`US005243992A
`.
`[11] Patent Number:
`5,243,992
`United States Patent 15
`Eckerle etal.
`[45] Date of Patent:
`Sep. 14, 1993
`
`
`
`
`4,239,048 12/1980 Steuer seesssscccssssssssseeeseseseens 128/666
`[54] PULSE RATE SENSOR SYSTEM
`4,307,728 12/1981 Walton....
`wee 128/687
`.
`4,353,372 10/1982 Ayer sun:
`ve 128/640
`[75]
`Inventors:
`Joseph S, Eckerle, Redwood City;
`4,409,983 10/1983 Albert .........
`vee 128/690
`Dale W. Ploeger, San Francisco;
`6/1984 Hiranoetal.
`4,456,959
`364/417
`Steven T. Holmes, Palo Alto; Thomas
`
`3/1987 Bergeret al.
`4,646,749
`128/678
`P. Low, Woodside; Rudolf Elbrecht,
`
`4,667,680 5/1987 Ellis ........0+
`we 128/672
`Los Altos; Philip R. Jeuck, Il,
`
`wore 364/417
`4,712,179 12/1987 Heimer...
`js
`:
`4,735,213 4/1988 Shirasaki .
`vue 128/681
`Menlo Fark; Boas Beene
`
`4.799.491
`1/1989 Eck
`7
`Menlo Park; Victor T, Newton, Jr.,
`ae a
`voy
`Menlo Park, all of Calif.
`wee 128/572
`4,802,488 2/1989 Eckerle.......
`?
`
`6/1989 Wenzel et al. oo... 128/672
`4,836,213
`[73] Assignee: Colin Electronics Co., Ltd., Aichi,
`Primary Examiner—Lee S. Cohen
`Japan
`Assistant Examiner—Marianne H. Parker
`[21] Appl. No.: 502,028
`Attorney, Agent, or Firm—Oliff & Berridge
`[22] Filed:
`Mar. 30, 1990
`(50)
`tnt, (CW csssucanacnacsecanmniset A6iB 5/02—[57]
`ABSTRACT
`[52] US. Ce seresssesssserens
`svsesseseees 128/690; 128/672
`A pulse rate sensor system is packaged in a wristwatch
`[58] Field of Search............... 128/687, 722, 699, 689,
`sized assembly and is worn by the user to provide an
`128/677, 681, 683, 690
`accurate determination of pulse rate. A tonometer sen-
`References Cited
`sor is provided to detect heartbeat pressure waves pro-
`U.S. PATENT DOCUMENTS
`duced by a superficial artery. A microcomputer manip-
`;
`ulates the unprocessed tonometersensor elementsignals
`3,999,537 12/1976 Noiles ........:ecceseeeeessseeeeeees 128/687
`using multiple algorithms to determine an accurate
`
`- 128/2.05T
`4,058,118 11/1977 Stupay et al......
`pulse rate
`4,086,916 5/1978 Freemanet al... 128/690
`‘
`
`1/1980 Mason et al... 128/689
`4,181,134
`4,202,350 5/1980 Walton 2.0... cscceceeeeeneses 128/690
`
`[56]
`
`36 Claims, 9 Drawing Sheets
`
`0001
`
`AppleInc.
`APL1047
`U.S. Patent No. 8,923,941
`
`Apple Inc.
`APL1047
`U.S. Patent No. 8,923,941
`
`0001
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 1 of 9
`
`5,243,992
`
`FIG.|
`
`0002
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 2 of 9
`
`5,243,992
`
`FLEXIBLE
`PC. BOARD 10
`
`SENSOR 2
`
`
`
`;
`
`SENSOR
`ADAPTER(2
`
`——AXIS #2 9a
`GIMBAL 9
`
`
`
`
`axis #1
`
`b
`
`MOUNT TO SPRING {1
`
`FIG.2
`
`MOUNTING
`
`PLANE FIG. 3A
`
`FIG 3B
`
`0003
`
`0003
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 3 of 9
`
`5,243,992
`
`<x
`oc
`
`FIG.4
`
`CONVERTER
`
`DIGITAL
`
`TO
`
`0004
`
`
`
`PROCESSING|_68=
`
`~=i
`
`J
`oO
`
`0004
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 4 of 9
`
`5,243,992
`
`
`SEGMENT#4 Nie||
`
`
`INITIALIZE
`
`CLOCK TICK
`OCCURED
`
`
`
`ecciescccePsnleeeenearecere
`
`
`
`
`
`
`Y
`
`AN
`
`—_—— ———_— = —
`
`| | !
