HK1125018B - Signal processing for pulse oximetry - Google Patents

Description

Signal processing for pulse oximetry

Cross Reference to Related Applications

This application claims priority from U.S. provisional application No. 60/722,257, filed on 30/9/2005.

Technical Field

The present application relates to determining the pulse rate of a biological pulsatile signal in the presence of noise. The techniques described herein are particularly useful for processing signals from pulse oximetry sensors.

Background

Pulse oximetry (Pulse oximetry) is a non-invasive diagnostic procedure for measuring oxygen saturation in a patient's arterial blood. The pulse oximetry method is based on the following principle: transmitting light energy from at least two wavelengths to a light-absorbing physiological medium, collecting reflected (or transmitted) emitted light in response to the light absorption, and calculating oxygen saturation level (oxygen saturation level) from the collected signal. A typical pulse oximeter (pulse oximeter) has two main components: a sensor attached to the skin of the patient for acquiring signals, and a processing unit for processing the acquired signals in order to determine the arterial oxygen saturation and the pulse rate. Unfortunately, conventional pulse oximetry systems are susceptible to noise, which can lead to unstable readings of pulse rate and oxygen saturation, inaccurate measurements, and false alarms. The signal level in the reflection pulse oximetry system is much lower than that in the transmission pulse oximetry system, and its noise is particularly difficult to deal with.

Disclosure of Invention

Each function contained in a preselected set of functions is compared to the input signal at a plurality of different time-shifts (time-shifts), and the function/time-shift combination that best matches the input signal is selected. The frequency of the best-matching function is then used as the best estimate of the frequency of the input signal.

Drawings

Fig. 1 is a block diagram illustrating a method for determining a pulse rate of a pulsatile signal.

Fig. 2A is a first example of a set of predefined frequency functions (S-PFF) for use with the method of fig. 1.

FIG. 2B is a second example of an S-PFF for use with the method of FIG. 1.

FIG. 2C is a third example of an S-PFF for use with the method of FIG. 1.

FIG. 3 is a fourth example of an S-PFF for use with the method of FIG. 1.

Fig. 4 shows an input signal and a matching function selected from the S-PFF.

Detailed Description

Although the preferred embodiments disclosed herein are described in the context of pulse oximetry, the invention may also be used in other contexts, including but not limited to signal processing of biological pulsatile signals obtained from sources other than oxygen probes (oximetry probes).

To calculate oxygen saturation (SpO) from signals acquired in a pulse oximetry system₂) It is generally preferred to first determine the pulse rate of the signal and locate the pulse position. It is important to obtain accurate estimates of pulse rate and pulse position for calculating the correct oxygen saturation in noisy signals.

Fig. 1 is a block diagram of a preferred method for determining the pulse rate of a pulsatile input signal 20 by solving an optimization problem involving a set of predefined frequency functions (S-PFF). The pulsed input signal 20 and the S-PFF 22 enter a mathematical operator 24 for processing. A mathematical operator 24 is used to transform the frequency estimation problem into an optimization problem. The output 26 of the system is the pulse rate determined for the input signal.

The input signal 20 may be a raw signal obtained from a source (e.g., an oxygen probe), or it may be pre-processed. In a preferred embodiment, DC and very low frequency components of the input signal are removed by a suitable pre-processing filter before entering the processing stage. Removing the DC component ensures that the processed signal is pulsed around 0 and has positive and negative peaks.

The S-PFF is a set of N periodic functions that are used to determine the frequency of the input signal. Preferably, the S-PFF function is chosen to be similar, at least approximately similar, to the signal expected in the desired application (without noise added). The frequency range of the S-PFF function, the number N of functions in a given S-PFF and the step size between successive frequencies are preferably selected based on the desired input signal and also based on the required accuracy of the obtained result. The pre-selection of the functions contained within a given S-PFF may be done with a suitable training process based on a data set representing the signals to be encountered in subsequent uses.

FIG. 2A is an example of a first S-PFF 30, which includes a set of N-period waveforms 31-33, each having a similar symmetrical shape but with a different frequency f₁-f_N. FIG. 2B is an example of a second S-PFF 40, which includes a set of N-period waveforms 41-43, each having a similar symmetrical shape but with a different frequency f₁-f_N. In an alternative embodiment, a digital S-PFF 50 may be used, as shown in FIG. 2C. The third S-PFF 50 also includes an N-cycle waveform f₁-f_N51-53, each having a similar pattern but a different frequency. In these embodiments, the functions 51-53 can only take 3 values: { -1,0, +1}. One advantage of using this type of S-PFF is that it makes it simpler and faster to perform mathematical operations than simulating S-PFF. It must be emphasized that the S-PFFs 30, 40, 50 shown in FIGS. 2A-2C are merely examples of usable S-PFFs and should not be considered as limitations of the present invention.

