Pulse oximetry system

Abstract

In one aspect, a computer-implemented method includes receiving signals corresponding to wavelengths of light detected by an optical sensor placed near a patient's body, and for each received signal: separating the signal into an AC signal and a DC signal; separating the AC signal into component signals; analyzing the component signals by fractional phase transform to identify a desired component signal and a harmonic signal associated with the desired component signal; smoothing the desired component signal, the harmonic signal, and the DC signal; and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulation signal. A modulation ratio signal is generated based on the modulation signal derived from the signal, and peripheral blood oxygen saturation (SpO2) of the patient's body is determined based on the modulation ratio signal.

Description

Background Art

[0001] The present disclosure relates to pulse oximetry.

[0002] Pulse oximetry is used to estimate the time-averaged percentage oxygen saturation (SpO2) present in the patient's peripheral blood based on the photoplethysmogram (PPG) signal received from the sensor. 2 ). Estimation of SpO via PPG 2 The basic principle relies on the relationship between the relative attenuation of photons transmitted through and / or reflected from blood volume in tissues at different wavelengths of light in response to the oxygenation level in the blood. Typically, two wavelengths of light are used, for example, wavelengths in the red and infrared (IR) parts of the spectrum. For each wavelength, a PPG is obtained at a location around the patient, such as a fingertip or earlobe, hereinafter referred to as a red PPG and an IR PPG, respectively, without loss of generality. The light source and detector are typically implemented using readily available light emitting diodes (LEDs) and photodiodes, respectively. A ratiometric metric is then derived from the parameters extracted from the two PPGs, resulting in a generally good estimate of the SpO due to the relative attenuation characteristics of the two wavelengths. 2 A monotonic mapping of , which is valid for a wide range of patients.

[0003] The PPG signal is primarily characterized by two components: a DC component corresponding to the bulk light transmission in the tissue, and an AC component corresponding to the transmission modulation due to blood flow (and changes in blood volume due to that flow) on each heart pulse. The raw transmission modulation is measured by detecting the raw amplitude of the AC component of the PPG signal. This raw amplitude measure is normalized with the measure of the DC component, which then yields a percentage modulation term.

[0004] The percentage modulation term is extracted from the red PPG signal and divided by the percentage modulation term of the IR PPG signal to form the value mapped to SpO 2 The normalized ratio R N . R N to SpO 2 The mapping of values ​​can be determined by a calibration procedure typically performed in a clinical setting (e.g., a hypoxia laboratory), where R N The value of arterial oxygen saturation (SaO 2 A set of (R N , SaO 2 ) value pairs, a set of controlled hypoxic states were induced in each subject (while being monitored for safety) to establish the corresponding SaO 2 Then, for example, using a lookup table or polynomial fit, (RN , SaO 2 ) value pairs. Apply this mapping to R N The value of SpO is then generated 2 The corresponding value of SaO 2 Typically, SpO 2 The values ​​of can be resolved with field acceptable accuracy in the range of approximately 70% to 100%.

[0005] The AC component of the PPG signal consists of pulses related to vascular blood flow, with peaks and periodic characteristics according to the cardiac cycle. Therefore, the pulse rate (PR) can be derived directly from this component. Many methods are available, from simple peak detection and interval counting to complex rate detection methods. Smoothing is usually applied to calculate the time-local average of the pulse rate.

[0006] SpO based only on the above basic principles 2 A working implementation of the metric can be performed in a fairly straightforward manner, for example, using classical filtering and energy detection methods to perform extraction of the AC and DC components and measurement of the AC amplitude. Such a basic approach will generally work as long as conditions are favorable, such as when the signal-to-noise ratio (SNR) is relatively high (more specifically, when the percentage modulation is relatively large compared to the background noise or interferers).

[0007] In practice, challenges may arise due to various conditions that reduce the SNR. For example, the patient's own breathing will modulate the PPG, producing a separate periodic interference component within the AC component signal, sometimes with greater raw power than the desired pulse-mediated modulation. In addition, motion at the tip associated with the optical sensing point can have a similar modulating effect on the PPG, producing periodic or transient interference terms depending on the motion. Furthermore, the patient can exhibit low perfusion, which reduces the percent modulation and therefore the amplitude of the AC signal relative to background noise and / or interference terms. Such conditions of reduced SNR tend to make the normalized ratio R N Bias 1 (usually in practice tends to make the mapped SpO 2 values ​​tend to be around 85%), which is important for maintaining SpO within the desired range in a wide range of patient populations. 2 Precision presents real performance challenges. Summary of the invention

[0008] The present disclosure describes systems and techniques related to pulse oximetry. The system processes PPG signals simultaneously through parallel channels. For each channel, the system performs input processing of the PPG signal, analytical wavelet group processing, two-phase (QP) domain harmonic detection and tracking, and QP domain harmonic state vector filtering and amplitude statistics processing. The input processing of the PPG signal includes: sample interpolation to increase the sampling rate of the system to support the frequency resolution required for harmonic processing, filtering to reduce the impact of out-of-band artifacts and generally improve SNR, and frequency band separation of the PPG signal into fundamental AC and DC components. The analytical wavelet group processing separates the AC component into n B The QP domain harmonic detection and tracking includes: QP transformation and state vector generation, periodicity detection based on harmonic pattern recognition in the QP state space for estimating the interval between each QP pulse, and periodicity detection for n H Harmonics track harmonic parameters in QP space. QP domain harmonic state vector filtering and amplitude statistical processing includes individual per-harmonic filtering of the parameters (QP-space linear smoothing) to form a robust motion metric of the harmonic amplitude, QP synchronized filtering of the DC component of the PPG signal, combination of the harmonic amplitudes to form a robust metric of the AC component pulse amplitude, combination of AC information with DC information to form a robust motion metric of the normalized percent fractional PPG pulse modulation and a motion metric of the normalized modulation ratio, and QP synchronized filtering to form the instantaneous pulse spacing. The system then performs statistical filtering of the motion metric of the normalized modulation ratio to form a smoothed motion estimate of the normalized modulation ratio, and performs statistical filtering of the instantaneous pulse spacing to form a smoothed motion estimate of the instantaneous pulse spacing. The system then maps the smoothed motion estimate of the normalized modulation ratio to the output SpO 2 , and maps a smoothed moving estimate of the instantaneous pulse interval to the output pulse rate.

[0009] By using the harmonic QP component, the system can provide accurate SpO for small percentage modulation amplitude levels caused by a) relatively low perfusion (low local blood volume) and b) relatively dark skin (large attenuation of the original signal). 2 The harmonic QP components are estimated to have individual high SNR and the harmonic QP parameters tracked averagely within the QP time to further improve the SNR. The system is resistant to cardiopulmonary coupling artifacts and motion artifacts by using periodic pattern detection to track the periodic components caused by pulses and suppress the components caused by respiratory artifacts and motion artifacts.

[0010] In general, in one innovative aspect, a system includes: an optical sensor configured to detect light; a display device; one or more data processing devices coupled to the optical sensor and the display device; and a non-transitory computer-readable storage medium encoded with a computer program. The program includes instructions that, when executed by the one or more data processing devices, cause the one or more data processing devices to perform operations. The operations include: receiving a first signal corresponding to a first wavelength of light detected by the optical sensor placed near a patient's body; and receiving a second signal corresponding to a second wavelength of light detected by the optical sensor. For each received signal of the first signal and the second signal, the operation includes: separating the signal into an AC signal and a DC signal; separating the AC signal into component signals, wherein each of the component signals represents a frequency-limited band; identifying a fundamental pulse rate from the component signals or from another source such as a heart rate detector based on an ECG; analyzing the component signals by fractional phase transformation to identify a desired component signal and harmonic signals associated with the desired component signal, all of which are locally associated with the fundamental pulse rate in time; smoothing the desired component signal, the harmonic signal, and the DC signal; and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulation signal. The operation includes: generating a modulation ratio signal based on the modulation signal derived from the first signal and the modulation signal derived from the second signal; determining the peripheral blood oxygen saturation (SpO2) of the patient's body based on the modulation ratio signal; and causing a display device to display a value representing the determined SpO2 of the patient's body.

[0011] In another aspect, an apparatus includes a non-transitory computer-readable storage medium encoded with a computer program. The program includes instructions that, when executed by one or more data processing devices, cause the one or more data processing devices to perform operations. The operations include: receiving a first signal corresponding to a first wavelength of light detected by an optical sensor placed near a patient's body; and receiving a second signal corresponding to a second wavelength of light detected by the optical sensor. For each received signal of the first signal and the second signal, the operations include: separating the signal into an AC signal and a DC signal; separating the AC signal into component signals, wherein each of the component signals represents a frequency-limited band; identifying a fundamental pulse rate from the component signals or from another source such as a heart rate detector based on an ECG; analyzing the component signals by fractional phase transform to identify a desired component signal and harmonic signals associated with the desired component signal, all of which are locally associated with the fundamental pulse rate in time; smoothing the desired component signal, the harmonic signal, and the DC signal; and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal. The operations include: generating a modulation ratio signal based on a modulation signal derived from the first signal and a modulation signal derived from the second signal; and determining a peripheral blood oxygen saturation (SpO2) of a patient's body based on the modulation ratio signal.

[0012] On the other hand, a method includes: receiving a first signal corresponding to a first wavelength of light detected by an optical sensor placed near a patient's body; and receiving a second signal corresponding to a second wavelength of light detected by the optical sensor. For each received signal in the first signal and the second signal, the operation includes: separating the signal into an AC signal and a DC signal; separating the AC signal into component signals, wherein each of the component signals represents a frequency-limited band; identifying a fundamental pulse rate from the component signals or from another source such as a heart rate detector based on an ECG; analyzing the component signals by fractional phase transformation to identify a desired component signal and a harmonic signal associated with the desired component signal, all of which are locally associated with the fundamental pulse rate in time; smoothing the desired component signal, the harmonic signal, and the DC signal; and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulation signal. The method includes: generating a modulation ratio signal based on a modulation signal derived from the first signal and a modulation signal derived from the second signal; and determining the peripheral blood oxygen saturation (SpO2) of the patient's body based on the modulation ratio signal.

