Apparatus and method for determining respiration rate of a subject
By analyzing the first harmonic phase of the heartbeat in the unfiltered PPG signal, and combining Fourier transform and resampling techniques, the problem of inaccurate respiratory rate measurement was solved, and a higher confidence level of respiratory rate determination was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KONINKLIJKE PHILIPS NV
- Filing Date
- 2021-10-20
- Publication Date
- 2026-06-09
AI Technical Summary
In existing technologies, respiratory rate measurement is not precise enough, especially since respiratory rate information is unclear in raw PPG signals, making it difficult to determine accurately.
By analyzing the most meaningful features in the unfiltered PPG signal, such as the first harmonic phase of the heartbeat, and combining Fourier transform and resampling techniques, characteristic signals describing the heartbeat phase sequence are extracted, and respiratory rate is determined using neural networks or conventional data processing methods.
It improves the accuracy and confidence of respiratory rate measurement, enabling more accurate determination of the subject's respiratory rate.
Smart Images
Figure CN116367774B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an apparatus, system, and method for determining the respiratory rate of a subject. Background Technology
[0002] Human vital signs are commonly used as indicators of a person's current condition and as strong predictors of serious medical events. In particular, the respiratory system is known to readily adapt to demands placed upon it. For this reason, a subject's respiratory rate can be used as an indicator of clinical deterioration and physiological conditions such as hypoxia, hypercapnia, and metabolic or respiratory acidosis. Therefore, accurate measurement of respiratory rate can facilitate early intervention.
[0003] Various methods have been developed to allow for accurate measurement of respiratory rate. Among other methods, photoplethysmography (PPG) can be used to determine the respiratory rate of an object.
[0004] PPG (Profilometry) is an optical measurement technique that evaluates changes in the temporal variation of light reflection or transmission in a region or volume of interest. PPG is based on the principle that blood absorbs more light than surrounding tissues, therefore changes in the blood volume with each heartbeat correspondingly affect transmission or reflection. Thus, the PPG waveform includes information about heart rate, but information attributable to respiration, particularly respiratory rate, can also be inferred from the PPG signal. (Arterial) oxygen saturation, often referred to as SpO2, can also be determined by evaluating transmittance and / or reflectance at different wavelengths (typically red and infrared).
[0005] PPG and calculating SpO2 values from PPG signals are existing techniques and have been used for decades. In earlier studies, it was observed that respiratory rate was often visible in the raw PPG signal, but not always clearly visible in the so-called plethysmography signal (which is the raw wave signal after inversion and filtering).
[0006] Many different methods have been developed to obtain a more accurate determination of respiratory rate.
[0007] For example, US 2016 / 0051205 A1 discloses a system and method for determining multi-parameter confidence in respiratory rate measurements using multiple physiological parameter inputs. The disclosed system and method are based on taking the raw PPG waveform as input. Summary of the Invention
[0008] The object of this invention is to provide an apparatus, system, and method for determining the respiratory rate of a subject. In particular, the object is to determine the respiratory rate based on a PPG signal with a high confidence level.
[0009] In a first aspect of the invention, as described in claim 1, an apparatus for determining the respiratory rate of a subject is presented.
[0010] In another aspect of the invention, a system for determining the respiratory rate of an object is presented, the system comprising:
[0011] - A PPG sensor configured to provide red and / or infrared light volumetric PPG signals, and
[0012] - A device used to determine the breathing rate of an object.
[0013] In another aspect of the invention, a corresponding method and computer program are provided, comprising a program code module for causing a computer to perform the steps of the method disclosed herein when the computer program is executed on the computer, and a non-transient computer-readable recording medium therein storing a computer program product that, when run by a processor, causes the execution of the method disclosed herein.
[0014] Preferred embodiments of the invention are defined in the dependent claims. It should be understood that the claimed methods, systems, computer programs, and media have preferred embodiments similar to and / or identical to the claimed devices, particularly as defined in the dependent claims and as disclosed herein.
[0015] This invention is based on the idea of using the most meaningful features of the PPG signal (particularly in the raw wave (PPG data), i.e., not in the plethysmography signal (which is an inverted and filtered raw wave signal)) to determine the respiratory rate of a subject. Because the raw wave is unfiltered, it typically contains more (useful) information than the plethysmography signal. Unexpectedly, the inventors discovered that extracting a signal from the (raw) PPG signal describing the phase of the first harmonic of the heartbeat relative to the start of said heartbeat can serve as a reliable basis for determining the respiratory rate of a subject. Therefore, the first feature can represent the phase position of the corresponding heartbeat signal relative to the start of said heartbeat. Thus, the first feature signal can provide an indication of how the first feature (i.e., the phase position of said first harmonic) changes over time. The inventors have discovered that changes in phase position over time can serve as a reliable basis for determining the respiratory rate of a subject.
[0016] In the first step, the device is configured to analyze the red and / or infrared PPG signals obtained from the input interface and locate the pulse trigger, i.e., the starting point of the pulse / heartbeat. Common algorithms can be used to do this. Preferably, the validity of each heartbeat is analyzed and determined based on the individual's pulsation history and current perfusion, pulse, and / or SpO2 values.
[0017] For all (valid) pulsations, all pulsation samples from the start to the end of the pulsation can be examined. Optionally, offset or drift removal can be applied, for example, by subtracting the fitted line through the first and last samples (i.e., the straight line) of the pulsation. The Fourier transform of the pulsation can then be analyzed. The Fourier transform is calculated, for example, by the feature signal unit. The phase and pulse start can be determined on a per-pulse basis. The feature signal unit then extracts the phase of the first harmonic of the corresponding pulsation associated with the start of the pulsation for all pulsations, thereby generating a phase sequence of the first harmonic associated with the start of the pulsation. This feature sequence generates a first feature signal. It is extracted from the PPG signal by the feature signal unit.
