may decompose the covariance matrix into eigenvalues and eigenvectors, which may be used to represent the dominant direction and magnitude of variation of the data. And sorting the third eigenvalues from large to small to obtain a sorted third eigenvalue sequence and a corresponding third eigenvector sequence. And cutting off the sequenced third characteristic value sequence to obtain a cut-off spliced characteristic value subsequence, and extracting a characteristic vector corresponding to the cut-off spliced characteristic value subsequence to obtain a spliced characteristic vector subsequence. The purpose of the truncation process is to remove eigenvalues that contribute less to the variation of the data, thereby simplifying the representation of the data while retaining the principal variation components. And (3) performing linear transformation on the spliced feature mapping vector through splicing the feature vector subsequences to obtain a transformed feature mapping vector. The linear transformation can project data of high dimensions into a space of low dimensions, making the data more representative and easy to analyze in the new space. And carrying out normalization processing and decentralization processing on the transformed feature mapping vector to obtain a fusion feature mapping vector. The normalization processing can eliminate scale differences among different features, so that the features are compared on the same scale; the decentralization processing is to eliminate the mean deviation of the data, so that the data is more concentrated at the origin, thereby improving the accuracy and stability of data analysis.

And 106, inputting the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector into a preset double-layer stacked support vector machine model to judge the atrial fibrillation state, and obtaining an atrial fibrillation state judging result.

Specifically, the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector are input into a preset double-layer stacking support vector machine model. The first layer support vector machine model of the two-layer stacked support vector machine model includes: a first support vector machine model, a second support vector machine model, a third support vector machine model, and a fourth support vector machine model, and the second layer support vector machine model comprises: a fifth support vector machine model, a sixth support vector machine model, a seventh support vector machine model, and an eighth support vector machine model. Classifying the first feature mapping vector through a first support vector machine model to obtain a first classification result, classifying the second feature mapping vector through a second support vector machine model to obtain a second classification result, and classifying the fusion feature mapping vector through a third support vector machine model to obtain a third classification result. And carrying out weighted fusion on the first classification result, the second classification result and the third classification result to obtain a first fusion classification result. The weight is set so as to fully consider the importance and the credibility of each classification result when the classification results are fused. And performing secondary classification on the first fusion classification result through a fourth support vector machine model to obtain a first secondary classification result. To ensure reliability of the classification results, a confidence level of the first secondary classification result is calculated and compared with a preset confidence threshold. If the confidence coefficient is larger than or equal to a preset confidence coefficient threshold value, outputting a first secondary classification result as a final atrial fibrillation state discrimination result; if the confidence level is less than the preset confidence threshold, further processing is required. And under the condition of insufficient confidence, carrying out feature enhancement processing on the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector to obtain an enhanced first feature mapping vector, an enhanced second feature mapping vector and an enhanced fusion feature mapping vector. The feature enhancement process may increase the accuracy of classification by increasing the dimension of the features or by some transformation to increase the distinguishing ability of the features. Classifying the enhanced first feature mapping vector through a fifth support vector machine model to obtain a fourth classification result, classifying the enhanced second feature mapping vector through a sixth support vector machine model to obtain a fifth classification result, and classifying the enhanced fusion feature mapping vector through a seventh support vector machine model to obtain a sixth classification result. And carrying out weighted fusion on the fourth classification result, the fifth classification result and the sixth classification result to obtain a second fusion classification result. And carrying out secondary classification on the second fusion classification result through an eighth support vector machine model according to the second fusion classification result to obtain a second secondary classification result. And calculating the confidence coefficient of the second classification result, and comparing the confidence coefficient with a preset confidence coefficient threshold value. If the confidence coefficient is larger than or equal to a preset confidence coefficient threshold value, outputting a second classification result as a final atrial fibrillation state discrimination result; if the confidence coefficient is still smaller than the preset confidence coefficient threshold value, the first secondary classification result is used as a final atrial fibrillation state discrimination result and output.

In the embodiment of the application, the PPG sensor and the pressure sensor are used for synchronous data acquisition, so that the time alignment of two signals is ensured, and the data deviation caused by time asynchronization is avoided. The preprocessing steps of baseline drift removal and high-frequency noise filtering are adopted, so that interference factors in signals are effectively removed, the purity of the signals is improved, time domain features and frequency domain features of the PPG waveform signals are extracted, and detailed time domain feature extraction is performed on pressure pulse signals, so that the change of the cardiovascular system is ensured to be captured from different dimensions. The extracted characteristic parameters cover a plurality of aspects of signals, enrich characteristic sets and improve the identification capability of atrial fibrillation states. The feature parameters of the PPG waveform signal and the pressure pulse signal are combined to form a high-dimensional feature vector, and then the feature vector is subjected to dimension reduction processing by methods such as principal component analysis and the like, so that redundant information is effectively reduced, and the most distinguished feature is reserved. The construction of the fusion characteristic fully utilizes the complementarity of the two signals, enhances the detection capability of the atrial fibrillation state and improves the discrimination performance of the model. The atrial fibrillation state is judged by adopting a double-layer stacked support vector machine model, and the robustness and generalization capability of the model are improved through multi-level classification and weighted fusion. In the classifying process, the accuracy and the reliability of atrial fibrillation detection are further improved through confidence calculation and feature enhancement processing, the atrial fibrillation state of a tested object is monitored in real time, and the data acquisition and the processing are carried out through portable equipment, so that the real-time judgment and the monitoring of the atrial fibrillation state are realized.