[0013] Various other functions and benefits of such a system will be apparent from the preceding detailed description and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a block diagram illustrating an example of a pulse oximetry system.

[0015] Figure 2 is a block diagram illustrating an example of an input processing and frequency band separation unit.

[0016] Figure 3a is a diagram showing the impulse response of one frequency band of the analysis wavelet group.

[0017] Figure 3b is a graph showing the frequency response of the real part of the analysis wavelet group.

[0018] Figure 4 An example of a pulse oximetry system including a patient-worn sensor in communication with a display device is shown.

[0019] Figure 5 is a flow chart illustrating an example of operations performed by a pulse oximetry system.

[0020] Figure 6 is a block diagram illustrating an example of a computing system that can be used to implement a pulse oximetry system.

[0021] Figure 7 is a block diagram illustrating an example of a computing system depicting a host system that may be used to implement a pulse oximetry system.

[0022] Like reference numbers and designations throughout the various drawings represent like elements. DETAILED DESCRIPTION

[0023] Figure 1 1 is a block diagram illustrating an example of a pulse oximetry system 100. The system 100 includes electronic components, such as hardware and / or software, that facilitate pulse oximetry. The system 100 estimates the time-averaged percentage oxygen saturation (SpO 2) and the time-local average of the pulse rate (PR) of the patient. In some implementations, a portion of the system 100 is included in a patient monitoring device that can continuously collect physiological vital sign data, such as electrocardiogram (ECG), heart rate, respiratory rate, blood oxygen saturation, blood pressure, temperature, and pulse rate. Examples of patient monitoring devices include chest sensors, such as those described in U.S. Publication No. 2015 / 0302539, entitled “Patient Care and Health Information Management Systems and Methods” filed on October 22, 2015, and U.S. Publication No. 2016 / 0228060, entitled “Patient Worn Sensor Assembly” filed on February 9, 2016; and other sensor assemblies connected to peripheral devices, such as the sensor assemblies described in U.S. Publication No. 2018 / 0213859, entitled “Sensor Connection Assembly for Integration with Garment” filed on January 25, 2018, the disclosures of which are incorporated herein by reference in their entirety. In some implementations, a portion of system 100 is included in a monitoring station that uses signals received from one or more peripheral devices, such as a chest sensor, a fingertip probe, an earlobe probe, or a non-invasive blood pressure (NIBP) cuff, to analyze and display physiological vital sign data. Examples of monitoring stations include bedside monitors and central server stations also described in US Publication No. 2015 / 0302539.

[0024] The system 100 includes an optical sensor 102 that generates a photoplethysmographic (PPG) signal (represented as signals I and R) corresponding to two wavelengths of light, such as an infrared portion and a red portion of the spectrum, sensed by the optical sensor 102 at a location peripheral to a patient, such as a fingertip or an earlobe. The optical sensor 102 may be attached to a fingertip or an earlobe of the patient, for example. The sensor 102 may include a light source and a detector, such as implemented using a light emitting diode (LED) and a photodiode, respectively. The sensor 102 may continuously acquire a PPG to facilitate continuous blood oxygen saturation and pulse rate monitoring. The sensor 102 may sample the PPG at a rate of 74 Hz to 76 Hz so that light interference at 50 Hz and 60 Hz may be filtered out.

[0025] System 100 processes I and R signals simultaneously via parallel channels. For each channel, the system 100 includes an input processing and band separation unit 104, an analysis wavelet group 106, a two-phase (QP) domain harmonic detection and tracking unit 108, and a QP domain harmonic state vector filtering and amplitude statistics processing unit 110. U.S. Patent No. 7,702,502, entitled "Apparatus for Signal Decomposition, Analysis and Reconstruction" filed on February 23, 2006, U.S. Patent No. 7,706,992, entitled "System and Method for Signal Decomposition, Analysis and Reconstruction" filed on February 23, 2006, U.S. Patent No. 8,086,304, entitled "Physiological Signal Processing to Determine a Cardiac Condition" filed on November 26, 2008, and U.S. Patent No. 8,086,304, entitled "Physiological Signal Processing to Determine a Cardiac Condition" filed on March 17, 2014, and U.S. Patent No. 8,086,304, entitled "Physiological Signal Processing to Determine a Cardiac Condition" filed on November 26, 2008, and U.S. Patent No. 8,086,304, entitled "Physiological Signal Processing to Determine a Cardiac Condition" filed on November 16, 2014, and U.S. Patent No. 8,086,304, entitled "Physiological Signal Processing to Determine a Cardiac Condition" filed on November 2 ... No. 9,530,425, entitled "Tracking" includes disclosure of techniques suitable for processing I and R signals when they are viewed as signals that generally have quasi-periodic components. The entire contents of U.S. Patent Nos. 7,702,502, 7,706,992, 8,086,304, and 9,530,425 are incorporated herein by reference.

[0026] For a system with ratiometric metrics, the two processing channels can have processing dynamics (e.g., filtering, delays, etc.) that are as similar as possible so that dynamic changes in the sensing system (e.g., optical sensor 102) will ripple through the processing channels in a time-aligned and phase-aligned manner. Therefore, when ripple effects arrive at the ratio point together, they tend to cancel and are thereby minimized at the output, providing robustness to transients. To this end, the system 100 can maintain symmetry between the parallel channels at each point to ensure delay synchronization between the I and R channels and between the AC and DC channels.

[0027] In some implementations, the input processing and band separation unit 104 performs sample interpolation to increase the sampling rate of the PPG signal to support the frequency resolution required for subsequent QP domain harmonic processing, performs filtering to reduce the impact of out-of-band artifacts and generally improve the SNR, and performs band separation of the PPG signal into AC and DC components. Figure 2 is a block diagram illustrating an example of the input processing and band separation unit 104. The input processing and band separation unit 104 reduces out-of-band interference and artifacts in the original PPG signal by band-pass filtering the AC component and performing multi-stage low-pass filtering on the DC component.

[0028] In some implementations, the input processing and band separation unit 104 includes a band-limited interpolation unit 202. The band-limited interpolation unit 202 includes an interpolation unit 204 and a low-pass filter 206 to perform band-limited interpolation of the PPG signal to attenuate high frequency (HF) interference and increase the input sampling rate by a factor N. ITP In some implementations, the interpolation unit is bypassed so that in practice N ITP =1.

[0029] As described in more detail below with respect to the QP domain harmonic detection and tracking unit 108, the system 100 performs extraction and tracking of time-varying harmonic components of the pulsatile modulated portion of the PPG signal. The highest frequency to be resolved, f HI The harmonics n (and hence the system passband) are determined by the product of: 1) the number of harmonics to be resolved H and 2) the highest pulse rate r for which the highest harmonic is to be resolved HI A low pass filter 206 having a corner substantially close to this frequency can attenuate out-of-band high frequency interferers and increase the system SNR.

[0030] At normal pulse rates, e.g., ranging from normal sinus rhythm (NSR) to mild exercise (~120 beats per minute (BPM)), most of the harmonic power (~90%) is contained essentially in the first three harmonics of the PPG signal. Above this range, the harmonic power steadily decreases with increasing pulse rate, so that above ~180 BPM, the signal is dominated by the first harmonic, i.e., the morphology becomes quasi-sinusoidal. Thus,

[0031] f HI =n H × HI =3×120BPM=360BPM=6Hz.

[0032] Frequency f HIis the highest signal frequency of interest, nominally defining the highest frequency wavelet band required in the analysis wavelet set 106 (described below) for extracting harmonic content from the PPG signal. As described in the patents incorporated herein by reference regarding techniques suitable for processing I and R signals, the analysis wavelet set 106 typically limits the scaling interval width to an integer number of samples to maintain high computational efficiency. This limitation may result in increasingly coarse frequency resolution at increasingly higher wavelet center frequencies, i.e., the wavelet band center frequencies are increasingly quantized, as the scale of the contraction becomes increasingly affected by the temporal quantization imposed by the integer-sample scale width. An added challenge is the desire to maintain sufficient resolution at higher frequencies to maintain separation of each harmonic component to be resolved.

[0033] F of the original PPG signal si =75 Hz will result in quantization of the center of the bands sufficient to produce inter-band response gaps that are large enough to prevent unambiguous resolution of harmonic amplitudes at the highest pulse rates. To address this issue, in some implementations, the input processing and band separation unit 104 interpolates samples for processing by the analysis wavelet group 106 by resampling the input signal stream (i.e., "upsampling"). The band-limited interpolation unit 202 provides band-limited interpolation achieved by the following combination: zero insertion performed by the interpolation unit 204, followed by synchronous low-pass filtering by the low-pass filter 206 to "fill" the added zeros. The corner frequency of the low-pass filter can be set to the highest expected signal frequency to further provide band limiting, thereby optimizing the SNR for high-frequency noise and artifacts. This upsampling is performed at the original input sampling rate F si Integer multiple N ITP Provide upsampling rate F s You can use N ITP =3 to provide the correct resolution f HI = The frequency resolution required for the third harmonic at 6 Hz, resulting in F s = Nominal internal wavelet processing sampling rate of 225 Hz.

[0034] Interpolation unit 204 may use zero insertion, which is a computationally efficient operation. Low pass filter 206 is also implemented with high computational efficiency by using a scaled integral kernel operator as described in U.S. Pat. No. 7,706,992. Interpolation unit 204 inserts (N) between each consecutive input sample. ITP -1) zero-valued samples to factor N ITP The raw PPG signal x is upsampled. This is followed by a synchronous low pass filter 206 implemented using a scaled integration kernel. As described in US Pat. No. 7,706,992, the frequency response of this digitally scaled integration kernel approximately follows a sine function in amplitude, with (F s / w), where w is the integrator scale parameter in samples, and F s is the upsampling rate. Set w to N ITP Integer multiples of will set the response to zero at the new sampling rate, thus eliminating any time domain ripple artifacts introduced by the zero interpolation process. The scale is therefore set according to the following formula:

[0035] w p =N ITP ·round(F s / (N ITP ·f HI / k 6dB )),

[0036] This yields the closest scale in the sample (at rate F s sampling), which is N ITP integer multiples of f to place the -6dB low-pass corner frequency as close to f as possible. HI , where factor k 6dB is the ratio of the -6 dB point of the low pass filter 206 to the first frequency response zero and depends on the order of the scaling integrator operator.