[0018] Since the first feature signal describes the phase sequence of the first harmonic of the heartbeat relative to the start of these heartbeats, its sampling location depends on the (variable) sequence of the object's heart. In fact, for each time point where a heartbeat is detected, there is one measurement, i.e., one measurement per heartbeat. Naturally, the heartbeat frequency is not constant, and so is the sampling location, and therefore the sampling rate. To obtain a constant sampling rate, the device may include a resampling unit configured to resample the first feature signal. Resampling produces a feature-time signal with a constant sampling rate (i.e., the phase of the first harmonic of the object's heartbeat relative to the start of the heartbeat). Thus, for example, resampling facilitates subsequent Fourier transforms. Preferably, the first feature signal is resampled using a resampling function. Subsequently, the extraction unit extracts the first respiratory rate from the transformed, resampled signal, i.e., the resampled first feature signal corresponding to the transformation of the phase sequence of the first harmonic of the heartbeat relative to the start of the heartbeat.
[0019] In one embodiment, the device further includes an analysis unit, wherein the extraction unit is configured to extract at least one additional first respiratory rate from the transformed resampled signal, and wherein the analysis unit is configured to determine the most probable respiratory rate based on the two first respiratory rates.
[0020] Typically, the respiratory rate can be calculated using different calculation methods. Therefore, the at least two first respiratory rates can be different. The analysis unit then determines the most probable respiratory rate, i.e., the respiratory rate closest to the subject's true respiratory rate, based on the respiratory rates already calculated in different ways. The analysis unit can be a conventional data processing unit. However, the analysis unit can also be a neural network. The neural network can be trained using training data, which can be previously processed data or simulated data.
[0021] In an embodiment, the feature signal unit is configured to extract a second feature sequence from the PPG signal and generate a second feature signal based on the second feature sequence, wherein the extraction unit is configured to extract at least a second respiratory rate from the second feature signal, wherein the second feature signal is based on any one of the following: (a) the amplitude of the processed PPG signal at the fundamental frequency of the heartbeat of the subject multiplied by the pulse width of the heartbeat, (b) the amplitude of the processed PPG signal at the fundamental frequency of the heartbeat, (c) single pulse perfusion, (d) pulse rate, (e) the amplitude ratio of the fundamental frequency to the first harmonic of the pulsation, (f) pulse amplitude variability, (g) pulse width variability, and (h) the surface of the pulsation.
[0022] Preferably, the second feature sequence comprises the amplitude sequence of the fundamental frequency of the object's heartbeat multiplied by the pulse width of the beat. To extract the amplitude at the fundamental frequency of the heartbeat, all raw wave samples (infrared or red) from the beginning to the end of a single beat are acquired. As described above, offset or drift removal can be applied, for example by subtracting a straight line, such as a fit through the first and last samples of the beat. Subsequently, the amplitude at the fundamental frequency of the beat is obtained by calculating, for example, a Fourier transform of a single beat through the feature signal unit. In this method, the number of points in the Fourier transform varies with the length of the pulse. For example, for a sampling rate of 125 Hz and an effective pulse rate range between 29.3 and 300 beats per minute (bpm), the number of samples per beat varies between 25 and 256 samples.
[0023] The table below presents results based on experimental data from several studies (2019, 2014, and 2018) regarding which different characteristic signals, for example, provide the most promising results in determining the respiratory rate of subjects. In Study 3 (2018), nearly 500 documents were analyzed. The criteria used to determine which characteristic signals provide the best results are NPI = a + b and NPI2 = 2 * a + b, with NPI2 being the better of the two. Generally, smaller NPI or NPI2 is better. The parameter “a” increases as the difference between the determined respiratory rate and the reference value increases, while the parameter “b” takes a fixed value if the algorithm cannot determine the respiratory rate, and is otherwise zero. Since displaying incorrect values is worse than not displaying any values, the “a” part is emphasized in NPI2. It is shown in the last column of the table. The lower the corresponding number, the better the result, i.e., the closer the determined most probable respiratory rate is to the true respiratory rate.
[0024]
[0025]
[0026]
[0027] In the table above, "PAV" refers to pulse amplitude variability, "PWV" refers to pulse width variability, "amp" refers to the amplitude at the fundamental frequency, "perf" refers to the single pulse perfusion, "phase" refers to the phase of the first harmonic of the pulsation, and "surface" refers to the surface of the pulsation, i.e., the area spanned between the original PPG signal and the straight line (not necessarily a horizontal line) from the maximum value of the pulsation to the maximum value of the next pulsation (see [reference]). Figure 13 ).
[0028] As can be seen, in all studies, the combination of the first harmonic of the subject's heartbeat relative to the phase at the start of the heartbeat and the amplitude at the fundamental frequency of the subject's heartbeat multiplied by the pulse width (i.e., pulse length) of the pulsation provided the best results.
[0029] However, the second characteristic signal can also be based on a single pulse perfusion, which is calculated based on a combination of red and infrared PPG signals. For greater accuracy, similar to patent WO 2006 / 097866 A1, but based on the red and infrared amplitudes of the fundamental frequency of a single pulse, the single pulse perfusion is calculated.
[0030] A pulse rate sequence can also be used as a second characteristic signal. Preferably, a person's pulse can be extracted based on an infrared PPG signal.
[0031] The ratio of the amplitude of the fundamental frequency to the amplitude of the first harmonic of the pulsation refers to the ratio of the amplitude of the processed PPG signal at the fundamental frequency and the amplitude at the first harmonic of the pulsation.
[0032] The pulse amplitude can be calculated by subtracting the minimum value of the pulse for each pulse from the maximum value of the pulse. The variability of the pulse amplitude can be used to determine the variability in the pulse amplitude.
[0033] Similarly, pulse width variability can be determined by comparing the widths of the pulses in the sequence. The width of a single pulse is determined by calculating the distance between the start and end of the pulse.
[0034] The pulsating surface is defined as the region spanned between the original PPG signal and a straight line (not necessarily a horizontal line) from the maximum value of the pulsation to the maximum value of the next pulsation (see [link to relevant documentation]). Figure 13 ).
