`(16) Patent No.:
`US 6,393,311 B1
`
`Edgar, Jr. et al.
`(45) Date of Patent:
`May 21, 2002
`
`U5006393311Bl
`
`(54) METHOD, APPARATUS AND SYSTEM FOR
`REMOVING MOTION ARTIFACTS FROM
`MEASUREMENTS 0F BODILY
`PARAMETERS
`
`(75)
`
`Inventors: Reuben W. Edgar, Jr.; August J. A110,
`Jr., both of San Antonio, TX (US);
`Jesus D. Martin, Wallingford, CT
`(US); John R- DelFavero, East
`Hampton, CT (US); Mlchael B' Jafi'e,
`Cheshire, CT (US)
`
`.
`(73) ASSignee:
`
`.
`.
`€513): Technology Inc., Wilmington, DE
`
`5,368,026 A
`5,368,224 A
`5,398,680 A
`
`11/1994 Swedlow et al.
`11/1994 Richardson et al.
`3/1995 Polson et al.
`
`(List continued on next page.)
`
`OTHER PUBLICATIONS
`
`Dowla, et al., Neural Networks and Wavelet Analysis in the
`Computer Interpretation of Pulse Oximetry Data, Neural
`Networks for Signal Processing VI—Proc.
`IEEE, 1996
`IEEE
`Signal
`Process,
`Soc.,
`IEEE Workshop,
`0—7803—3550—3 (1996).
`
`( * ) Notice:
`
`Subject. to any disclaimer, the term of this
`patent is extended or adjusted under 35
`U'S'C' 154(b) by 0 days.
`
`Primary Examiner—Eric F. Winakur
`Assistant Examiner—Matthew Kremer
`(74) Attorney, Agent, or Firm—TraskBritt
`
`(21) Appl. No.2 09/410,991
`
`(22)
`
`Filed:
`
`Oct. 1, 1999
`
`(60)
`
`Related US. Application Data
`Provisional application No. 60/104,422, filed on Oct. 15,
`1998-
`
`Int. Cl.7 .................................................. A61B 5/00
`51
`)
`(
`(52) US. Cl.
`....................... 600/323; 600/336; 600/310;
`600/324
`(58) Field of Search ................. 600/309—311, 322—326,
`600/330—331, 336, 473’ 476; 356/39_41;
`369/60.01
`
`(56)
`
`References Cited
`U.S. PATENT DOCUMENTS
`
`4,800,495 A *
`4,942,877 A *
`4,955,379 A *
`5,025,791 A
`39:22??? 2
`,
`,
`5,299,120 A *
`5,349,952 A *
`5,351,685 A
`
`1/1989 Smith ......................... 600/322
`
`7/1990 Sakai et al.
`600/323
`........................... 600/366
`9/1990 Hall
`6/1991 lea
`3/133: £0153? et a: 1
`we ow e a.
`3/1994 Kaestle ....................... 600/310
`9/1994 McCarthy et al.
`.......... 600/473
`10/1994 Potratz
`
`(57)
`
`ABSTRACT
`
`A method for removing motion artifacts from devices for
`sensing bodily parameters and apparatus and system for
`effecting same. The method includes analyzing segments of
`measured data representing bodily parameters and possibly
`noise from motion artifacts. Each segment of measured data
`may correspond to a single light signal
`transmitted and
`detected after transmission or reflection throu h bodil
`g
`y
`tissue. Each data segment is frequency analyzed to deter-
`mine dominant frequency components. The frequency com-
`pcnent which represents at least one bodily parameter of
`interest is selected for further processing. The segment of
`data is subdivided into subsegments, each subsegment rep-
`resenting one heartbeat. The subsegments are used to cal-
`culate a modified average pulse as a candidate output pulse.
`The candidate output pulse is analyzed to determine whether
`it is a valid bodily parameter and, if yes, it is output for use
`in calculating the at least one bodily parameter of interest
`without any substantial noise degradation. The above
`method may be applied to red and infrared pulse oximetry
`signals prior to calculating pulsatile blood oxygen concen-
`tration. Apparatus and systems disclosed incorporate meth-
`.
`.
`.
`.