[0037] Then, by the integer factor (w p / N ITP ) size (i.e., the number of input samples within the scale of the integrator) and the order of the scaling integrator to control the overall smoothness and out-of-band HF suppression of the final interpolated result. A second-order integrator kernel can provide sufficient smoothness and out-of-band HF suppression. For F si = 75 Hz nominal input sampling rate, the following parameters can be used for the band-limited interpolator scaling integral:

[0038] N ITP =3, F s =225Hz, f HI =6Hz

[0039] <![CDATA[w p ]]> <![CDATA[N i ]]> <![CDATA[D ITP ]]> 18 2 17

[0040] Table 1

[0041] Among them, for N i =2, k 6dB =0.44295, and D ITP At the rate F s The inherent delay of the next sample.

[0042] The AC component of the upsampled and band-limited PPG signal is separated by a high pass filter 208. The high pass filter 208 is designed to remove the large DC offset associated with the raw light transmission in the PPG signal, give an output centered around a zero baseline, and suppress as much low frequency artifacts as practical below the lower nominal pulse rate band edge (in some implementations, 30 BPM or 0.5 Hz, corresponding to a 2 second beat-to-beat pulse interval). In some implementations, the high pass filter 208 may have a linear phase response. Given the size of the time scale, the high pass filter 208 may have a minimum phase response to minimize the time lag of the system due to the inherent group delay of the filter. Using an IIR filter topology in recursive form may produce a memory efficient filter architecture. A second order response may be used to provide good stability and adequate rejection in the stop band. The high pass filter 208 is specified according to the following analog domain transfer function:

[0043]

[0044] Among them, ω c controls the corner frequency (in radians per second), and Q controls the resonance of the filter at the corners. The value of Q can be set, for example, to Q = 0.720 to maximize the flatness of the response in the passband and the sharpness of the response at the corners, while minimizing ringing in the impulse response. For this value of Q, the formula ω c =2πk Q f c At frequency f c = 0.5Hz and factor k Q =1.325 to set the -6dB point of the filter.

[0045] High pass filter 208 may be implemented digitally using a second order difference equation of the form: 0 y HP (n) = b 0 x(n)+b 1 x(n-1)+b 2 x(n-2)-a 1 y HP (n-1)-a 2 y HP (n-2)

[0046] Among them, x(n) and y HP (n) are the signal input and output of the high-pass filter 208 at sampling time n. The filter response is given by the coefficient a i and b jThe high pass filter 208 may employ a bilinear transformation to produce a digital filter that approximates the analog model, as it can provide a very close match of the analog filter response, particularly in its passband and transition band. The resulting digital filter coefficients are as follows:

[0047] <![CDATA[a 0 ]]> 1.0 <![CDATA[a 1 ]]> -1.97429463444953 <![CDATA[a 2 ]]> 0.974632543411999 <![CDATA[b 0 ]]> 0.987231794465383 <![CDATA[b 1 ]]> -1.97446358893077 <![CDATA[b 2 ]]> 0.987231794465383

[0048] Table 2

[0049] The high pass filter 208 additionally provides a complementary low pass filter output formed as the path difference between the input and output of the high pass filter 208, producing y in the sampled domain. I_HP (n) = x(n) - y HP (n). Although a low pass filter formed in this manner generally does not have a particularly steep cutoff characteristic in its stop band, its group delay tends to match the group delay of high pass filter 208 in the transition band, thereby tending to maintain time alignment of the corresponding processing stages along the AC and DC signal paths for both the R and I signals.

[0050] The output y of the high pass filter 208 HP is tapped to produce x ac , thereby forming a filtered AC component of the original PPG signal (both R and I), from which the modulation amplitude is estimated. Since this waveform contains the pulsatile component of the PPG signal, it can also be used for estimation of the pulse rate PR. The signal itself can also be used to display a clean, zero-centered PPG waveform (e.g., on a display 120, such as a bedside monitor or central nursing station described in U.S. Publication No. 2015 / 0302539). Typically, the filtered AC component generated by the I signal will be used for this purpose because the amplitude of the I signal is typically larger.

[0051] The complementary low-pass output y I_HP is tapped to form x LP , thereby forming a time-aligned, locally separated DC component and thus providing a higher resolution than f with almost no additional computational cost. c The AC component of VS is advantageously used as an input to the low pass filter 210 with some initial low pass attenuation.

[0052] A dedicated linear phase low-pass smoothing filter 210 separates the DC component of the upsampled / band-limited PPG signal. The low-pass filter 210 is designed to remove all residual pulsatile and transient components from the raw PPG signal to provide a moving average, time-local value representing the raw light transmission in the PPG signal. In addition, its time delay is aligned to match the time delay of the corresponding processing path of the AC component. The low-pass filter 210 is implemented using a scaling integrator kernel, resulting in a computationally efficient implementation.

[0053] like Figure 1 and Figure 2 As shown, the AC processing path includes an AC component high-pass filter 208, an analysis wavelet group 106, and a QP domain harmonic detection and tracking unit 108 before the input of the QP domain harmonic state vector filtering and amplitude statistics processing unit 110, wherein the processing of the AC component occurs together with the separated DC component. According to the previously described principle of minimizing transients due to potential signal changes, the inherent delays of the AC and DC processing paths to the QP domain harmonic state vector filtering and amplitude statistics processing unit 110 are designed to be equal. These delays, respectively, are from Figure 2 The input x of the interpolation unit 204 is Figure 1 The QP domain harmonic state vector filtering and amplitude statistics processing unit 110 is D ac and D dc ,as follows:

[0054] D ac =D ITP +D HP +D BNK

[0055] D dc =D ITP +D HP +D LP

[0056] Among them, D BNK is the intrinsic delay of the analysis wavelet group 106, D HP is the group delay of the high pass filter 208 of the input processing and band separation unit 104, and D LP is the inherent delay of the DC component low pass filter 210 of the input processing and band separation unit 104. The DC path delay includes D HP , because as mentioned above, the input of the DC component low-pass filter 210 is taken from the complementary low-pass output x of the high-pass filter 208 LP The QP domain harmonic detection and tracking unit 108 is a memoryless operation and thus has virtually zero intrinsic delay. ac and D dc Equal, D LP =D BNK The size of the DC component low-pass filter 210 that meets this requirement is determined according to the delay by the following formula:

[0057] w p =2·D LP / N i +1

[0058] Only 2·D LP YesNi The scale is an integer value, and if N i =2 and D LP As will be discussed below, in one example, the analysis wavelet group 106 has an inherent delay D BNK =1833. For F s =225 Hz processing sampling rate, the parameters of the low-pass filter 210 are as follows:

[0059] <![CDATA[w p ]]> <![CDATA[N i ]]> <![CDATA[D LP ]]> 1834 2 1833

[0060] Table 3

[0061] The -6dB low pass cutoff of the low pass filter 210 is F c =k 6dB ·F s / w, or 0.0543 Hz in this example, which can result in almost complete attenuation of the pulsating component and very effective DC smoothing (note that for N i =2, k 6dB =0.44295). The output of the low pass filter 210 is a mobile estimate of the original DC transfer component of the PPG signal (both R and I), time aligned with the mobile modulation amplitude estimate for the purposes of normalization and subsequent processing.

[0062] Reference Figure 1 , the analysis wavelet group 106 processes the AC component of the PPG signal. The analysis wavelet group 106 performs the AC component to n B Separation of component frequency bands, each with significantly increased SNR. As described in U.S. Pat. No. 7,706,992 and U.S. Pat. No. 9,530,425, the analysis wavelet group 106 separates the input AC signal into quasi-periodic components, each of which contains information that is significantly bandwidth-limited relative to the input signal bandwidth. Thus, those frequency bands that contain information associated with local periodicity contained in the input signal will present that information with increased SNR.

[0063] The analysis wavelet group 106 can be designed as a group of bandpass filters with adjacently spaced frequency band centers. Each bandpass filter responds to the periodicity present in the group input signal by different amounts according to its frequency response, and has a maximum output amplitude for the frequency closest to the center frequency of the bandpass. For a group input signal containing (quasi) periodic components, the filters that essentially have the maximum output amplitude at a certain point in time will be those frequency bands whose centers are closest to the frequencies of those signal components, relative to adjacent filters. Tracking the frequency band amplitude as a function of the frequency band frequency at that point in time will reveal the "peaks" at those associated frequencies. Finding and tracking these amplitude peaks as a function of time, as well as the basic information in the frequency bands indexed by these peaks, forms the basic principle for designing the analysis wavelet group 106. Another advantage of using bandpass filters is that each filter in the group has a relatively narrow bandwidth and therefore typically has a higher SNR relative to the associated frequency components.

[0064] The design consideration of the analysis wavelet group 106 is the so-called "time-frequency frame" of the system 100 controlled by the "Q" (relative bandwidth) of the bandpass filter. Closely related to the time-frequency frame is the frequency spacing of the bands. Excessive band overlap reduces the selectivity between adjacent bands, which tends to "blur" the sharpness of the spectral peaks used to identify the associated components. However, insufficient band overlap may result in excessive amplitude loss in the crossover region between the band centers, leaving "holes" in the spectral representation of the signal, and in extreme cases, causing blurring of the spectral peak pattern. A higher Q will produce a narrower filter bandwidth, allowing the filters to be more closely spaced in frequency for a given amount of band overlap, thereby providing increased frequency resolution. However, achieving a narrower filter bandwidth may require having an impulse response that is wider in time, which in turn translates into a lower time resolution. Therefore, the choice of Q represents a trade-off between frequency resolution and time resolution.

[0065] As described in U.S. Patent No. 9,530,425, processing in QP space allows time-frequency information to be captured with greater time and frequency accuracy than can be conveyed by simply using the center frequency of a given band as a rate estimate. This feature completely relaxes the frequency resolution requirements between the centers of the bands to resolve frequencies. In theory, a minimum of two bands per octave is required to isolate the fundamental harmonic from the second harmonic of the input periodic component, although in practice more bands may be required to achieve low crossover losses between band responses.

