Physiological signal feature selection method and physiological signal feature selection device

Abstract

This paper provides a method and apparatus for selecting physiological signal features. The method includes receiving a signal comprising a plurality of pulses; determining whether a second pulse among the plurality of pulses is similar to a first pulse among the plurality of pulses; after the second pulse is determined to be similar to the first pulse, selecting the second pulse as one of a plurality of reference pulses in a reference waveform sequence; averaging the plurality of reference pulses into a template pulse; determining whether each of the plurality of pulses matches the template pulse; and selecting a plurality of features from each pulse among the plurality of pulses that matches the template pulse. This method can filter the waveform of a signal to improve the accuracy of signal analysis.

Description

Technical Field

[0001] This invention relates to the field of physiological signal selection, and more particularly to a method and apparatus for selecting physiological signal features. Background Technology

[0002] Photoplethysmography (PPG) is a non-invasive method for detecting changes in blood volume, thereby measuring physiological indicators such as heart rate and blood oxygen saturation. The sensing elements in PPG primarily consist of light-emitting diodes (LEDs) and photodiodes. The LEDs emit light; the photodiodes receive the light signal passing through the blood vessels. The alternating current (AC) component reflects changes in blood volume caused by heartbeats, while the direct current (DC) component reflects the unchanged light absorption of subcutaneous tissue and venous blood.

[0003] Optical signals measured by the optical volume change mapping method are extremely susceptible to external noise interference. Low-frequency noise mostly comes from motion and breathing interference and causes amplitude drift in the DC component, while high-frequency noise mostly comes from ambient light sources and causes disturbance in the AC component.

[0004] Furthermore, the dense network of capillaries in the ear canal has a higher blood vessel perfusion value than that in the hand, meaning that blood pressure measurements taken from the earlobe are more accurate. However, it is not visually easy to tell whether a current measuring device is not properly fitted, and the sensor may not be properly seated in the user's ear canal, leading to light scattering or the photodiode receiving noise from the mixed external light source.

[0005] In view of this, there is a real need to improve the existing technology. Summary of the Invention

[0006] Therefore, the main objective of this invention is to provide a method and apparatus for selecting physiological signal features that can filter the waveform of a signal to improve the accuracy of signal analysis.

[0007] This invention discloses a method for selecting physiological signal features, comprising receiving a signal, wherein the signal includes a plurality of pulses; determining whether a second pulse among the plurality of pulses is similar to a first pulse among the plurality of pulses; after the second pulse is determined to be similar to the first pulse, selecting the second pulse as one of a plurality of reference pulses in a reference waveform sequence; averaging the plurality of reference pulses into a template pulse; determining whether the plurality of pulses match the template pulse respectively; and selecting a plurality of features from each pulse among the plurality of pulses that matches the template pulse.

[0008] This invention discloses a physiological signal feature selection device, comprising a processing circuit and a storage circuit. The processing circuit is used to execute a code. The storage circuit is coupled to the processing circuit and is used to store the code, wherein the code includes receiving a signal, wherein the signal comprises a plurality of pulses; determining whether a second pulse among the plurality of pulses is similar to a first pulse among the plurality of pulses; after the second pulse is determined to be similar to the first pulse, selecting the second pulse as one of a plurality of reference pulses in a reference waveform sequence; averaging the plurality of reference pulses into a template pulse; determining whether each of the plurality of pulses matches the template pulse; and selecting a plurality of features from each pulse among the plurality of pulses that matches the template pulse. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of a signal feature selection device according to an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of the preprocessing signals according to an embodiment of the present invention.

[0011] Figure 3 This is a schematic diagram of the signal processing flow according to an embodiment of the present invention.

[0012] Figures 4 to 6 These are schematic diagrams of pulses according to embodiments of the present invention.

[0013] Figure 7 and Figure 8 The following are flowcharts of methods according to embodiments of the present invention.

[0014] Explanation of symbols in the attached drawings:

[0015] 10. Signal feature selection device;

[0016] 100. Input module;

[0017] 100S, signal;

[0018] 102. Pre-processing module;

[0019] 102S, 200C, 200N, preprocessing signals;

[0020] 104. Regional extreme value detection module;

[0021] 106. Waveform segmentation module;

[0022] 106L1~106L5, long pulse duration;

[0023] 106S1~106S5, 107S1~107S5, 108f1~108fn, 110i1~110i5, pulse;

[0024] 107L, long normalized pulse duration;

[0025] 108. Template initialization module;

[0026] 108V, sample pulse;

[0027] 110. Template matching module;

[0028] 112. Calculation module;

[0029] 1122. Feature extraction unit;

[0030] 1124. Calculation Unit;

[0031] 114. Output module;

[0032] 70, 80, Methods;

[0033] Dt1, diastolic duration;

[0034] P1, P2, sampling points;

[0035] P3, P4, troughs;

[0036] P5, peak;

[0037] S1~S4, area;

[0038] S700~S714, S800~S826, Steps;

[0039] St1, duration of systole;

[0040] WL1, Window Length. Detailed Implementation

[0041] Figure 1 This is a schematic diagram of a signal feature selection device 10 according to an embodiment of the present invention. The signal feature selection device 10 can filter the waveform of the received signal to ensure that the signal waveform used for subsequent interpretation is good / correct, and avoid using abnormal waveforms for signal analysis, thereby improving the accuracy of signal analysis.