`Dds dISCIOSed accordmg to the Invennon‘
`
`34 Claims, 11 Drawing Sheets
`
`Start
`
`1
`OVYquQlO
`Acqu/reosegmsnt
`
`
`v
`
`mar
`
`,
`
`120/
`
`
`“0’” hammer-H
`
`v
`“gig“
`
`component
`v
`130 f\, Output average
`pulsestgnal
`
`
`
`
`
`v
`
`VES
`
`140 TV “$9332?
`NO
`
`End
`
`1
`
`APPLE 1005
`
`APPLE 1005
`
`1
`
`
`
`US 6,393,311 B1
`
`Page 2
`
`US. PATENT DOCUMENTS
`
`5,431,170
`5,448,991
`5,482,036
`5,490,505
`5,555,882
`5,588,427
`5,632,272
`5,645,060
`5,685,299
`5,713,355
`5,743,263
`5,769,785
`
`>>>>>>>>>>>>
`
`7/1995
`9/1995
`1/1996
`2/1996
`9/1996
`12/1996
`5/1997
`7/1997
`11/1997
`2/1998
`4/1998
`6/1998
`
`Matthews
`Polson et a1.
`Diab et a1.
`Diab et a1.
`Richardson et a1.
`Tien
`Diab et a1.
`Yorkey
`Diab et a1.
`Richardson et a1.
`Baker, Jr.
`Diab et a1.
`
`5,800,348
`5,803,910
`5,820,550
`5,852,638
`5,853,364
`5,885,213
`5,919,134
`5,934,277
`6,067,462
`6,098,038
`6,122,535
`
`>>>>>>>>>>>
`
`*
`
`9/1998
`9/1998
`10/1998
`12/1998
`12/1998
`3/1999
`7/1999
`8/1999
`5/2000
`9/2000
`9/2000
`
`Kaestle
`Potratz
`Polson et a1.
`................. 375/344
`Chen et a1.
`Baker, Jr. et a1.
`Richardson et a1.
`Diab
`Mortz
`Diab et a1. .................. 600/310
`Hermansky et a1.
`........ 704/226
`Kaestle et a1.
`.............. 600/322
`
`* cited by examiner
`
`2
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 1 0f 11
`
`US 6,393,311 B1
`
`100
`
`110
`
`120
`
`130
`
`Acquire a segment
`of raw data
`
`Analyze frequency
`components
`
`Determine best
`
`frequency
`component
`
`Output average
`pulse signal
`
`
`
`
`
`
`
`
`
`140 /\/
`
`New segment
`of raw data?
`
`FIG.
`
`1
`
`3
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 2 0f 11
`
`US 6,393,311 B1
`
`
`
`Raw Infrared Data Segment
`
`
`303360505 +
`
`'E
`3 3.50E+05
`0.
`
`
`
`
`
`
`i 3 4OE+O5
`T; 3 30E+05
`2
`(-5 3 20E+05
`Eu
`(/3
`
`+
`
`H
`
`3'10E 0510.0
`
`12.0
`
`18.0
`16.0
`14.0
`Time (Seconds)
`
`20.0
`
`
`
`4
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 3 0f 11
`
`US 6,393,311 B1
`
`
`
`FIG. 3
`
`Power Spectrum of Data from FIG. 1
`
`
`
`
`
`50000
`
`
`
`
`
`
`
`
`40000
`
`a; 30000
`
`a c
`
`? 20000
`
`5
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 4 0f 11
`
`US 6,393,311 B1
`
`
`
`FIG. 4
`
`Bandpass Filtered and Raw IR Data
`
`£30905
`
`
`
`WUMIIIIIMIIIIIII,” ..
`
`
`
`3
`
`F; 3.3E+05
`
`(D
`i 3.2E+05
`
`1
`
`.