[0066] A wavelet support (impulse response length) of approximately 3 cycles long may provide a good combination of frequency isolation of the components and a sufficiently fast response time to resolve the expected temporal dynamics associated with changes in pulse rate. Inter-band crossover rolloff levels of less than -0.5 dB from the center of the band may provide sufficient amplitude accuracy in representing component amplitudes for pulse oximetry. Therefore, the band spacing may be set to 9 bands per octave.

[0067] As described in U.S. Pat. No. 7,706,992, the analysis wavelet set 106 can be implemented using an enhanced form of the Kovtun-Ricci wavelet transform with a parallel form of the analysis component band. The wavelet consists of a cascaded scaled derivative kernel and a scaled integral kernel. The scaled derivative kernel has a z-domain transfer function:

[0068]

[0069] For a frequency band p with a scale parameter k p and the order parameter N dp .

[0070] In addition to the scaled integral kernel and the scaled difference kernel, the analysis wavelet set 106 may include an additional kernel type in the kernel cascade, referred to herein as a scaled comb kernel, having a z-domain transfer function:

[0071]

[0072] For a frequency band p with a scale parameter k p and c p And the order parameter N cp The intrinsic latency of this core is D cp =N cp ·k p ·(C p -1). cp = 1, the frequency response of the scaled comb kernel is a Dirichlet function, which closely approximates a sine function at its first main lobe around DC, and then p The controlled spacing repeats these main lobes in frequency (specifically, the frequency spacing is F s / (2k p ), where F s is the processing sampling rate). At this value k p By design it is set to the scaling value k that appears in the expression for the scaled difference kernel p This centers the first repeating mainlobe of the scaled comb core on the first mainlobe of the scaled difference core, defining a new, more selective bandpass function centered on the peak of the mainlobe.

[0073] The scaled comb kernel is characterized by zero responses on either side of the main lobe of each repetition, according to the scale parameter c p is spaced from each center so that for c p = 2, the zero value is halfway between the center of the main lobe, and for larger c p , the zeros are moved proportionally closer to the center of the main lobe. These zeros have the effect of narrowing the bandwidth of the main lobe, effectively increasing the Q of the filter. Q also increases with the order N cp The core is therefore characterized by two parameters by which Q can be controlled. The performance can be further enhanced because even for c p = 2, the increased Q obtained due to the increased side zero value results in less impulse response broadening than when similar Q is obtained by simply increasing the scaled derivative filter order. On the other hand, increasing the order may reduce the height of the side lobes between the main lobe images, so the order N cp becomes a means of controlling stopband suppression in the region just beyond the first passband zero.

[0074] In a broadband sense, the scaled comb kernel acts as a comb filter, i.e., it has lobes around DC and scales according to k. p Repeating lobes. To obtain a pure bandpass response, the first repetition of the main lobe response above DC can be isolated, which is at frequency F s / (2k p ) is centered. To suppress the lobe at DC, a scaled difference kernel is used, since it has a main response zero at DC. A scaled integration kernel is then used to suppress the mainlobe image above the first, as well as to provide an overall smoothing of the impulse response. In practice, using these two kernels in the role of stopband suppression often does not require a high order (e.g., an order of 4 usually provides sufficient out-of-band suppression).

[0075] Therefore, the parameter used to program the bandpass wavelet response is k p 、c p 、N cp 、N dp 、w p and N ip For the Pth passband, k p controls the center frequency, and Q is determined (primarily) by c p and (minor) N cp and the scaling derivative order N dp Parameter N cp and N dp Controls the stopband rejection near DC and scales the integration scale w p and order Nip control the stopband rejection above the passband.

[0076] As described in U.S. Pat. No. 7,706,992, the scale parameter k of the transform waveletp controls the frequency of the peak of the fundamental bandpass lobe, while the scale parameter w p Determines the frequency of the first zero of the lowpass. Select w p ≈2k p The nominal value of / 3 places the first zero approximately at H cp The peak of the second bandpass lobe of (z), i.e., the first repetition of the comb, effectively isolates the basic bandpass response. The integration order parameter N ip Controls the degree of stopband attenuation for low-pass operation; value N ip =4 can provide effective performance under low computational load.

[0077] Derivative order parameter N dp Controls the bandwidth of the bandpass function. For filter banks, it controls the amount of overlap between the band responses. You can choose N dp The value of is such that the sum of the band responses is approximately flat over the power band, allowing the total power of the complex to be represented in a relatively unbiased manner with respect to frequency.

[0078] This family of bandpass wavelets is computationally efficient and characterized by a linear phase response. As described in US Pat. No. 7,706,992, the transform is accomplished by time alignment across the frequency bands to preserve the timing information of the base component as its rate varies across the frequency bands.

[0079] Based on the above description and derivation, and for the sampling period as previously specified for the PPG signal, the bandpass wavelets of the analysis wavelet group 106 may be implemented using the following design parameters:

[0080]

[0081]

[0082] Table 4

[0083] Value F Cp is the center frequency of the band, defined as the frequency at which the Pth band exhibits its peak amplitude response. The frequency range covers sufficient range to resolve the range required for the fundamental pulse rate (30 to 240 BPM) as well as the coverage required to resolve the third harmonic at up to 120 BPM. As described above, the bands are nominally spaced at 9 bands per octave, so that complete coverage of the desired rate range is achieved with 34 bands. The time domain impulse response produced by this wavelet order is characterized by an effective time length of approximately 3 cycles, and therefore represents good time resolution of local periodic characteristics.

[0084] Figure 3ais a graph 300 showing the impulse response resulting from implementing the above parameters for the first band of the filter bank. Since the wavelet is analytic, the impulse response is complex valued. The "real" and "imag" traces are the real and imaginary parts of the impulse response, respectively. The "mag" trace is the instantaneous complex amplitude at each time point. Figure 3b is a graph 350 showing the frequency response of the real part of the analysis wavelet group.

[0085] To ensure calibration of the relative responses across the filter bank, the band output signals are normalized to unity gain at the center of the band. Cp The real and imaginary magnitude responses are evaluated (separately) and the multiplicative inverse is taken to determine the calibration factor K CR and K CI The real and imaginary filter outputs can be multiplied by their respective calibration factors to obtain a calibrated filter bank. The calibration factor for each respective frequency band is thus:

[0086]

[0087]

[0088] Table 5

[0089] Reference Figure 1 The QP domain harmonic detection and tracking unit 108 performs QP transformation and state vector generation, based on harmonic pattern recognition in the QP state space for estimating each QP pulse interval PI QP Robust periodicity detection for n H Harmonic tracking of harmonic parameters in QP space. Through the QP transform, the detection and tracking unit 108 analyzes the quasi-periodic components of the pulsating signal to extract critical amplitude and local phase / frequency parameters occurring at QP update points, thereby generating a time-frequency QP state vector. QP update points are time points associated with updates to inherent information in the signal and occur naturally at a rate much lower than the fundamental sampling rate of the signal.

[0090] As described in U.S. Pat. No. 7,706,992, the detection and tracking unit 108 detects QP transitions on each component signal and forms an associated QP object at each transition. For the purpose of PPG processing, the QP object contains the phase value (represented by the phase label) at the time point of the QP detection, the time value and component amplitude, and the index of the frequency band for the basic component. As described in U.S. Pat. No. 7,706,992, the current QP at the current time point is collected into an array across frequency bands to form a QP state vector that is indexed by frequency band and updated when any QP object on any frequency band is updated. The concentrated time-frequency information contained in this state vector forms the basis for being able to identify and track the desired signal component and to make a high-precision estimate of its instantaneous frequency and amplitude.

[0091] The detection and tracking unit 108 further enhances the SNR by identifying, selecting, and tracking QP state information associated with the desired signal. The QP state vector provides instantaneous amplitude information in both frequency and time (e.g., time-frequency analysis), which can be analyzed for patterns that can be used to identify and distinguish the desired signal from artifacts. The current states of the associated QP amplitudes acquired across the frequency band form a QP state spectrum, whose peaks are associated with time-localized quasi-periodic components. Such component peaks associated with the desired signal include the pulse rate and its harmonics (pulse harmonic sequence), while other peaks may be associated with other quasi-periodic artifacts, such as those generated by respiration (at the respiratory rate) and / or by repetitive motion.

[0092] The detection and tracking unit 108 analyzes the pattern of peaks in the QP state spectrum to identify only those peaks associated with the desired signal by taking advantage of the typical harmonic pattern of the desired signal and the fact that other quasi-periodic artifacts tend to have harmonics of low or unstable amplitudes above the fundamental. Transients in the input signal appear simultaneously in multiple frequency bands and / or tend to be relatively short in time and are therefore suppressed either entirely or by filtering during the tracking process.

[0093] After peak detection, the detection and tracking unit 108 sorts the peaks by size to form a metric on them. To save computation time, only the first three peaks may be retained for harmonic sequence analysis, since any peaks outside this range may come from harmonics (and therefore already be part of the harmonic sequence) or from lower level artifacts. For similar reasons, only those peaks of these top three with amplitudes at least a minimum fraction of the first (largest) peak qualify for further analysis. This fraction may be set to 0.35. Under normal high SNR conditions, only one peak may qualify, so this peak may be used directly as a detection cycle.

[0094] Otherwise, for each remaining qualified peak, the QP magnitude spectrum is sampled at the band corresponding to the peak and the harmonic frequencies of the peak. This sequence of QP magnitudes is then summed to form a harmonic sequence magnitude metric (HTA), one for each qualified peak. As described above, the third harmonic may be used in the calculation. In the event that the band index of a harmonic falls above the provided band, as may occur for a higher frequency fundamental peak, the magnitude contributions of those harmonics are set to zero. These HTAs are again sorted according to the metric, with only the largest two retained.

