WO2015107268A1

WO2015107268A1 - Method and device for the detection of respiratory rate

Info

Publication number: WO2015107268A1
Application number: PCT/FI2015/050023
Authority: WO
Inventors: Niku OKSALA; Sasu LIUHANEN
Original assignee: Aboa Legis Oy
Priority date: 2014-01-16
Filing date: 2015-01-15
Publication date: 2015-07-23
Also published as: EP3094248A1; EP3094248A4; EP3094248B1; JP6310086B2; JP2017502782A

Abstract

A method for estimating respiratory rate comprises gathering data related to pulse wave in an electrical form, extracting a good quality beat series with proper noise removal and interpolated beats for missing beats, extracting four primary sub-signals and two derived sub-signals from the beat series, performing both time and frequency domain analysis independently for each of the aforementioned six sub-signals and combining the results of the time and frequency domain analysis independently for each sub-signal thus obtaining a group of estimates (2 x 6) for the respiratory rate. The estimating of the respiratory rate is then calculated by removing sub-signal estimates with poor signal-to-noise ratio and those that are statistical outliers.

Description

METHOD AND DEVICE FOR THE DETECTION OF RESPIRATORY RATE

TECHN ICAL FIELD OF THE INVENTION

The invention relates to a method and device for the detection of respiratory rate. Especially the invention relates to continuous estimation of the respiratory rate.

BACKG ROUND OF THE INVENTION

Respiratory rate is a sensitive surrogate of critical conditions that develop in patients hospitalized due to acute illness or healthy subjects developing life- threatening conditions. Respiratory rate is conventionally detected by physical examination, capnography, electrocardiography (ECG) based impedance variation, microphone-based technologies or by impedance tomography. There are however some disadvantages related to the known techniques. Physical examination is not suitable for continuous monitoring, for example. Capnography requires wearing a mask, microphone-based technologies are sensitive to external noise and impedance tomography requires electrodes to be placed on the chest. These challenges limit the usability of existing technology especially for long-term monitoring of ambulatory patients or in sports or other personalized applications.

SUMMARY OF THE INVENTION

An object of the invention is to alleviate and eliminate the problems relating to the known prior art. Especially the object of the invention is to provide a device for continuous monitoring of respiratory rate in a reliable, easy and fast way.

The object of the invention can be achieved by the features of independent claims. The invention relates to a method for estimating respiratory rate according to claim 1 , as well as to a device for estimating respiratory rate according to claim 1 2 and computer program product of claim 1 6.

According to an embodiment a method for estimating respiratory rate comprises steps of:

- gathering data, such as spectrum data, related to peripheral pulse wave in an electrical form,

extracting a good quality beat series with proper noise removal and interpolated beats for missing beats, which may comprise in an exemplary method extracting e.g. four primary sub-signals and two derived sub-signals from the beat series of said gathered data, which again can be used for selective averaging of said sub-signals, performing both time and frequency domain analysis independently for each of the aforementioned six sub-signals,

- combining the results of the time and frequency domain analysis independently for each sub-signal thus obtaining a group of estimates (2 x 6) for the respiratory rate,

- calculating the final estimate for the respiratory rate by removing sub- signal estimates with poor signal-to-noise ratio and those that are statistical outliers.

Optionally, or alternatively, the method may comprise also:

- determining averaged power spectrums and maximum power for each of said averaged power spectrums of said four sub-signals,

- combining said averaged power spectrums and maximums of said averaged power spectrums of said four sub-signals and determining a combined power spectrum,

- determining maximums in said combined power spectrum and frequency of sequence said determined maximums, and

- estimating respiratory rate based on said frequency determined from said sequential maximums of said combined power spectrum.

The (spectrum) data related to peripheral pulse wave is advantageously measured by peripheral sensors located for example at the wrist and fingers. The sensor may be for example a pulse oximeter, but also other suitable sensors can also be used, such as pressure sensors or the like. An initial target of an embodiment of the invention is to obtain a reliable beat series by the following four steps, namely:

- pre-processing the spectrum data related to peripheral pulse wave, comprising e.g. a low pass filtering to remove high frequency noise, - detection of peaks and troughs from the pre-processed data,

- filtering the peaks and troughs, and

- inter-beat-analysis with corrections for noise, ectopic and missing beats.