`
` AVERAGE FROM
`All. ELEMENTS
`ie —_— oe ee eee ee ee ee es i
`ACCUMULATE
`
`CORRELATION
`SUMMATIONS
`
`
`loon’
`
`I
`|
`
`MARKING
`TIME AFTER
`SYSTOLIC
`?
`
`LL
`
`SEGMENT#5
`
`|
`
`5e
`
`0005
`
`0005
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 5 of 9
`
`5,243,992
`
` Y
`
`
`
`FOUND SYSTOLIC SET] 6¢
`SIGNAL VALID SET
`ara
`
`AS MAX SLOPE}
`
`6"
`
`el
`
`DOWNHILL FLAG
`
`SET FLAG THAT
`
`.
`
`ExcerneD
`
`(c)
`
`SLOPE THRESHOLD =
`
`scam
`
`SAVE SYSTOLIC
`AND DIASTOLIC
`
`RESET ABSOLUTE
`MINIMUM AND
`MAXIMUM STORAGE
`
`6h
`
`6i
`
`0006
`
`0006
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 6 of 9
`
`5,243,992
`
`alan healamiiettaming’ <r 7
`GMENT
`
`@ FERMe
`
`MOIa eeeSEGMENT#8|
`
`||j | | | ||D
`
`| ! |1 | | Il||
`
`| | | | | | | | | |1 | | | | | ||I
`
`| | | |
`
`onl 8a
`PERIOD >0.7
`
`
`
`CURRENT PERIOD
`=CURRENT PERIOD
`+ PREVIOUS PERIOD
`
`
`
`
`| | ||| |
`
`;|
`
`0007
`
`(PULSE AMPLITUDE)
`
`ETERMINE ELEMENT
`WITH MAXIMUM
`POSITIVE PULSE
`AMPLITUDE IF
`
`AVAILABLE
`
`
`SET ELEMENT OF
`
`
`INTEREST TOELEMENT|%
`WITH MOST MAXIMUMS
`INLAST 5 TRIGGERS
`
`| ||
`
`| |
`
`|
`
`
`
`
`
`AVERAGE PULSE
`RATE = (OLD AVERAGE)
`
`7b |
`
`0007
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 7 of 9
`
`5,243,992
`
`(B)
`
`1 i
`
`VALID=
`
`GNAL||
`
`Pos cagingacm eee) me eee =
`[_——— —1— ——_" “SEGMENT
`||CALCULATE
`
`||CORRELATION Jia mle
`faecal| COEFFICIENT
`
`| | | |
`
`{2c
`
`n
`[SIGNAL
`ALID=+74(B)
`FALSE|
`|
`Lo.Mans he ee Gente al
`SET | MENT
`VALID
`
`
`PULSE IN LAST SN_|DISPLAY||!|CALCULATE |
`
`14a
`
`|
`
`30SEC.? f-BLANK];|PULSERATE Hehe
`
` SEGMENT 1
`
`
`
`
`| 1}
`
`SET POST SYSTOLIC
`Po 1 |DELAY TO 0.25
`|OF PULSE TIME
`;
`
`{3b
`
`13c
`
`| |
`
`| |
`
`(RESET SYSTOLIC
`
`O ! TIMER
`
`|UPDATE DISPLAY
`713d
`|IF VALID PULSE FOUND
`|
`Le veered mae a eee a
`
`0008
`
`0008
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 8 of 9
`
`5,243,992
`
`($30D)
`
`
`
`
`
`(4091)¥713409WNSNWNSGdé
`02002O0b02
`02oseeSeesOLS-OSESépSsG-
`
`
`“WWNOISGSONVHN3“TWWNOIS
`
`ATIVILNSYSSSIO~G3I9VYESAV
`SIObSI-O}0Ol
`29I4
`9Is
`(YOOIVT3H09=HOI
`0}I(}-4$0)
`AVY8#INJW3I14
`OS£O€-3d
`G2€G-2Olt
`
`‘TWNOIS
`
`0009
`
`0009
`
`
`
`U.S. Patent
`
`Sep. 14, 1993
`
`Sheet 9 of 9
`
`5,243,992
`
`
`
`0010
`
`0010
`
`
`
`5,243,992
`
`1
`
`PULSE RATE SENSOR SYSTEM
`
`FIELD OF THE INVENTION
`
`The present invention relates to pulse monitors with
`visual readouts of pulse rate and more particularly to a
`tonometer sensor pulse rate monitor which employs
`multiple noise and motion artifact rejection methods to
`determine an accurate pulse rate.