[0042] For example, the signal feature selection device 10 can be used in an ear-hook physiological measurement device (e.g., a Thor ear-hook device). The ear-hook physiological measurement device can contact the ear canal and may include a photoplethysmography module for transmitting and receiving optical signals to generate a signal 100S. Because cardiac contraction and relaxation cause pulsation and changes in blood vessel volume, and the oxygenated and deoxygenated hemoglobin in the blood also change, this affects light absorption, resulting in different light intensities received by the photoplethysmography module, and the signal 100S output by the photoplethysmography module contains an AC component. The signal feature selection device 10 can screen the waveform of the signal 100S to ensure that the signal waveform used for subsequent analysis is good rather than abnormal, thereby enabling accurate estimation of blood pressure using the signal 100S.

[0043] like Figure 1 As shown, the signal feature selection device 10 may include an input module 100, a preprocessing module 102, a region extremum detection module 104, a waveform segmentation module 106, a template initialization module 108, a template matching module 110, a calculation module 112, and an output module 114.

[0044] The input module 100 can receive the signal 100S from the ear-hook physiological measurement device, for example, by transmitting the signal 100S through wireless communication such as Bluetooth or WiFi via a transmission module or through a transmission line.

[0045] The preprocessing module 102 can process the signal 100S to generate a preprocessed signal 102S. In one embodiment, the preprocessing module 102 can normalize the signal 100S according to the average value or standard deviation of the signal 100S, so that the displacement / magnitude of the signal 100S falls between -1 and 1.

[0046] In one embodiment, the accelerometer of the ear-hook physiological measurement device can confirm whether the subject is in a stationary state. The preprocessing module 102 can then perform corresponding signal processing, such as discarding pulse waveforms where the subject is not in a stationary state and not selecting their pulse wave characteristics, or the preprocessing module 102 can not process signal 100S and instead cause the output module 114 to request the subject to re-measure.

[0047] In one embodiment, the preprocessing module 102 may include a bandpass filter to bandpass filter the signal 100S. The preprocessing module 102 can filter out high-frequency noise or ambient light noise, resulting in a smoother waveform and improved signal-to-noise ratio (SNR), thereby enabling the signal feature selection device 10 to perform more accurate calculations. In one embodiment, the bandpass filter may be a Butterworth filter, and the cutoff frequency band may be designed to be 0.5–5 Hz.

[0048] In one embodiment, the preprocessing module 102 can process the signal 100S and output a preprocessed signal 102S containing only AC components. Figure 2 This is a schematic diagram of one embodiment of preprocessing signals 200c and 200n, where the sampling rate can be 50 points per second, such that the interval between two adjacent sampling points can be 20 milliseconds (ms). Preprocessing signal 102S can be preprocessing signal 200c or 200n. Preprocessing signals 200c, 200n, and 102S (or signal 100S) can be periodic pulse waves and may contain multiple pulses. Preprocessing signals 200c and 200n can be continuous waveforms measured on the same subject at different times, such as... Figure 2 As shown, the waveform of preprocessing signal 200c is better than that of preprocessing signal 200n. The waveform of preprocessing signal 200n may be affected by interference and contain anomalies. Therefore, preprocessing signal 200c is more suitable for selecting pulse wave features, while only a portion of the waveform of preprocessing signal 200n is suitable for selecting pulse wave features. The signal feature selection device 10 of the present invention can remove abnormal pulse waveforms from preprocessing signal 200n and select pulse wave features from the good pulse waveforms in preprocessing signal 200n.

[0049] The regional extremum detection module 104 can use a window of variable length to search for regional extrema. The module 104 can initially use a window length (e.g., window length WL1), and then dynamically adjust the window length based on the found relatively low values ​​(e.g., sampling point P1) and relatively high values ​​(e.g., sampling point P2) to obtain sampling points for the relative minimum values ​​of the waveform (e.g., waveform troughs P3 and P4) and sampling points for the relative maximum values ​​(e.g., waveform peaks P5). In one embodiment, the window length WL1 can correspond to a typical heart rate, and the window length WL1 is, for example, 1 second, but is not limited thereto.