The obtained series is then used for the actual analysis, where the preliminary step is to extract new sub-signals comprising:

- amplitude (AM),

- baseline,

- inter-beat-interval (FM) (used as squared), and

- pulse width (used as squared). Two composite signals are also created comprising:

- amplitude - baseline (phase difference), and

- inter-beat-interval + pulse width (used as squared).

Each sub-signal (n=6, for example) is resampled at constant intervals using cubic interpolation, as an example. Thereafter each sub-signal is subjected to both time and frequency domain analysis. Time domain analysis per signal comprises:

- peak and trough detection,

- detrending, and

- zero crossing detection. An exemplary frequency estimate comprises the following steps:

1) f1 = ((n-peaks + n-troughs) / 2) / delta-time, and

2) f2 = (n-zero-crossings / 2) / delta-time

Time domain estimate f = min(f1 , f2), when the two estimates differ, the difference is mostly caused by noise, whereupon the minimum is used.

An exemplary frequency domain analysis per signal comprises:

- windowing, - FFT (Fast Fourier Transform) ,

- power spectrum calculation, and

- frequency estimation from the obtained power spectrum.

Signal-to-noise ratio is advantageously calculated for each time and frequency domain estimate. The time and frequency domain estimates of the sub-signals are then combined to calculate the final respiratory frequency estimate.

An exemplary combining the time and frequency domain estimates: comprises:

- 2 x (4 + 2) estimates,

- drop time domain estimates with poor estimated SN R (signal-to noise ratio), and

- drop frequency domain estimates with poor estimated SN R.

Use remaining estimates plus previous final estimate comprises:

- drop statistical outliers, and

- choose median = final estimate.

As in more details and according to an exemplary embodiment the method comprises the steps of low pass filtering the raw data. The data is detrended by subtracting 1 s (example and configurable) average from each sample (detrended). The data is smoothed by calculating 1 /5s (example and configurable) average and subtracting the 1 s average (smoothed). The local minima and maxima are found by using both of the above mentioned (detrended and smoothed). For each minima and maxima of the smoothed curve (smoothed to reduce noise), the corresponding minima and maxima of the detrended data are detected and used to create beat candidates.

The exemplary method comprises also filter for the beat candidates as follows:

if multiple candidates exist for one minimum / maximum of the averaged curve, select the best:

- if no statistical data exists:

o calculate a score based on distance between the smoothed and the detrended miminum / maximum + peripheral flow index

(pfi),

- otherwise: a weighted average of the following standard deviations: pulse width, amplitude, pfi, distance of the maximum and distance of the minimum

Next, inter-beat-interval analysis may be used for the remaining candidates as following:

- if a candidate is statistically an ectopic beat:

o remove the candidate and replace it with a synthetic beat

(location midway between previous and following)

(width average of the previous and following)

- if a beat is statistically missed (either truly missed or missed during analysis)

o add a synthetic beat (see above)

- if a beat is statistically noise, mark it noise

A corrected beat series is thus obtained and used for further analysis. The present invention offers advantages over the known prior art, such as very reliable measuring results due to using four sub-signals in the estimation process. In addition the invention offers possibility to use signals from a number of different sensors simultaneously and for same estimation process. Moreover the data processing steps of the raw data as described is not very vulnerable to environmental noise, for example. Especially it is to be noted that the current invention enables continuous estimation process for the respiratory rate.

BRIEF DESCRIPTION OF THE DRAWINGS

Next the invention will be described in greater detail with reference to exemplary embodiments in accordance with the accompanying drawings, in which:

Figure 1 A illustrates a principle of an exemplary method for estimating respiratory rate according to an advantageous embodiment of the invention,

Figure 1 B illustrates a principle of another exemplary method for estimating respiratory rate according to an advantageous embodiment of the invention, Figure 2 illustrates an exemplary raw data gathered by a measuring device, such as pulse oximeter, according to an advantageous embodiment of the invention,

Figures 3A-3D illustrate examples of resampled and extracted sub- samples used for estimating respiratory rate according to an advantageous embodiment of the invention and

Figure 4 illustrates an exemplary device and arrangement for estimating respiratory rate according to an advantageous embodiment of the invention.