`
`13.0
`
`17.0
`15.0
`Time (Seconds)
`
`19.0
`
`2 "
`
`331505110
`
`# Raw IR Data — Filtered IR Data
`
`
`6
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 5 0f 11
`
`US 6,393,311 B1
`
` FIG. 5
`
`12 Superimposed Pulse Subsegments
`
`1.5E+04
`
`
`
` 76
`
`5 1.0E+O4
`I
`3%
`j
`\
`,
`I}:
`IAN/fwfiv'm
`g 5.0E+O3 WW/ v «V
`erg; 0.0E+OO QWVA/e/fighfimtflfiwg'Q/W/‘m
`E —5 OE+03 “\AW’AQANQAQQWWRJ/Aé
`
`g
`-
`y W"/ \w '
`«34.0304
`-
`
`D -1.5E+04
`1O 20 30 4O 5O 60 70 80 90100
`
`Data Points
`
`
`
`
`7
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 6 0f 11
`
`US 6,393,311 B1
`
`
`
`r
`
`FIG 6
`
`Modified Average Pulse
`
`
`_ 6000 —
`(D
`5 4000
`_J
`
`E 2000 —
`.9
`c0
`
`0
`
`—
`
`
`
`
`
`
`
`
`
`
`
`_
`
`'1
`
`8'
`
`o -2000
`
`C e
`
`*5 -4000
`D 6000 _
`0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
`Time (Seconds)
`L____________#__fi
`
`8
`
`
`
`
`
`FIG. 7
`
`9
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 8 0f 11
`
`US 6,393,311 B1
`
`10
`
`
`
`
`Processor
`
`FIG. 8
`
`16
`
`
`
`
`Moiion Ariifcci Circuit Card
`
`FIG. 9
`
`10
`
`10
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 9 0f 11
`
`US 6,393,311 B1
`
`Processor
`
`FIG. 10
`
`11
`
`11
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 10 0f 11
`
`US 6,393,311 B1
`
`200
`
`21 0
`
`220
`
`2 30
`
`240
`
`Bandpass Filter Data
`
`Segment with HR
`
`Filter
`
`on Newest Pulse
`
`9
`
`Subsegment Data
`
`Measure Dispersion
`
`of Saturation Based
`
`
`
` YES
`Is the
`
`
`Dispersion Low?
`
`
`NO
`
`Use Newest Pulse
`
`
`Pulse
`Instead of
`Computing Average
`
`
`Compute Average
`Pulse
`
`250
`
`FIG.
`
`1 l
`
`12
`
`12
`
`
`
`US. Patent
`
`May 21, 2002
`
`Sheet 11 0f 11
`
`US 6,393,311 B1
`
`
`
`Evaluate Pulse
`
`260
`
`
`
`
`
`
`Shape Against Rules
`to Obtain Quality
`
`Measure
`
`+1120
`
`
`
`270
`
`
`
`Minimize the
`
`
`Dispersion of
`Saturation Values
`
`
`
`280
`
`290
`
`
`
`
`
`
`Last Candidate
`
`Frequency/2
`
`
`
`
`
`Select Candidate
`Frequency with Best
`
`
`Quality Measure
`
`
`
`FIG. 12
`
`13
`
`13
`
`
`
`US 6,393,311 B1
`
`1
`
`METHOD, APPARATUS AND SYSTEM FOR
`REMOVING MOTION ARTIFACTS FROM
`MEASUREMENTS OF BODILY
`PARAMETERS
`
`CROSS-REFERENCE TO RELATED
`APPLICATION
`
`This utility patent application claims the benefit of US.
`provisional patent application Ser. No. 60/104,422, filed
`Oct. 15, 1998.
`
`BACKGROUND OF THE INVENTION
`
`1. Technical Field
`
`This invention relates to the field of signal processing.
`More particularly, this invention relates to processing mea-
`sured signals to remove unwanted signal components caused
`by noise and especially noise caused by motion artifacts.
`2. State of the Art
`
`The measurement of physiological signals can often be
`difficult because the underlying physiological processes may
`generate very low level signals. Furthermore, interfering
`noise is inherent in the body and the interface between the
`body and sensors of physiological processes, Examples of
`physiological measurements include: measurement of elec-
`trocardiogram (ECG) signals based on the electrical depo-
`larization of the heart muscle, blood pressure, blood oxygen
`saturation, partial pressure of C02, heart rate, respiration
`rate, and depth of anesthesia. ECG signals, for example, are
`typically detected by surface electrodes mounted on the
`chest of a patient. ECG signals are weak at the signal source
`(i.e., the heart) and are even weaker at the surface of the
`chest. Furthermore, electrical interference from the activity
`of other muscles (e.g., noise caused by patient breathing,
`general movement, etc.) causes additional interference with
`physiological signals such as an ECG. Thus, considerable
`care must be taken in the design and use of physiological
`processors to enhance the quality of the true signal and
`reduce the effects of interfering noise signals.