`BACKGROUND OF THE INVENTION
`
`10
`
`2
`but the pressure sensitive element(s)itself is located off
`the artery. In this case the subject’s pulse can push up on
`the housing and lessen the pressure on the pressure
`sensitive element. The result is that a pulse waveform is
`still received, but it is inverted and shows a negative
`relative pressure. Pulse measuring devices which rely
`on pressure measurements but can correctly interpret
`only positive pressure waveforms must be placed and
`held accurately on the artery, creating additional de-
`mands on attachment of the device and/or lowering
`comfort to the user.
`Additionally, pronounced dichrotic notches can be
`found in the pulse of many people. When dichrotic
`notches are present there are two rises and twofalls in
`blood pressure during a single heartbeat. These can be
`mistakenly interpreted as two heartbeats, leading to a
`major inaccuracy in pulse rate measurement.
`The present invention overcomes the problems en-
`countered with other pulse rate sensors by applying the
`principles of arterial tonometry for signal acquisition
`for a pulse rate sensor. In the invention, multiple algo-
`rithms are used in signal processing and pulse rate cal-
`culation to compensate for multiple signal errors which
`could occur during pulse rate measurement.
`Theprinciples of arterial tonometry are described in
`several U.S. Patents including: U.S. Pat. Nos. 3,219,035;
`4,799,491 and 4,802,488. These principles are also de-
`scribed in several publications including an article enti-
`tled “Tonometry, Arterial,” in Volume 4 of the Ency-
`clopedia of Medical Devices and Instruments. (J. G.
`Webster, Editor, John Wiley & Sons, 1988). All of these
`references discuss arterial tonometry as used for the
`measurement of blood pressure.
`For blood pressure measurement, it is desirable to
`flatten a section ofthe arterial wall as described in these
`references. Flattening is produced by exerting an appro-
`priate hold down force on the tonometer sensor. For
`pulse sensing, significant flattening of the arterial wall is
`not necessary and a lower hold downforce can be used.
`This results in greater comfort for the wearer.
`SUMMARYOF THE INVENTION
`
`The present invention relates generally to a method
`and apparatus for measuring and displaying pulse rate
`with increased accuracy. Morespecifically, the present
`invention provides a method for increasing the accu-
`racy ofa pulse rate sensing system by meansof a novel
`pressure sensing array and multiple methodsfor identifi-
`cation and elimination of artifacts.
`Other methods and apparatus are known for measur-
`ing pulse rates and for rejecting pulse artifacts. For
`example, U.S. Pat. No. 4,409,983 shows a pulse measur-
`ing device which employs multiple transducers con-
`nected to averaging circuits and differential amplifiers.
`This invention helps separate signals corresponding to
`motion artifacts from the signal corresponding to a
`heartbeat pulse. Other apparatus and methodsfor re-
`moving motion artifacts are disclosed in U.S. Pat. Nos.
`4,307,728, 4,202,350, 4,667,680, 4,239,048, 4,181,134 and
`4,456,959. Methodsused to reducesignal errors include
`the use of windowing and averaging techniques and
`auto correlation algorithms.
`The pulse rate sensor systems described above are
`subject to several sources of inaccuracies. First, it is
`difficult to reject motion and noise artifacts in many of
`these systems. This is especially true for systems em-
`ploying a single sensor element. (See U.S. Pat. Nos.
`4,202,350 and 4,239,048.) These systems have no physi-
`cal meansfor receiving both a pulse-plus-artifact signal
`and a separate artifact signal. Other means are required
`to compensate for, or eliminate, the error caused by
`artifacts such as motion artifacts. Signal processing
`techniques such as filtering and windowing are often
`used.
`Even those systems or methods which employ multi-
`ple sensor elements inaccurately measure pulse rate
`because only a single method is used for enhancedsig-
`nal processing. For example, different types of motion
`artifacts can occur simultaneously, and with other per-
`tubations, on the pulse sensor. It is also not unusual for
`signal errors to be interpreted as pulses or for actual
`pulses to be missed by the pulse sensor. Methods of
`pulse rate determination which do not compensate for
`these errors are inherently inaccurate under real-world
`conditions ere artifacts are present.
`For example, if a pulse rate system detects a “pulse”
`caused by noise, several adverse results may be seen.
`The pulse rate system could use the noise as the basis for
`windowing the signal. The pulse rate system could
`simply use this “pulse” as part of the overall pulse rate
`calculation. In addition, the pulse rate system could
`recognize the noise as noise and subtract out the noise,
`in some cases subtracting out a valid signal as well.