In a specific embodiment, the process of executing step 101 may specifically include the following steps:

s11: setting sampling frequencies of the PPG sensor and the pressure sensor, setting the sampling frequency of the PPG sensor to be f1, and setting the sampling frequency of the pressure sensor to be f2, wherein f1 and f2 are preset sampling frequencies which are more than 500 Hz;

S12: according to the set PPG sensor, carrying out first PPG waveform signal acquisition on the finger of the tested object, wherein the acquisition time length is t1, so as to obtain a first acquired PPG waveform signal, and meanwhile, according to the set pressure sensor, carrying out first pressure pulse signal acquisition on the wrist artery of the tested object, wherein the acquisition time length is t1, so as to obtain a first acquired pressure pulse signal;

s13: aligning sampling starting time points of the PPG waveform signal acquired for the first time and the pressure pulse signal acquired for the first time to obtain a PPG waveform signal and a pressure pulse signal after the first synchronization;

S14: according to the set PPG sensor, carrying out second PPG waveform signal acquisition on the finger of the tested object, wherein the acquisition time length is t2, so as to obtain a second acquired PPG waveform signal, and meanwhile, according to the set pressure sensor, carrying out second pressure pulse signal acquisition on the wrist artery of the tested object, and the acquisition time length is t2, so as to obtain a second acquired pressure pulse signal;

S15: aligning sampling starting time points of the PPG waveform signals acquired for the second time and the pressure pulse signals acquired for the second time to obtain PPG waveform signals and pressure pulse signals after the second synchronization;

S16: repeating the step S14 and the step S15 until the acquisition of the PPG waveform signal and the pressure pulse signal for N times of preset acquisition times is completed, and obtaining the PPG waveform signal and the pressure pulse signal after N times of synchronization;

S17: performing time sequence splicing on the PPG waveform signals after N times of synchronization to obtain spliced PPG waveform signals, and performing time sequence splicing on the pressure pulse signals after N times of synchronization to obtain spliced pressure pulse signals;

S18: calculating a signal-to-noise ratio SNR1 of the spliced PPG waveform signals, and calculating a signal-to-noise ratio SNR2 of the spliced pressure pulse signals;

S19: if SNR1 is smaller than a preset signal-to-noise ratio threshold value η1 or SNR2 is smaller than a preset signal-to-noise ratio threshold value η2, positions of the PPG sensor and the pressure sensor are adjusted, and steps S11 to S18 are re-executed until SNR1 is greater than or equal to η1 and SNR2 is greater than or equal to η2, and the spliced PPG waveform signal and the spliced pressure pulse signal obtained at this time are used as a final initial PPG waveform signal and an initial pressure pulse signal.

Specifically, the PPG sensor and the pressure sensor are set to ensure that their sampling frequencies are f1 and f2, respectively, and are both greater than 500Hz. The acquired signals are ensured to have sufficient time resolution to be able to capture subtle changes in heart activity. In this setup procedure, the formula f=1/T may be used, where f is the sampling frequency and T is the sampling period. By adjusting the value of T, both f1 and f2 are made greater than 500Hz. And acquiring a first PPG waveform signal of the finger of the tested object according to the set PPG sensor, wherein the acquisition duration is t1, and obtaining the first acquired PPG waveform signal. Meanwhile, according to the set pressure sensor, the wrist artery of the tested object is subjected to first pressure pulse signal acquisition, the acquisition duration is t1, and the pressure pulse signal acquired for the first time is obtained. In order to ensure the synchronism of the two signals, the sampling starting time points of the PPG waveform signal acquired for the first time and the pressure pulse signal acquired for the first time are aligned, so that the PPG waveform signal and the pressure pulse signal after the first synchronization are obtained. By setting a common time stamp, it is ensured that both sensors start sampling at the same point in time. And acquiring a second PPG waveform signal of the finger of the tested object according to the set PPG sensor, wherein the acquisition duration is t2, and obtaining a second acquired PPG waveform signal. Meanwhile, according to the set pressure sensor, the wrist artery of the tested object is subjected to second pressure pulse signal acquisition, the acquisition duration is also t2, and the pressure pulse signal acquired for the second time is obtained. And aligning sampling starting time points of the PPG waveform signal acquired for the second time and the pressure pulse signal acquired for the second time again to obtain a PPG waveform signal and a pressure pulse signal after the second synchronization. And repeating the acquisition and alignment processes until the acquisition of the preset N times of PPG waveform signals and pressure pulse signals is completed, and obtaining the N times of synchronized PPG waveform signals and pressure pulse signals. In order to better perform subsequent analysis, performing time sequence splicing on the PPG waveform signals after N times of synchronization to obtain spliced PPG waveform signals; And similarly, performing time series splicing on the N-time synchronized pressure pulse signals to obtain spliced pressure pulse signals. The signals collected each time are connected in time sequence to form a continuous signal sequence with long time. And calculating the signal-to-noise ratio SNR1 of the spliced PPG waveform signals, and calculating the signal-to-noise ratio SNR2 of the spliced pressure pulse signals. The formula of the signal-to-noise ratio is snr=10xlog 10 (p_signal/p_noise), where p_signal is the power of the signal and p_noise is the power of the noise. By this formula, the ratio of the useful component to the noise component in the signal can be measured to evaluate the signal quality. If SNR1 is smaller than a preset signal-to-noise threshold η1 or SNR2 is smaller than a preset signal-to-noise threshold η2, the positions of the PPG sensor and the pressure sensor need to be adjusted, and the above steps are re-performed. Adjusting the position of the sensors can be achieved by changing their contact position or pressure with the skin, thereby optimizing the signal acquisition conditions. Steps S11 to S18 are continued until SNR1 is greater than or equal to η1 and SNR2 is greater than or equal to η2, at which time the obtained spliced PPG waveform signal and spliced pressure pulse signal will be taken as final initial PPG waveform signal and initial pressure pulse signal. For example, it is assumed that sampling frequencies of the PPG sensor and the pressure sensor are set to f1=600 Hz and f2=700 Hz, respectively, and a time period per acquisition is set to t1=t2=10 seconds. During the first acquisition, the PPG waveform signal and the pressure pulse signal are obtained and time aligned. Also, in the second acquisition, the same steps are repeated and the initial time points of the signals of the two acquisitions are ensured to be aligned. And acquiring N times to obtain a plurality of synchronized signal sequences, and splicing the synchronized signal sequences to obtain a PPG waveform signal and a pressure pulse signal with long time sequences. The signal-to-noise ratios of these spliced signals are calculated assuming that the resulting SNR1 and SNR2 are 20dB and 18dB, respectively. If the preset signal-to-noise ratio thresholds eta 1 and eta 2 are 25dB and 20dB respectively, the position of the sensor is adjusted to improve the signal-to-noise ratio, and the acquisition and processing steps are repeatedly executed until the signal-to-noise ratio meets the requirements. Finally, the resulting stitched PPG waveform signal and pressure pulse signal will be used for further atrial fibrillation detection analysis.