[0035] In this embodiment, the analysis unit can be configured to determine the most probable respiratory rate based on two first respiratory rates (and at least one second respiratory rate). However, another first or second respiratory rate can be determined, and therefore its most probable respiratory rate can also be determined.
[0036] In another embodiment, the device further includes a computing unit configured to calculate the first derivative of the first feature signal. Similarly, the first derivative of the second feature signal can be calculated.
[0037] Regarding the resampling unit, the second feature signal or the derivative of the second feature signal can be resampled using the resampling unit. The extraction unit can be configured to extract the first respiratory rate from the resampled first derivative.
[0038] In another embodiment, the device further includes a filter unit configured to filter the resampled signal to generate a filtered resampled signal, wherein the extraction unit is configured to extract the first respiratory rate from the filtered resampled signal, and / or to filter the first derivative of the resampled signal to generate a filtered first derivative of the resampled signal, wherein the extraction unit is configured to extract the first respiratory rate from the filtered first derivative of the resampled signal. Similarly, the resampled second feature signal and / or the first derivative of the resampled second feature signal can be filtered, and a second respiratory rate can be determined.
[0039] In a preferred embodiment of the device, the filter unit is configured to adaptively filter based on an expected respiratory rate corresponding to the resampled signal and / or the first derivative of the resampled signal. For example, an adaptive low-pass filter may be applied to the first characteristic signal (and / or the second characteristic signal). The cutoff frequency of the low-pass filter may be at the expected respiratory rate plus 5 breaths per minute, for example, wherein the expected respiratory rate is preferably derived by the extraction unit based on the (unfiltered) amplitude of the processed PPG at the fundamental frequency of the subject's heartbeat multiplied by the pulse width of the heartbeat. However, the cutoff frequency of the low-pass filter may also be higher or lower.
[0040] The transform unit can be configured to calculate the Fourier transform of the resampled second feature signal (and / or the Fourier transform of the first derivative of the resampled second feature signal), wherein the extraction unit is configured to extract the second respiratory rate from the signal(s). Preferably, the extraction unit comprises two subunits, wherein the first subunit is configured to extract the respiratory rate from the signal in the time domain, and wherein the second subunit is configured to extract one or more respiratory rates from the signal in the frequency domain.
[0041] In another embodiment, the device further includes a scoring unit configured to assign a score describing the probability that the first respiratory rate represents the object's true respiratory rate to the at least two first respiratory rates, wherein the analysis unit is configured to use the score to determine the most probable respiratory rate. Similarly, the scoring unit may also be configured to assign a score to the second respiratory rate, wherein the analysis unit is configured to separately compare the scores corresponding to the first and second respiratory rates, or to compare the score for the first respiratory rate with the score for the second respiratory rate to determine the most probable respiratory rate.
[0042] In another embodiment, the device includes a grouping unit configured to group the at least two first respiratory rates into groups of similar respiratory rates, wherein similarity is assumed if the respiratory rate difference is less than a predetermined rate difference, specifically wherein the predetermined rate difference does not exceed two breaths per minute. However, the predetermined respiratory rate difference can also be higher or lower. If no group exists within a given predetermined rate difference window, a new group is created. The output of the grouping unit has at least two groups. Similarly, the grouping unit can be configured to also group the second respiratory rate into such groups (alone or mixed with the first respiratory rates).
[0043] In a preferred embodiment of the device, the grouping unit is further configured to sum the scores of at least two first respiratory rates in a group to generate a group score, wherein the analysis unit is configured to determine the most probable respiratory rate based on the group with the highest group score. Preferably, the grouping unit generates at least two groups (each feature signal) and determines their corresponding group scores. The two groups with the highest scores are then provided to the analysis unit, which is configured to determine the most probable respiratory rate based on the groups. However, the grouping unit may also be configured to sum the scores of second respiratory rates in a group, or, if mixed groups exist, to sum the scores of the first and second respiratory rates of the group to generate a group score, and the analysis unit may be configured to determine the most probable respiratory rate based on the group with the highest group score.
[0044] In another preferred embodiment of the device, the analysis unit is further configured to determine a confidence value for the most probable respiratory rate based on the number of respiratory rates and the group score of the corresponding group (i.e., the group to which the most probable respiratory rate belongs). Specifically, the confidence value is based on a comparison of the number of respiratory rates and the group score in the group of the most probable respiratory rate with the number of respiratory rates and the group score in a second group (or even another group). For greater precision, the confidence value is based on the difference between group size and group score. If the scores and / or sizes are similar between different groups, the resulting confidence value is small. However, if one group is dominant, the confidence value is high. The confidence value (or other values calculated therefrom, such as signal quality indices) can then be used by clinical staff as an indicator of how reliable the most probable respiratory rate is. For example, even if the respiratory rate given is the most probable, its confidence may be low. In this case, the most probable respiratory rate does not reliably represent the true respiratory rate.
[0045] The analysis unit can also determine the second most likely respiratory rate and the third most likely respiratory rate, etc. (either based solely on the group score or also based on the number of respiratory rates in the group).
[0046] In another embodiment, the first feature signal covers a time span of at least 60 seconds. However, the time span can also be shorter and can be dynamically adjusted depending on the stability of the signal. For example, a shorter time span can be specifically used to obtain a higher score for a particular respiratory rate. Alternatively, the first feature signal may not cover a specific time span, but rather a specific (predetermined) number of heartbeats. This can also be applied to the second feature signal.
[0047] All units of the device mentioned above can be used to process the first feature signal, but can also be used to process the second feature signal in the same way. In fact, additional feature signals can be extracted from the PPG signal by the feature signal unit; these additional feature signals describe the same features as the first and / or second feature signals, or other features. Attached Figure Description
[0048] These and other aspects of the invention will become apparent from, and are illustrated with reference to, the embodiments described below. In the following figures:
[0049] Figure 1 A schematic diagram of a first embodiment of a device for determining the respiratory rate of an object according to the present invention is shown;
[0050] Figure 2A graph of a pulsation-to-pulsation signal, derived according to the present invention based on infrared and / or red PPG signals, is shown.