[0050] The waveform segmentation module 106 can segment a waveform using the positions of troughs (or peaks), and extract a pulse from a periodic pulse wave. For example, Figure 3This is a schematic diagram of the signal processing flow according to an embodiment of the present invention. Figure 3 As shown, the waveform segmentation module 106 can select pulses 106S1 to 106S5 from the preprocessing signal 102S, where the sampling rate can be 50 points per second, so that the interval between two adjacent sampling points can be 20 milliseconds.

[0051] In one embodiment, since the pulse durations 106L1 to 106L5 of the preprocessing signal 102S pulses 106S1 to 106S5 may not be of equal length (for example, the pulse duration 106L3 of pulse 106S3 may be longer than the pulse duration 106L4 of pulse 106S4), in order to facilitate subsequent processing, the waveform segmentation module 106 can adjust the pulses 106S1 to 106S5 to pulses 107S1 to 107S5 respectively, so that the pulses 107S1 to 107S5 each have the same normalized pulse duration 107L. In one embodiment, the waveform segmentation module 106 can make the pulse duration 107L of pulses 107S1 to 107S5 equal to the median of pulse durations 106L1 to 106L5, or other non-extremely high or non-extremely low pulse durations, such as the arithmetic mean, geometric mean, harmonic mean, or root mean square of pulse durations 106L1 to 106L5. In another embodiment, the pulse duration 107L of pulses 107S1 to 107S5 can be made equal to the mode of pulse durations 106L1 to 106L5.

[0052] In one embodiment, the waveform segmentation module 106 can normalize pulses 106S1 to 106S5 to adjust the pulse duration 106L1 to 106L5 of pulses 106S1 to 106S5. For longer pulses (e.g., pulse 106S3), the waveform segmentation module 106 can downsample (reduce the number of sampling points), and for shorter pulses (e.g., pulse 106S1), it can upsample (e.g., interpolation or linear interpolation), so that the pulse duration 107L of pulses 107S1 to 107S5 is equal in length (or pulses 107S1 to 107S5 have the same number of sampling points). In one embodiment, the waveform segmentation module 106 can use linear interpolation to extract two adjacent sampling points in a pulse (e.g., pulse 106S1) to calculate a linear equation, and can insert any number of sampling points that conform to the linear equation, so that the pulse duration of pulses 107S1 to 107S5 is consistent with the desired normalized pulse duration 107L. In one embodiment, the waveform segmentation module 106 can avoid downsampling to remove more important sampling points (e.g., peaks), and can remove less important sampling points.

[0053] Although the preprocessing module 102 can filter out high-frequency noise, there may still be noise with a frequency band similar to the heart rate (such as noise generated by exercise). Therefore, the present invention uses the template initialization module 108 or the template matching module 110 to screen each pulse of the periodic pulse wave one by one in order to select a good pulse waveform.

[0054] The template initialization module 108 can select a portion of pulses from multiple pulses to form a reference waveform sequence, and average the pulses of the reference waveform sequence to output a template pulse. The averaged template pulse can improve the stability and fault tolerance of the algorithm. When an undesirable waveform is incorrectly used to form the reference waveform sequence, averaging can reduce bias and distribute risk. For example, the template initialization module 108 can select pulse 107S3 (used as pulse 108f1) to form a reference waveform sequence with other selected pulses 108f2 to 108fn, and average pulses 108f1 to 108fn to obtain the template pulse 108v, making the waveform of the output template pulse 108v more ideal. Even though the waveform of pulse 108f1 is flatter near its peak compared to pulses 108f2 to 108fn, averaging the waveforms of pulses 108f1 to 108fn makes the sample pulse 108v representative of pulses 108f1 to 108fn. Therefore, the sample pulse 108v can be considered the standard pulse waveform of the test subject. Averaging the waveforms of pulses 108f1 to 108fn means averaging the amplitudes of pulses 108f1 to 108fn at each sampling point to obtain the amplitude of the sample pulse 108v at all sampling points.

[0055] In one embodiment, the template initialization module 108 can select pulses based on similarity. Based on the periodicity of the pulse wave, the template initialization module 108 can compare each previous pulse (e.g., pulse 107S2 selected in the previous sampling period) with the pulse to be judged (e.g., pulse 107S3 selected in the current sampling period). If the previous pulse (e.g., pulse 107S2) and the pulse to be judged (e.g., pulse 107S3) have a high similarity, it means that the waveform of the pulse to be judged is likely stable and representative, and therefore can be selected into the reference waveform sequence. In one embodiment, if the cross-correlation coefficient between the previous pulse (which may be called the first pulse) and the pulse to be judged (which may be called the second pulse) is greater than or equal to a similarity threshold (e.g., 0.9, but not limited thereto), then it can be determined that the previous pulse (e.g., pulse 107S2) and the pulse to be judged (e.g., pulse 107S3) are similar, and the pulse to be judged (e.g., pulse 107S3) can be added to the reference waveform sequence. A higher similarity threshold results in more accurate calculations, but requires the test subject to wear the device correctly, and the measurement waiting time will also be longer. In one embodiment, the similarity threshold can be 0.8 to 0.9 to provide a sufficient degree of screening.