In a specific embodiment, the process of executing step 102 may specifically include the following steps:

S21: performing first median filtering on the initial PPG waveform signal, wherein the size of a filtering window is W1, so as to obtain a PPG waveform signal after first filtering, and meanwhile, performing first median filtering on the initial pressure pulse signal, and the size of the filtering window is W1, so as to obtain a pressure pulse signal after first filtering;

S22: extracting a baseline drift component of the PPG waveform signal through a self-adaptive threshold method according to the PPG waveform signal after the first filtering to obtain the baseline drift component of the PPG waveform signal, and extracting the baseline drift component of the pressure pulse signal through the self-adaptive threshold method according to the pressure pulse signal after the first filtering to obtain the baseline drift component of the pressure pulse signal;

S23: performing primary spline interpolation on the baseline wander component of the PPG waveform signal to obtain a PPG baseline wander component after primary interpolation, and performing primary spline interpolation on the baseline wander component of the pressure pulse signal to obtain a pressure pulse baseline wander component after primary interpolation;

S24: subtracting the first filtered PPG waveform signal from the first interpolated PPG baseline wander component to obtain a first baseline wander removed PPG waveform signal, and subtracting the first filtered pressure pulse signal from the first interpolated pressure pulse baseline wander component to obtain a first baseline wander removed pressure pulse signal;

S25: performing wavelet transformation on the PPG waveform signal after the first baseline drift is removed, and performing wavelet transformation on the pressure pulse signal after the first baseline drift is removed to obtain wavelet coefficients of the PPG waveform signal and the pressure pulse signal respectively;

s26: performing soft threshold processing on the wavelet coefficient of the PPG waveform signal to obtain a denoised PPG wavelet coefficient, and performing soft threshold processing on the wavelet coefficient of the pressure pulse signal to obtain a denoised pressure pulse wavelet coefficient;

S27: performing wavelet inverse transformation on the denoised PPG wavelet coefficients to obtain a PPG waveform signal after the first high-frequency noise removal, and performing wavelet inverse transformation on the denoised pressure pulse wavelet coefficients to obtain a pressure pulse signal after the first high-frequency noise removal;

S28: performing a second median filtering on the PPG waveform signal after the first high-frequency noise removal, wherein the size of a filtering window is W2, so as to obtain a second filtered PPG waveform signal, and performing a second median filtering on the pressure pulse signal after the first high-frequency noise removal, wherein the size of the filtering window is W2, so as to obtain a second filtered pressure pulse signal;

S29: and repeating the steps S22 to S28 until the baseline drift component and the high-frequency noise component of the PPG waveform signal are removed to obtain a target PPG waveform signal, and until the baseline drift component and the high-frequency noise component of the pressure pulse signal are removed to obtain a target pressure pulse signal.

Specifically, an initial PPG waveform signal and an initial pressure pulse signal are initially processed. Median filtering is a commonly used denoising technique that can effectively remove impulse noise from a signal. And (3) by setting the size of the filtering window to be W1, sorting the data points in each window according to the size, and taking the intermediate value as the representative value of the data points in the window. And extracting the baseline drift component of the PPG waveform signal by a self-adaptive threshold method according to the PPG waveform signal after the first filtering to obtain the baseline drift component of the PPG waveform signal. Meanwhile, according to the pressure pulse signals after the first filtering, the baseline drift component of the pressure pulse signals is extracted through a self-adaptive threshold method, and the baseline drift component of the pressure pulse signals is obtained. Baseline wander is typically caused by low frequency factors such as respiration, body movement, etc., which can be effectively identified and extracted by adaptive thresholding. Performing primary spline interpolation on the baseline drift component of the PPG waveform signal to obtain a PPG baseline drift component after primary interpolation; and similarly, performing first spline interpolation on the baseline wander component of the pressure pulse signal to obtain a pressure pulse baseline wander component after the first interpolation. Spline interpolation is a smoothing technique, and baseline wander components can be fitted through interpolation functions, so that the baseline wander components are smoother and more continuous, and subsequent processing is facilitated. Subtracting the first filtered PPG waveform signal from the first interpolated PPG baseline drift component to obtain a first baseline drift removed PPG waveform signal; And similarly, subtracting the pressure pulse signal after the first filtering from the pressure pulse baseline wander component after the first interpolation to obtain the pressure pulse signal after the first baseline wander removal. Performing wavelet transformation on the PPG waveform signal after the first baseline drift is removed, and performing wavelet transformation on the pressure pulse signal after the first baseline drift is removed to obtain wavelet coefficients of the PPG waveform signal and the pressure pulse signal respectively. Wavelet transformation is a time-frequency analysis technique that can decompose signals onto different frequency scales, thereby extracting multi-scale features of the signals. Performing soft threshold processing on the wavelet coefficient of the PPG waveform signal to obtain a denoised PPG wavelet coefficient; And performing soft threshold processing on the wavelet coefficient of the pressure pulse signal to obtain a denoised pressure pulse wavelet coefficient. Soft thresholding is a denoising technique that reduces or even zero coefficients below a threshold by setting the threshold, thereby reducing noise components. Performing wavelet inverse transformation on the denoised PPG wavelet coefficient to obtain a PPG waveform signal after the first high-frequency noise removal; and performing wavelet inverse transformation on the denoised pressure pulse wavelet coefficient to obtain a pressure pulse signal after the first high-frequency noise removal. The inverse wavelet transform converts the processed wavelet coefficients back to a time domain signal such that high frequency noise components in the signal are removed. Performing a second median filtering on the PPG waveform signal after the first high-frequency noise removal, wherein the size of a filtering window is W2, and obtaining a PPG waveform signal after the second filtering; and similarly, performing a second median filtering on the pressure pulse signal after the first high-frequency noise removal, wherein the size of a filtering window is also W2, and obtaining the pressure pulse signal after the second filtering. The second median filtering may further remove residual noise from the signal, making the signal smoother and cleaner. Repeatedly executing the steps until the baseline drift component and the high-frequency noise component of the PPG waveform signal are removed, and obtaining a target PPG waveform signal; likewise, the baseline wander component and the high-frequency noise component of the pressure pulse signal are removed until the target pressure pulse signal is obtained. The process gradually extracts pure signals through repeated iterative filtering and denoising.