`It is convenient to characterize a measured signal as being
`a composite signal composed of a true signal component and
`a noise signal component. The terms “measured signal” and
`“composite signal” will be used interchangeably hereinafter.
`Signal processors are frequently used to remove noise signal
`components from a composite measured signal in order to
`obtain a signal which closely, if not identically, represents
`the true signal. Conventional filtering techniques such as
`low pass, band pass, and high pass filtering can be used to
`remove noise signal components from the measured com-
`posite signal where the noise signal component occupies a
`frequency range outside the true signal component. More
`sophisticated techniques for conventional noise filtering
`include multiple notch filters, which are suitable for use
`where the noise signal component exists at multiple, distinct
`frequencies, all outside the true signal frequency band.
`However, it is often the case that the frequency spectrum
`of the true and noise signal components overlap and that the
`statistical properties of both signal components change with
`time. More importantly, there are many cases where little is
`known about the noise signal component. In such cases,
`conventional filtering techniques are ineffective in extract-
`ing the true signal.
`The measurement of oxygen saturation in the blood of a
`patient is a common physiological measurement the accu-
`racy of which may be compromised by the presence of
`noise. Knowledge of blood oxygen saturation can be critical
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`2
`during surgery. There are means of obtaining blood oxygen
`saturation by invasive techniques, such as extracting and
`testing blood removed from a patient using a co-oximeter.
`But, such invasive means are typically time consuming,
`expensive, and uncomfortable for the patient. Fortunately,
`non-invasive measurements of blood oxygen saturation can
`be made using known properties of energy attenuation as a
`selected form of energy passes through a bodily medium.
`Such non-invasive measurements are performed routinely
`with a pulse oximeter.
`The basic idea behind such energy attenuation measure-
`ments is as follows. Radiant energy is directed toward a
`bodily medium, where the medium is derived from or
`contained within a patient, and the amplitude of the energy
`transmitted through or reflected from the medium is then
`measured. The amount of attenuation of the incident energy
`caused by the medium is strongly dependent on the thick-
`ness and composition of the medium through which the
`energy must pass, as well as the specific form of energy
`selected. Information about a physiological system can be
`derived from data taken from the attenuated signal of the
`incident energy transmitted or reflected. However, the accu-
`racy of such information is reduced where the measured
`signal includes noise. Furthermore, non-invasive measure-
`ments often do not afford the opportunity to selectively
`observe the interference causing the noise signal component,
`making it difficult to remove.
`A pulse oximeter is one example of a physiological
`monitoring system which is based upon the measurement of
`energy attenuated by biological
`tissues and substances.
`More specifically, a pulse oximeter measures the variable
`absorption caused by arterial blood volume changes. Pulse
`oximeters transmit electromagnetic energy at two different
`wavelengths, typically at 660 nm (red) and 940 nm (infrared,
`hereinafter IR) into the tissue and measure the attenuation of
`the energy as a function of time. The output signal of a pulse
`oximeter is sensitive to the pulsatile portion of the arterial
`blood flow and contains a component which is a waveform
`representative of the patient’s arterial pulse. This type of
`signal, which contains a component related to the patient’s
`pulse, is called a plethysmographic waveform or plethys-
`mogram.
`Pulse oximetry measurements typically use a digit, such
`as a finger, or an ear lobe or other element of the body, where
`blood flows close to the skin as the medium through which
`light energy is transmitted. The finger, for example,
`is
`composed of various tissues and substances including skin,
`fat, bone, muscle, blood, etc. The extent to which each of
`these biological tissues and substances attenuate incident
`electromagnetic energy is generally known. However, the
`effect of motion can cause changes in the optical coupling of
`the sensor (or probe)
`to the finger,
`the underlying
`physiology, the local vasculature, optical properties of tis-
`sues due to changing optical path length as well as combi-
`nations and interactions of the all of the above. Thus, patient
`motion may cause erratic energy attenuation.
`A typical pulse oximeter includes a sensor, cabling from
`the sensor to a computer for signal processing and visual
`display,
`the computer and visual display typically being
`included in a patient monitor. The sensor typically includes
`two light emitting diodes (LEDs) placed across a finger tip
`and a photodetector on the side opposite the LEDs. Each
`LED emits a light signal at different frequencies. The
`detector measures both transmitted light signals once they
`have passed through the finger. The signals are routed to a
`computer for analysis and. display of the various parameters
`measured.