DETAILED DESCRI PTION

Referring to the Figure 1 A the method 1 00 for estimating respiratory rate according to an advantageous embodiment of the invention comprises the following steps. According to an embodiment of the invention the respiratory rate of a patient is determined based on the signals measured 1 01 by peripheral sensors located for example at the wrist and fingers. The device for the detection of the respiratory rate of a patient advantageously continuously and non-invasively comprises either a photo-plethysmography (PPG) sensor, an infrared (I R) sensor, charge coupled device (CCD), optically by CMOS sensor, pulse oximeter or pressure sensor, such as an Electromechanical film (EMFi) sensor or a capacitive pressure sensor, tonometer or impedance or dielectric spectroscopy configured to record pulse wave as a raw data at the wrist or finger. It is to be noted that plurality of different sensors can be used, and that plurality of different raw data signals can be provided. An example of the raw data is described in Figure 2.

According to an embodiment low-pass filtering (e.g. 5 Hz) is performed 1 02 first to the measured raw data and the first and second derivatives of the resulting low-pass filtered signal are calculated 1 03. The low-pass filtering is done for removing high-frequency artefacts. The signal is then detrended 1 04 to remove any DC-component. The local minima and maxima are identified 1 05 from the detrended signal using the previously calculated derivatives and multi-phase adaptive thresholding is performed to identify peaks and troughs of true pulse waves. Thereafter, four sub-signals are calculated 1 06a, 1 06b, 1 06c, 1 06d from (each) raw data signal. Examples of derived four sub-signals are described in Figures 3A-3D.

For these four sub-signals the width of the pulse waves is calculated at first ((sub signal 1) 106a). The onset and end of each pulse wave can be detected using the previously calculated derivatives and set of rules based on their behaviour. Second, the pulse wave amplitudes are calculated ((sub signal 2) 106b); amplitude modulation of the original signal. Third, pulse rate variability is calculated ((sub signal 3) 106c); frequency modulation of the original signal. And fourth, the degree of baseline variation is calculated ((sub signal 4) 106d). Each of the four sub-signals is then resampled 1 07 at constant intervals using cubic interpolation. The sub-signals are Hann-windowed 1 08 and subjected to Fast Fourier Transformation (FFT) 1 09. The results are high-pass filtered 1 1 0 in order to eliminate low frequency bands and a power spectrum is obtained for each. Welch's method 1 1 1 is used to reduce noise. The maximum power is estimated per spectrum. It is to be noted that the order of determining said sub-signals can vary.

In the advantageous embodiment of the invention all these four sub-signals are used for determining the respiratory rate. In practice it might be that one of these four sub-signal might not be reliable readable, whereupon the respiratory rate may either be derived using at least three of those. Additionally readability of these four sib-signals can be used as a quality check for reliability of the derived respiratory rate.

According to an embodiment the sub-signal related to the pulse wave amplitudes or changes of the pulse wave amplitudes is processed also separately for determining a volemia (condition of the volume of plasma or blood circulating in the body). Namely, as an example, if the sub-signal related to the pulse wave amplitudes or changes of the pulse wave amplitudes of the pulse wave diminish (or disappears), the volume of plasma or blood circulating in the body is very high and reliability of the measurements of the respiratory rate by the method degrades. According to an example this can be used as a quality control check for the reliability of the measurements.

Additional criteria are calculated 1 1 2 using the four spectra: frequency corresponding to the maximum power per spectrum, thresholded maximum power per spectrum, maximum power within broader, clinically relevant 0.2 Hz frequency windows per spectrum, occurrence of maximum power within these windows and the cross-correlation between the spectra. Using these criteria, the four power spectra are selectively averaged and maximum power of the averaged spectrum is located 1 1 3. As an end result, the system provides an estimation of the respiratory rate 1 1 4.