[0056] In one embodiment, such as Figure 3 As shown, the reference waveform sequence may include pulses 108f1 to 108fn (which may be referred to as reference pulses). In one embodiment, the upper limit of the number of pulses in the reference waveform sequence can be set to a preset number of pulses, for example, equal to 5, i.e., n = 5, but not limited to this, the preset number of pulses may be between 3 and 20. When the number of pulses selected by the template initialization module 108 reaches, for example, 5, the template initialization module 108 can pause the comparison between the previous pulse and the pulse being judged, and average the 5 reference pulses constituting the reference waveform sequence, and output the template pulse 108v. The larger the number of pulses in the reference waveform sequence, the longer the measurement waiting time for the subject, therefore the number of pulses in the reference waveform sequence does not need to be too large.

[0057] In one embodiment, pulses 108f1 to 108fn are not necessarily consecutive pulses in the preprocessing signal 102S. Pulses 108f1 to 108fn have a high degree of similarity to the preceding or even earlier pulse in the preprocessing signal 102S, but pulses 108f1 to 108fn may not be consecutive pulses.

[0058] In one embodiment, the template initialization module 108 can compare the previous pulse (e.g., the pulse 107S2 selected in the previous sampling period) with the pulse to be judged (e.g., the pulse 107S3 selected in the current sampling period), or the template initialization module 108 can also compare an earlier pulse (e.g., pulse 107S1) with the pulse to be judged (e.g., pulse 107S3).

[0059] In one embodiment, the pulse durations of pulses 108f1 to 108fn and the sample pulse 108v are equal to the normalized pulse duration 107L. That is, the signal feature selection device 10 can calculate the cross-correlation coefficient after the pulse duration of the previous pulse (e.g., pulse 107S2) is equal to the pulse duration of the pulse to be judged (e.g., pulse 107S3) for comparison or subsequent averaging.

[0060] The template matching module 110 can compare each pulse (e.g., pulses 107S1 to 107S5) during the measurement period with the template pulse 108v. If the compared pulse has a high similarity to the template pulse 108v, feature extraction can be performed on the pulse. In one embodiment, the template matching module 110 can calculate the cross-correlation coefficient between the pulse (e.g., pulse 107S1) and the template pulse 108v. If the cross-correlation coefficient is greater than or equal to a matching threshold (e.g., 0.9, but not limited thereto), the pulse can be determined to match the template pulse 108v, and feature extraction can be performed on the pulse. In one embodiment, the matching threshold can be 0.8 to 0.9 to provide a sufficient degree of screening. For example, Figure 4 This is a schematic diagram of pulses 110i1 and 110i2 in an embodiment of the present invention. The template matching module 110 can compare pulses 110i1 and 110i2 with the template pulse 108v respectively. Since pulse 110i1 and template pulse 108v have a high similarity (e.g., the peaks are roughly aligned and the slopes are roughly similar), feature extraction can be performed on pulse 110i1.

[0061] If the template matching module 110 determines that the similarity between the pulse being compared and the template pulse 108v is low, it may not perform feature extraction on the pulse or may filter it out. For example, in Figure 4 The template matching module 110 determines that the similarity between pulse 110i2 and template pulse 108v is low, therefore feature extraction of pulse 110i2 is not required. For example, Figure 5This is a schematic diagram of pulses 110i3 to 110i5 in an embodiment of the present invention, wherein the sampling rate can be 50 points per second, allowing a 20-millisecond interval between two adjacent sampling points. Pulse 110i3 is an abnormally high spike, pulse 110i4 is a double peak caused by arrhythmia, and pulse 110i5 is a waveform with an excessively short systolic phase. In one embodiment, because pulses 110i3 to 110i5 (relative to pulses before or after them) have abnormal waveforms in a continuous waveform, their cross-correlation coefficient (or correlation coefficient) with the template pulse 108v (or the waveform of other pulses) is lower than the matching threshold and they are filtered out for feature extraction. The waveforms of other pulses may be used to select pulse wave features. In other words, the signal feature selection device 10 of the present invention can eliminate abnormal pulse waveforms and select pulse wave features from other good pulses in the signal.