`
`14
`
`14
`
`
`
`US 6,393,311 B1
`
`3
`The underlying physical basis of a pulse oximeter is
`Beer’s law (also referred to as Beer-Lambert’s or Bouguer’s
`law) which described attenuation of monochromatic light
`traveling through a uniform medium which absorbs light
`with the equation:
`Itransmitted=1.
`tnct'dente
`
`(1)
`
`4mg»)
`
`:
`
`where Itmmmmed is the intensity of the light transmitted
`through the uniform medium,
`Iincidem is the intensity of
`incident light, d is the distance light is transmitted through
`the uniform medium, c is the concentration of the absorbing
`substance in the uniform medium, expressed in units of
`mmol L'l, and (X.()\.) is the extinction or absorption coeffi-
`cient of the absorbing substance at wavelength A, expressed
`in units of L/(mmol cm). The properties of Beer’s law are
`valid even if more than one substance absorbs light in the
`medium. Each light absorbing substance contributes its part
`to the total absorbance.
`
`Each LED emits light at different wavelengths, typically
`red (centered at about 660 nm) and IR (centered at about 940
`nm) frequency bands. The intensity of light
`transmitted
`through tissue, Immmmed, is different for each wavelength of
`light emitted by the LEDs. Oxyhemoglobin (oxygenated
`blood) tends to absorb IR light, whereas deoxyhemoglobin
`(deoxygenated blood) tends to absorb red light. Thus, the
`absorption of IR light relative to the red light increases with
`oxyhemoglobin. The ratio of the absorption coefficients can
`be used to determine the oxygen saturation of the blood.
`To estimate blood oxygen saturation, SpOz, a two-solute
`concentration is assumed. A measure of functional blood
`
`oxygen saturation level, SpOz, can be defined as:
`
`co
`cr+co
`
`SO:
`p 2
`
`,
`
`(2)
`
`where c0 represents oxyhemoglobin solute concentration,
`and Cr represents reduced or deoxyhemoglobin solute con-
`centration.
`
`Noise signal components in a measured pulse oximetry
`light signal can originate from both AC and DC sources. DC
`noise signal components may be caused by transmission of
`electromagnetic energy through tissues of relatively con-
`stant thickness within the body, e.g., bone, muscle, skin,
`blood, etc. Such DC noise signal components may be easily
`removed with conventional filtering techniques. AC noise
`signal components may occur when tissues being measured
`are perturbed and, thus, change in thickness while a mea-
`surement is being made. Such AC noise signal components
`are difficult
`to remove with conventional filtering tech-
`niques. Since most materials in and derived from the body
`are easily compressed, the thickness of such matter changes
`if the patient moves during a non-invasive physiological
`measurement. Thus, patient movement can cause the prop-
`erties of energy attenuation to vary erratically. The erratic or
`unpredictable nature of motion artifacts induced by noise
`signal components is a major obstacle in removing them.
`Various approaches to removing motion artifacts from
`measured physiological signals, and particularly for use in
`pulse oximeters, have been proposed. US. Pat. Nos. 5,482,
`036, 5,490,505, 5,632,272, 5,685,299, and 5,769,785, all to
`Diab et al., and US. Pat. No. 5,919,134 to Diab, disclose
`methods and apparatuses for removing motion artifacts
`using adaptive noise cancellation techniques. The basic
`proposition behind the Diab et al. approach is to first
`generate a noise reference signal from the two measured
`signals, and then use the noise reference signal as an input
`
`4
`to an adaptive noise canceler along with either or both of the
`measured signals to remove the reference noise signal from
`the measured signals, thus approximating the actual para-
`metric signals of interest. The Diab et al. approach appears
`to require the use of both measured input signals to generate
`a noise reference signal.
`Another approach to noise artifact elimination is disclosed
`in US. Pat. No. 5,588,427 to Tien. Tien uses fractal dimen-
`sion analysis to determine the complexity of waveforms in
`order to determine the proper value of the ratio of true
`intensities based on signal complexity. The Tien approach
`employs a fractal analyzer to determine values for two
`ratios, (X and [3, based on the measured time varying intensity
`of the transmitted red and IR light signals including noise.