Figure 1 B illustrates a principle of another exemplary method 200 for estimating respiratory rate according to an advantageous embodiment of the invention, which advantageously comprises number of similar and same features and step with the method 1 00 described in Figure 1 A, such as especially steps 1 01 -1 05, 1 06a-1 06d, 1 07-1 1 1 and 1 1 3-1 1 4. The method 200 describes an additional or alternative step 1 04a for smoothening the loss pass filtered data and step 1 04b for detecting local maxima and minima of the smoothed data. In addition the method 200 describes the steps after 1 05 in more details, where in step 1 05a the corresponding maxima and minima of the detrended data is found for each maximum and minimum of the smoothed data. The in step 1 05b beat candidates are found, in step 1 05c the best candidates are selected, in step 1 05d inter-beat-analysis is performed and in step 1 05e beat series is divided into sub-signals.

The method 200 also comprises steps 1 06e and 1 06f for calculating composite sub-signals (sub-signal 5 and sub-signal 6).

After resampling of the sub-signals 1 07 the method continues to steps for two alternative routes, either via step 1 08a of frequency domain analysis to steps 1 08-1 1 1 , or via step 1 1 5 of time domain analysis to steps 1 1 6a and/or 1 1 6b of estimating frequency (f1 ) based on zero-crossing and/or of estimating frequency (f2) peaks and troughs. The next step 1 1 7 in this route is to perform and set time domain estimate as min(f 1 , f2).

The branches of the method 200 continue in step 1 1 8, where estimates with poor signal-to-noise ratio is discarded and step 1 1 9 where statistical outliers are dropped, and where the previous estimate can be used as an additional estimate 1 20. After this the method is continued as in example of Figure 1 A with steps 1 1 3 and 1 1 4.

Figure 4 illustrates an exemplary device 401 and arrangement 400 for estimating respiratory rate according to an advantageous embodiment of the invention, where the device 401 comprises a data gathering means 402 for gathering a spectrum data related to pulse wave in an electrical form, and a data processing means 403 configured for:

- extracting four sub-signals of said spectrum data for selective averaging of said four sub-signals,

Advantageously the device 401 comprises or said data gathering means for gathering a spectrum data related to pulse wave in an electrical form comprises a photo-plethysmography (PPG) sensor 404, an infrared (I R) sensor 405, charge coupled device (CCD) 406, optically by CMOS sensor, pulse oximeter 407, or pressure sensor, such as an Electromechanical film (EMFi) sensor or a capacitive pressure sensor 408, tonometer or impedance or dielectric spectroscopy configured to record said pulse wave as a raw data. According to an example the device 401 is implemented as a wristband device with possible auxiliary means, whereupon the sensors to be used are advantageously configured to be positioned at the wrist or finger of the patient the respiratory rate of which is to be estimated. However, it is to be noted that the data processing can be implemented by the wristband device by the data processing means 403, or alternatively the wristband device may send (e.g. wireless way) the measuring signals to the external data processing backend 41 0 for data calculation, which comprises the data processing means 41 1 . Thus the data processing means 403 is optional in the device 401 . For communicating 41 2 with the backend 41 0, the device 401 advantageously comprises data communication device 409, most advantageously wireless communication device, such as implemented by the Bluetooth or the like known by a skilled person.

The data processing backend 41 0 may comprise e.g. could server 41 2, any computer or mobile phone application 41 3-41 5 and according to an example it can send the calculated results or otherwise processed data e.g. for displaying back to the wristband device or other data displaying device, such as a computer or the like in data communication network or to a smartphone of the user.

The invention has been explained above with reference to the aforementioned embodiments, and several advantages of the invention have been demonstrated. It is clear that the invention is not only restricted to these embodiments, but comprises all possible embodiments within the spirit and scope of the inventive thought and the following patent claims.

Claims

1 . A method for estimating respiratory rate,

wherein the method comprises steps of:

- gathering data related to pulse wave in an electrical form,

- extracting a good quality beat series with proper noise removal and interpolated beats for missing beats,

- extracting four primary sub-signals and two derived sub-signals from the beat series,

- performing both time and frequency domain analysis independently for each of the aforementioned six sub-signals

- combining the results of the time and frequency domain analysis independently for each sub-signal thus obtaining a group of estimates (2 x 6) for the respiratory rate

2. A method of claim 1 , wherein the method comprises:

3. A method of claim 2, wherein said selective averaging of said four sub- signals comprises:

- determining location of maximum power,

- thresholding maximum power per sub-signal to identify peaks and troughs of true pulse waves,

- determining spectral power within predetermined windows,

- determining location of maximum power window,

- determining maximum power within maximum power window, and - determining cross-correlation between the spectra.