[0051] Figure 3A A schematic diagram of a second embodiment of a device for determining the respiratory rate of an object according to the present invention is shown;
[0052] Figure 3B A schematic diagram of a third embodiment of a device for determining the respiratory rate of an object according to the present invention is shown;
[0053] Figure 4 A schematic diagram of a fourth embodiment of a device for determining the respiratory rate of an object according to the present invention is shown;
[0054] Figure 5 A first schematic diagram of a method for determining the respiratory rate of an object according to the present invention is shown;
[0055] Figure 6 A second schematic diagram of a method for determining the respiratory rate of an object according to the present invention is shown;
[0056] Figure 7 A schematic diagram of a system for determining the respiratory rate of an object according to the present invention is shown;
[0057] Figure 8A and Figure 8B Each shows a diagram of the (second) characteristic signal according to the present invention;
[0058] Figure 9A and Figure 9B Each diagram illustrates a characteristic signal according to the present invention;
[0059] Figure 10 A first plot of the amplitude spectrum of the Fourier transform of the resampled characteristic signal according to the present invention is shown;
[0060] Figure 11 A graph of pulse amplitude variability (PAV) in the time domain according to the present invention is shown;
[0061] Figure 12 A second plot of the amplitude spectrum of the Fourier transform of the processed PPG signal corresponding to a single beat, according to the present invention, is shown; and
[0062] Figure 13 A graph illustrating a pulsation-to-pulsation signal derived according to the invention, based on infrared and / or red PPG signals showing different characteristics of the pulsation. Detailed Implementation
[0063] Figure 1A schematic diagram of a first embodiment of a device 10 for determining the respiratory rate of a subject according to the present invention is shown. The device 10 includes an input interface 12, a feature signal unit 14, an extraction unit 16, and optionally an analysis unit 18. The input interface 12 is configured to acquire, for example, red and / or infrared PPG signals 11 from a PPG sensor attached to the patient. The feature signal unit 14 analyzes the acquired PPG signals and extracts a first feature sequence from the signals. Specifically, a heartbeat is extracted from the signals, and for the heartbeat, the phase of the first harmonic of the subject's heartbeat relative to the start of the heartbeat is determined. These phases associated with the start of the heartbeat represent the feature sequence, and the feature signal unit generates a first feature signal based on the sequence. From the first feature signal 13a, the extraction unit 16 extracts at least a first respiratory rate 151a. Optionally, the extraction unit 16 may extract additional respiratory rates 152a. In a preferred embodiment, three or four first respiratory rates are presented. Then, in an optional subsequent step, based on one or more respiratory rates, the analysis unit may determine a high-probability respiratory rate group including at least one high-probability respiratory rate. Preferably, a less likely group of respirations can also be identified, which includes at least one respiration rate that is less likely than at least one high-probability respiration rate.
[0064] Figure 2 A graph showing a pulse-to-pulse signal derived from infrared and / or red PPG signals is illustrated. Similar pulse-to-pulse signals can be derived from infrared or red PPG signals alone. The signal shows three pulses: a first pulse i-1, a second pulse i, and a third pulse i+1. The points shown in the signal positively mark the pulse trigger for each pulse, i.e., the start of each pulse. The pulse trigger is calculated as the highest slope of the falling edge shown in the graph. In practice, the device 10 of the present invention, preferably the feature signal unit 14 of the device 10, is configured to locate the pulse trigger in the PPG signal. Thus, the start time of the pulse is known, and the phase of the first harmonic of the pulse can be calculated relative to the start of the pulse. The start of the heartbeat can be determined, for example, based on curve characteristics such as maximum or minimum values, inflection points, maximum slope, etc.
[0065] Figure 3A A schematic diagram of a second embodiment of a device for determining a subject's respiratory rate is shown. In this embodiment, device 10 is configured to calibrate the cutoff frequency of filter unit 22. Device 10 includes an input interface 12, a feature signal unit 14, a calculation unit 26, a resampling unit 20, a filter unit 22, and an extraction unit 16. The feature signal used for calibration is preferably a second feature signal, more preferably the amplitude of PPG processed at the fundamental frequency of the subject's heartbeat multiplied by the corresponding pulse width. Figure 3A In the diagram, the signal that bypasses the computing unit 26 corresponds to the dashed line (this also applies to...). Figure 3B and 4 ).
[0066] The third second respiratory rate 053b, also referred to as the second respiratory rate three, is determined by first resampling the second feature signal 13b generated by the feature signal unit 14 based on the incoming PPG signal 11. Resampling is performed by the resampling unit 20. Resampling can be performed, for example, using a trapezoidal resampling function. The resampled (first feature) signal 19b is then provided to the extraction unit 16, which derives the third second respiratory rate 053b. Optionally, an additional second respiratory rate can be derived from said signal.
[0067] The fourth second respiratory rate 054b, also known as the second respiratory rate four, is determined by first providing the second feature signal 13b to the calculation unit 26. The calculation unit 26 then calculates the first derivative 25b of the first feature signal 13b. Subsequently, the resampling unit 20 resamples the first derivative 25b, thus generating a resampled first derivative 27b with a constant sampling rate. The resampled first derivative 27b is then provided to the extraction unit 16, which is configured to extract the fourth second respiratory rate 054b.
[0068] Therefore, in this embodiment, the second respiratory rate is determined based on the signal in the time domain.
[0069] Subsequently, the breathing rate is used to determine the cutoff frequency of the filter in filter unit 22. Specifically, the third second breathing rate 053b is used to configure (i.e. calibrate) the bandpass filter, while the fourth second breathing rate 054b is used to configure (i.e. calibrate) the lowpass filter.