`(X is defined as the ratio of the time varying true intensity of
`light transmitted from the red LED and the time varying true
`intensity of the light transmitted from the IR LED. 11 is a
`similar ratio relating the noise introduced during the mea-
`surement of the light transmitted by the red LED and the
`noise introduced during the measurement of the light trans-
`mitted by the IR LED. According to Tien, a fractal analyzer
`then determines values for (X and [3 and provides (ot,[3) pairs
`to a statistical analyzer. The statistical analyzer performs
`analysis of one or more (ot,[3) pairs to determine the best
`value for (X, which is then provided to a look-up table. The
`look-up table provides a value corresponding to the arterial
`oxygen saturation in the patient. While the Tien approach
`appears to be an innovative use of fractal analysis, it also
`appears to be computationally complex.
`Yet another approach to noise artifact elimination is
`disclosed in US. Pat. Nos. 5,885,213, 5,713,355, 5,555,882
`and 5,368,224, all to Richardson et al. The basic proposition
`behind the Richardson et al. approach is to switch operative
`frequencies periodically based on evaluating the noise level
`associated with various possible frequencies of operation in
`order to select the frequency of operation that has the lowest
`associated noise level. It would appear that data measured at
`a noisy frequency, using the Richardson et al. approach may
`be invalid or useless for calculating arterial oxygen satura-
`tion. Furthermore, Richardson et al. requires a computa-
`tional overhead to constantly monitor which frequency of
`operation provides the least noise.
`Another approach to noise artifact elimination is disclosed
`in US. Pat. No. 5,853,364 to Baker, Jr et al. The Baker, Jr.
`et al. approach first calculates the heart rate of the patient
`using an adaptive comb filter, power spectrum and pattern
`matching. Once the heart rate is determined, the oximetty
`data is adaptively comb filtered so that only energy at integer
`multiples of the heart rate are processed. The comb filtered
`data and the raw oximetry data are filtered using a K nan
`filter to adaptively modify averaging weights and averaging
`times to attenuate motion artifact noise. The adaptive filter-
`ing of the Baker, Jr. et al. approach appears to add significant
`computational complexity to solve the problem of motion
`artifact rejection.
`Still another approach to noise artifact elimination is
`disclosed in US. Pat. No. 5,431,170 to Mathews. Mathews
`couples a conventional pulse oximeter light transmitter and
`receiver with a transducer responsive to movement or vibra-
`tion of the body. The transducer provides an electrical signal
`varying according to the body movements or vibrations,
`which is relatively independent of the blood or other fluid
`flow pulsations. Mathews then provides means for compar-
`ing the light signals measured with the transducer output and
`performing adaptive noise cancellation. An apparent disad-
`vantage of the Mathews approach is the need for a secondary
`sensor to detect motion.
`
`15
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`15
`
`
`
`5
`Thus, a need in the art exists for a method, apparatus and
`system to eliminate motion-induced noise artifacts from
`light signals, that is relatively simple computationally, and
`that does not require more than one sensor.
`SUMMARY OF THE INVENTION
`
`The present invention includes methods, apparatuses and
`systems for removing noise in physiological measurements
`caused by motion or other similar artifacts. The methods,
`apparatuses and systems of the present invention eliminate
`noise from light signals using a single conventional sensor
`and are relatively simple computationally.
`In a method embodiment, a segment of data from a
`measured pulse oximetry signal is conventionally filtered,
`and frequency analyzed for major frequency components.
`The frequency components with the largest power spectral
`density are selected for subdividing into subsegments, each
`comprising an individual heartbeat. The subsegments are
`averaged and then analyzed to determine if the averaged
`subsegment is a valid pulse oximetry signal. Additionally,
`various quality or confidence measures may be used to
`evaluate the validity of such signal. Valid averaged subseg-
`ments become outputs for further processing to calculate
`physiological parameters such as blood oxygen saturation
`levels.
`
`includes a processor and
`A circuit card embodiment
`memory for storing a computer program capable of execut-
`ing instructions embodying the above method.
`Asystem embodiment includes an input device, an output
`device, a memory device and a motion artifact rejection
`circuit card capable of executing instructions stored in the
`memory device implementing the methods described herein.
`Finally, a system embodiment includes an input device,
`and output device, a memory device and a processor, which
`may be a digital signal processor, capable of executing
`instructions stored in the memory device implementing the
`methods described herein.