4. A method of any of previous claims, wherein the spectrum data is gathered by at least one peripheral sensor located at the wrist and finger area.

5. A method of any of previous claims, wherein said gathered spectrum data is at first preprocessed by removing high-frequency artifacts using low pass filter.

6. A method of any of previous claims, wherein a first and second derivate are determined from said gathered spectrum data, and said data is detrended to remove DC-component, and local minima and maxima are identified from said detrended signal using the previously calculated derivatives.

7. A method of any of previous claims, wherein before the combination each of said four sub-signals is resampled at constant intervals using cubic interpolation, Hann-windowed, subjected to Fast Fourier Transformation (FFT), high-pass filtered to eliminate low frequency bands and thereby deriving a maximum power spectrum for each of said four sub-signals.

8. A method of any of previous claims, wherein Welch's method is used to reduce noise from data.

9. A method of any of previous claims, wherein said four sub-signals comprises:

- a first signal relating to width of the pulse wave,

- a second signal relating to pulse wave amplitude,

- a third signal relating to pulse rate variability, and

- a fourth signal relating to degree of baseline variation.

1 0. A method of any of previous claims, wherein the sub-signal related to pulse wave amplitudes or changes of the pulse wave amplitudes of the pulse wave is processed for determining a volemia and thereby reliability of the measurements so that if the sub-signal related to the width of the pulse wave diminish below a threshold value, the volume of plasma or blood circulating in the body is determined to be very high and reliability of the measurements of the respiratory rate is determined to be unvalid.

1 1 . A method of any of previous claims, wherein said estimation process is continuous and non-invasive estimation process for the respiratory rate.

12. A device for estimating respiratory rate,

wherein the device comprises:

- data gathering means for gathering data related to pulse wave in an electrical form,

- data processing means configured for:

o gathering data related to pulse wave in an electrical form, o extracting a good quality beat series with proper noise removal and interpolated beats for missing beats,

o extracting four primary sub-signals and two derived sub-signals from the beat series,

o performing both time and frequency domain analysis independently for each of the aforementioned six sub-signals o combining the results of the time and frequency domain analysis independently for each sub-signal thus obtaining a group of estimates (2 x 6) for the respiratory rate

13. A device of claim 12, wherein the data processing means is configured for:

o determining averaged power spectrums and maximum power for each of said averaged power spectrums of said four sub- signals,

o combining said averaged power spectrums and maximums of said averaged power spectrums of said four sub-signals and determining a combined power spectrum,

o determining maximums in said combined power spectrum and frequency of sequence said determined maximums, and o estimating respiratory rate based on said frequency determined from said sequential maximums of said combined power spectrum.

1 4. A device of any of previous device claims 12-13, wherein the device is configured to process the sub-signal related to pulse wave amplitudes or changes of the pulse wave amplitudes of the pulse wave for determining a volemia and thereby reliability of the measurements so that if the sub-signal related to the width of the pulse wave diminish below a threshold value, the volume of plasma or blood circulating in the body is determined to be very high and reliability of the measurements of the respiratory rate is determined to be unvalid.

15. A device of any of previous device claims 12-14, wherein the device comprises a photo-plethysmography (PPG) sensor, an infrared (IR) sensor, charge coupled device (CCD), optically by CMOS sensor, pulse oximeter, or pressure sensor, such as an Electromechanical film (EMFi) sensor or a capacitive pressure sensor, tonometer or impedance or dielectric spectroscopy configured to record said pulse wave as a raw data, and wherein said sensor is advantageously configured to be positioned at the wrist or finger of the patient the respiratory rate of which is to be estimated.

16. A computer program product for estimating respiratory rate, characterized in that it comprises program code means stored on a computer-readable medium, which code means are arranged to perform all the steps of the method defined in claims 1 -10, when the program is run on a data processing means, such as on a device of any of claims 12-15.