[0095] The detection and tracking unit 108 then applies a set of peak-to-amplitude metric rules to the pair of HTAs to select the one that best correlates with the desired signal. In one implementation, the rule set is as follows:

[0096] ● If the larger peak has a higher frequency band

[0097] →Accept this peak (smaller, slower rate peaks may be artifacts)

[0098] ● Otherwise (peak maximum at lower rate: check absolute, relative rate rules)

[0099] ○ If the largest peak has a frequency band with a rate above the upper respiratory rate band limit

[0100] →Accept the peak (the rate is too high and becomes a breathing artifact)

[0101] ○ Otherwise, check the harmonic relationship of the peak pairs at the center of the frequency band

[0102] ■Calculate peak centroid (interpolation peak)

[0103] ■ Compare the interpolated band centroids of the harmonic patterns

[0104] ●Frequency ratio is within 2 / 1 or 3 / 1 range

[0105] ●Ratio within 1.2 bands @ 9 bands per octave

[0106] ■If the frequency band relationship is harmonic

[0107] Compare a pair of harmonic series to measure amplitude

[0108] ●If the smaller peak < 0.9 * the larger peak

[0109] → Accept the larger peak (the smaller peak may be the second harmonic)

[0110] Otherwise

[0111] → Accept the smaller peak (the larger peak may be an artifact at PR / 2)

[0112] ■ Otherwise (the frequency band relationship is not harmonic)

[0113] ●If the smaller peak > 0.62*larger peak

[0114] → Accept smaller peaks (peaks large enough to be pulses)

[0115] Otherwise

[0116] → Accept larger peaks (smaller peaks may be artifacts)

[0117] In the above rule, the upper respiratory rate band limit may correspond to band 12 at 65 BPM (corresponding to the rate of the upper limit of hyperventilation). In the case where only one peak appears in the spectrum, the peak is accepted by default. In the case where no peak appears in the spectrum, the frequency band with the largest amplitude in the spectrum may be accepted as the peak.

[0118] To calculate the peak centroid in the above rule, the detection and tracking unit 108 uses the centroid (centroid) calculation to calculate the interpolated band index. The center band is calculated as the band-by-band sum of the product of the band index and the corresponding band amplitude divided by the band-by-band sum of the band amplitude. The bands used in the calculation include the peak band and the cross-band. If the cross-band meets the symmetry criterion, they are all used together with the peak band. If the symmetry criterion is not met, the maximum cross-band is used together with the peak band.

[0119] The symmetry criterion is defined as the absolute difference in amplitude across the frequency bands divided by the average of these amplitudes (i.e., the amplitude similarity metric is equivalent to the Bland-Altman metric). This metric is compared to a threshold, which can be set to a value of 0.1 (i.e., + / - 10% similarity), and similarity is declared if the metric is not greater than the threshold.

[0120] The detection and tracking unit 108 uses the received HTA to select the frequency band on which it is based, which in turn is associated with the fundamental of the pulse period. The detection and tracking unit 108 then uses the QP state information associated with this frequency band and its harmonic frequency bands to update the dedicated tracking of these n H QP harmonic state vector of QP information of harmonics. Harmonics with band indices exceeding the upper limit are marked for appropriate processing in the QP domain harmonic state vector filtering and amplitude statistics processing unit 110.

[0121] Updates to the QP harmonic state information are indicated by an advance of the QP time index accompanied by a change in the phase label in any included harmonic (indicating a phase advance). These updates are detected for each harmonic in each of the I and R channels, and the corresponding amplitude and interval statistics are updated at these points. In order to form stable amplitude statistics, the QP amplitudes are individually smoothed by a scaled integration kernel operating in the QP space. The integration scale is set to a number of QPs that naturally match the fundamental wavelet scale support of the fundamental cardiac cycle. For system 100, this translates to 2.5 cardiac cycles, or 10 QPs at 4 QPs per cycle for the fundamental cycle. For the harmonics, the scale increases proportionally with the number of harmonics. This causes the harmonics to all closely follow the same time support, resulting in effectively synchronous smoothing on the same time scale for all harmonics. When a harmonic is marked as out of group-wide QP updates, the input to the amplitude smoothing is set to zero. A second-order scaled integration kernel can provide sufficient smoothing at this stage.

[0122] The tracked frequency band corresponding to the lowest harmonic has frequency information corresponding to the fundamental period of the PPG pulse wave and therefore to the pulse interval (PI). The detection and tracking unit 108 of the I channel determines the pulse interval estimate by calculating the instantaneous period of the fundamental harmonic, called the QP "atomic period". Extracting this atomic period provides the pulse interval PI at each QP update QP As described in U.S. Patent No. 7,706,992, this is based on calculating the time difference and phase difference from one QP update to the next QP update for that particular component, which in turn yields information about the period of the fundamental frequency component with high accuracy. The QP atomic period in this case is represented as follows:

[0123] PI QP =dT QP / dP QP

[0124] Value PI QP corresponds to the (instantaneous) atomic pulse spacing in seconds per cycle. For the fundamental harmonic, the value dT QP Corresponds to the time interval from the previous QP update to the current QP update, in seconds. QP corresponds to the phase advance value associated with the number of two-phase tags traversed from the previous QP state update to the current QP state update, and may be evaluated according to the following table:

[0125]

[0126] Table 6

[0127] As described in U.S. Pat. No. 9,530,425, the number of QP phase labels traversed corresponds to the increments encountered along the cyclic sequence ... A→B→C→D→A... in the current QP state transition. A base phase advance (e.g., increment 1) is the most common phase change. A zero phase change is possible if the current QP update is associated with a change in the selected base band that happens to be updated to the same phase label as the phase label of the selected base band at the previous QP. A 2 phase label change is equally possible with a change in the selected base band, but with the phase reversed from the previous band to the next. Phase reversals are typically associated with transients in the signal, or with bands that are far away from any significant periodic components and therefore exhibit low relative band amplitudes. The system 100 can be designed to mitigate both of these undesirable conditions, making phase reversals a very unlikely type of transition to encounter.

[0128] Then, the detection and tracking unit 108 can update the atomic pulse interval PI at each QP update of the fundamental harmonic according to the following conditions and rules: QP :

[0129] ●If PI QP Is an invalid floating point number (Inf, NaN) or a negative number → PI QP Forced to a period corresponding to the center frequency of the selected basic wavelet band

[0130] ●Otherwise, if PI QP corresponding to the center frequency of at least one wavelet frequency band above or below the selected basic wavelet frequency band

[0131] → Respectively QP Force to a period corresponding to the center frequency of a wavelet band above or below the selected base wavelet band

[0132] ●Otherwise, PI QP In line with the above calculation results

[0133] Tracking updates can be defined to occur when the QP of the fundamental harmonic is updated.

[0134] like Figure 1 As shown, the detection and tracking unit 108 of the I channel provides PI to the detection and tracking unit 108 of the R channel. QPand associated band index information. The basic pulse-related periodicity information based on the pulse modulation is essentially common to both PPG signals, since the timing of the modulation is primarily related to the cardiac cycle and perfusion mechanisms. Therefore, the identification of the frequency bands of the harmonic pattern can be calculated on one channel (the "master"), and the associated band selection information can then be passed to the other channel (the "slave") and used for harmonic tracking purposes. In the system 100, the I channel is the master channel because it typically exhibits a higher SNR. Therefore, the frequency bands of the QP state vector associated with the desired pulse harmonic sequence contain concentrated, high SNR time-varying amplitude and phase / frequency information that is sufficient to represent (and / or reconstruct) each harmonic of the pulse modulation component of the PPG signal, and by extension, the modulation component of the PPG signal itself.

[0135] The process of periodicity detection and periodic band selection occurs at each QP state update of the I channel, as does the update of its QP harmonic state vector. The QP harmonic state vector of the R channel is updated at each QP state update of the R channel, and the QP harmonic state vector of the R channel uses the current periodic band selected at the update time. The detection and tracking unit 108 provides these two QP harmonic state vectors to the QP domain harmonic state vector filtering and amplitude statistics processing unit 110.

[0136] System 100 includes performing QP synchronous filtering to generate instantaneous pulse interval PI inst The pulse interval (PI) filter 112 is used to filter the pulse interval (PI) of the reported pulse rate. To maintain time alignment in the reported pulse rate, the pulse interval PI is processed with the same QP time-based filtering / smoothing operation. QP , for example, the pulse spacing PI is processed by smoothing with a scaled integral kernel with the same parameters as those applied to the fundamental harmonic QP amplitude QP , as used in the QP domain harmonic state vector filtering and amplitude statistics processing unit 110, is updated at each QP update of the primary harmonic to produce an instantaneous pulse interval PI synchronized with the smoothed harmonic amplitude inst .

[0137] The QP domain harmonic state vector filtering and amplitude statistics processing unit 110 performs individual per-harmonic filtering of parameters (QP-space linear smoothing) to form a robust motion metric of harmonic amplitudes, performs QP synchronous filtering of the DC component of the PPG signal, combines the harmonic amplitudes to form a robust metric of AC component pulse amplitudes, and performs combining of the AC component pulse amplitudes with the DC information to form a robust motion metric of the normalized percent fractional PPG pulse modulation (A of the I signal path). NI , R signal path A NRThe QP domain harmonic state vector filtering and amplitude statistics processing unit 110 filters the QP amplitude parameters of each harmonic individually to provide statistical time-based smoothing for general suppression of variability and noise at the output of the tracking process. The separated DC component I of the PPG signal is processed by a parallel scaled integration kernel with the same parameters as those applied to the fundamental harmonic QP amplitude. dc and R dc Smoothing is performed and updated at the same QP update. This produces a smoothed DC component I that is synchronized with the smoothed harmonic amplitudes. dcs and R dcs .

[0138] By filtering each harmonic individually, the amplitude associated with each fine aspect of the morphology can be independently smoothed, thereby increasing independent robustness to interference that may occur in frequency bands near specific harmonics. In addition, because these harmonic state filters are independently updated at the QP update of each individual harmonic, computational savings relative to per-sample filtering can be achieved, especially at lower harmonics because the QP update rate scales proportionally with frequency. In order to maintain time alignment between the AC and DC components, unit 110 can also sample the DC component of the PPG signal at each QP update of the first harmonic (fundamental) and process them using the same QP time-based filtering / smoothing operation.