[0070] Figure 3B A schematic diagram of a third embodiment of a device for determining the respiratory rate of a subject is shown. The device 10 includes an input interface 12, a feature signal unit 14, a calibration unit 26, a resampling unit 20, a filter unit 22, a transformation unit 24, an extraction unit 16, a scoring unit 28, a grouping unit 30, and an analysis unit 18.
[0071] A first respiratory rate 151a, also known as the first respiratory rate one, is determined by first resampling a first characteristic signal 13a generated by the characteristic signal unit 14 based on the incoming PPG signal 11. Resampling is performed by the resampling unit 20. The resampled (first characteristic) signal 19a is then provided to the transformation unit 24, which is configured to calculate the Fourier transform of the resampled signal 19a, i.e., to generate a Fourier-transformed resampled signal 23a. The Fourier-transformed resampled signal 23a is then provided to the extraction unit 16, which derives the first respiratory rate 151a from the signal. Thus, the first respiratory rate 151a is determined in the frequency domain. Other first respiratory rates, such as sixth and seventh first respiratory rates 156a and 157a, can also be determined from the signal.
[0072] The second first respiratory rate 152a, also known as the first respiratory rate second, is determined by first providing the first feature signal 13a to the calculation unit 26. The calculation unit 26 calculates the first derivative 25a of the first feature signal 13a. Subsequently, the resampling unit 20 resamples the first derivative 25a, thus generating a resampled first derivative 27a with a constant sampling rate. The resampled first derivative 27a is then Fourier transformed by the transformation unit 24, resulting in a resampled first derivative 29a. Using the resampled first derivative 29a, the extraction unit 16 extracts the second first respiratory rate 152a. Therefore, the second first respiratory rate 152a is also determined based on the signal in the frequency domain.
[0073] Optionally, the resampling unit 20 and the filter unit 22 can be used to extract the third first respiratory rate 153a, also referred to as the first respiratory rate three. In the first step, the first feature signal 13a is resampled by the resampling unit 20 and then filtered by the filter unit 22, which generates a filtered resampled signal 21a. In this embodiment, the filter unit 22 is adapted to the desired respiratory rate (upper limit), which is preferably derived from the (unfiltered) second feature signal, for example, from the amplitude sequence at the fundamental frequency of the subject's heartbeat multiplied by the pulse width of the heartbeat (see...). Figure 3A Based on the filtered resampled signal 21a, the extraction unit 16 determines the first respiratory rate 153a. That is, in this embodiment, the filtered resampled signal 21a is analyzed in the time domain.
[0074] Optionally, the calculation unit 26, the resampling unit 20, and the filter unit 22 can be used to extract the fourth first respiratory rate 154a, also referred to as the first respiratory rate four. The calculation unit 26 calculates the first derivative 25a of the first feature signal 13a. After calculating the first derivative, the corresponding signal 25a is provided to the resampling unit, which is configured to generate the resampled first derivative 27a. After filtering the resampled first derivative, i.e., generating the filtered resampled first derivative 33a, the extraction unit 16 determines the first respiratory rate four 154a based on the filtered resampled first derivative 33a. That is, in this embodiment, the filtered resampled first derivative 33a is analyzed in the time domain.
[0075] Regarding the extraction of respiratory rates 153a and 154a, filter unit 22 may optionally also be provided with respiratory rates 053b and 054b, such as in another device or as per the description of respiratory rates 053b and 054b. Figure 3A The respiratory rates 053b and 054b, as determined in the previously described calibration steps, can be specifically used for filter calibration. However, filter calibration can also be performed in advance.
[0076] Typically, extraction unit 16 is configured to extract an additional first respiratory rate from the signal in the frequency domain. Preferably, extraction unit 16 extracts not only first respiratory rates one and two from the resampled signal 23a after Fourier transform and the resampled first derivative 29a, but also first respiratory rates 156a, 157a, 158a, and 159a, as shown below. Figure 3B As shown.
[0077] The extraction unit 16 can be divided into two subunits, wherein the first (extraction) subunit is configured to process signals in the time domain, and the second (extraction) subunit is configured to process signals in the frequency domain. Therefore, in this embodiment, the first extraction subunit can be configured to extract first respiratory rates 153a and 154a, and the second extraction subunit can be configured to extract first respiratory rates 151, 152a, 156a, 157a, 158a, and 159a.
[0078] The first respiratory rate is also processed by scoring unit 28. Scoring unit 28 assigns a score to each of the respiratory rates, thereby producing scored respiratory rates 161a, 162a, etc. The assigned score describes the probability of respiratory rates 151a, 152a, etc., to represent the true respiratory rate of the object. For this purpose, scoring unit 28 can indicate a quality metric for each of the first respiratory rates. That is, the higher the score of the respiratory rate, the higher the probability that the rate is a true respiratory rate. In particular, with respect to low respiratory rates, the respiratory rate determined by the derivative in the frequency domain is more reliable than the respiratory rate determined by the derivative in the time domain. Therefore, the respiratory rate extracted from the derivative in the frequency domain can be assigned a higher weight than the respiratory rate extracted from the derivative in the time domain. The corresponding weight may then affect the score of the respiratory rate.
[0079] Furthermore, for signals in the frequency domain, the presence of significant peaks typically corresponds to a high respiratory rate score. However, since the spectrum displays several peaks, it is not always clear which peak represents the correct respiratory rate. Therefore, the score also depends on the prominence of the highest peak and its height relative to the other peaks.
[0080] The scored first respiratory rates 161a, 162a, etc., are then provided to the grouping unit 30, which is configured to group the respiratory rates into groups of similar respiratory rates. In this embodiment, the respiratory rates are divided into at least two groups: a first group 171 with a high group score and a second group 172 with a low group score. Subsequently, both groups 171 and 172 are provided to the analysis unit 18 to extract the most probable respiratory rate 17. Generally, the respiratory rate with the highest score is selected by the analysis unit 18 as the most probable respiratory rate 17.