[0062] The computation module 112 can use statistical or machine learning methods (such as linear regression or neural networks) to select parameters and build a model, for example, for blood pressure estimation. The computation module 112 may include a feature extraction unit 1122 and a computation unit 1124.

[0063] The feature extraction unit 1122 can perform feature extraction on certain pulses according to the instructions of the template matching module 110. For example, Figure 6 This is a schematic diagram of pulse 110i6 according to an embodiment of the present invention. The feature extraction unit 1122 can select features from the waveform of pulse 110i6 such as systolic pulse wave area (i.e., the sum of areas S1 and S2), diastolic pulse wave area (i.e., the sum of areas S3 and S4), pulse wave area (i.e., the sum of areas S1 to S4), heart rate, systolic time St1, diastolic time Dt1, maximum pulse wave amplitude, minimum pulse wave amplitude, maximum slope during systole, ratio of maximum amplitude to minimum amplitude, or heart rate variability.

[0064] In one embodiment, the subject's pulse wave may not have a dicrotic wave, therefore the feature extraction unit 1122 may not select the dicrotic wave feature from the waveform of pulse 110i6. A dicrotic wave is primarily formed when the aortic valve suddenly closes during early ventricular diastole, causing blood to flow backward and impact the aorta, resulting in a rebound that raises aortic pressure. The dicrotic wave can be used as an alternative measure of pulse transit time (PTT), but diastolic time Dt1 can also be used as an alternative measure.

[0065] The calculation unit 1124 can calculate a physiological index based on the features selected by the feature extraction unit 1122. The calculation unit 1124 can use these features to build a model and use this model to calculate the physiological index.

[0066] In one embodiment, the computing unit 1124 can use linear regression to estimate blood pressure. Linear regression is a regression model that uses known blood pressure and corresponding features to train an equation that minimizes error (e.g., calculates the slope and intercept). This equation can be used to approximate the trend distribution of data points to predict blood pressure. In one embodiment, blood pressure measured by a cuff-style blood pressure monitor can be used as known blood pressure, and features selected from waveforms measured by autoscopy can be used to train the regression model. In one embodiment, the equation can be Y = α0 + α1x1 + α2x2 + ... + α n x n Where blood pressure is denoted as Y, and the feature labels input to the calculation unit 1124 are x1 to x2. n , α1~α n This represents the coefficients of the equation obtained under the least square error.

[0067] In one embodiment, the calculation unit 1124 may select certain features through repeated trials to minimize the error of the linear regression. In one embodiment, the features selected by the calculation unit 1124 are pulse transit time and heart rate. The relationship between blood pressure, pulse transit time, and heart rate can be expressed as BP = aPTT + bHR + c, where BP represents blood pressure, PTT represents pulse transit time, and HR represents heart rate. Pulse transit time can be determined from the time difference between the peak of an electrocardiogram (ECG) waveform and the peak of an optometry waveform. Since the peak of an ECG waveform is related to ventricular contraction, while the peak of an optometry waveform is related to vascular contraction, pulse transit time (i.e., the transit time of blood from the heart to the measured vessel) is related to blood pressure. In one embodiment (e.g., without ECG information), diastolic time can be used instead of pulse transit time; in other words, the features selected by the calculation unit 1124 are diastolic time and heart rate. The estimated relationship between blood pressure, diastolic time, and heart rate can be: in This represents the estimated blood pressure, and Dt represents the diastolic time.

[0068] In one embodiment, the calculation unit 1124 can establish prediction models for diastolic and systolic blood pressure respectively. In one embodiment, the calculation unit 1124 can obtain a set of linear equations corresponding to coefficients a, b, and c for diastolic blood pressure, and another set of linear equations corresponding to coefficients a, b, and c for systolic blood pressure. By substituting the features into the corresponding linear equations, the estimated diastolic or systolic blood pressure can be obtained.

[0069] In one embodiment, the computing unit 1124 can utilize machine learning to find potential correlations between features and blood pressure. During the training phase, the computing unit 1124 can input known first data (e.g., features selected by the feature extraction unit 1122 that have a corresponding known blood pressure) into an untrained model and compare the model's output with the known blood pressure of the first data. This allows for re-evaluation and optimization of the model's parameters to train the model and minimize errors. During the inference or prediction phase, the computing unit 1124 can apply information from the trained model to infer results. Accordingly, when unknown second data to be interpreted (e.g., features selected by the feature extraction unit 1122 that do not have a corresponding known blood pressure) is input into the model, the model can infer from the second data based on its optimized parameters to output a prediction.

[0070] The output module 114 can output physiological indicators (such as systolic or diastolic blood pressure), such as outputting images on a screen, flashing lights, or sound through a speaker.