[0139] Unit 110 combines the smoothed harmonic amplitudes to form a robust motion estimate of the amplitude of the AC component (i.e., the pulse amplitude of the PPG signal). At each tracking update, the smoothed statistics may be combined as follows. The smoothed harmonic amplitudes of each PPG channel are linearly combined to form a motion estimate of the AC component corresponding to the PPG pulse amplitude. IR and a R The linear combination weights applied to the smoothed harmonic amplitudes need not be uniform. The weight on the first harmonic is reduced to suppress the effects of artifacts, which tend to affect the lowest harmonics most at low SNRs, without significantly affecting performance at higher SNRs. The weights of the linear combination can be in order of harmonic number as follows: 0.125, 1.0, 1.0.

[0140] Then the motion estimate a is calculated by the corresponding smoothed DC component value IR and a R Normalization is performed to obtain a normalized percentage fraction PPG pulse modulation robust movement metric, such as Figure 1 A in NI and A NR Normalized percentage fraction PPG pulse modulation A NI and A NR The movement metric of is calculated at each tracking update as follows:

[0141] A NI =a IR / I dcs

[0142] A NR =a R / R dcs

[0143] System 100 includes a divide unit 114 that provides A NI and A NR The mobility metric R of the normalized modulation ratio N . Normalized modulation ratio R N =A NR / A NI , from which SpO is formed 2 The instantaneous movement metric being mapped.

[0144] System 100 includes statistical filters 116a and 116b. Statistical filter 116a forms a normalized modulation ratio R N The smooth moving estimate R PPG The statistical filter 116b forms the instantaneous pulse interval PI inst Smooth moving estimate PI PPG Each statistical filter performs statistically based smoothing based on a moving stack of filters to provide a final smoothing function and remove any remaining large, far-away transient excursions from the final output. The same smoothing filter is applied in parallel to R N and PI inst The two are roughly time-aligned to produce the final mobile PPG metric R PPG and PI PPG .

[0145] The statistical filters 116a and 116b are designed to remove any remaining large, off-center transient excursions from the final output and to substantially stabilize the output readings. stat = 2.5 seconds may produce an acceptable level of smoothness while keeping the time lag as low as possible. Therefore, at each tracking update, the interval T before (and including) the current tracking update is used. stat R appears within seconds N and PI instThe average of each statistical filter 116a and 116b is updated with the corresponding value of the median. Rather than performing a simple moving average, the smoothed median is calculated by taking the average of the central values ​​in the sense of percentiles within a range around the pure median. The central range average can be set to include those values ​​within the interquartile range centered on the median to maintain robustness to outliers while also producing a well-smoothed output over time. By using the same parameters and update time, synchronization can be maintained between the two smoothed output statistics.

[0146] System 100 includes mapping units 118a and 118b. Mapping unit 118a performs R PPG Mapping to output SpO 2 The mapping unit 118b performs PI inst Mapping to output pulse rate PR value.

[0147] Normalized modulation ratio R PPG can be mapped to SpO by a polynomial function such as a second-order polynomial 2 The polynomial coefficients can be determined by a calibration procedure, for example, by collecting SaO from subjects following a standard ramp-down protocol in a hypoxia laboratory. 2 and the corresponding R PPG The second-order fit is then performed to form the SpO 2 A mapping of R can be supported by various probes, such as finger probes and ear probes, by using each probe to collect R PPG values, and then form a separate corresponding map for each probe.

[0148] The polynomial is evaluated as follows:

[0149]

[0150] Among them, R PPG is the input value x, and the output y is SpO 2 In one implementation, for the finger probe and the ear probe, the value a n as follows:

[0151] N <![CDATA[a n (Finger Probe)]]> <![CDATA[a n (Ear Probe)]]> 0 112.02185 115.49644 1 -21.632549 -27.554501 2 -4.2294321 -3.3103607

[0152] Table 7

[0153] To prevent the output value from drifting too far, the SpO 2 The output value range is limited to an upper limit value of 100% and a lower limit value of 0%.

[0154] Instantaneous pulse interval PI PPG Usually mapped to pulse rate PR by the following formula:

[0155] PR=60·F s / PI PPG

[0156] The pulse interval is expressed in samples and PR is expressed in BPM. To prevent the output value from being excessively offset, the PR output value range can be limited to an upper limit of 240 BPM and a lower limit of 30 BPM.

[0157] A partial loss of signal amplitude may occur due to complete signal loss, such as due to lead detachment, or due to lack of pulse modulation, such as due to extremely low perfusion or due to cardiac arrest or asystole. 2 and PR are not well defined under these conditions, so the system 100 includes an output qualification unit 122 that uses the amplitude level of the normalized modulation signal to qualify the pulse modulation amplitude to obtain a level sufficient to support the confidence level of the desired accuracy. The amplitude level can be based on the formation of the AC component I ac The magnitude of the movement estimate I ama and from the DC component I dc Time-aligned motion estimation I md , and the ratio of these values ​​is compared to a threshold located at the point where a significant drop in performance is experimentally observed.

[0158] By smoothing the AC components I separately with a moving average filter ac and DC component I dc The absolute value of the moving AC amplitude estimate I ama and DC component estimation I md The filter can be implemented as a cascade of scaled integrators with the same parameters, resulting in a time-aligned linear phase between the estimates. The smoothing can be set to second order with a time scale of 3 seconds to adequately smooth the pulsatile components from the estimates while still preserving the time response to the amplitude dips and rises. For F s = 225 Hz processing sampling rate, the following scaling integral parameters can be used for the smoothing filter:

[0159] <![CDATA[w sm ]]> <![CDATA[N i ]]> <![CDATA[D sm ]]> 675 2 674

[0160] Table 8

[0161] Logo f sig Used to limit the signal level of pulse oximetry, this flag f sigIt is valid for normalized modulation signal amplitude levels that are generally above the confidence threshold, and invalid for normalized modulation signal amplitude levels that are below the confidence threshold. In the case where the signal level moves slowly through the threshold, superimposing small ripples, jitter of the qualifying flag may result. This can be mitigated by adding hysteresis to the threshold (e.g., making the threshold anti-correlated with the qualifying state). In addition, due to the time lag inherent in the various processing stages, a time delay can be added to the validation and invalidation of the flag so that the indication of signal quality is aligned in time with the output reading of the pulse oximetry.

[0162] The output qualification unit 122 can meet these two goals by implementing the signal qualification process using a state machine. The state definition and state transition rules are as follows:

[0163] Initialization: state = NOT_QUAL / / f sig = False

[0164]

[0165] Table 9

[0166] The threshold can be set as follows:

[0167] Threshold (up, down) value threshU 0.00028 threshD 0.00014

[0168] Table 10 The delay time can be set as follows:

[0169] Delay(up, down) Value (seconds) TdelayU 22.5 TdelayD 5

[0170] Table 11

[0171] The output limiting unit 122 sets the signal limiting flag f sig Provided to output interface 124 as a means of suppressing display of false or erroneous pulse oximetry output during adverse signal conditions. The delay may be selected to cause the flag to be displayed before signal conditions deteriorate. sig Invalid, and after the signal condition is restored, the flag f sig Revalidate only when flag f sig Seeks to ensure favorable signal conditions when effective.

[0172] The system 100 may continuously display the average SpO on the display 120. 2 The display 120 may be a display of a bedside monitor, a central server station, or a mobile device as described in U.S. Publication No. 2015 / 0302539. If an alarm state is detected for a patient, the display 120 may alert the user of the display 120. The display 120 may alert the user by, for example, flashing, changing color, becoming larger, or otherwise visually indicating an alarm state. For example, if the patient's SpO2 If the level increases above an upper acceptable limit or decreases to a lower acceptable limit, the display 120 may flash to indicate an alarm state to a user of the display 120. The system 100 may alert the user by triggering an audible alarm. In some implementations, different audible alarms may be used to indicate different alarm states or severity of alarm states. The volume of the audible alarm may also be used to indicate the relative severity of the alarm state.

[0173] The system 100 may also alert the user of an alarm state by flashing one or more portions of the display 120, sounding an audible alarm, or sending an alarm to other devices. For example, a bedside monitor may send an alarm to a central server, which may then identify one or more caregivers to direct the alarm associated with the patient to these caregivers. The central server may then send an alarm to a device associated with the identified caregiver.

[0174] The pulse oximetry system 100 may be used in any setting where the patient's oxygenation is unstable, including critical care, surgery, rehabilitation, emergency and hospital ward settings, pilots on unpressurized aircraft to assess any patient's oxygenation, and determine the effectiveness or need for supplemental oxygen. The system 100 may provide continuous and instantaneous blood oxygen saturation values, which are critical in emergency medicine, and are also very useful for patients with respiratory or cardiac problems, particularly chronic obstructive pulmonary disease (COPD), or for the diagnosis of some sleep disorders such as apnea and hypopnea.

[0175] In some implementations, the system 100 can be implemented as a portable battery-powered pulse oximeter. Such a portable battery-powered pulse oximeter is useful for pilots working on non-pressurized aircraft above 10,000 feet (12,500 feet in the United States) where supplemental oxygen is required. Portable pulse oximeters are also useful for climbers and athletes whose oxygen levels may decrease at high altitudes or when exercising. A portable pulse oximeter can use the system 100 to monitor a patient's blood oxygen and pulse, serving as a reminder to check blood oxygen levels.

[0176] Figure 4A system 400 is shown including a patient-worn sensor 402 that communicates with a display device 404, such as a bedside monitor, a central server station, or a display of a mobile device as described in U.S. Publication No. 2015 / 0302539. The display device 404 displays information for the patient. The display device 404 can display a user interface 406, which includes information received from the patient-worn sensor 402 and / or information associated with the patient associated with the patient-worn sensor 402. The patient-worn sensor 402 can be, for example, a chest sensor as described in U.S. Publication No. 2015 / 0302539. The patient-worn sensor 402 can include contacts for attaching to the patient's skin and recording various patient vital signs such as blood pressure, body temperature, respiratory rate, body impedance, blood oxygenation, heart rhythm (via ECG) and heart rate.