In a specific embodiment, the process of executing step 103 may specifically include the following steps:

S31: performing peak detection on the target PPG waveform signal to obtain a peak point sequence of the PPG waveform signal, and calculating the time interval between adjacent peak points to obtain a peak interval of the PPG waveform signal; performing peak detection on the target pressure pulse signals to obtain a peak point sequence of the pressure pulse signals, and calculating time intervals between adjacent peak points to obtain peak intervals of the pressure pulse signals;

S32: calculating the average peak interval, the standard deviation of the peak interval and the variation coefficient of the peak interval of the PPG waveform signal according to the peak interval of the PPG waveform signal to obtain the time domain characteristic parameter of the PPG waveform signal; calculating the average peak interval, the standard deviation of the peak interval and the variation coefficient of the peak interval of the pressure pulse signal according to the peak interval of the pressure pulse signal to obtain the time domain characteristic parameter of the pressure pulse signal;

s33: performing fast Fourier transform on the peak interval of the PPG waveform signal to obtain a frequency domain signal of the PPG waveform signal, and calculating fundamental wave frequency, fundamental wave energy, first harmonic frequency, first harmonic energy, second harmonic frequency and second harmonic energy of the PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal;

s34: performing spectral entropy analysis on the target PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal; performing spectral entropy analysis on the target pressure pulse signal to obtain a time domain characteristic parameter of the pressure pulse signal;

S35: carrying out Poincare mapping on the target PPG waveform signal to obtain a Poincare mapping diagram of the PPG waveform signal, and calculating the ratio of the major axis to the minor axis of the PPG waveform signal to obtain a time domain characteristic parameter of the PPG waveform signal; carrying out Poincare mapping on the target pressure pulse signals to obtain Poincare mapping diagrams of the pressure pulse signals, and calculating the ratio of the major axis to the minor axis of the pressure pulse signals to obtain time domain characteristic parameters of the pressure pulse signals;

s36: carrying out association dimension and Lyapunov exponent analysis on the PPG waveform signal according to the peak point sequence of the PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal; according to the peak point sequence of the pressure pulse signals, carrying out association dimension and Lyapunov index analysis on the pressure pulse signals to obtain time domain characteristic parameters of the pressure pulse signals;

S37: combining the time domain characteristic parameters and the frequency domain characteristic parameters of the PPG waveform signals extracted in the steps S32 to S36 to obtain a first characteristic parameter set of the PPG waveform signals, and combining the time domain characteristic parameters of the pressure pulse signals extracted in the steps S32, S34 to S36 to obtain a second characteristic parameter set of the pressure pulse signals.

Specifically, peak detection is performed on the target PPG waveform signal and the target pressure pulse signal. Peak detection is a common signal processing technique for identifying local maxima in a signal, which points generally correspond to the peaks of the heart beat. By calculating the time interval between adjacent peak points, the peak interval of the PPG waveform signal and the peak interval of the pressure pulse signal are obtained, and the intervals reflect the periodicity of the heart activity. And calculating the average peak interval, the standard deviation of the peak interval and the variation coefficient of the peak interval of the PPG waveform signal according to the peak interval of the PPG waveform signal to obtain the time domain characteristic parameters of the PPG waveform signal. The formula is as follows:

wherein, Represents the i-th peak interval, N represents the total peak logarithm,Representing the average peak interval. These time domain characteristic parameters may reflect the average length of the cardiac cycle, the volatility, and the relative degree of variability. Similarly, according to the peak interval of the pressure pulse signal, calculating the average peak interval, the standard deviation of the peak interval and the variation coefficient of the peak interval of the pressure pulse signal to obtain the time domain characteristic parameters of the pressure pulse signal. And performing fast Fourier transform on the peak intervals of the PPG waveform signals to obtain frequency domain signals of the PPG waveform signals, and calculating fundamental wave frequency, fundamental wave energy, first harmonic frequency, first harmonic energy, second harmonic frequency and second harmonic energy of the PPG waveform signals, so as to obtain frequency domain characteristic parameters of the PPG waveform signals. The formula of the fast fourier transform is as follows:

；

wherein, Representing the frequency domain signal,Represents a time domain signal, N represents a length of the signal,Representing the frequency. The time domain signal is converted into the frequency domain by the fast fourier transform, thereby analyzing its frequency components. Performing spectral entropy analysis on the target PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal; and similarly, performing spectral entropy analysis on the target pressure pulse signal to obtain a time domain characteristic parameter of the pressure pulse signal. Spectral entropy is an index that measures the complexity of a signal and can reflect the randomness and uncertainty of the signal. The calculation formula of the spectral entropy is as follows:

Spectral entropy ；

Wherein,Representing the normalized power spectral density of the spectrum. And carrying out Poincare mapping on the target PPG waveform signal to obtain a Poincare mapping diagram of the PPG waveform signal, and calculating the ratio of the major axis to the minor axis of the PPG waveform signal to obtain the time domain characteristic parameter of the PPG waveform signal. Poincare mapping is a nonlinear dynamics analysis method, and geometric features of heart rate variability can be analyzed by drawing a scatter diagram of adjacent peak intervals. And carrying out Poincare mapping on the target pressure pulse signals to obtain Poincare mapping diagrams of the pressure pulse signals, and calculating the ratio of the major axis to the minor axis of the pressure pulse signals to obtain time domain characteristic parameters of the pressure pulse signals. And carrying out correlation dimension and Lyapunov exponent analysis on the PPG waveform signal according to the peak point sequence of the PPG waveform signal to obtain frequency domain characteristic parameters of the PPG waveform signal. The correlation dimension and Lyapunov index are important indexes for measuring the chaotic characteristics of signals, and can reflect nonlinear dynamic behaviors of a system. The calculation formula of the association dimension is as follows:

；

wherein, Indicating a radius ofIs a logarithmic point in the neighborhood of (a). And carrying out association dimension and Lyapunov index analysis on the pressure pulse signals according to the peak point sequence of the pressure pulse signals to obtain time domain characteristic parameters of the pressure pulse signals. Combining the time domain characteristic parameters and the frequency domain characteristic parameters of the PPG waveform signals extracted in the steps S32 to S36 to obtain a first characteristic parameter set of the PPG waveform signals; and combining the time domain characteristic parameters of the pressure pulse signals extracted in the steps S32, S34 and S36 to obtain a second characteristic parameter set of the pressure pulse signals.

In a specific embodiment, the process of executing step 104 may specifically include the following steps:

S41: constructing a Gaussian kernel matrix of the PPG waveform signal according to the first characteristic parameter set; constructing a Gaussian kernel matrix of the pressure pulse signal according to the second characteristic parameter set;

S42: performing feature decomposition on the Gaussian kernel matrix of the PPG waveform signal to obtain a first feature value and a first feature vector of the PPG waveform signal, and sequencing the first feature value from large to small according to the first feature value to obtain a sequenced first feature value sequence and a corresponding first feature vector sequence; performing feature decomposition on the Gaussian kernel matrix of the pressure pulse signal to obtain a second feature value and a second feature vector of the pressure pulse signal, and sequencing the second feature value from large to small according to the second feature value to obtain a sequenced second feature value sequence and a corresponding second feature vector sequence;

S43: according to a preset first eigenvalue threshold value, carrying out truncation processing on the first eigenvalue sequence to obtain a truncated PPG eigenvalue subsequence, and extracting eigenvectors corresponding to the truncated PPG eigenvalue subsequence to obtain a PPG eigenvector subsequence; cutting off the second characteristic value sequence according to a preset second characteristic value threshold value to obtain a cut-off pressure pulse characteristic value subsequence, and extracting a characteristic vector corresponding to the cut-off pressure pulse characteristic value subsequence to obtain a pressure pulse characteristic vector subsequence;

s44: normalizing the PPG characteristic vector subsequence to obtain a normalized PPG characteristic vector subsequence, and normalizing the pressure pulse characteristic vector subsequence to obtain a normalized pressure pulse characteristic vector subsequence;

S45: according to the normalized PPG feature vector subsequence, mapping the first feature parameter set to a high-dimensional space through a kernel function to obtain a kernel matrix of the first feature parameter set, and according to the normalized pressure pulse feature vector subsequence, mapping the second feature parameter set to the high-dimensional space through the kernel function to obtain a kernel matrix of the second feature parameter set;

S46: performing singular value decomposition on the nuclear matrix of the first characteristic parameter set to obtain a left singular matrix, a singular value matrix and a right singular matrix of the first characteristic parameter set, and performing singular value decomposition on the nuclear matrix of the second characteristic parameter set to obtain a left singular matrix, a singular value matrix and a right singular matrix of the second characteristic parameter set;

S47: performing dimension reduction processing on the left singular matrix of the first characteristic parameter set according to the singular value matrix of the first characteristic parameter set to obtain a left singular matrix of the first characteristic parameter set after dimension reduction, and performing dimension reduction processing on the left singular matrix of the second characteristic parameter set according to the singular value matrix of the second characteristic parameter set to obtain a left singular matrix of the second characteristic parameter set after dimension reduction;

S48: performing linear transformation on the first characteristic parameter set through the left singular matrix of the first characteristic parameter set after the dimension reduction to obtain a characteristic mapping matrix of the first characteristic parameter set, and performing linear transformation on the second characteristic parameter set through the left singular matrix of the second characteristic parameter set after the dimension reduction to obtain a characteristic mapping matrix of the second characteristic parameter set;

s49: and carrying out vectorization processing on the feature mapping matrix of the first feature parameter set to obtain a first feature mapping vector, and carrying out vectorization processing on the feature mapping matrix of the second feature parameter set to obtain a second feature mapping vector.

Specifically, a first characteristic parameter set of the PPG waveform signal and a second characteristic parameter set of the pressure pulse signal are utilized to construct respective Gaussian kernel matrices. A gaussian kernel matrix is a matrix based on gaussian kernel functions for mapping data to a new feature space, revealing the structure and internal relationships of the data. The formula of the gaussian kernel function is as follows:

；

wherein, Representing the output of the gaussian kernel function,AndRespectively representing two feature vectors of the model,Representing the bandwidth parameters of the gaussian kernel. And constructing a Gaussian kernel matrix by calculating the Gaussian kernel function value between each pair of eigenvectors. And carrying out feature decomposition on the Gaussian kernel matrix of the PPG waveform signal to obtain a first feature value and a first feature vector of the PPG waveform signal. Feature decomposition is a process of decomposing a matrix into a set of feature vectors and feature values that can be used to represent the main features and trends of the data. The formula of the feature decomposition is as follows:

；

wherein, The gaussian kernel matrix is represented as such,The feature vector is represented by a vector of features,Representing the characteristic value. And solving the eigenvalue problem to obtain the eigenvalue and eigenvector of the Gaussian kernel matrix. And sorting the first eigenvalues from large to small to obtain a sorted first eigenvalue sequence and a corresponding first eigenvector sequence. And similarly, carrying out feature decomposition on the Gaussian kernel matrix of the pressure pulse signal to obtain a second feature value and a second feature vector of the pressure pulse signal, and sequencing from large to small according to the second feature value to obtain a sequenced second feature value sequence and a corresponding second feature vector sequence. And cutting off the first eigenvalue sequence according to a preset first eigenvalue threshold value to obtain a cut-off PPG eigenvalue subsequence, and extracting eigenvectors corresponding to the cut-off PPG eigenvalue subsequence to obtain a PPG eigenvector subsequence. Similarly, according to a preset second eigenvalue threshold value, the second eigenvalue sequence is truncated to obtain a truncated pressure pulse eigenvalue sub-sequence, and eigenvectors corresponding to the truncated pressure pulse eigenvalue sub-sequence are extracted to obtain a pressure pulse eigenvector sub-sequence. The purpose of the truncation process is to remove features with smaller eigenvalues and lower contribution to the data, thereby simplifying the model and improving computational efficiency. Carrying out normalization processing on the PPG feature vector subsequence to obtain a normalized PPG feature vector subsequence; and similarly, carrying out normalization processing on the pressure pulse characteristic vector subsequence to obtain a normalized pressure pulse characteristic vector subsequence. The normalization processing can eliminate dimension differences among different features, so that the features are compared on the same scale, and the accuracy and consistency of feature mapping are improved. According to the normalized PPG feature vector subsequence, mapping the first feature parameter set to a high-dimensional space through a kernel function to obtain a kernel matrix of the first feature parameter set; and similarly, according to the normalized pressure pulse feature vector subsequence, mapping the second feature parameter set to a high-dimensional space through a kernel function to obtain a kernel matrix of the second feature parameter set. Nonlinear data is converted into a linearly separable form through kernel function mapping, so that subsequent feature processing and analysis are facilitated. Performing singular value decomposition on the nuclear matrix of the first characteristic parameter set to obtain a left singular matrix, a singular value matrix and a right singular matrix of the first characteristic parameter set; and similarly, performing singular value decomposition on the nuclear matrix of the second characteristic parameter set to obtain a left singular matrix, a singular value matrix and a right singular matrix of the second characteristic parameter set. Singular value decomposition is a process of decomposing a matrix into three sub-matrices, which represent the row space, column space, and singular value space of the original matrix, respectively, and can be used to extract the main feature components of the data. Performing dimension reduction processing on the left singular matrix of the first characteristic parameter set according to the singular value matrix of the first characteristic parameter set to obtain a left singular matrix of the first characteristic parameter set after dimension reduction; And similarly, performing dimension reduction processing on the left singular matrix of the second characteristic parameter set according to the singular value matrix of the second characteristic parameter set to obtain the left singular matrix of the second characteristic parameter set after dimension reduction. The dimension reduction processing is to reduce the dimension of the data, thereby reducing the computational complexity and the storage requirement, and simultaneously preserving the main characteristic information of the data. Performing linear transformation on the first characteristic parameter set through the left singular matrix of the first characteristic parameter set after dimension reduction to obtain a characteristic mapping matrix of the first characteristic parameter set; and similarly, linearly transforming the second characteristic parameter set through the left singular matrix of the second characteristic parameter set after the dimension reduction to obtain a characteristic mapping matrix of the second characteristic parameter set. Linear transformation is an operation of transforming data from one space to another, by which the data can be more easily analyzed and processed in the new space. Vectorizing the feature mapping matrix of the first feature parameter set to obtain a first feature mapping vector; and similarly, carrying out vectorization processing on the feature mapping matrix of the second feature parameter set to obtain a second feature mapping vector. The vectorization processing is to convert the data in the matrix form into the vector form, so that the subsequent machine learning algorithm is convenient to process and analyze.

In a specific embodiment, the process of executing step 105 may specifically include the following steps:

S51: splicing the first feature mapping vector and the second feature mapping vector to obtain a spliced feature mapping vector, and calculating a covariance matrix of the spliced feature mapping vector;

s52: performing feature decomposition on the covariance matrix to obtain a third feature value and a third feature vector of the covariance matrix, and sorting the covariance matrix from large to small according to the third feature value to obtain a sorted third feature value sequence and a corresponding third feature vector sequence;

S53: cutting off the sequenced third characteristic value sequence to obtain a cut-off spliced characteristic value subsequence, and extracting a characteristic vector corresponding to the cut-off spliced characteristic value subsequence to obtain a spliced characteristic vector subsequence;

S54: performing linear transformation on the spliced feature mapping vector through splicing the feature vector subsequences to obtain a transformed feature mapping vector;

S55: and carrying out normalization processing and decentralization processing on the transformed feature mapping vector to obtain a fusion feature mapping vector.

Specifically, the first feature mapping vector and the second feature mapping vector are spliced. Assume that the first feature mapping vector isAnd the second feature mapping vector isThey are horizontally spliced to form a new feature vector. The spliced feature map vector contains all feature information from both signals. And calculating a covariance matrix of the spliced feature mapping vector. The covariance matrix is a matrix describing the interrelationship between a plurality of variables and has the following formula:

；

wherein, The covariance matrix is represented by a matrix of covariance,A matrix of feature vectors is represented,The mean vector representing the feature vector and n representing the number of samples. And (3) by calculating a covariance matrix, knowing the linear relation among all the features in the feature vector. And carrying out feature decomposition on the covariance matrix to obtain a third feature value and a third feature vector of the covariance matrix. Feature decomposition is a process of decomposing a matrix into a set of feature vectors and feature values that can be used to represent the main features and trends of the data. The formula of the feature decomposition is as follows:

；

wherein, The covariance matrix is represented by a matrix of covariance,The feature vector is represented by a vector of features,Representing the characteristic value. And solving the eigenvalue problem to obtain eigenvalues and eigenvectors of the covariance matrix. And sorting the third eigenvalues from large to small to obtain a sorted third eigenvalue sequence and a corresponding third eigenvector sequence. And cutting off the third characteristic value sequence according to a preset characteristic value threshold value to obtain a cut-off spliced characteristic value subsequence, and extracting a characteristic vector corresponding to the cut-off spliced characteristic value subsequence to obtain a spliced characteristic vector subsequence. The purpose of the truncation process is to remove features with smaller eigenvalues and lower contribution to the data, thereby simplifying the model and improving computational efficiency. And (3) performing linear transformation on the spliced feature mapping vector through splicing the feature vector subsequences to obtain a transformed feature mapping vector. Linear transformation is an operation of transforming data from one space to another, by which the data is made more representative and easy to analyze in the new space. And carrying out normalization processing and decentralization processing on the transformed feature mapping vector to obtain a fusion feature mapping vector. The normalization processing can eliminate scale differences among different features, so that the features are compared on the same scale; the decentralization processing is to eliminate the mean deviation of the data, so that the data is more concentrated at the origin, thereby improving the accuracy and stability of data analysis.