In the case of pulse oximetry, the input signal that is processed is the physiological pulsatile signal originating from the optical sensor plus a lot of noise. The frequency of the pulsations corresponds to the heart rate or pulse rate of the patient. Since the desired pulse rate is between 30 and 240 beats per minute (bpm) for almost all patients, a suitable range of frequencies for S-PFF for pulse oximetry is from 1/2Hz (corresponding to 30bpm) to 4Hz (corresponding to 240 bpm). Since many medical applications require the heart rate to be determined with an accuracy and resolution of 1bpm, a suitable S-PFF for pulse oximetry would be a set of 211 different frequencies (N ═ 211) in 1bpm steps ranging from 30bpm (1/2Hz) to 240bpm (4 Hz). If higher resolution or accuracy is required, N may be increased beyond 211. Optionally, the step size between different frequencies can be set to be non-linear, with smaller bpm steps used at lower frequencies and larger bpm steps used at higher frequencies (e.g., varying in a logarithmic series to achieve the same percentage accuracy).

In an oximetry procedure, the shape of a healthy patient's physiological pulse signal will typically have a faster rise time and a slower fall time. As a result, an S-PFF suitable for pulse oximetry would be the S-PFF 40 shown in FIG. 2B, which also has a faster rise time and a slower fall time, and thus approximately matches the desired signal. As explained above, 211 frequencies ranging from 1/2Hz to 4Hz in 1bpm steps (i.e., N ═ 211) are examples of suitable frequency distributions for pulse oximetry applications. However, in some cases, the shape of the physiological pulse waveform will be different (e.g., based on the particular physiological condition), which will degrade performance if the input waveform does not match well with the waveform in the S-PFF. It is therefore advantageous to incorporate a variety of different waveforms into the S-PFF to account for this difference.

Fig. 3 is an example of a preferred S-PFF 60 for pulse oximetry applications, optimized to process the input signal with different waveforms. The S-PFF 60 includes a subset of functions 61-63 with 211 different frequencies ranging from 30bpm to 240bpm, with faster rise times and slower fall times to match the desired physiological pulsatile signal from most healthy patients. The S-PFF 60 also includes a second subset of functions 64-66 having each of the 211 different frequencies with a slower rise time and a faster fall time to match a correspondingly shaped physiological pulsatile signal from a patient whose waveform deviates from the more prevalent fast rising waveform. The S-PFF 60 also includes a third subset of functions 67-69 having each of the 211 different frequencies with matched rise and fall times to match the correspondingly shaped physiological pulse signal from a patient whose waveform deviates from normal. Thus, the preferred S-PFF 60 includes a total of 633 functions (3 × 211). It should be noted that a suitable duration for each of the functions 61-69 for pulse oximetry in the S-PFF 60 is 3-5 seconds long.

Another preferred S-PFF for pulse oximetry applications comprises: a subset of functions 61-63 having 211 different frequencies ranging from 30bpm to 240bpm, with faster rise times and slower fall times (to match the expected physiological pulsatile signals from most healthy patients); and a second subset of functions 67-69 having each of the 211 different frequencies and having matching rise and fall times. This preferred S-PFF therefore comprises a total of 422 functions (2X 211). 3-5 seconds is also a suitable duration for each function 61-63, 67-69 within the S-PFF.

Optionally, additional subsets of functions (not shown) may be added to match particular physiological conditions. For example, an additional set of 211 functions may be added to one of the plurality of S-PFFs that match the physiological pulsatile signal expected from a patient with a mitral valve that is incompetent. (waveforms from such patients are expected to be different from those from healthy patients because blood flow initially flows from the left ventricle into the left atrium.

Alternatively, different S-PFFs may be used for different types of patients, such as adults, children, infants, and fetuses. For example, for infants, the frequency range of the S-PFF can be shifted to higher frequencies (compared to the 30-240bpm range described above).

Returning now to fig. 1, the input signal 20 includes the underlying physiological pulsating signal plus noise. The input signal 20 and the S-PFF 22 enter a processing stage that includes a mathematical operator F24 designed to estimate the frequency of the input signal. The mathematical operator F converts the frequency estimation problem into an optimization problem, and its specific implementation depends on practical applications. Since in the case of S-PFF, the mathematical operator F can be optimized and tested during the training period applied to a typical data set.