[0177] The patient-worn sensor 402 can wirelessly communicate with the display device 404 via a wireless connection 408 using a wireless communication protocol such as Bluetooth, WiFi, or a cellular protocol. The patient-worn sensor 402 can send vital sign information about the patient to the display device 404 via the wireless connection 408. In some implementations, the patient-worn sensor 402 can process the collected vital sign information before sending the information to the display device 404, and in some implementations, the patient-worn sensor 402 can send the original vital sign information instead of the processed information or in addition to the processed information to the display device 404. The display device 404 is a touch screen device capable of receiving touch screen input, such as a tablet computer. In some implementations, the display device 404 can receive input from a keyboard, a mouse, an input button, or one or more devices capable of recognizing voice commands. In some implementations, the display device 404 can be controlled using a device that wirelessly communicates with the display device 404, such as a mobile phone. In some implementations, the display device 404 is a "headless" device that does not include direct user input and / or output functionality, but instead functions only as a processing device for processing raw vital sign information received from the patient-worn sensors 402, detecting alarm conditions, sending alerts to other devices in communication with the display device 404, and sending patient information to one or more central servers. In such a case, the display device 404 would not include a display.

[0178] The user interface 406 displays information about the patient associated with the patient-worn sensors 402. For example, electrodes of the patient-worn sensors 402 may be in contact with the patient's skin and collect vital sign information, which is transmitted to the display device 404 via the wireless connection 408, processed by the display device 404, and displayed as part of the user interface 406. The user interface 406 shows various vital sign waves and numeric levels.

[0179] For example, user interface 406 shows a patient's heart rate waveform 410 and a patient's digital heart rate or pulse rate value 412. In the example shown, the patient's pulse rate value 412 is 80 beats per minute. User interface 406 indicates at 414 that the patient's acceptable heart rate level falls between 50 and 120 beats per minute. Since the patient's current heart rate of 80 beats per minute falls within the indicated acceptable range, there is currently no alarm state about the patient's heart rate. This is indicated by a bell-shaped icon 416 superimposed with an "X" symbol. Icon 416 indicates that the patient's current heart rate is within an acceptable range. In the case of the patient's heart rate not being within an acceptable level, icon 416 can change to indicate an alarm state. For example, "X" can disappear from icon 416 and icon 416 can light up or flash to indicate an alarm state. In addition, display device 404 can emit an audible alarm to warn a nearby caregiver about the patient's alarm state. In some implementations, other parts of user interface 406 can flash or otherwise indicate an alarm state. For example, when the patient's heart rate exceeds an acceptable level, the displayed pulse rate value 412 may flash. In some implementations, the icon 416 (or other portions of the user interface 406) may flash at varying rates to indicate the severity of a particular alarm state. For example, the further the patient's heart rate is from an acceptable range, the faster the icon 416 flashes.

[0180] The user interface 406 also shows the patient's blood oxygenation waveform 418 and a numeric blood oxygenation value 420. The user interface 406 also shows the patient's acceptable blood oxygenation range at 422. The user interface 406 also includes an icon 424 indicating that the patient's blood oxygenation level is within the acceptable range (indicated by an "X" symbol superimposed on a bell, indicating that the alarm is "off").

[0181] Figure 5 5 is a flow chart illustrating an example of operation 500 performed by a pulse oximetry system. Operation 500 may be performed by a system of one or more computers, such as Figure 1 Operation 500 may include the details discussed above.

[0182] At 502, the system receives signals corresponding to wavelengths of light detected by an optical sensor placed near a patient's body. The signals include a first signal corresponding to a first wavelength of light detected by the optical sensor and a second signal corresponding to a second wavelength of light detected by the optical sensor.

[0183] Operations 504 to 514 are performed on each received signal of the first signal and the second signal. At 504, the signal is optionally upsampled to increase the sampling rate. At 506, the signal is separated into an AC signal and a DC signal.

[0184] At 508, the AC signal is separated into component signals, wherein each of the component signals represents a frequency limited band. Separating the AC signal into component signals may include: filtering the AC signal using a bandpass filter bank having adjacently spaced frequency band centers; and normalizing the output of the bandpass filter bank to generate the component signals.

[0185] At 510, the system analyzes the component signals by fractional phase transformation to identify the desired component signals and the harmonic signals associated with the desired component signals. To identify the desired component signals and the harmonic signals, the system can determine the atomic period by calculating the instantaneous period of the lowest harmonic signal derived from the first signal. For the first signal, the desired component signals and the harmonic signals can be identified based on the atomic period and by analyzing the relationship of the time-aligned amplitude values ​​of the component signals and the amplitudes of the harmonic signals. Based on the atomic period, the system can determine the atomic pulse interval, and the desired component signals and the harmonic signals of the second signal can be identified based on the atomic pulse interval.

[0186] At 512, the system smoothes the desired component signal, the harmonic signal, and the DC signal. At 514, the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal are combined to generate a modulated signal. Combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal may include: linearly combining the smoothed desired component signal and the smoothed harmonic signal to generate a combined signal; and normalizing the combined signal using the smoothed DC signal to generate the modulated signal.

[0187] At 516, the system generates a modulation ratio signal based on the modulation signal derived from the first signal and the modulation signal derived from the second signal. At 518, the system determines the peripheral blood oxygen saturation (SpO2) of the patient's body based on the modulation ratio signal. 2 ). Determine the patient's body's SpO 2 The method may include: generating a smoothed motion estimate of the modulation ratio signal; and mapping the smoothed motion estimate to an SpO2 representing the patient's body. 2The system can also determine the patient's body pulse rate based on the atomic pulse interval determined above.

[0188] Figure 6 6 is a block diagram illustrating an example of a computing system 600 executing a computer program that implements the above-described operations, data structures, and / or programs contained on a computer-readable medium 602 and executed by a central processing unit (CPU) 604. The CPU 604 may be an x86 series CPU or other CPU known in the art. In some implementations, input for executing the program may be implemented through a keyboard 606. In some implementations, input for executing the program may be implemented through a peripheral device such as a mouse, a light pen, a touch screen, a touch pad, or any other input device known in the art. In some implementations, the described operations may be called by another program to perform and process oximetry data. Once processed, the data processed using the above-described operations, data structures, and / or programs may be output to a display 608, or sent to another user terminal 612 via the Internet 610 or other wireless communication channels. In some implementations, the output may be placed in a database 614 or another database located off-site.

[0189] Figure 7 is a block diagram showing an example of a computing system depicting a host computer 700, which in some implementations may include a set of units or modules, such as a communication module 702 and associated communication link 704 for communicating with sensors worn by a patient, a microprocessor unit 706 (MPU) including a computer readable medium 708 containing a set of application instructions, an associated storage module 710 for storing patient data, a human-machine interface 712 through which an operator (e.g., a physician, technician, or patient) can interact with the MPU 706 and may include a visual display, an auditory display, a keyboard, a mouse, a touch screen, a tactile device, or other means of interaction. The host computer 700 also includes an input / output module 714 for connecting to and communicating with computer peripherals (e.g., printers, scanners, and / or data acquisition or imaging devices) in general, and a network communication module 716 for communicating with other hosts on a computer network such as Ethernet and / or a wireless network or WiFi.

[0190] The features described in the present disclosure may be implemented in digital electronic circuit systems, or in computer hardware, firmware, software, or a combination thereof. The apparatus may be implemented in a computer program product tangibly embodied in an information carrier such as a machine-readable storage device for execution by a programmable processor, and the method steps may be performed by a programmable processor executing an instruction program to perform the functions of the described implementation by operating on input data and generating output. The described features may advantageously be implemented in one or more computer programs, which are executable on a programmable system including at least one programmable processor, which is coupled to receive data and instructions from a data storage system, at least one input device, and at least one output device and to send data and instructions to the data storage system, at least one input device, and at least one output device. A computer program is a set of instructions that can be used directly or indirectly in a computer to perform a specific activity or produce a specific result. A computer program may be written in any form of programming language, including a compiled or parsed language, and it may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0191] Suitable processors for executing instruction programs include, for example, both general-purpose and special-purpose microprocessors, and the sole processor or one of multiple processors of any kind of computer. Typically, the processor will receive instructions and data from a read-only memory or a random access memory or both. The basic elements of a computer are a processor for executing instructions and one or more memories for storing instructions and data. Typically, a computer will also include one or more mass storage devices for storing data files, or be operably coupled to communicate with one or more mass storage devices for storing data files; such devices include disks, such as internal hard disks and removable disks; magneto-optical disks; and optical disks. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of non-volatile memory, including, for example, semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and memory may be supplemented by or incorporated into an ASIC (Application Specific Integrated Circuit). To provide interaction with a user, the features may be implemented on a computer having a display device such as a CRT (cathode ray tube) or LCD (liquid crystal display) monitor for displaying information to the user and a keyboard and pointing device such as a mouse or trackball through which the user can provide input to the computer.

[0192] Features can be implemented in a computer system that includes a back-end component such as a data server, or includes a middleware component such as an application server or an Internet server, or includes a front-end component such as a client computer with a graphical user interface or an Internet browser, or any combination thereof. The components of the system can be connected by digital data communications of any form or medium, such as a communication network. Examples of communication networks include LANs, WANs, and computers and networks that form the Internet, for example. A computing system may include a client and a server. The client and server are typically remote from each other and typically interact through a network such as the one described. The relationship between the client and the server is generated by means of computer programs that run on respective computers and have a client-server relationship with each other.

[0193] Although this specification contains many specific implementation details, these implementation details should not be interpreted as limitations on any invention or the scope that can be claimed, but as descriptions of features specific to a particular embodiment of a particular invention. Certain features described in the context of different embodiments in this specification may also be implemented in a combined manner in a single embodiment. Conversely, the individual features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments. In addition, although the features may be described as acting in a certain combination as described above and even initially claimed as such, in some cases, one or more features from the claimed combination may be deleted from the combination, and the claimed combination may be directed to a sub-combination or a variation of the sub-combination.

[0194] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that such operations be performed in the particular order shown or in a continuous order, or that all of the operations shown be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all implementations, and it should be understood that the described program components and systems can generally be integrated together in a single software or hardware product or packaged into multiple software or hardware products.

[0195] Thus, specific embodiments of the present subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired results. Additionally, the processes depicted in the accompanying drawings do not necessarily require the particular order or sequential order shown to achieve the desired results. In certain implementations, multitasking and parallel processing may be advantageous.

