Optimal area selection chest / abdomen respiration signal extraction method based on human body pressure distribution

Abstract

The invention belongs to the technical field of sleep health monitoring, and particularly relates to an optimal area selection chest / abdomen respiration signal extraction method based on human body pressure distribution. The method comprises the steps that data signals of human body pressure distribution changing along with time during sleeping are collected; the collected original pressure data are preprocessed, and clear and accurate pressure distribution data are obtained; for the preprocessed data, based on calculation of parameters such as the frequency, the power and the signal-to-noise ratio of the chest and abdomen area, an optimal breathing area selection algorithm is used for calculating respective band-shaped areas, most suitable for breathing signal extraction, of the chest / abdomen area; for the selected area, respiratory waveforms are extracted through a pressure signal fusion method based on weighted summation, and thoracic / abdominal respiratory signal monitoring is achieved; the sleep breathing monitoring device can accurately monitor sleep breathing in the form of a non-contact mattress under the condition that sleep quality is not interfered, and has a wide clinical application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of sleep health monitoring, and particularly relates to a method for extracting sleep respiration signals. Background Art

[0002] In sleep health management, sleep respiration quality is an important part of sleep health monitoring, which is related to the diagnosis of related sleep diseases. Among them, respiratory rate is an important vital sign and is closely related to human health. The respiratory rate of a normal adult at rest is about 12 - 20 BPM. A respiratory rate higher than 20 BPM is called tachypnea, which may be related to diseases such as fever and pneumonia. A respiratory rate lower than 10 BPM is called bradypnea, which may be related to diseases such as apnea. In addition, respiratory rate can be used to judge the deterioration assessment of clinical symptoms and predict cardiac arrest.

[0003] Currently, the traditional medical sleep monitoring method is based on a polysomnography (PSG) monitor, which can monitor physiological signals such as electroencephalogram, electrocardiogram, chest and abdominal respiratory movements, body position, and blood oxygen saturation. As a medical - specific device, the polysomnography monitor has problems such as complex instrument use, the need for professional technical personnel, high price, and affecting normal sleep quality.

[0004] Current sleep respiration monitoring research mainly focuses on long - term real - time monitoring in a home environment. Since this application scenario involves home use, long - term monitoring, sleep, and privacy environment, relatively high requirements are put forward for aspects such as the stability, real - time performance, ease of use, comfort, privacy protection, and cost of the monitoring system.

[0005] In existing research, the methods for respiratory monitoring can be mainly divided into 4 categories according to the type of hardware system, including methods based on wearable devices, methods based on cameras, methods based on radio frequency, and methods based on pressure - sensing mattresses. The method based on wearable devices requires close contact between the human body and the device, which may cause discomfort and has the problem of motion artifacts, affecting accuracy; the method based on cameras is greatly affected by environmental factors and has a risk of privacy leakage; the method based on radio frequency has a relatively high price and is also easily affected by environmental factors; the pressure - sensing system extracts the respiratory rate through the regular pressure changes generated by the chest and abdominal movements during breathing on the mattress, does not cause discomfort to the human body, is less affected by the environment, has a relatively low cost, and has no privacy problem.

[0006] In existing research, the algorithms for extracting respiratory signals based on human body pressure distribution can be mainly divided into two categories. The first category does not use a reference signal, such as the superposition method, which superimposes all pressure sensor data into a dynamic pressure signal to reflect the respiratory signal. Due to the interference of body movement and the noise of the sensor itself in the respiratory action, this method has the defect of low signal-to-noise ratio; the second category uses a reference signal and performs phase correction and weighted summation on the pressure signals of each sensor based on the reference signal. Compared with the first category, this method considers the quality of different sensor signals and optimizes them through weighted summation. The effect is better than the first category, but there is a problem of large computational complexity, and speed and accuracy cannot be achieved at the same time. Moreover, both of the above methods only consider extracting a single respiratory signal from the overall pressure signal, while the traditional gold standard polysomnography monitor monitors the respiratory movements of the human chest and abdomen respectively through chest / abdominal straps. However, some respiratory diseases can destroy the normal breathing pattern and lead to thoracic and abdominal asynchrony (Thoracoabdominal Asynchrony, TAA). TAA is a sign of respiratory disease, which occurs when the chest and abdomen are not parallel or even opposite in movement (respiratory paradox), such as emphysema, severe asthma, chronic obstructive pulmonary disease, etc. These diseases can cause damage to the basic ventilation function of the lungs, induce respiratory muscle fatigue and paradoxical breathing, and abnormal respiratory movements such as periodic alternation of abdominal and chest breathing, disorder and non-parallel abdominal pressure breathing. Therefore, it can be seen that independent and synchronous monitoring of chest / abdominal breathing rather than monitoring the overall respiratory movement alone can help us diagnose and prevent these related diseases, which is very necessary.