In a preferred embodiment of the invention, the mathematical operator F is defined as a non-linear operator as follows:

(1)F＝MAX_i，j{a₁*f₁(S-PFFj(t)，I(t-i))+a₂*f₂(S-PFFj(t)，I(t-i))+a₃*f₃(S-PFFj(t)，I(t-i))...+a_n*f_n(S-PFFj(t)，I(t-i))}

wherein the content of the first and second substances,

f₁-f_nis a set of n linear or non-linear functions;

S-PFF is a set of predefined frequency functions, as described above;

j is a reference numeral representing an S-PFF function selected from N (note that N is independent of N) defined S-PFF functions;

a₁-a_nis a selected weighting constant;

is a multiplication operator;

i is an input signal;

t is an index that represents a particular sample of S-PFFj and I (i.e., these functions at time t); and

i is the selected offset value representing the current offset between the function S-PFFj and the input signal I.

The mathematical operator F represents an optimization problem in which a maximum is searched for, as expressed by equation (1). In this way, each function contained in the S-PFF is compared to the input signal at a number of different time offsets, and the function/time offset combination that best matches the input signal is selected (i.e., by selecting the combination at which the maximum occurs). A suitable granularity of time offset (granularity) between different comparisons is the offset corresponding to one sample. The frequency of the best matching function is then used as the best estimate of the frequency of the input signal. For example, if in solving for F, a maximum is found at (j-12, i-5), this would mean S-PFF₁₂Is a function of the result that provides the maximum. Thus, the function S-PFF₁₂Will be selected as the frequency of the input signal I for output. In the case of pulse oximetry, the frequency will be the pulse rate (i.e., heart rate).

In addition to optimizing the S-PFF for the desired signal, as described above, F is a function of a portion of F₁-f_nAnd a weighting constant a₁-a_nCan be defined and optimized during the training phase to best suit the particular application and the signal characteristics desired in that application. The optimization can be performed using a suitable standard optimization method such as MLS (mean least squares) or maximum likelihood.

In a preferred embodiment adapted to determine the pulse rate using a preferred 633 function S-PFF 60 (shown in fig. 3) in a pulse oximetry system, n-3, so the mathematical operator F comprises only 3 functions F₁、f₂And f₃. In this embodiment, the function f₁(I, j) is the sum of products obtained by multiplying the function S-PFFj by the input signal I shifted by I samples, and the weighting constant a₁Is +1, resulting in the following equation: f. of₁(i，j)＝∑_tS-pffj (t) I (t-I). Function f₂(I, j) are successive values I (t) of the input signal I_k) The sum of the differences between them, and a weighting constant a₂Is-1. Value t_kIndex representing the maximum value of the function S-PFFj (t). Function f₃(I, j) is the continuous value of the input signal I (t)_m) The sum of the differences between them, and a weighting constant a₃Is also-1. Value t_mIndex representing the minimum value of the function S-PFFj (t). In a particular set of these three functions, f₁Is primarily responsible for estimating the frequency or pulse rate of the input signal. Function f₂And f₃It is mainly responsible for preventing harmonic frequencies of the input signal from interfering with the estimation of the frequency.

Once the best function within the S-PFF is selected, the index t_kDenotes the position of the maximum value of the pulse with the index t_mRepresents the position of the pulse minimum. By locating at t_kThe value of the acquired signal at the index minus the value at t_mThe value at the index, the amplitude of the pulse is obtained. These values can be used to calculate oxygen saturation using standard oximetry methods.

Optionally, the pulse rate and/or the location of the maxima and minima of the pulse can be used to extract a specific segment of the pulse wave for further processing. For example, in a pulse oximetry system, once the maxima, minima, and pulse rate are identified, the different phases of the cardiac cycle can be located (since the rising portion corresponds to the beginning phase of the cardiac cycle and the falling portion corresponds to the end phase of the cardiac cycle). So that when the sample corresponding to the rising portion is used for SpO₂Improved results are obtained in the calculation.

Returning now to fig. 4, this method can be further extended by taking a large number of samples in each cardiac cycle. In the illustrated example, the incoming input signal 70 has a frequency of 1Hz, which corresponds to 60bpm, and has a relatively fast rising phase and a slower falling phase. By applying the signal processing techniques described above using the S-PFF 60 (shown in FIG. 3), a matching 60bpm (1Hz) function 75 will be selected as described above. If the sampling is fast enough to provide 12 SpO per second₂Measuring, then overlaying the samples 71 of the input signal 70 to a function 75 selected from the S-PFF would enable differentiation between different phases of the cardiac cycle. For example, samples at a rising phase of the input signal 70 may be identified by selecting samples corresponding to the rising portion 76 of the selected function 75. In the illustrated example, the sample 71' will be identified as corresponding to a rising portion of the input signal 70. This sample 71' will then be used to perform SpO₂And (4) calculating.