Claims

1. A non-transitory computer-readable storage medium encoded with a computer program, the program comprising instructions that, when executed by one or more data processing devices, cause the one or more data processing devices to perform operations comprising: receiving a first signal corresponding to a first wavelength of light detected by an optical sensor positioned proximate to a patient's body; receiving a second signal corresponding to a second wavelength of light detected by the optical sensor; For each received signal of the first signal and the second signal: Separating the received signal into an AC signal and a DC signal; separating the AC signal into component signals, wherein each of the component signals represents a frequency limited band, analyzing the component signals by fractional phase transform to identify a desired component signal and harmonic signals associated with the desired component signal, smoothing the desired component signal, the harmonic signal and the DC signal, and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal; generating a modulation ratio signal based on the modulation signal derived from the first signal and the modulation signal derived from the second signal; and The peripheral blood oxygen saturation (SpO2) of the patient's body is determined based on the modulation ratio signal.

2. The non-transitory computer-readable storage medium of claim 1, wherein: The analyzing step generates a fractional phase state vector corresponding to the desired component signal and the harmonic signal, The smoothing step includes smoothing the fractional phase state vector, and The combining step includes combining the smoothed DC signal and the smoothed fractional phase state vector to generate the modulated signal.

3. The non-transitory computer-readable storage medium of claim 2, wherein: Each of the fractional phase state vectors is updated at each transition between fractional phases, and Each of the smoothed DC signal, the smoothed fractional phase state vector, and the modulated signal is updated when one of the fractional phase state vectors is updated.

4. The non-transitory computer-readable storage medium of claim 1, wherein: The operations further include: for each received signal of the first signal and the second signal, converting the received signal having a first sampling rate into an upsampled signal having a second sampling rate, the second sampling rate being higher than the first sampling rate; and Separating the received signal into the AC signal and the DC signal includes separating the upsampled signal into the AC signal and the DC signal.

5. The non-transitory computer-readable storage medium of claim 1, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes analyzing time-aligned amplitude values ​​of the component signals.

6. The non-transitory computer-readable storage medium of claim 1, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes analyzing a relationship of amplitudes of the harmonic signals.

7. The non-transitory computer-readable storage medium of claim 1, wherein: The operations also include determining an atomic period by calculating an instantaneous period of the desired component signal.

8. The non-transitory computer-readable storage medium of claim 7, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes identifying the harmonic signal based on the atomic period.

9. The non-transitory computer-readable storage medium of claim 7, wherein: The operations also include determining an atomic pulse interval based on the atomic period.

10. The non-transitory computer-readable storage medium of claim 9, wherein: For the second signal, identifying the desired component signal and the harmonic signal is based on the atomic pulse spacing.

11. The non-transitory computer-readable storage medium of claim 9, wherein: The operations also include determining a pulse rate of the patient's body based on the determined atomic pulse interval.

12. The non-transitory computer-readable storage medium of claim 1, wherein: Separating the AC signal into the component signals comprises: filtering the AC signal using a bandpass filter bank having adjacently spaced frequency band centers; and The outputs of the bandpass filter bank are normalized to generate the component signals.

13. The non-transitory computer-readable storage medium of claim 1, wherein: Combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal comprises: linearly combining the smoothed desired component signal and the smoothed harmonic signal to generate a combined signal; and The combined signal is normalized using the smoothed DC signal to generate the modulated signal.

14. The non-transitory computer-readable storage medium of claim 1, wherein: Determining the SpO2 of the patient's body based on the modulation ratio signal includes: generating a smoothed motion estimate of the modulation ratio signal; and The smoothed motion estimate is mapped to an output value representative of SpO2 of the patient's body.

15. A computer-implemented method comprising: receiving a first signal corresponding to a first wavelength of light detected by an optical sensor positioned proximate to a patient's body; receiving a second signal corresponding to a second wavelength of light detected by the optical sensor; For each received signal of the first signal and the second signal: Separating the received signal into an AC signal and a DC signal; separating the AC signal into component signals, wherein each of the component signals represents a frequency limited band, analyzing the component signals by fractional phase transform to identify a desired component signal and harmonic signals associated with the desired component signal, smoothing the desired component signal, the harmonic signal and the DC signal, and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal; generating a modulation ratio signal based on the modulation signal derived from the first signal and the modulation signal derived from the second signal; and The peripheral blood oxygen saturation (SpO2) of the patient's body is determined based on the modulation ratio signal.

16. The computer-implemented method of claim 15, wherein: The analyzing step generates a fractional phase state vector corresponding to the desired component signal and the harmonic signal, The smoothing step includes smoothing the fractional phase state vector, and The combining step includes combining the smoothed DC signal and the smoothed fractional phase state vector to generate the modulated signal.

17. The computer-implemented method of claim 16, wherein: Each of the fractional phase state vectors is updated at each transition between fractional phases, and Each of the smoothed DC signal, the smoothed fractional phase state vector, and the modulated signal is updated when one of the fractional phase state vectors is updated.

18. The computer-implemented method of claim 15, further comprising: For each received signal of the first signal and the second signal, converting the received signal having a first sampling rate into an upsampled signal having a second sampling rate, the second sampling rate being higher than the first sampling rate; and Wherein, separating the received signal into the AC signal and the DC signal includes separating the up-sampled signal into the AC signal and the DC signal.

19. The computer-implemented method of claim 15, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes analyzing time-aligned amplitude values ​​of the component signals.

20. The computer-implemented method of claim 15, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes analyzing a relationship of amplitudes of the harmonic signals.

21. The computer-implemented method of claim 15, further comprising: The atomic period is determined by calculating the instantaneous period of the desired component signal.

22. The computer-implemented method of claim 21, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes identifying the harmonic signal based on the atomic period.

23. The computer-implemented method of claim 21 , further comprising: An atomic pulse interval is determined based on the atomic period.

24. The computer-implemented method of claim 23, wherein: For the second signal, identifying the desired component signal and the harmonic signal is based on the atomic pulse spacing.

25. The computer-implemented method of claim 23, further comprising: A pulse rate of the patient's body is determined based on the determined atomic pulse intervals.

26. The computer-implemented method of claim 15, wherein: Separating the AC signal into the component signals comprises: filtering the AC signal using a bandpass filter bank having adjacently spaced frequency band centers; and The outputs of the bandpass filter bank are normalized to generate the component signals.

27. The computer-implemented method of claim 15, wherein: Combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal comprises: linearly combining the smoothed desired component signal and the smoothed harmonic signal to generate a combined signal; and The combined signal is normalized using the smoothed DC signal to generate the modulated signal.

28. The computer-implemented method of claim 15, wherein: Determining the SpO2 of the patient's body based on the modulation ratio signal includes: generating a smoothed motion estimate of the modulation ratio signal; and The smoothed motion estimate is mapped to an output value representative of SpO2 of the patient's body.

29. A blood oximetry system comprising: an optical sensor configured to emit light; Display device; one or more data processing devices coupled to the optical sensor and the display device; as well as A non-transitory computer-readable storage medium encoded with a computer program, the program comprising instructions that, when executed by the one or more data processing devices, cause the one or more data processing devices to perform operations comprising: receiving a first signal corresponding to a first wavelength of light detected by the optical sensor positioned proximate to a patient's body; receiving a second signal corresponding to a second wavelength of light detected by the optical sensor; For each received signal of the first signal and the second signal: Separating the received signal into an AC signal and a DC signal; separating the AC signal into component signals, wherein each of the component signals represents a frequency limited band, analyzing the component signals by fractional phase transform to identify a desired component signal and harmonic signals associated with the desired component signal, smoothing the desired component signal, the harmonic signal and the DC signal, and combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal; generating a modulation ratio signal based on the modulation signal derived from the first signal and the modulation signal derived from the second signal; determining the peripheral blood oxygen saturation (SpO2) of the patient's body based on the modulation ratio signal; and The display device is caused to display a value representing the determined SpO2 of the patient's body.

30. The system of claim 29, wherein: The analyzing step generates a fractional phase state vector corresponding to the desired component signal and the harmonic signal, The smoothing step includes smoothing the fractional phase state vector, and The combining step includes combining the smoothed DC signal and the smoothed fractional phase state vector to generate the modulated signal.

31. The system of claim 30, wherein: Each of the fractional phase state vectors is updated at each transition between fractional phases, and Each of the smoothed DC signal, the smoothed fractional phase state vector, and the modulated signal is updated when one of the fractional phase state vectors is updated.

32. The system of claim 29, wherein: The operations further include: for each received signal of the first signal and the second signal, converting the received signal having a first sampling rate into an upsampled signal having a second sampling rate, the second sampling rate being higher than the first sampling rate; and Separating the received signal into the AC signal and the DC signal includes separating the upsampled signal into the AC signal and the DC signal.

33. The system of claim 29, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes analyzing time-aligned amplitude values ​​of the component signals.

34. The system of claim 29, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes analyzing a relationship of amplitudes of the harmonic signals.

35. The system of claim 29, wherein: The operations also include determining an atomic period by calculating an instantaneous period of the desired component signal.

36. The system of claim 35, wherein: For the first signal, identifying the desired component signal and the harmonic signal includes identifying the harmonic signal based on the atomic period.

37. The system of claim 35, wherein: The operations also include determining an atomic pulse interval based on the atomic period.

38. The system of claim 37, wherein: For the second signal, identifying the desired component signal and the harmonic signal is based on the atomic pulse spacing.

39. The system of claim 37, wherein: The operations also include determining a pulse rate of the patient's body based on the determined atomic pulse interval.

40. The system of claim 29, wherein: Separating the AC signal into the component signals comprises: filtering the AC signal using a bandpass filter bank having adjacently spaced frequency band centers; and The outputs of the bandpass filter bank are normalized to generate the component signals.

41. The system of claim 29, wherein: Combining the smoothed desired component signal, the smoothed harmonic signal, and the smoothed DC signal to generate a modulated signal comprises: linearly combining the smoothed desired component signal and the smoothed harmonic signal to generate a combined signal; and The combined signal is normalized using the smoothed DC signal to generate the modulated signal.

42. The system of claim 29, wherein: Determining the SpO2 of the patient's body based on the modulation ratio signal includes: generating a smoothed motion estimate of the modulation ratio signal; and The smoothed motion estimate is mapped to an output value representative of SpO2 of the patient's body.

Images