`Anothersource of inaccuracy that occurs using pulse
`measuring devices that measure pressure variations
`caused by a subject’s pulse (see U.S. Pat. No. 4,409,983
`for example)is inverted pulse waveforms. An inverted
`waveform can occur when the housing that holds the
`pressure sensitive element(s) is located on the artery,
`
`40
`
`50
`
`55
`
`Accordingly, the present invention has been devel-
`oped to overcomethe foregoing shortcomingsofexist-
`ing pulse rate sensor systems.
`It is therefore an object of the present invention to
`_ provide a method and an apparatus for measuring pulse
`rates using arterial tonometry techniques including a
`sensor array with multiple sensing elements disposed in
`an array, in order to provide increased accuracyin the
`determination ofpulse rate.
`Another object of the present inventionis to increase
`the accuracy of the displayed pulse rate by calculating
`the displayed pulse rate using only pulse rates deter-
`mined to be valid.
`A further object of the present inventionis to deter-
`mine whether pulses detected are valid, based on the
`correlation between the present pulse and the previous
`pulse.
`A still further object of the present invention is to
`remove motion artifacts from the sensor elementsignals
`by subtracting a value from all these signals based on a
`spatially weighted average of these signals.
`Anadditional object of the present invention is to
`cancelout artifacts from a sensor element which exceed
`a level predetermined to be the maximum level of a
`valid blood pressure signal.
`
`60
`
`0011
`
`0011
`
`
`
`5,243,992
`
`4
`FIGS. 8A and B are graphical representations of
`normal and inverted blood pressure waveforms, respec-
`tively.
`DESCRIPTION OF THE PREFERRED
`EMBODIMENTS
`
`3
`Still another object of the present invention is to
`accurately process inverted waveforms caused by mis-
`alignment or shifting of the sensor elements relative to
`an underlying artery.
`These and other objects and advantages are achieved
`in accordance with the present invention by the steps
`Referring to FIG. 1, a pulse rate monitor 1 in accor-
`of: sensing at least one blood pressure waveform signal
`dance with the preferred embodiment of the present
`at a predetermined sampling period using a tonometer
`invention is shown having a tonometer sensor 2 and a
`sensor having a plurality of sensor elements disposed in
`case 3 housing processing and display circuitry. The
`an array; producing a plurality of sensor elementsig-
`tonometer sensor 2 is mounted in a gimbal assembly 9
`nals, at least one of the sensor element signals corre-
`which in turn is connected to case 3 by a spring 4.
`sponding to theat least one blood pressure signal; cor-
`Spring 4 includes a sensor position adjustment 5 section
`recting the sensor element signals using a correction
`at the connection point between gimbal assembly 9 and
`factor based on one characteristic of the sensor element
`spring 4.
`signals; calculating a plurality of slopes based on the
`Furtherdetails of this arrangement are shown in FIG.
`corrected sensor elementsignals; selecting a corrected
`2, which uses the same componentdesignations found in
`sensor element signal corresponding to one of the sen-
`FIG. 1, where possible. The tonometer sensor2is at-
`sor elements, the selected sensor element signal having
`tached to a sensor adapter 12 which, in turn is pinned to
`slopes greater than a predetermined slope threshold;
`the gimbal assembly 9 at axis number 2. A spring mount-
`determining a plurality of pulse rates based on the se-
`ing pad 11 is pinned along axis number 1 of gimbal
`lected sensor element signal; computing a value corre-
`assembly 9, mounting pad 11 being the point at which
`spondingto the autocorrelation of the corrected sensor
`sensor position adjustment 5 connects to gimbal assem-
`element signal over a predetermined time period; and
`bly 9. Flexible printed circuit 10 connects the tonometer
`calculating a display pulse rate based on at least two of
`sensor 2 to circuitry (shown in FIG. 4) in case 3.
`the pulse rates, each of the two pulse rates having the
`Tonometer sensor 2 is an array of pressure or force
`value within a predetermined range.
`sensitive elements fabricated into a single structure.
`These and other objects and advantages are achieved
`Standard photolithographic manufacturing techniques
`in accordance with the preferred embodiment of the
`can be used to construct the tonometer sensor. Experi-
`present invention comprising: a tonometer sensor means
`mental testing indicates that about 3 to 6 individual
`having a plurality of pressure sensing elements disposed
`sensor elements are necessary for good accuracy in
`in an array, for sensing a blood pressure waveform onat
`pulse rate measurementbut even a single sensor element
`least one of the pressure sensing elements and produc-
`adapted for use with the method and apparatus of the
`ing a plurality of sensor element signals, at least one of
`present invention will produce more accurate pulse rate
`the sensor elementsignals being indicative of the blood
`determinations.