[0081] Figure 4 A schematic diagram of a fourth embodiment of a device for determining the respiratory rate of a subject is shown. The fourth embodiment includes the same units as the third embodiment. Furthermore, in this embodiment, the feature signal unit 14 is configured to generate a first feature signal 13a and a second feature signal 13b based on the PPG signal 11. While the first feature signal 13a may include a phase sequence of the first harmonic of the subject's beats relative to the start of these beats, the second feature signal 13b may include a sequence of amplitudes at the fundamental frequency of the subject's heartbeats multiplied by the pulse width of these beats. However, it is also conceivable that the first feature signal 13a includes a sequence of amplitudes at the fundamental frequency of the heartbeats, and the second feature signal 13b includes a sequence of single pulse perfusions. In any case, the feature signal 13b is processed only as in this embodiment. Specifically, in this embodiment, four first respiratory rates 151a, 152a, 153a, and 154a are determined, and four second respiratory rates 151b, 152b, 153b, and 154b are determined.
[0082] The first respiratory rate 151a and the second respiratory rate 152b are determined in the same manner as the first respiratory rate 151a in the third embodiment. The first respiratory rate 152a and the second respiratory rate 152b are determined in the same manner as the first respiratory rate 152a in the third embodiment, and rates 153a and 153b and 154a and 154b are determined in the same manner as the first respiratory rate 153a and the first respiratory rate 154a in the third embodiment.
[0083] In summary, four pairs of respiratory rates were extracted, of which two pairs were extracted from the signal in the frequency domain and two pairs were extracted from the signal in the time domain.
[0084] The respiratory rate is provided to the scoring unit 28, which assigns a score to the rate. The higher the score corresponding to the respiratory rate, the higher the probability that the rate is the true respiratory rate. After obtaining the respiratory rates and their corresponding scores, the grouping unit 30 groups the respiratory rates in different groups. Specifically, the respiratory rates of the first respiratory rates 151a, 152a, 153a, and 154a are grouped together if their differences do not exceed a predetermined respiratory rate difference, for example, two breaths per minute (rpm). The second respiratory rates 151b, 152b, 153b, and 154b are grouped in the same manner. Then (by the grouping unit) the scores of all respiratory rates in the groups are added together, and preferably also weighted to produce a group score. Thus, the grouping unit provides the analysis unit 18 with four groups 171a, 171b, 172a, and 172b, for example, and their corresponding group scores. In the analysis unit 18, the group scores are analyzed, and it is assumed that the group with the highest group score includes the most likely respiratory rate. The most likely respiratory rate 17 is the rate that occurs most frequently in the group with the highest group score in this embodiment. However, there are other ways that can be conceived to determine the most likely respiratory rate 17.
[0085] In fact, the most likely respiratory rate 17 can be selected not only from the group with the highest group score, but also from the number of respiratory rates in the group. For example, if the group score is high, but there is only one respiratory rate in the corresponding group, the most likely respiratory rate 17 can be selected from another group with a higher number of respiratory rates.
[0086] The most probable respiratory rates corresponding to the first and second respiratory rates can also be determined. These most probable respiratory rate candidates for all features can then be clustered, yielding the final respiratory rate. Based on this step, a confidence value can also be calculated. Preferably, the confidence value is based on the difference between group size and weight. If the weights and / or sizes are similar between different groups, the confidence value is small. However, if one group is dominant, the confidence value is high. A signal quality index can also be derived regarding the final respiratory rate. The signal quality index can be based on the confidence level of the final respiratory rate and the history of the output values.
[0087] In addition, the grouping unit 30 can group the first and second respiratory rates regardless of their affiliation with a specific characteristic signal. Therefore, the first and second respiratory rates can also be mixed within a group.
[0088] Regarding the first and second respiratory rates, the analysis unit 18 of the device 10 can also be used to determine the second most probable respiratory rate (along with their corresponding group size and group weight) or the second most probable respiratory rate for both respiratory rates, i.e., all features of the extracted PPG signal.
[0089] Figure 5 A first figure illustrates a method for determining the respiratory rate of an object according to the present invention. The method includes step S11 of providing a PPG signal, and in steps S13a, S13b, ..., S13N determining N (distinct) feature signals based on the PPG signal. A Fourier transform or derivation can then be performed on the N feature signals in an optional intermediate step, not shown in the figure. In steps S151a, S152a, ..., S151b, S152b, ..., S151N, S152N..., at least one respiratory rate is extracted from each of the N processed feature signals. Preferably, at least two respiratory rates are extracted from the signal in the frequency domain. Therefore, too many respiratory rates can be extracted for each feature. Similarly, if the feature is considered only in the time domain, only one breath can be extracted. In steps S161a, S162a, ..., S161b, S162b, S163b, ..., S161N, S162N, ..., scores are assigned to the feature signals. In steps S171a, S171b, ..., S171N, the scored respiratory rates for each feature signal are grouped into different groups. In this embodiment, two groups are generated for each feature signal. Specifically, for each feature (signal), a group with a high group score and a group with a low group score are generated. However, further groups with similar scores can also be generated. In step S17, the analysis unit determines which respiratory rate is the most likely respiratory rate.
[0090] Figure 6 A second diagram illustrates a method for determining the respiratory rate of a subject. In the first step S11 of the method, a PPG signal is provided. In step S13a, a feature sequence is extracted from the PPG signal, and a first feature signal is generated. The feature signal is then processed in four different ways.