[0071] In one embodiment, if the measurement time exceeds a preset time length (e.g., more than 20 seconds, but not limited thereto) and a template pulse 108v still cannot be established (e.g., the number of reference pulses in the reference waveform sequence cannot reach the preset number of pulses within the preset time length), or if the number of pulses matching the template pulse 108v is less than a threshold value (e.g., less than 4 pulses, but not limited thereto), the output module 114 can output a remeasurement message to remind the subject to remeasure blood pressure, and can remind the subject to relax, remain still, or adjust the wearing position to remeasure. Accordingly, the signal feature selection device 10 can re-receive a signal from the ear-hook physiological measurement device.

[0072] In one embodiment, the input module 100, preprocessing module 102, region extremum detection module 104, waveform segmentation module 106, template initialization module 108, template matching module 110, calculation module 112, feature extraction unit 1122, calculation unit 1124, or output module 114 may include / correspond to a circuit. In one embodiment, the connection method, order, or number of the input module 100, preprocessing module 102, region extremum detection module 104, waveform segmentation module 106, template initialization module 108, template matching module 110, calculation module 112, feature extraction unit 1122, calculation unit 1124, and output module 114 can be adaptively adjusted.

[0073] Furthermore, the calculation method for the cross-correlation coefficient is detailed below. First, the cross-correlation coefficient ρ XY [n] represents the cross-correlation coefficient between the previous pulse Y (e.g., pulse 107S2) and the pulse X being judged (e.g., pulse 107S3), which satisfies Among them, X NORM This represents the normalized pulse being judged (e.g., pulse 107S3), Y NROM This indicates the previous pulse after normalization (e.g., pulse 107S2).

[0074] To calculate the similarity ratio, it is necessary to compare the maximum possible cross-correlation coefficient (i.e., the autocorrelation coefficient max(ρ) of the previous pulse (e.g., pulse 107S2)). YY [n]), which satisfies Cross-correlation coefficient SQI XCORR Can be defined as The cross-correlation coefficient between the pulse X being judged (e.g., pulse 107S3) and the previous pulse Y (e.g., pulse 107S2) is compared with the autocorrelation coefficient of the previous pulse Y (e.g., pulse 107S2). The closer the value is to 1, the more similar they are; conversely, the less similar they are, the less similar they are.

[0075] As can be seen from the above, if the cross-correlation coefficient between the previous pulse and the pulse to be judged is greater than the similarity threshold, the occurrence of poor waveforms being analyzed can be reduced. Therefore, the present invention can reduce the false judgment rate.

[0076] The signal feature selection device 10 of the present invention is merely an embodiment of the present invention, and those skilled in the art can make different changes and modifications accordingly. For example, the above embodiment is mainly described using blood pressure, but in other embodiments, other physiological indicators such as blood oxygen concentration can also be calculated. The above embodiment is mainly described using the signal 100S sensed by photoplethysmography, but in other embodiments, the signal feature selection device 10 can also process signals measured by various contact sensors (e.g., electrocardiogram), signals measured by various non-contact sensors (e.g., radar sensor), or other various signals.

[0077] Figure 7 This is a flowchart of a method 70 according to an embodiment of the present invention. Method 70 may include the following steps:

[0078] Step S700: Begin.

[0079] Step S702: Receive a signal 100S, wherein the signal 100S contains a plurality of pulses.

[0080] Step S704: Determine whether a second pulse among the plurality of pulses is similar to a first pulse among the plurality of pulses.

[0081] Step S706: After the second pulse is determined to be similar to the first pulse, the second pulse is selected as one of a plurality of reference pulses in a reference waveform sequence.

[0082] Step S708: Average the plurality of reference pulses into a single plate pulse.

[0083] Step S710: Determine whether each of the multiple pulses matches the template pulse.

[0084] Step S712: Select a plurality of features from each pulse that matches the template pulse from the plurality of pulses.

[0085] Step S714: End.

[0086] In one embodiment, the second pulse follows the first pulse. A start point or an end point of the first pulse may or may not coincide with an end point or a start point of the second pulse. In one embodiment, an end point of the first pulse may coincide with a start point of a pulse series, and an end point of the pulse series may coincide with a start point of the second pulse, the pulse series comprising a plurality of consecutively connected pulses.

[0087] In one embodiment, after determining whether the second pulse is similar to the first pulse, it is determined whether a third pulse among the plurality of pulses is similar to the second pulse. In another embodiment, it is determined one by one whether any pulse among the plurality of pulses is similar to another pulse among the plurality of pulses until the number of the plurality of reference pulses in the reference waveform sequence reaches a preset number of pulses.