`pressure acting on at least one of the pressure sensing
`Referring again to FIG.1., the remainderofthe pulse
`elements; means for pivoting the tonometer sensor
`rate monitor1 will be described in terms of wearing the
`means about a pair of axes; means for pressing the to-
`pulse rate monitor 1. In the preferred embodiment,the
`nometer sensor means against a radial artery of a sub-
`pulse rate monitor is worn on the operator’s wristlike a
`ject; meansfor anchoring the pressing means on a dorsal
`wrist watch. Whenthe wearer donsthe pulse rate moni-
`side of the subject; central processing means for deter-
`tor 1, the tonometer sensor 2 is positioned above a radial
`mining a pulse rate based on at least one of the sensor
`artery and the case 3 is positioned on the opposite side
`element signals received from the tonometer sensor
`of the wrist. The case 3, which anchors one end of
`means; means for displaying the pulse rate; and means
`spring 4, is held in place by strap 8 by cinching an flexi-
`for holding the anchoring means on the subject, the
`ble portion 8a of a strap 8 and locking the strap in posi-
`holding meansat no time contacting the pressing means.
`tion by means of a buckle 7. Strap 8 is prevented from
`directly contacting the tonometer sensor 2, the gimbal
`BRIEF DESCRIPTION OF THE DRAWINGS
`assembly 9 and the spring 4 by a protective band 6
`whichis part ofthe strap 8. Protective band 6 has a box
`shaped cutout section which allowsit to fit around the
`tonometer sensor 2,
`the gimbal assembly 9 and the
`spring 4, without contacting any of these elements.
`The tonometer sensor2 is held against the artery with
`a thin cantilever spring 4 which is attached to the case
`3. The spring reaches around the wrist to position the
`tonometer sensor 2. A low profile gimbal assembly 9
`(See FIGS.1 and 2) connects the tonometersensor 2 to
`the spring 4 and allows the tonometer sensor2 to pivot
`so as to lie flat against the wrist while being worn. Gim-
`bal assembly 9 allows about 20 degrees of rotation about
`each of its two axes. The position on spring 4 of the
`tonometer sensor-gimbal assembly 2, 9 is adjustable by
`means of the sensor position adjustment 5 section of
`spring 4.
`Sensor hold down force must be controlled to pro-
`vide enough pressure to partially flatten the radial ar-
`tery but not enough pressure to cause discomfort to the
`wearer. The optimum hold down force is unique to
`
`Thepreferred embodiments are described with refer-
`ence to the drawings in which:
`FIG. 1 is a general arrangementofa pulse rate sensor
`system illustrating a pulse rate sensor connected to a
`case assembly;
`FIG. 2 is a perspective view of the tonometer sensor
`supported in a gimbal assembly;
`FIGS. 3A, B and C show low, medium and high
`curvature spring profiles, respectively.
`FIG.4 is a block diagram ofthe pulse rate processing
`circuitry in accordance with the preferred embodiment
`of the present invention;
`FIGS. 5A, B, C and D are a flowchart describing a
`pulse rate measurement and compensation method in
`accordance with one embodimentofthe present inven-
`tion;
`FIG. 6 is a tableillustrating the effect of differential
`enhancement on raw sensor elementsignal levels;
`FIG. 7 is a table illustrating the calculations of the
`correlation algorithm; and
`
`35
`
`40
`
`45
`
`55
`
`65
`
`0012
`
`0012
`
`
`
`5
`each individual but ranges from about 100 grams to
`about 500 grams. Some wearers may require a higher
`hold down force to obtain a reliable pulse signal from
`the tonometer sensor 2. Other wearers may be sensitive
`to the hold down force of the tonometer sensor 2
`against their wrists and desire the lowest possible hold
`downforce.
`Ideally, since the hold down force of the tonometer
`sensor 2 is controlled only by the deflection of the
`spring 4 and since the size and shape of the wrist can
`vary greatly between individuals, each spring 4 would
`have to be custom fit for each wearer. In general, three
`configurations for spring 4, shown in FIGS. 3A, B and
`C, will suffice to cover the majority of the population.
`Springs 4A, B and C, shown in FIGS. 3A, B and C,
`provide adequate length and curvature adjustment to
`cover the general population. The springs 4A, 4B and
`4C differ only in their radius of curvature.