In a specific embodiment, the process of executing step 106 may specifically include the following steps:

S61: inputting the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector into a preset double-layer stacking support vector machine model, wherein a first layer support vector machine model in the double-layer stacking support vector machine model comprises: the first support vector machine model, the second support vector machine model, the third support vector machine model and the fourth support vector machine model, and the second layer support vector machine model comprises: a fifth support vector machine model, a sixth support vector machine model, a seventh support vector machine model, and an eighth support vector machine model;

S62: classifying the first feature mapping vector through a first support vector machine model to obtain a first classification result, classifying the second feature mapping vector through a second support vector machine model to obtain a second classification result, and classifying the fusion feature mapping vector through a third support vector machine model to obtain a third classification result; the first classification result, the second classification result and the third classification result are subjected to weighted fusion to obtain a first fusion classification result, wherein the weight of the first classification result is w1, the weight of the second classification result is w2, the weight of the third classification result is w3, and the weights of w1, w2 and w3 meet the following conditions: w1+w2 +w3= 1,0< w1<1,0< w2<1,0< w3<1;

s63: performing secondary classification on the first fusion classification result through a fourth support vector machine model to obtain a first secondary classification result;

S64: calculating the confidence coefficient of the first secondary classification result, comparing the confidence coefficient with a preset confidence coefficient threshold value, taking the first secondary classification result as a final atrial fibrillation state discrimination result if the confidence coefficient is larger than or equal to the preset confidence coefficient threshold value, and executing step S65 if the confidence coefficient is smaller than the preset confidence coefficient threshold value;

S65: performing feature enhancement processing on the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector to obtain an enhanced first feature mapping vector, an enhanced second feature mapping vector and an enhanced fusion feature mapping vector;

S66: classifying the enhanced first feature mapping vector through a fifth support vector machine model to obtain a fourth classification result, classifying the enhanced second feature mapping vector through a sixth support vector machine model to obtain a fifth classification result, and classifying the enhanced fusion feature mapping vector through a seventh support vector machine model to obtain a sixth classification result;

s67: and carrying out weighted fusion on the fourth classification result, the fifth classification result and the sixth classification result to obtain a second fusion classification result, wherein the weight of the fourth classification result is w4, the weight of the fifth classification result is w5, the weight of the sixth classification result is w6, and the weights of w4, w5 and w6 meet the following conditions: w4+w5+w6=1, 0< w4<1,0< w5<1,0< w6<1;

s68: performing secondary classification on the second fusion classification result through an eighth support vector machine model according to the second fusion classification result to obtain a second secondary classification result;

S69: calculating the confidence coefficient of the second secondary classification result, comparing the confidence coefficient with a preset confidence coefficient threshold value, taking the second secondary classification result as a final atrial fibrillation state discrimination result if the confidence coefficient is larger than or equal to the preset confidence coefficient threshold value, and taking the first secondary classification result as the final atrial fibrillation state discrimination result and outputting the first secondary classification result if the confidence coefficient is smaller than the preset confidence coefficient threshold value.

Specifically, the dual-layer stacked support vector machine model is an advanced classifier, and by combining the outputs of multiple support vector machine models, the classification accuracy and robustness can be improved. The first layer contains four support vector machine models and the second layer contains four support vector machine models. Inputting the first feature mapping vector into a first support vector machine model for classification to obtain a first classification result; inputting the second feature mapping vector into a second support vector machine model for classification to obtain a second classification result; and inputting the fusion feature mapping vector into a third support vector machine model for classification to obtain a third classification result. To integrate these classification results, they are weighted fused. And inputting the fusion classification result into a fourth support vector machine model for secondary classification to obtain a first secondary classification result. To evaluate the reliability of this result, its confidence is calculated. Confidence is an indicator of how reliable the classification results are. Comparing the calculated confidence coefficient with a preset confidence coefficient threshold value, and if the confidence coefficient is larger than or equal to the preset confidence coefficient threshold value, taking the first secondary classification result as a final atrial fibrillation state discrimination result. If the confidence level is less than the preset confidence threshold, further processing is required. At this time, feature enhancement processing is performed on the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector, so as to obtain an enhanced first feature mapping vector, an enhanced second feature mapping vector and an enhanced fusion feature mapping vector. The feature enhancement process may increase the accuracy of classification by increasing the dimensionality of the features or by some transformation to increase the distinguishing ability of the features. Classifying the enhanced first feature mapping vector through a fifth support vector machine model to obtain a fourth classification result; classifying the enhanced second feature mapping vector through a sixth support vector machine model to obtain a fifth classification result; and classifying the enhanced fusion feature mapping vector through a seventh support vector machine model to obtain a sixth classification result. And carrying out weighted fusion on the fourth classification result, the fifth classification result and the sixth classification result to obtain a second fusion classification result. And carrying out secondary classification on the second fusion classification result through an eighth support vector machine model according to the second fusion classification result to obtain a second secondary classification result. To evaluate the reliability of this result, its confidence is again calculated. And comparing the calculated confidence coefficient with a preset confidence coefficient threshold value. And if the confidence coefficient is greater than or equal to a preset confidence coefficient threshold value, taking the second classification result as a final atrial fibrillation state discrimination result. If the confidence coefficient is smaller than the preset confidence coefficient threshold value, the first secondary classification result is used as a final atrial fibrillation state discrimination result and is output.