[0088] In one embodiment, one of the plurality of features is a first time length between a first trough and a peak of one of the plurality of pulses, a second time length between the peak and a second trough of the pulse, the reciprocal of a third time length between the first trough and the second trough, the amplitude of the peak of the pulse, or a maximum slope between the first trough and the peak, but is not limited thereto.

[0089] Figure 8 This is a flowchart of a method 80 according to an embodiment of the present invention. Method 80 may include the following steps:

[0090] Step S800: Start.

[0091] Step S802: Perform preprocessing on signal 100S.

[0092] Step S804: Search for the regional extreme values ​​of the preprocessed signal 102S after preprocessing.

[0093] Step S806: Based on the regional extreme values, segment the waveform of the preprocessing signal 102S to select multiple pulses.

[0094] Step S808: Determine whether the previous pulse is similar to the pulse being judged. If yes, proceed to step S810; if no, replace the pulse being judged with another pulse and proceed to step S808 again.

[0095] Step S810: Add the pulse to be judged to the reference waveform sequence, and determine whether the number of pulses in the reference waveform sequence is equal to a preset number of pulses. If yes, proceed to step S812; if no, proceed to step S822.

[0096] Step S812: Average the pulses of the reference waveform sequence to 108V for the sample pulse.

[0097] Step S814: Determine whether a pulse during the measurement period matches the sample pulse 108V. If yes, proceed to step S816; if no, replace the pulse being determined with another pulse and proceed to step S814 again.

[0098] Step S816: Determine whether the number of pulses matching the template pulse 108V is greater than a threshold value. If yes, proceed to step S818; otherwise, proceed to step S824.

[0099] Step S818: Perform feature extraction on the pulse that matches the sample pulse 108V.

[0100] Step S820: Calculate physiological indicators using the selected features.

[0101] Step S822: Determine whether the number of pulses in the reference waveform sequence exceeds a preset time length or is still less than a preset number of pulses. If yes, proceed to step S824; otherwise, proceed to step S808.

[0102] Step S824: Remind the subject to repeat the measurement.

[0103] Step S826: End.

[0104] Method 70 or 80 can be used Figure 1 The signal feature selection device 10. Method 70 or 80 can be compiled into code and executed by a processing circuit, and stored in a storage circuit. One or more of the steps S702 to S712 in method 70 or steps S802 to S824 in method 80 can be selectively omitted. Furthermore, the order of one or more of the steps S702 to S712 in method 70 or steps S802 to S824 in method 80 can be interchanged. The quality of the preprocessed signal 102S after preprocessing can also be checked before step S804.

[0105] In summary, the signal feature selection device of the present invention can perform signal processing before feature extraction, such as screening waveforms measured by the photoplethysmography (PPG) module. The present invention can establish waveform templates based on the similarity of continuous periodic pulse waves, and use these templates to screen signal waveforms, ensuring that the selected waveforms are of good quality. This guarantees that the waveform features used for subsequent blood pressure estimation are correct, thereby improving the accuracy of blood pressure measurement using PPG. The present invention can utilize the PPG module and the signal feature selection device to monitor whether blood pressure is rising or falling.

[0106] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be included in the scope of the present invention.

Claims

1. A method for selecting physiological signal features, characterized in that, Includes: Receive a signal, wherein the signal contains a plurality of pulses; Determine whether a second pulse among the plurality of pulses is similar to a first pulse among the plurality of pulses, wherein the second pulse follows the first pulse, and a start point or an end point of the first pulse coincides with an end point or a start point of the second pulse; After the second pulse is determined to be similar to the first pulse, the second pulse is selected as one of a plurality of reference pulses in a reference waveform sequence, and it is determined whether a third pulse in the plurality of pulses is similar to the second pulse. The multiple reference pulses are averaged into a single plate pulse; Determine whether each of the multiple pulses matches the template pulse; as well as From each of the plurality of pulses that matches the template pulse, a plurality of features are selected.

2. The physiological signal feature selection method according to claim 1, characterized in that, Also includes: Based on the duration of the multiple pulses, a normalized pulse duration is calculated. as well as By using downsampling or upsampling, the duration of each of the multiple pulses is adjusted to the duration of the normalized pulse. The normalized pulse duration is either the median or the mode of the multiple pulse durations.

3. The method for selecting physiological signal features according to claim 1, characterized in that, The step of determining whether the second pulse is similar to the first pulse includes: Calculate a first cross-correlation coefficient between the second pulse and the first pulse; Based on the fact that the first cross-correlation coefficient is greater than or equal to a similarity threshold, it is determined that the first pulse is similar to the second pulse; as well as Based on the fact that the first cross-correlation coefficient is less than the similarity threshold, it is determined that the first pulse and the second pulse are not similar.