[0007] This method proposes a new breathing extraction method based on body pressure distribution data, namely, a chest / abdomen breathing signal extraction method based on optimal region selection, which can accurately and synchronously extract the breathing action signals of the chest and abdomen regions. Summary of the invention

[0008] The purpose of the present invention is to address the shortcomings of the current traditional sleep breathing monitoring methods, such as expensive equipment, the need for professional technicians, and the impact on normal sleep quality. Considering the shortcomings of existing respiratory signal monitoring system research and related respiratory signal extraction methods, a new respiratory extraction method based on body pressure distribution data is designed, that is, a chest / abdomen respiratory signal extraction method based on optimal area selection. This method is based on body pressure distribution data and can ensure the low cost, stability, comfort and privacy of signal acquisition; the respiratory extraction algorithm used can realize fast and real-time extraction of respiratory signals during sleep, and the respiratory rate calculation has high accuracy.

[0009] The optimal region selection method for chest / abdomen respiration signal extraction based on human body pressure distribution provided by the present invention includes preprocessing the collected original pressure data to obtain relatively clear and accurate pressure distribution data; for the preprocessed data, aiming at the chest and abdomen regions, based on calculating parameters such as its frequency, power, and signal-to-noise ratio, using the optimal respiration region selection algorithm to calculate the strip regions in the chest / abdomen regions that are most suitable for respiration signal extraction respectively; for the selected regions, extracting the respiration waveform through the pressure signal fusion method based on weighted summation to realize chest / abdomen respiration signal monitoring; the specific steps are as follows:

[0010] (1) Acquisition of pressure distribution image data of the upper body of the human body during sleep; based on the flexible array pressure sensing mattress system (H.K. Diao et al., "Deep Residual Networks for Sleep Posture Recognition With Unobtrusive Miniature Scale Smart Mat System," IEEE Transactions on Biomedical Circuits and Systems, vol. 15, no. 1, pp. 111 - 121, Feb 2021), a pressure sensing system covering the entire upper body area of the user is used to collect the pressure distribution image of the upper body of the human body during sleep at a sampling frequency of 1 Hz. The sleep pressure distribution data recorded once is a two-dimensional array of N×N, where the value of each element represents the relative pressure magnitude collected at the corresponding point in the array sensor; for the entire sleep process time T, the data volume collected is N×N×T;

[0011] (2) Preprocessing the pressure distribution image data collected in step (1) to obtain the processed upper body pressure distribution data; the preprocessing operations include crosstalk correction and threshold filtering;

[0012] (3) Dividing the pressure distribution data of the chest region and the abdomen region and determining the optimal strip regions;

[0013] For the upper body pressure distribution data, calculate its centroid position. Based on its coordinates, divide the upper body pressure into the chest region and the abdomen region. For the two regions, the following operations are performed respectively;

[0014] (1) Obtain the dynamic pressure signal S of each sensor within this time range from the pressure distribution data i ;

[0015] (2) For each S i , calculate its signal frequency f i , signal power power iand the signal-to-noise ratio SNR i , and the corresponding calculation method is as follows:

[0016] The signal frequency f i is S i the frequency corresponding to the maximum peak value of the single-sided spectrum after fast Fourier transform;

[0017] The signal power power i :

[0018]

[0019] where L is the length of S i and S ij is the value of the j-th sampling point of S i , and S i is the mean value of S

[0020] S i The signal-to-noise ratio SNR i is defined as the ratio of the respiratory signal power power i of S i,s to the noise power power i,n , and is roughly estimated as the square of the amplitude i of f in the single-sided spectrum to the mean square of the amplitudes except f i in the single-sided spectrum, where N is the number of FFT points. The calculation formula of SNR is: i The calculation formula of SNR

[0021]

[0022] (3) Select S i that satisfies 0.15 < f i < 0.35 and 0.1 < power i < 10 and SNR i > 0.5·max(SNR i ), and form a list S′ with them;

[0023] (4) For S i that does not satisfy the conditions in step (3), set the corresponding weight coefficient w i to 0;

[0024] For S′ that satisfies the conditions in step (3), the calculation formula of its corresponding weight coefficient w i is:

[0025]

[0026] (5) Based on the weight w of each sensori Obtain sensor weight distribution data of size (N, N), use a window of size (2, N), and slide the window with a step size of 1 to calculate the sum W of w within the window for each slide i within the window j to form a list W;

[0027] (6) Select the sliding window corresponding to the maximum W in the list W j as the optimal band region within this period of time;

[0028] (IV) For the selected region, extract the chest / abdomen respiration signal;

[0029] (1) To reduce the mutual interference between signals of different phases, select the sensor signal with the maximum weight w in the optimal band region as the reference signal, and calculate the covariance cov between it and other sensors max For sensors with cov i not greater than 0, set its w i to -w i ; i

[0030] (2) Select the S i in the optimal band region and the corresponding w i to calculate the output signal

[0031] (3) Perform band-pass filtering on the output signal described in step (2) to obtain the filtered signal, which is the extracted respiration signal waveform;

[0032] (4) Perform Fourier transform on the respiration signal described in step (3) to obtain the corresponding spectrogram, and use the frequency corresponding to the maximum peak in the spectrogram as the required respiration frequency.