`Thediscussion below of the computation of the pulse
`rate can best be understood byfirst understanding the
`important features of a normal blood pressure wave-
`form. Referring to FIG. 8A a normal blood pressure
`waveform 100 with an average blood pressure 102 is
`shown. The point 104 on the waveform 100 where
`blood pressure is maximum is referred to as systole,
`while the point 106 where blood pressure is a minimum
`is referred to as diastole. It is known from scientific
`studies that the maximum absolute value of the slope of
`wave form 100 occurs just prior to systole, in region
`108, when the blood pressureis rising from diastole. A
`dichrotic notch 110 is present in the blood pressure
`waveform 100 of some subjects. Referring to FIG. 8B,
`an inverted waveform 100’
`is shown having systole
`(104), diastole (106),
`the region (108) and dichrotic
`notch (110).
`Computation of pulse rate is performed by electronic
`circuitry shown in FIG. 4 and located in case 3,
`in
`accordance with the flow chart shown in FIGS. 5A,B,
`C and D.
`Referringfirst to FIG.4, circuitry to process tonome-
`ter signals and compute pulse rate in accordance with
`the present invention comprises: a preamplifier 58; a
`high passfilter 60; a low pass filter 62; an amplifier 64;
`an analog-to-digital digital converter (ADC) 66; a cen-
`tral processing unit (CPU) 68; random access memory
`(RAM)70; read only memory (ROM)72; input/output
`unit (I/O) 74 and display 76. As shown in FIG. 4, the
`preamplifier 58, filters 60 and 62, amplifier 64 and ADC
`66 each include inputs corresponding to the individual
`sensor elements.
`During operation, the outputofall sensor elements of
`tonometer sensor 2 are routed to preamplifier 58 which
`amplifies the signals beforefiltering. The outputof pre-
`amplifier 58 is the input to high pass filter 60 which
`removes the DC and very low frequency components
`ofthe signals. The outputofthe high passfilter 60 is the
`input to the low pass filter 62 which removes some of
`the high frequency components ofthesignal. (High pass
`filter 60 coupled with low pass filter 62 effectively act as
`a band pass filter.) The preferred bandwidth of the
`effective bandpassfiler is about 0.1-30 Hz. After filter-
`ing, the signals are amplified by amplifier 64 to a level
`compatible with the ADC 66. ADC 66 multiplexes the
`signals and converts them to digital data whichis sent to
`I/O 74. CPU 68 reads the digital data form I/O 74 and
`stores the digital data in RAM 70.
`RAM 70 is segmented to form a plurality of data
`buffers for storage of digital data representing sensor
`
`20
`
`30
`
`35
`
`40
`
`45
`
`60
`
`5,243,992
`
`6
`element signals, counts, flag settings and calculated
`values. A ROM 72 contains the operating program
`software for CPU 68. CPU 68, through I/O 74, receives
`digital data from ADC 66 and outputspulse rate calcu-
`lations to display 76.
`The operation of CPU 68 is best understood by refer-
`ring to the flow chart of FIGS. 5A, B, C and D.
`When the pulse rate sensor system is turned on or
`reset by the wearer, the system goes throughan initial-
`ization routine to clear stored data. At this point, a
`clock (not shown)starts and a clock signal is provided
`to the CPU 68 to set a predetermined sampling period.
`Referring to FIG. 5A, while performing segment 1 of
`the program, sub-step 1a checks to see if a clock tick has
`occurred. If a clock tick is not detected at sub-step 1a,
`the cycle repeats. If a clock tick is detected, the CPU 68
`executes sub-step 15 and samples the outputs ofali sen-
`sor elements of the tonometer sensor 2. In addition,
`system timers are updated. Control then passes to pro-
`gram segment2.
`During program segment 2, each data element for
`each sensor element is checked to see if it exceeds a
`predetermined maximum value. If an actual value ex-
`ceeds the predetermined maximum value,i.e. the prede-
`termined maximum amplitude, the data from that sensor
`elementis set to zero. The program also sets a flag so
`that all data from that sensor elementis set to zero for
`the next five seconds. This eliminates unusually large
`signals which are usually noise. Program control then
`passes to program segment3.
`Differential enhancementof the signal occurs during
`execution of program segment 3. During sub-step 3a,all
`signals from sensor elements not previously set to zero
`are averaged to producea spatially weighted average
`signal. In the preferred embodiment, the weighted fac-
`tor is about 1.0, but can be adjusted as described below.