The embodiments described above provide an accurate and efficient method for determining the frequency of a pulsating input signal. This method successfully calculates the pulse rate even for very low SNR (signal-to-noise ratio) values that occur when the acquired reflected signal amplitude is low and the noise level is high. Although the method is described above in the context of determining heart rate and pulse position from pulse oximetry derived signals, it should be appreciated that the methods may also be used to process other physiological pulsatile signals. Furthermore, the functions, frequencies, resolutions, durations and frequency ranges described above are merely examples and thus should not be considered as limitations of the present invention.

Claims (23)

1. A method for determining a pulse rate of a living subject from a pulse oximetry input signal, the method comprising the steps of:

comparing the input signal at each of a plurality of different time offsets to each member of a set of functions, wherein the set includes functions having a plurality of different frequencies between 30 and 240 cycles per minute, wherein the set includes at least a first subset and a second subset, the first subset including functions in which the rise time is shorter than the fall time and the second subset including functions in which the rise time is about the same as the fall time or the rise time is longer than the fall time, wherein the shapes of all functions in the first subset are similar, and wherein the shapes of all functions in the second subset are similar;

selecting a function from the set that best matches the input signal based on the results obtained in the comparing step; and

using the frequency of said selected function as an estimate of the pulse rate of the living subject.

2. The method of claim 1, further comprising the step of removing DC and very low frequency components from the input signal, wherein the removing step is performed before the comparing step.

3. The method of claim 1, wherein the functions in the set each have a duration between 3 seconds and 5 seconds.

4. The method of claim 3, wherein the set includes at least 211 functions.

5. The method of claim 1, wherein at least one function in the first subset and at least one function in the second subset have the same frequency.

6. The method of claim 1, wherein the frequencies of the functions in the first subset and the second subset are distributed in frequency to provide a resolution of at least one cycle per minute.

7. The method of claim 6, wherein the set further includes a third subset of functions, wherein the rise time is longer than the fall time, and the shape of all functions in the third subset is similar.

8. The method of claim 1, wherein the comparing and selecting steps are performed based on the following equation:

F＝MAX_i，j{a₁＊f₁(S-PFFj(t)，I(t-i))+a₂＊f₂(S-PFFj(t)，I(t-i))

+a₃＊f₃(S-PFFj(t)，I(t-i))...+a_n＊f_n(S-PFFj(t)，I(t-i))}

wherein the content of the first and second substances,

f₁-f_nis a set of n linear or non-linear functions;

S-PFF is a set of predefined frequency functions;

j is a reference numeral representing an S-PFF function selected from the N S-PFF functions defined;

a₁-a_nis the resulting weighted constant;

is a multiplication operator;

i is an input signal;

t is time; and

i is the selected offset value representing the current offset between the function S-PFFj and the input signal I.

9. The method of claim 1, wherein the comparing and selecting steps are performed based on the following formula:

F＝MAX_i，j{f₁(S-PFFj(t)，I(t-i))-f₂(S-PFFj(t)，I(t-i))-f₃(S-PFFj(t)，I(t-i))}

wherein the content of the first and second substances,

letter f₁(I, j) is the sum of the products obtained by multiplying the function S-PFFj by the input signal I offset by I samples, resulting in the equation f₁(i，j)＝∑_tS-PFFj(t)＊I(t-i)；

Function f₂(I, j) are successive values I (t) of the input signal I_k) The sum of the differences between, where the value t_kIndex representing the maximum value of the function S-PFFj (t);

function f₃(I, j) are successive values of the input signal I (t)_m) The sum of the differences between, where the value t_mIndex representing the minimum value of the function S-PFFj (t);

S-PFF is a set of predefined frequency functions;

j is a reference numeral representing an S-PFF function selected from the N S-PFF functions defined;

i is an input signal;

t is time; and

i is the inverted offset value representing the current offset between the function S-PFFj and the input signal I.

10. A method of processing a pulse oximetry input signal, the method comprising the steps of:

comparing the input signal at each of a plurality of different time offsets to each member of a set of functions, wherein the set includes functions having a plurality of different frequencies between 30 and 240 cycles per minute, wherein the set includes at least a first subset and a second subset, the first subset including functions in which the rise time is shorter than the fall time and the second subset including functions in which the rise time is about the same as the fall time or the rise time is longer than the fall time, wherein the shapes of all functions in the first subset are similar, and wherein the shapes of all functions in the second subset are similar;

selecting a function from the set that best matches the input signal based on the results obtained in the comparing step;

identifying a rising portion of the input signal based on a temporal alignment with a rising portion of the selected function; and

calculating oxygen saturation based on one or more samples from the portion of the input signal identified in the identifying step.