[0033] In step (I) of the present invention, it is preferred that N is 10 - 64, and more preferably 64.

[0034] In step (IV-3) of the present invention, the band-pass filtering uses a 3rd-order Butterworth filter.

[0035] Compared with the existing sleep respiration monitoring methods, the advantages and implementation effects of the present invention are mainly as follows:

[0036] Using the pressure distribution data collected by the flexible pressure sensor mattress to perform the sleep monitoring task has high stability, comfort, low cost, and privacy; compared with the previous research on sleep respiration extraction methods based on body pressure distribution, the chest / abdomen respiration signal extraction method based on the optimal region can extract chest / abdomen respiration signals simultaneously while taking into account accuracy and speed, achieving a sleep respiration acquisition form similar to the PSG chest / abdomen respiration strap, and has high application prospects. ​Description of the Drawings

[0037] Figure 1 is the overall process of respiratory monitoring of the present invention.

[0038] Figure 2 is the flowchart of the method for extracting chest / abdominal respiration signals by optimal region selection based on human pressure distribution of the present invention.

[0039] Figure 3 is the calculation process of the method for extracting chest / abdominal band region respiration signals based on optimal region calculation. Detailed Description of the Invention

[0040] The present invention will be described in detail below with reference to the drawings and specific embodiments.

[0041] The present invention proposes a method for sleep respiratory monitoring using body pressure distribution data collected by a flexible array type pressure sensitive mattress and based on an independent extraction algorithm for chest / abdominal respiration signals extracted from an optimal region. The overall process is as Figure 1 shown, in which the extraction of respiratory signals in the sleep state is as Figure 2 shown, and the calculation process of the method for extracting chest / abdominal band region respiration signals based on optimal region calculation is as Figure 3 shown. The whole process is as follows:

[0042] (1) Use a flexible pressure sensitive mattress to collect body pressure distribution data of the subject during the whole night's sleep. The subject lies on the bed according to the experimental instructions and wears the chest / abdominal respiration straps of the polysomnography device. The mattress samples at a frequency of 1 Hz, and the PSG samples synchronously. The subject is not restricted during sleep, allowing body movement and body posture change.

[0043] (2) Collect the data of the whole sleep process, and the time is between 6 - 8 hours. In this embodiment, the time series data of the whole night's human pressure distribution is collected by the mattress device, and the whole night's respiration signal waveform is collected by the PSG chest / abdominal straps;

[0044] (3) Preprocess the body pressure distribution data, including crosstalk correction and threshold filtering, to obtain the preprocessed body pressure distribution data;

[0045] (4) For the upper body pressure distribution data, calculate its centroid position, and divide the upper body pressure into the chest region and the abdominal region based on its coordinates. For the two regions, subsequent operations are taken respectively;

[0046] (5) Obtain the dynamic pressure signal of each sensor within this time range from the pressure distribution data;

[0047] (6) For each dynamic pressure signal, calculate its signal frequency, signal power, and signal-to-noise ratio;

[0048] (7) Screen the dynamic pressure signals that meet the conditions that the signal frequency is between 0.15 - 0.35, the signal power is between 0.1 - 10, and the signal-to-noise ratio is greater than 0.5 times the maximum signal-to-noise ratio.

[0049] (8) For the sensors that are not screened, their corresponding weights are 0; for the sensors that enter the screening, their corresponding weights are the ratio of the square root of the power of this signal to the noise power.

[0050] (9) Obtain the sensor weight distribution array based on the weights of each sensor. Use a strip window and slide it from top to bottom with a sliding step of 1, calculate the sum of the weights within the window for each sliding, and take the window with the maximum sum of weights as the optimal strip region.

[0051] (10) To reduce the mutual interference between signals of different phases, take the sensor signal with the maximum weight in the optimal strip region as the reference signal, calculate the covariance between it and other sensors. For the sensors with covariance not greater than 0, set their weights to the opposite of the original value.

[0052] (11) Perform weighted summation based on the dynamic pressure signals and corresponding weights of each sensor within the optimal region, and perform band-pass filtering on the output signal to obtain the chest / abdomen region respiration signal to be extracted.