The method for measuring atrial fibrillation based on PPG waveform signals and pressure pulse signals in the embodiment of the present application is described above, and the atrial fibrillation measuring device based on PPG waveform signals and pressure pulse signals in the embodiment of the present application is described below, referring to fig. 2, an embodiment of the atrial fibrillation measuring device based on PPG waveform signals and pressure pulse signals in the embodiment of the present application includes:

The acquisition module 201 is configured to acquire synchronous data of a subject by using a PPG sensor and a pressure sensor, so as to obtain an initial PPG waveform signal and an initial pressure pulse signal;

The removing module 202 is configured to perform baseline drift removal and high-frequency noise filtering on the initial PPG waveform signal and the initial pressure pulse signal, respectively, to obtain a target PPG waveform signal and a target pressure pulse signal;

The extracting module 203 is configured to perform time domain feature extraction and frequency domain feature extraction on the target PPG waveform signal to obtain a first feature parameter set of the PPG waveform signal, and perform time domain feature extraction on the target pressure pulse signal to obtain a second feature parameter set of the pressure pulse signal;

The mapping module 204 is configured to perform feature mapping and vector conversion on the first feature parameter set and the second feature parameter set, to obtain a corresponding first feature mapping vector and a corresponding second feature mapping vector;

The combination module 205 is configured to perform feature combination and principal component dimension reduction on the first feature mapping vector and the second feature mapping vector to obtain a fused feature mapping vector;

And the judging module 206 is used for inputting the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector into a preset double-layer stacked support vector machine model to judge the atrial fibrillation state and obtain an atrial fibrillation state judging result.

By the cooperative cooperation of the above components, the synchronous data acquisition is performed by using the PPG sensor and the pressure sensor, so that the time alignment of the two signals is ensured, and the data deviation caused by time asynchronization is avoided. The preprocessing steps of baseline drift removal and high-frequency noise filtering are adopted, so that interference factors in signals are effectively removed, the purity of the signals is improved, time domain features and frequency domain features of the PPG waveform signals are extracted, and detailed time domain feature extraction is performed on pressure pulse signals, so that the change of the cardiovascular system is ensured to be captured from different dimensions. The extracted characteristic parameters cover a plurality of aspects of signals, enrich characteristic sets and improve the identification capability of atrial fibrillation states. The feature parameters of the PPG waveform signal and the pressure pulse signal are combined to form a high-dimensional feature vector, and then the feature vector is subjected to dimension reduction processing by methods such as principal component analysis and the like, so that redundant information is effectively reduced, and the most distinguished feature is reserved. The construction of the fusion characteristic fully utilizes the complementarity of the two signals, enhances the detection capability of the atrial fibrillation state and improves the discrimination performance of the model. The atrial fibrillation state is judged by adopting a double-layer stacked support vector machine model, and the robustness and generalization capability of the model are improved through multi-level classification and weighted fusion. In the classifying process, the accuracy and the reliability of atrial fibrillation detection are further improved through confidence calculation and feature enhancement processing, the atrial fibrillation state of a tested object is monitored in real time, and the data acquisition and the processing are carried out through portable equipment, so that the real-time judgment and the monitoring of the atrial fibrillation state are realized.

The present application also provides a computer device, where the computer device includes a memory and a processor, where the memory stores computer readable instructions that, when executed by the processor, cause the processor to execute the steps of the atrial fibrillation measurement method based on the PPG waveform signal and the pressure pulse signal in the foregoing embodiments.

The present application also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and may also be a volatile computer readable storage medium, where instructions are stored in the computer readable storage medium, when the instructions are executed on a computer, cause the computer to perform the steps of the atrial fibrillation measurement method based on the PPG waveform signal and the pressure pulse signal.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, systems and units may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals, the method comprising:

2. The atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals according to claim 1, wherein the synchronous data acquisition is performed on the tested object by using a PPG sensor and a pressure sensor respectively to obtain an initial PPG waveform signal and an initial pressure pulse signal, and the method comprises the following steps:

3. The atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals according to claim 1, wherein the performing baseline drift removal and high-frequency noise filtering on the initial PPG waveform signals and the initial pressure pulse signals to obtain target PPG waveform signals and target pressure pulse signals respectively includes:

s21: performing a first median filtering on the initial PPG waveform signal, wherein the size of a filtering window is W1, so as to obtain a first filtered PPG waveform signal, and simultaneously performing a first median filtering on the initial pressure pulse signal, and the size of the filtering window is W1, so as to obtain a first filtered pressure pulse signal;

4. The atrial fibrillation measurement method based on PPG waveform signal and pressure pulse signal according to claim 1, wherein the performing time domain feature extraction and frequency domain feature extraction on the target PPG waveform signal to obtain a first feature parameter set of PPG waveform signal, and performing time domain feature extraction on the target pressure pulse signal to obtain a second feature parameter set of pressure pulse signal includes:

5. The method for atrial fibrillation measurement based on PPG waveform signals and pressure pulse signals according to claim 4, wherein the performing feature mapping and vector conversion on the first feature parameter set and the second feature parameter set to obtain corresponding first feature mapping vector and second feature mapping vector respectively includes:

6. The atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals according to claim 1, wherein the performing feature combination and principal component dimensionality reduction on the first feature map vector and the second feature map vector to obtain a fused feature map vector comprises:

7. The atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals according to claim 1, wherein the inputting the first feature mapping vector, the second feature mapping vector and the fusion feature mapping vector into a preset double-layer stacked support vector machine model for atrial fibrillation state discrimination to obtain an atrial fibrillation state discrimination result comprises:

8. Atrial fibrillation measurement device based on PPG waveform signal and pressure pulse signal for performing an atrial fibrillation measurement method based on PPG waveform signal and pressure pulse signal as claimed in any one of claims 1-7, the device comprising:

9. A computer device, the computer device comprising: a memory and at least one processor, the memory having instructions stored therein;

The at least one processor invokes the instructions in the memory to cause the computer device to perform the atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals as in any one of claims 1-7.

10. A computer readable storage medium having instructions stored thereon, which when executed by a processor, implement the atrial fibrillation measurement method based on PPG waveform signals and pressure pulse signals as in any one of claims 1-7.