11. The method of claim 10, wherein the input signal has a sampling rate of at least 12 samples per second.

12. The method of claim 10, further comprising the step of removing DC and very low frequency components from the input signal, wherein the removing step is performed before the comparing step.

13. The method of claim 10, wherein the functions in the set each have a duration between 3 seconds and 5 seconds.

14. The method of claim 10, wherein the set includes at least 211 functions.

15. The method of claim 10, wherein at least one function in the first subset and at least one function in the second subset have the same frequency.

16. The method of claim 10, wherein the frequencies of the functions in the first subset and the second subset are distributed in frequency to provide a resolution of at least one cycle per minute.

17. The method of claim 16, wherein the set further includes a third subset of functions, wherein the rise time is longer than the fall time, and the shape of all functions in the third subset is similar.

18. The method of claim 10, wherein the comparing and selecting steps are performed based on the following equation:

F＝MAX_i，j{a₁＊f₁(S-PFFj(t)，I(t-i))+a₂＊f₂(S-PFFj(t)，I(t-i))

+a₃＊f₃(S-PFFj(t)，I(t-i)...+a_n＊f_n(S-PFFj(t)，I(t-i))}

wherein the content of the first and second substances,

f₁-f_nis a set of n linear or non-linear functions;

S-PFF is a set of predefined frequency functions;

j is a reference numeral representing an S-PFF function selected from the N S-PFF functions defined;

a₁-a_nis a selected weighting constant;

is a multiplication operator;

i is an input signal;

t is time; and

i is the selected offset value representing the current offset between the function S-PFFj and the input signal I.

19. The method of claim 10, wherein the comparing and selecting steps are performed based on the following formula:

F＝MAX_i，j{f₁(S-PFFj(t)，I(t-i))-f₂(S-PFFj(t)，I(t-i)-f₃(S-PFFj(t)，I(t-i))}

wherein the content of the first and second substances,

function f₁(I, j) is the sum of the products obtained by multiplying the function S-PFFj by the input signal I offset by I samples, resulting in the equation f₁(i，j)＝∑_tS-PFFj(t)＊I(t-i)；

Function f₂(I, j) are successive values I (t) of the input signal I_k) The sum of the differences between, where the value t_kIndex representing the maximum value of the function S-PFFj (t);

function f₃(I, j) are successive values of the input signal I (t)_m) The sum of the differences between, where the value t_mIndex representing the minimum value of the function S-PFFj (t);

S-PFF is a set of predefined frequency functions;

j is a label indicating a selected S-PFF function from the defined N S-PFF functions;

i is an input signal;

t is time; and

i is the selected offset value representing the current offset between the function S-PFFj and the input signal I.

20. A method of determining the frequency of a pulsatile input signal, the method comprising the steps of:

comparing the input signal at each of a plurality of different time offsets to each member of a set of functions, wherein the set includes functions having a plurality of different frequencies within a desired frequency range for the input signal, wherein the set includes at least a first subset and a second subset, the first subset including functions in which a rise time is shorter than a fall time and the second subset including functions in which a rise time is about the same as a fall time or a rise time is longer than a fall time, wherein the shapes of all functions in the first subset are similar, and wherein the shapes of all functions in the second subset are similar;

selecting a function from the set that best matches the input signal based on the results obtained in the comparing step; and

using the frequency of said selected function as an estimate of the frequency of the input signal.

21. The method of claim 20, further comprising the step of removing DC and very low frequency components from the input signal, wherein the removing step is performed before the comparing step.

22. The method of claim 20, wherein at least one function in the first subset and at least one function in the second subset have the same frequency.

23. The method of claim 20, wherein the comparing and selecting steps are performed based on the following equation:

F＝MAX_i，j{a₁＊f₁(S-PFFj(t)，I(t-i))+a₂＊f₂(S-PFFj(t)，I(t-i))

+a₃＊f₃(S-PFFj(t)，I(t-i))...+a_n＊f_n(S-PFFj(t)，I(t-i))}

wherein the content of the first and second substances,

f₁-f_nis a set of n linear or non-linear functions;

S-PFF is a set of predefined frequency functions;

j is a reference numeral representing an S-PFF function selected from the N S-PFF functions defined;

a₁-a_nis a selected weighting constant;

is a multiplication operator;

i is an input signal;

t is time; and

i is the inverted offset value representing the current offset between the function S-PFFj and the input signal I.