[0091] In the first approach, the first feature signal is resampled in step S20 and subjected to a Fourier transform in step S24. In parallel, i.e., in the second approach, the first derivative of the first feature signal is calculated in step S26, the derivative is resampled in step S20, and a Fourier transform is performed on the resampled derivative in step S24. In the third approach, the first feature signal is resampled in step S20 and then filtered later in step S22. In the fourth approach, the first derivative of the first feature signal is derived (step S26), then resampled (step S20) and filtered (step S22). Therefore, there are two signals in the frequency domain and two signals in the time domain. From these signals, the first respiratory rate is extracted in steps S151a, S152a, S153a, and S154a, respectively. Even more respiratory rates can be extracted from the signals in the frequency domain. For example, three respiratory rates can be extracted from each signal in the frequency domain. Then, in steps S161a, S162a, S163a, and S164a, scores are assigned to the extracted respiratory rates. Subsequently, in step S171a, similar respiratory rates are grouped into corresponding groups, and the scores of the respiratory rates in the groups are summed to form a so-called group score. In this embodiment, two groups are generated, wherein the first group includes respiratory rates with a high probability of representing the true respiratory rate, and wherein the second group includes respiratory rates with a low probability. In step S18, depending on the composition of the groups (i.e., the number of respiratory rates in the groups) and / or the group score and / or the scores of the different respiratory rates in the groups, it is determined which is the most likely and second most likely respiratory rate of the subject.
[0092] Figure 7 A schematic diagram of an embodiment of a system for determining the respiratory rate of a subject is shown. System 100 includes a PPG sensor 50 configured to provide red and / or infrared light volumetric plethysmography (PPG) signals, and a device 10 for determining the respiratory rate of a subject 200. Device 10 of the system includes an input interface 12, a feature signal unit 14, an extraction unit 16, an analysis unit 18, and a display unit 32.
[0093] In this embodiment, the PPG sensor 50 is designed as a patch attached to the forehead of the subject. However, the PPG sensor 50 can also be attached to other body areas of the subject 200. In fact, using a remote PPG sensor is also possible. The sensor 50 and the input interface 12 of the device 10 are connected via a wired connection; however, the PPG sensor 50 and the input interface 12 can also operate via a wireless communication interface.
[0094] PPG sensor 50 is configured to measure the patient's PPG signal 11 and provide the signal 11 to the input interface 12 of device 10. Device 50 processes the PPG signal 11, and the analysis unit 18 of device 10 is configured to determine the most probable respiratory rate 17 of subject 200. The most probable respiratory rate 17 is then displayed on the display unit 32 of device 10. In this way, for example, clinical staff can identify clinical deterioration at an early stage.
[0095] Figure 8A and Figure 8B Each diagram illustrates a (second) characteristic signal according to the invention. Specifically, Figure 8A A graph showing the amplitude sequence at the fundamental frequency of each heartbeat in the time domain is presented, and Figure 8B A graph showing a sequence of single pulse perfusions in the time domain is illustrated. In an exemplary embodiment, two features can be analyzed over 60 seconds, as marked in the figure.
[0096] Figure 9A and Figure 9B Each diagram illustrates a characteristic signal according to the invention. In particular, Figure 9A A diagram showing the phase sequence of the first harmonic of a heartbeat relative to the start of the heartbeat is presented. Figure 9B This shows a sequence of amplitudes at the fundamental frequency of the heartbeat multiplied by the pulse length.
[0097] Figure 10A graph of the amplitude spectrum of the Fourier transform of the resampled characteristic signal is shown. Typically, a (first or second) respiratory rate can be derived from such a Fourier transform. To derive the respiratory rate (RR), a respiratory rate range [minRR, maxRR] can be defined, within which the respiratory rate is most likely to lie. The range can be particularly dependent on the patient's pulse rate, which can also be extracted from the PPG signal (e.g., via the characteristic signal unit). In an exemplary embodiment of the device, the lower end of the range, minRR, can be defined by (pulse rate) / 15, and the higher end of the range, maxRR, can be defined by (pulse rate) / 2. However, other ranges can also be used. The highest peak in the range is then selected as the respiratory rate, i.e., the most likely respiratory rate calculated in this way. The second peak is the second most likely respiratory rate calculated in this way. The (most likely) respiratory rates calculated in different ways are then compared, and the most likely respiratory rate is determined by analysis unit 18. However, according to... Figure 10 The Fourier transform shown cannot derive the respiratory rate. In fact, the highest peak value indicating the most probable respiratory rate cannot be inferred from the graph shown. There are actually three peaks with almost identical heights, and therefore a reliable respiratory rate cannot be derived. In other words, to determine the score of the respiratory rate, the prominence of the peaks in the graph and the height of those peaks are determined. For example, if two peaks in the spectrum have similar heights, the determined respiratory rate is unreliable and therefore has a low score. Figure 10 In the spectrum shown, there are actually three peaks of similar height. Therefore, the corresponding respiratory rate scores are quite low, and the results are therefore unreliable.
[0098] Figure 11 A graph of pulse amplitude variability (PAV) in the time domain is shown. A 60-second window and a 50-second window are labeled. The time window can be adjusted depending on the score corresponding to the (first or second) respiratory rate, as determined by extraction unit 16. In particular, if the score for the respiratory rate for the 60-second time window is small (and therefore the derived respiratory rate is unreliable), the time window can be reduced to, for example, 50 seconds.
[0099] Figure 12 A graph showing the amplitude spectrum of the Fourier transform of a single beat of the infrared primitive wave is presented. From this Fourier transform, the fundamental frequency and harmonics can be derived. Specifically, in... Figure 12The fundamental frequency (f0), first harmonic (f1), and second harmonic (f2) are labeled. Therefore, the amplitude at the fundamental frequency or the phase of the first harmonic can be derived from such a Fourier transform. To obtain these values, the following procedure is followed. First, all raw red or infrared PPG samples are obtained from the beginning to the end of a single beat. Then, a straight line, fitted through the first and last samples of the beat, is subtracted. Next, the Fourier transform of this signal corresponding to a single beat is calculated. With a sampling rate of 125 Hz and an effective pulse rate range of 29.3 to 300 beats per minute, the number of samples per beat varies between 25 and 256. Therefore, the number of points in the Fourier transform can vary with the length of the pulse.