`At sub-step 35, the spatially weighted averagesignalis
`subtracted from each of the actual sensor elementsig-
`nals.
`The differential enhancementalgorithm reduces mo-
`tion artifacts in the tonometer sensorsignals. This algo-
`rithm can only be used when multiple sensor elements
`are employed simultaneously. Since all of the sensor
`elementsignals are often affected equally by a motion
`artifact, the differential enhancement algorithm aids in
`distinguishing between artifacts and blood pressure
`signals. For example, motion artifacts such as footsteps
`usually affect all sensor elements in the same way.
`Blood pressure signals, on the other hand, affect only
`sensor elements whichare directly over or very near to
`the artery.
`Differential enhancement addsall of the signals from
`all of the sensor elements together and formsa spatially
`weighted average signal. If each sensor elementis af-
`fected in the same way by a motion,
`the spatially
`weighted averagesignal will be an accurate representa-
`tion of the motion artifact. This spatially weighted aver-
`age signal is then subtracted from eachindividual sensor
`elementsignal to form a differentially enhanced signal
`for each sensor elementsignal. (I.e. an approximation of
`the motion artifact is subtracted from the raw sensor
`element signals to produce corrected sensor element
`signals.) The raw sensor elementsignals are expected to
`be either motionartifact signals or motionartifact sig-
`nals plus blood pressure signals. For example, FIG. 6
`showsthe effect of the differential enhancement algo-
`rithm on the signals from a three elementsensorarray.
`Signals for only one clock tick are shown.
`
`0013
`
`0013
`
`
`
`5,243,992
`
`8
`as valid pulses, but rejects or ignores pulses that have
`correlation coefficients outside this predetermined
`range. Ofcourse, different acceptance ranges for COR
`may be used without departing from the teachings of
`this disclosure.
`Thetraditional mathematical definition of the corre-
`lation coefficient applies only when the two waveforms
`being correlated (CORELA and CORELB of the
`above example) have exactly the same duration. In
`other words, Eq. (1) is mathematically rigorous only
`when both heartbeats’ durations are equal. The present
`invention departs from this mathematical constraint and
`allows COR to be computed even if the durationsof the
`two heartbeats are not exactly equal. Eq. (1) can be used
`to advantage in the case where the two waveforms do
`not have equal duration since Eq. (1) can be used very
`effectively to recognize movement artifacts even
`thoughit is not used in a mathematically rigorous way.
`In the preferred embodiment, the array, CORELB,of
`the above illustration is not used. Instead, a running
`summation of the numerator of Eq.(1), called NSUM,
`and of the denominator of Eq. (1), called DSUM,are
`updated after each sample of sensor data is obtained.
`Specifically, if BP(ELEMENT)is the pressure mea-
`sured by a selected elementat time ICOR,the following
`sequence is executed:
`
`10 DSUM = DSUM + (BPR(ELEMENT) x BRELEMENT))
`
`20 NSUM = NSUM + BRELEMENT) x CORELA(UICOR)
`
`30 CORELA(ICOR) = BR(ELEMENT)
`
`The correlation coefficient is then simply calculated
`after systole is found as COR=NSUM/DSUM. This
`procedure works because in line 20 of the above code
`CORELA(ICOR)still contains the pressure data from
`the previous heartbeat. Line 30 updates CORELA for
`the calculation on the next (future) heartbeat. Use of
`this running summation (in place of the two arrays,
`CORELA and CORELB,of Eq.(1)) is advantageous
`because it reduces computer speed and memoryre-
`quirements. However, Eg. (1) defines the correlation
`coefficient so it may be calculated by other procedures
`without departing from the teachingsofthis disclosure.
`The table shown in FIG.7 gives an example of what
`the relevant variables hold for a few samples of hypo-
`thetical data. If systole occurred at ICOR =3 (in reality
`the waveform normally extends for many more samples
`before a systole is found) the correlation coefficient
`would be calculated as
`
`COR = NSUM/DSUM= 3850/5325=0.723 L Feed
`
`and this heartbeat would be accepted as valid by the
`acceptance criterion described above.
`.