[0053] (12) Perform Fourier transform on the respiration signal described in step (11) to obtain the corresponding spectrogram, and take the frequency corresponding to the maximum peak in the spectrogram as the required respiration frequency.

[0054] Table 1 shows the evaluation indicators obtained by comparing the respiration rate calculated from the extracted chest / abdomen respiration signal with the respiration rate of the PSG gold standard respiration signal, including Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Root Mean Square Percentage Error (RMSPE), Pearson Correlation Coefficient (PCC), and single operation time. It can be seen that the respiration extracted by the method of the present invention has extremely small errors and high correlations with the PSG gold standard respiration. In addition, the calculation speed is fast, meeting the requirements of real-time monitoring.

[0055] Table 1 Respiration Signal Evaluation Indicators

[0056] Extracted signal MAE MAPE / % RMSE RMSPE / % PCC Single operation time / s Chest breathing 0.27 1.98 0.68 5.09 0.94 0.073 Abdominal breathing 0.31 2.25 0.81 5.74 0.91 0.067 。

[0057] Compared with the existing sleep apnea monitoring methods, the method of the present invention realizes accurate and rapid monitoring of sleep apnea signals on the premise of ensuring monitoring comfort, stability, privacy, etc. When applied to actual life scenarios, its related performance has also been tested, and it has good application value and prospects.

Claims

1. A method for extracting chest / abdomen respiratory signals based on optimal region selection of human body pressure distribution, characterized in that: The specific steps are as follows: (i) Collection of pressure distribution image data of the upper body of a human body during sleep; specifically, a flexible array pressure sensing mattress system is used to collect pressure distribution images of the upper body of a human body during sleep at a sampling frequency of 1 Hz. The sleep pressure distribution data recorded at a single time is an N×N two-dimensional array, in which the value of each element represents the relative pressure size collected by the corresponding point in the array sensor; for the entire sleep process time T, the amount of data collected is N×N×T; (2) preprocessing the pressure distribution image data collected in step (1) to obtain processed upper body pressure distribution data; Preprocessing operations include crosstalk correction and threshold filtering; (iii) dividing the pressure distribution data of the chest area and the abdominal area and determining the optimal belt area; For the upper body pressure distribution data, calculate its center of mass position, and based on its coordinates, divide the upper body pressure into the chest area and the abdomen area. For the two areas, perform the following operations respectively; (1) Obtain the dynamic pressure signal S of each sensor within the time range from the pressure distribution data i ; (2) For each S i , calculate its signal frequency f i , signal power i And the signal-to-noise ratio SNR i , the corresponding calculation method is as follows: Signal frequency f i For S i The frequency corresponding to the maximum peak of the single-sided spectrum after fast Fourier transform; Signal power i : Where L is S i Length, S ij For S i The value of the jth sampling point, For S i The mean of S i Signal-to-noise ratio SNR i Defined as S i The breathing signal power i,s and noise power i,n The ratio of f i The square of the magnitude Divide f by the single-sided spectrum i The amplitude of the other The ratio of N to FFT points; SNR i The calculation formula is: (3) Screening meets 0.15 <f i <0.35, and 0.1 <power i <10 and SNR i >0.5·max(SNR i ) i , and form them into a list S'; (4) For S that does not meet the conditions in step (3) i , and its corresponding weight coefficient w i Set to 0; For S' that satisfies the conditions in step (3), its corresponding weight coefficient w i The calculation formula is: (5) Based on the weight w of each sensor i Get the sensor weight distribution data of size (N,N), slide the window with a window size of (2,N) and a sliding step of 1, and calculate w in each sliding window i The sum of W j , forming a list W; (6) Take the list W with the largest W j The corresponding sliding window is used as the optimal strip area within this period of time; (iv) extracting chest / abdomen respiratory signals for the selected region; (1) Take the optimal strip area with the maximum weight w max The sensor signal is used as the reference signal, and the covariance cov between other sensors and it is calculated. i , for cov i For sensors that are not greater than 0, i Set to -w i ; (2) Take the optimal strip area S i The corresponding w i , calculate the output signal (3) performing bandpass filtering on the output signal of step (2) to obtain a filtered signal, which is the extracted respiratory signal waveform; (4) Performing Fourier transform on the respiratory signal in step (3) to obtain a corresponding spectrum diagram, and taking the frequency corresponding to the maximum peak in the spectrum diagram as the desired respiratory frequency.

2. The method for extracting chest / abdomen respiratory signals by selecting optimal regions based on human body pressure distribution according to claim 1, characterized in that: In step (1), the pressure distribution image N is 64.

3. The method for extracting chest / abdomen respiratory signals by selecting optimal regions based on human body pressure distribution according to claim 1, characterized in that: The bandpass filtering described in step (four-3) adopts a third-order Butterworth filter.

Images