[0100] Figure 13 A graph showing pulsation-to-pulsation signal is presented, derived from infrared and red PPG signals illustrating different characteristics of the pulsation. Specifically, Figure 13 The diagram shows three pulsations originating from the original infrared wave. The surface of pulsation i is defined as the region spanned by the original signal and the straight line from the maximum value of the pulsation to the maximum value of the next pulsation i+1. Figure 13 The characteristics PAV (i.e., pulse height) and PWV (i.e., pulse width) are also shown.
[0101] Although the invention has been illustrated and described in detail in the accompanying drawings and foregoing description, such illustrations and descriptions are to be considered illustrative or exemplary rather than restrictive; the invention is not limited to the disclosed embodiments. Other variations of the disclosed embodiments can be understood and implemented by those skilled in the art in practicing the claimed invention by studying the drawings, specification, and appended claims.
[0102] In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality. A single element or other unit can perform the functions of several items recited in the claims. Although specific measures are recited in different dependent claims, this does not indicate that combinations of these measures cannot be advantageously used.
[0103] Computer programs can be stored / distributed on suitable non-transient media, such as optical storage media or solid-state media supplied together with or as part of other hardware, but computer programs can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.
[0104] Any reference numerals in the claims should not be construed as limiting the scope.
Claims
1. A device for determining the respiratory rate of a subject, the device comprising: The input interface is used to acquire red and / or infrared photoplethysmography (PPG) signals. A feature signal unit is configured to extract a first feature sequence from the PPG signal and generate a first feature signal based on the feature sequence. A resampling unit is configured to resample the first feature signal using a resampling function to generate a resampled signal with a constant sampling rate and / or resample the first derivative to generate a resampled first derivative with a constant sampling rate. A transformation unit is configured to compute a Fourier transform of the resampled signal to produce a transformed resampled signal, or to compute a Fourier transform of the first derivative of the resampled signal to produce a transformed first derivative of the resampled signal. An extraction unit is configured to extract a first respiratory rate from the transformed resampled signal or to extract the first respiratory rate from the first derivative of the transformed resampled signal. Wherein, the first feature sequence is the phase sequence of the first harmonic of the heartbeat of the object relative to the start of the heartbeat, and wherein the first feature signal corresponds to the phase sequence.
2. The device according to claim 1, It also includes an analysis unit. in, The extraction unit is configured to extract at least one additional first respiratory rate from the transformed resampled signal, and the analysis unit is configured to determine the most probable respiratory rate based on the at least two first respiratory rates.
3. The device according to claim 1 or 2, in, The feature signal unit is configured to extract a second feature sequence from the PPG signal and generate a second feature signal based on the second feature sequence. The extraction unit is configured to extract at least a second respiratory rate from the second feature signal. Wherein, the second characteristic signal corresponds to any one of the following: (a) the amplitude of the heartbeat at its fundamental frequency multiplied by the pulse width of the heartbeat, (b) the amplitude of the heartbeat at its fundamental frequency, (c) single pulse perfusion, (d) pulse rate, (e) the amplitude ratio of the fundamental frequency to the first harmonic of the heartbeat, (f) pulse amplitude variability, (g) pulse width variability, and (h) the surface of the heartbeat.
4. The device according to claim 1 or 2, It also includes a computing unit configured to calculate the first derivative of the first feature signal.
5. The device according to claim 1 or 2, It also includes a filter unit configured to filter the resampled signal to produce a filtered resampled signal, wherein, The extraction unit is configured to extract the first respiratory rate from the filtered resampled signal and / or filter the first derivative of the resampled signal to generate a filtered first derivative of the resampled signal, wherein the extraction unit is configured to extract the first respiratory rate from the filtered first derivative of the resampled signal.
6. The device according to claim 5, in, The filter unit is configured to adaptively filter based on the expected breathing rate corresponding to the resampled signal or the first derivative of the resampled signal.
7. The device according to claim 2, It also includes a scoring unit configured to assign a score describing the probability that the first respiratory rate represents the true respiratory rate of the object to the at least two first respiratory rates. in, The analysis unit is configured to use the score to determine the most likely respiratory rate.
8. The device according to claim 7, It also includes a grouping unit configured to group the at least two first respiratory rates into groups with similar respiratory rates, wherein, If the difference in respiratory rates is less than a predetermined rate difference, then similarity is assumed.
9. The device according to claim 8, in, The predetermined rate difference shall not exceed two breaths per minute.
10. The device according to claim 8, in, The grouping unit is further configured to sum the scores of the at least two first respiratory rates in the group to generate a group score, wherein the analysis unit is configured to determine the most likely respiratory rate based on the group with the highest group score.
11. The device according to claim 10, in, The analysis unit is also configured to determine a confidence value for the most likely respiratory rate based on the number of respiratory rates and the group score of the corresponding group.
12. A system for determining the respiratory rate of an object, the system comprising: A PPG sensor, configured to provide red and / or infrared light volumetric recording of PPG signals, and The device according to any one of the preceding claims is used to determine the breathing rate of a subject.
13. A computer-implemented method for determining the respiratory rate of an object, the method comprising: Obtain red and / or infrared photoplethysmography (PPG) signals. A first feature sequence is extracted from the PPG signal, and a first feature signal is generated based on the feature sequence. The first feature signal is resampled using a resampling function to generate a resampled signal with a constant sampling rate and / or the first derivative is resampled to generate a resampled first derivative with a constant sampling rate. Calculate the Fourier transform of the resampled signal to produce the transformed resampled signal, or calculate the Fourier transform of the first derivative of the resampled signal to produce the first derivative of the transformed resampled signal. The first respiratory rate is extracted from the transformed resampled signal or from the first derivative of the transformed resampled signal. Wherein, the first feature sequence is the phase sequence of the first harmonic of the heartbeat of the object relative to the start of the heartbeat, and wherein the first feature signal corresponds to the phase sequence.
14. A computer program product comprising a computer program, the computer program including program code modules configured to cause the computer to perform the steps of the method according to claim 13 when the computer program is executed on the computer.