`Finally, the correlation coefficient in other applica-
`tions is sometimes also defined as:
`
`COR =(CORELA(/)* CORELBY)/(AMPA)x(AMPB)
`
`where AMPA=sqrt(DSUM), AMPB=sqrt(DSUM’),
`and DSUM'is the value of DSUM for the previous
`heartbeat. This definition makes the correlation coeffi-
`cient independent of any overall gain change between
`one heartbeat and the next,
`After executing sub-step 4a, the program checks to
`see if a systolic peak has been detected within a prede-
`
`7
`Differential enhancementis part of a larger class of
`algorithms wherethe processed output for each valueis
`a weighted sum of the unprocessed values. For exam-
`ple, differential enhancement of three raw values pro-
`duces a processed output for value 1 as follows:
`
`Processed Value 1 = (2/3Xraw value 1) +
`(=—1/3raw value 2) + (—1/3)(raw value 3).
`
`Other extensions of this procedure are possible and
`different weighting factors than those used by the basic
`differential enhancementalgorithm are possible without
`departing from the teachings of this disclosure. For
`example, it may be advantageous to assign large nega-
`tive weights to sensor element signals far from a se-
`lected sensor element and large positive weights to
`sensor elementsignals located near to a selected sensor
`element.
`Returning to FIG. 5A, after performing program
`segment 3, program control passes to program segment
`4 where the differentially enhanced signal data is pro-
`cessed by a correlation algorithm.
`The correlation algorithm computes a quantitative
`measure of the similarity between sensor waveforms
`over a predetermined time period of about two consec-
`utive heartbeats. In principle, the shape of a person’s
`blood pressure waveform will be fairly constant from
`one heartbeat to the next. The correlation algorithm
`compares the blood pressure waveform for the current
`heartbeat with a previously recorded waveform for the
`preceding heartbeat.
`The processor executes the correlation algorithm as
`follows. A variable, ICOR, is used as a counter to keep
`track of the number of clock “ticks” (i.e. the elapsed
`time) that the processor has spent looking for a systole.
`Atthe start, ICORis set equal to 1. ICORis then incre-
`mented by one for each subsequentclock tick (i.e. each
`time data is sampled from the sensor) When ICORisset
`equal to 1, the variables DSUM and NSUM,described
`below, are both set equal to zero. The data from the
`selected element of the tonometer sensor between the
`time when the processorstarts looking for a systole and
`the time whenit actually finds a (presumed valid) sys-
`tole is stored in an array, called CORELA.
`ICORis
`used as a pointer for the array CORELA,i.e. the value
`of the data from the selected element after ICOR clock
`ticks have occurred (after the start of looking for a
`systole) is stored in CORELA(ICOR).
`For example, suppose a similar procedure had been
`used for the previous heartbeat, and the values from the
`selected element had been stored in another array, CO-
`RELB.After the systole is found for the present heart-
`beat, the correlation coefficient, COR, may be calcu-
`lated, The correlation coefficient between the pressure
`waveforms for the current and previous heartbeats is
`mathematically defined as:
`
`.
`(CORELA(/)*CORE.
`COR =} {EORELIG"CORELA
`
`(1)
`
`is over all values of elapsed
`,
`where the summation,
`time, j, from the start of looking for the systole. If the
`current heartbeat is exactly identical to the previous
`heartbeat, the array, CORELB, will be equal to the
`array, CORELA,andthecorrelation coefficient, COR,
`will be equal to 1. If COR differs greatly from 1, the
`two waveformsare notsimilar and at least one of them
`is probably distorted by a movementartifact. The pro-
`gram accepts values of COR between about 0.6 and 2.0
`
`5
`
`20
`
`30
`
`35
`
`55
`
`60
`
`65
`
`0014
`
`0014
`
`
`
`5,243,992
`
`9
`termined. numberof sampling periods,e.g., the last 0.25
`seconds, at sub-step 45. If the answer is yes, program
`segment 4 loops back to program segment 1 to accumu-
`late sensor elementsignals. If the answeris no, program
`control passes to program segment5.
`A normal, non-inverted pulse waveform will have a
`fairly large negative slope as the blood pressure drops
`from its peak value at systole. There can also be a pro-
`nounced dichrotic notch which has a local minimum
`pressure. See FIG. 8A. These features can be mistak-
`enly interpreted by the program as an inverted wave-
`form (i.e. a waveform containing a large negative slope
`followed by a local minimum). Similar problems can
`occur with a true inverted waveform, with the dich-
`rotic notch just after systole being mistakenly inter-
`preted as a new, non-inverted pulse. See FIG. 8B. Since
`the program must be able to correctly handle both nor-
`mal and inverted waveforms, the best strategy is to turn
`off normal processing for a short period of time after
`systolic pressure has been identified. This presents prob-
`lems for the correlation algorithm, which works best
`whenit has full waveforms to process. The compromise
`solution is to only collect data and perform correlation
`processing during the