4. The method for selecting physiological signal features according to claim 1, characterized in that, Determine whether any pulse in the plurality of pulses is similar to another pulse in the plurality of pulses until the number of the plurality of reference pulses in the reference waveform sequence reaches a preset number of pulses.

5. The method for selecting physiological signal features according to claim 1, characterized in that, Determining whether each of the multiple pulses matches the template pulse includes: Calculate a second cross-correlation coefficient between one of the complex pulses and the template pulse; Based on the fact that the second cross-correlation coefficient is greater than or equal to a matching threshold, it is determined that the pulse matches the template pulse; and If the second cross-correlation coefficient is less than the matching threshold, it is determined that the pulse does not match the template pulse.

6. The method for selecting physiological signal features according to claim 1, characterized in that, One of the plurality of features is a first time length between a first trough and a peak of the plurality of pulses, a second time length between the peak and a second trough of the pulse, the reciprocal of a third time length between the first trough and the second trough, the amplitude of the peak of the pulse, or the maximum slope between the first trough and the peak.

7. The method for selecting physiological signal features according to claim 1, characterized in that, Also includes: To build a model using these multiple features; and This model is used to calculate a physiological indicator. The model is trained using machine learning to optimize at least one of its parameters.

8. The method for selecting physiological signal features according to claim 1, characterized in that, If it is determined that the number of pulses matching the template pulse in the plurality of pulses is less than a threshold value, or the number of the plurality of reference pulses in the reference waveform sequence cannot reach a preset number of pulses within a preset time length, another signal is required to be transmitted.

9. A physiological signal feature selection device, characterized in that, Includes: A processing circuit used to execute a piece of code; as well as A storage circuit, coupled to the processing circuit, is used to store the code, wherein the code includes: Receive a signal, wherein the signal contains a plurality of pulses; Determine whether a second pulse among the plurality of pulses is similar to a first pulse among the plurality of pulses, wherein the second pulse follows the first pulse, and a start point or an end point of the first pulse coincides with an end point or a start point of the second pulse; After the second pulse is determined to be similar to the first pulse, the second pulse is selected as one of a plurality of reference pulses in a reference waveform sequence, and it is determined whether a third pulse in the plurality of pulses is similar to the second pulse. The multiple reference pulses are averaged into a single plate pulse; Determine whether each of the multiple pulses matches the template pulse; as well as From each of the plurality of pulses that matches the template pulse, a plurality of features are selected.

10. The physiological signal feature selection device according to claim 9, characterized in that, The code also includes: Based on the duration of the multiple pulses, a normalized pulse duration is calculated. as well as By using downsampling or upsampling, the duration of each of the multiple pulses is adjusted to the duration of the normalized pulse. The normalized pulse duration is either the median or the mode of the multiple pulse durations.

11. The physiological signal feature selection device according to claim 9, characterized in that, The step of determining whether the second pulse is similar to the first pulse includes: Calculate a first cross-correlation coefficient between the second pulse and the first pulse; Based on the fact that the first cross-correlation coefficient is greater than or equal to a similarity threshold, it is determined that the first pulse is similar to the second pulse; as well as Based on the fact that the first cross-correlation coefficient is less than the similarity threshold, it is determined that the first pulse and the second pulse are not similar.

12. The physiological signal feature selection device according to claim 9, characterized in that, Determine whether any pulse in the plurality of pulses is similar to another pulse in the plurality of pulses until the number of the plurality of reference pulses in the reference waveform sequence reaches a preset number of pulses.

13. The physiological signal feature selection device according to claim 9, characterized in that, Determining whether each of the multiple pulses matches the template pulse includes: Calculate a second cross-correlation coefficient between one of the complex pulses and the template pulse; Based on the fact that the second cross-correlation coefficient is greater than or equal to a matching threshold, it is determined that the pulse matches the template pulse; and If the second cross-correlation coefficient is less than the matching threshold, it is determined that the pulse does not match the template pulse.

14. The physiological signal feature selection device according to claim 9, characterized in that, One of the plurality of features is a first time length between a first trough and a peak of the plurality of pulses, a second time length between the peak and a second trough of the pulse, the reciprocal of a third time length between the first trough and the second trough, the amplitude of the peak of the pulse, or the maximum slope between the first trough and the peak.

15. The physiological signal feature selection device according to claim 9, characterized in that, The code also includes: To build a model using these multiple features; and This model is used to calculate a physiological indicator. The model is trained using machine learning to optimize at least one of its parameters.

16. The physiological signal feature selection device according to claim 9, characterized in that, If it is determined that the number of pulses matching the template pulse in the plurality of pulses is less than a threshold value, or the number of the plurality of reference pulses in the reference waveform sequence cannot reach a preset number of pulses within a preset time length, another signal is required to be transmitted.

Images