`
`BRIEF DESCRIPTION OF THE DRAWING
`FIGURES
`
`is currently
`In the drawings, which illustrate what
`regarded as the best mode for carrying out the invention and
`in which like reference numerals refer to like parts in
`different views or embodiments:
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`FIG. 1 is a high-level flowchart of a method embodiment
`of the invention.
`
`45
`
`FIG. 2 is a graph of a segment of data representing
`transmitted infrared (IR) light data signal received by a pulse
`oximetry sensor suitable for signal processing in accordance
`with the invention.
`
`FIG. 3 is a graph of the power spectrum of the IR data
`segment in FIG. 2 in accordance with the invention.
`FIG. 4 illustrates a bandpass filtered IR data segment
`superimposed on the original IR data segment of FIG. 2 in
`accordance with the invention.
`
`FIG. 5 is a graph of each pulse subsegment detrended and
`superimposed upon one another in accordance with the
`second method of computing an average pulse.
`FIG. 6 shows the modified average pulse as computed
`from the detrended pulse subsegments shown in FIG. 5.
`FIG. 7 illustrates the acquisition of a new pulse segment
`of raw IR data and the removal of the oldest segment of raw
`IR data in accordance with the invention.
`
`FIG. 8 is a block diagram of a motion artifact rejection
`circuit card configured to remove noise artifacts from signals
`representing bodily parameters in accordance with the
`invention.
`
`50
`
`55
`
`60
`
`65
`
`US 6,393,311 B1
`
`6
`FIG. 9 is a block diagram of a pulse oximetry system
`including a motion artifact rejection circuit card capable of
`removing noise from pulse oximetry data in accordance with
`the invention.
`
`FIG. 10 is a block diagram of a pulse oximetry system
`including a processor device programmed to remove noise
`from pulse oximetry data in accordance with the invention.
`FIGS. 11 and 12 are a detailed flowchart of the method of
`
`10
`
`determining the best frequency component, block 120, of
`FIG. 1.
`
`DETAILED DESCRIPTION OF THE
`INVENTION
`
`The following detailed description discloses methods,
`apparatuses and systems for removing motion artifacts from
`measured plethysmographic waveforms, particularly, but
`without limitation, those used in pulse oximetry. A system
`embodiment of the invention includes pulse oximetry hard-
`ware and associated software to perform the motion artifact
`suppression. A method embodiment of this invention
`includes a series of steps which exploits certain character-
`istics of plethysmographic waveforms. The methods, appa-
`ratuses and systems described below are suitable for light
`transmitted or reflected through bodily tissues and sub-
`stances. For convenience, the following detailed description
`will assume measurement of light which has been transmit-
`ted through a finger of a human. The terms “signal” and
`“waveform” are used interchangeably herein.
`FIG. 1 is a high-level flowchart of a method embodiment
`of the invention. The method steps include acquiring a
`segment of raw data 100, either red or IR, analyzing the data
`segment for dominant frequency components 110, determin-
`ing the frequency component which represents a valid
`plethysmographic pulse 120, computing an average pulse
`based on the correct frequency component 130 and repeating
`for new raw data segments 140. In order to calculate blood
`oxygen concentration, SpOz, the method embodiment of the
`invention may be applied to both red and IR data signals to
`eliminate or reduce noise from the data signals prior to
`calculating SpOz.
`The method of this invention begins with acquiring a
`segment of data (e.g., five or more pulses or approximately
`ten seconds) measured from a single light source transmitted
`through a finger and detected with a sensor on the opposite
`side of the finger. Acquiring a data segment is illustrated by
`block 100 of FIG. 1. For convenience, a 10.24 second
`segment of data will be used to illustrate the method. This
`corresponds to 1024 data points with a sampling rate of 100
`data points per second. It should be readily apparent that the
`method of the invention is not limited to data segments of
`this size. FIG. 2 is an example of such a data segment for an
`IR light source. The signal processing steps described herein
`may be performed on both red and IR data segments
`independently and simultaneously. Thus, while the steps of
`the method may be illustrated with data from an IR light
`signal, the same steps are applicable to data from a red light
`signal and vice versa. The terms “data segment”, “input
`waveform”, “data signal” and “signal” are used interchange-
`ably herein.
`A segment of data may be received from a sensor that
`converts transmitted or reflected light signals into electrical
`signals. Once a segment of data from a single electrical
`signal has been acquired, it may be filtered to reduce spectral
`leakage resulting from frequency analysis. There are several
`window filters which may be suitable for such purposes. For
`example, and not by way of limitation, a Hanning window
`16
`
`16
`
`
`
`US 6,393,311 B1
`
`7
`may be used to reduce spectral leakage. It will be readily
`apparent to one of skill in the art of digital signal processing
`that other window filters and methods of filtering data to
`reduce spectral leakage may be selected for reducing spec-
`tral leakage. As methods of filtering and various filters are
`known to one of skill in the art of signal processing, they will
`not be further detailed herein. The filtered data is then
`frequency analyzed to determine the dominant frequency
`components, see block 110 of FIG. 1. FIG. 3 illustrates the
`power spectrum of the IR data segment of FIG. 2 after
`filtering.
`Signal processing as described herein is generally per-
`formed in the frequency domain. The segment of data is
`converted into the frequency domain by, for example, per-
`forming the well-known Fast Fourier Transform (FFT) on
`the data. Other common techniques of converting time-
`domain data to the frequency domain may also be used, e.g.,
`classical methods using the FFT such as the periodogram or
`correlogram, autoregressive methods, Prony’s method,
`minimum variance methods, maximum likelihood methods.
`Additionally,
`time domain data may be converted to the
`frequency domain using transforms such as discrete cosine
`transform, wavelet transform, discrete Hartley transform,
`and Gabor transform.
`
`Both transient and periodic noise artifacts can induce
`peaks in the frequency domain that may be larger than the
`peak caused by the patient’s heart rate. The frequency peak
`which actually represents the patient’s heart rate must then
`be determined, see block 120 of FIG. 1. One approach to
`determining the correct frequency is to order the frequencies
`by peak amplitude from largest to smallest, F1 to F”, where
`F1 through F” are not harmonics of each other, and analyze
`them one by one to find the correct frequency,
`i.e., the
`patient’s heart rate. For purposes of illustration, only the
`frequencies associated with the two largest power spectrum
`amplitude (peaks), F1 and F2, will be used to explain the
`signal processing in accordance with the invention. It will be
`readily apparent that the signal processing described herein
`may be extended from 2 to n candidate frequencies. For
`convenience of notation, F1 is the frequency of the largest
`amplitude peak, and F2 is the next largest peak, which is not
`a harmonic of F1.
`Where this is not the first analysis cycle, an additional
`check is made to determine if one of the two potential
`frequencies, F1 and F2,
`is similar to a known valid
`frequency, F0, of the patient’s heart rate as determined
`during the previous analysis cycle. Otherwise, the signal
`processing proceeds as described in the next paragraph.
`Historical trends in heart rate may be used to select the
`proper frequency peak.
`In the case where none of the
`candidate frequencies is similar to the previous heart rate
`frequency, e.g., both frequencies F1 and F2 are large ampli-
`tude noise frequencies, the smaller amplitude frequency of
`the two potential frequencies is discarded, and the previous
`heart rate frequency is selected as the second potential
`frequency. Thus, the method of the invention prevents the
`situation where there is no correct frequency to choose
`because of multiple large amplitude noise frequencies.
`Once candidate frequencies F1 and F2 have been selected,
`each is processed separately to determine which is more
`likely to be the frequency of the patient’s heart rate. Ana-
`lyzing Fl first, the data segment may be optionally filtered
`with a narrow bandstop filter at (or near) F2 to “notch out”
`that frequency’s influence on the data. Alternatively, each
`candidate frequency is analyzed in turn without “filtering
`out” the effects of the other frequencies. The signal process-
`ing continues by determining the beginning and end of each
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`8
`heartbeat pulse within the data segment in order to subseg-
`ment the data into its consecutive heartbeat pulses.
`The period of rhythmic contraction of the heart by which
`blood is driven through the aorta and pulmonary artery is
`known as systole. Maximum light absorbance occurs during
`the systole of a cardiac cycle and is indicated on a plethys-
`mogram by a low point or systolic valley. Conversely, the
`period of rhythmic relaxation and dilation of the heart
`cavities occurs during diastole when blood is drawn into the
`heart cavities. Minimum light absorbance occurs during the
`diastole of a cardiac cycle and is in