United States Patent
`Pardey et al.
`
`[19]
`
`US005999846A
`[11] Patent Number:
`[45] Date of Patent:
`
`5,999,846
`Dec. 7, 1999
`
`[54] PHYSIOLOGICAL MONITORING
`
`[75] Inventors: James Pardey; Mark Jeremy Laister,
`both of Oxford; Michael Richard
`DadsWell, Oxon; Lionel Tarassenko,
`Oxford, all of United Kingdom
`
`[73] Assignee: Oxford Medical Limited, Oxon,
`United Kingdom
`
`Robert G. Norman, et al., A Likelihood Based Computer
`Approach to Conventional Scoring of Sleep, Proceeding of
`the Annual International Conference of the IEEE Engineer
`ing in Medical and Biology Society, Oct. 29—Nov. 1, 1992.
`
`Gregory Belenky, et al., Discrimination of Rested From
`S leep—Deprived EEG inAwake Normal Humans byArti?cial
`Neural Network, IEEE, 1994.
`
`[21] Appl. No.: 08/745,780
`[22] Filed:
`Nov. 8, 1996
`[30]
`Foreign Application Priority Data
`
`Primary Examiner—Robert L. Nasser
`Attorney, Agent, or Firm—Hoffmann & Baron, LLP
`
`Nov. 8, 1995 [GB]
`
`United Kingdom ................. .. 9522872
`
`[57]
`
`ABSTRACT
`
`[51] Int. Cl.6 ...................................................... .. A61B 5/04
`[52] US. Cl. ............................................................ .. 600/544
`[58] Field of Search ................................... .. 600/509, 513,
`600/544, 555, 546, 545
`
`[56]
`
`References Cited
`
`U.S. PATENT DOCUMENTS
`
`4,776,345 10/1988 Cohen et a1. .
`5,047,930
`9/1991 Martens et a1. ....................... .. 600/544
`5,222,503
`6/1993 Ives et al. ....... ..
`..
`5,299,118
`3/1994 Martens et a1. ....................... .. 600/544
`5,447,166
`9/1995 Gevins et a1. ........................ .. 600/544
`5,450,855
`9/1995 Rosenfeld ............................. .. 600/545
`
`OTHER PUBLICATIONS
`
`S. Roberts et al., New Method ofAutomated Sleep Quanti
`?cation, Medical & Biological Engineering & Computing,
`Sep. 30, 1992.
`
`An insomnia or vigilance monitor comprising one or more
`electrodes (1a,1b) for obtaining an electrical signal from a
`subject over a period of epochs, the electrical signal being
`related to the sleep or wakefulness stage type being expe
`rienced by the subject; and a processor (5) adapted to
`analyze the electrical signal and assign a sleep or Wakeful
`ness stage type to each epoch to generate a hypnogram.
`Methods of monitoring sleep or vigilance using the mastoid
`site are also disclosed. Further disclosures relate to a method
`of training and testing a ?rst neural network for use in a
`physiological monitor, and a method of assigning a class to
`an epoch of a physiological signal obtained from a subject
`as a set of samples.
`
`13 Claims, 11 Drawing Sheets
`
`ELECTRODES
`FROM
`SUBJECT'S
`HEAD
`1
`C
`
`C‘RCU'
`g
`
`1b
`
`I
`
`LCD DISPLAY
`u
`l
`
`ISOLATED RS232
`CONNECTION
`12
`
`IMPEDANCE
`MEASUREMENT
`oIRcuITRY ‘
`m
`
`FILTEF!
`a
`
`HOLD
`Q
`
`coNvERTER
`Q
`
`5
`—
`
`J
`
`I
`
`19 J
`To EXTERNAL
`ooMPuTER oR PRINTER
`7
`SELECT
`1_5
`ENTER
`
`RECORD
`11
`
`ON/OFF
`m
`
`F
`
`+
`
`+
`
`ALARM
`g
`
`WATCHDOG
`g
`
`PRSSNFIQQ'yE
`6
`
`_
`
`+
`NoN VOLATILE
`DATA
`MEMORY
`
`7
`
`‘
`
`+
`REAL TIME
`CLOCK
`8
`
`_
`
`13
`SWITCHES
`BATTERY AND
`POWER SUPPLY _>
`g
`j
`
`18
`TO OTHER
`CIRCUITRY
`
`1 of 20
`
`FITBIT EXHIBIT 1012
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 1 0f 11
`
`5,999,846
`
`2 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 2 0f 11
`
`5,999,846
`
`‘\ 0
`
`y0
`
`OUTPUT
`UNITS
`
`1.0
`
`0.5
`
`-10'.0
`
`-5‘.0
`
`0.0
`
`5:0
`
`10.0
`
`3 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 3 0f 11
`
`5,999,846
`
`Fig.4A. 1'
`
`WAKE
`
`O “
`
`l
`
`1
`
`1
`
`l
`
`l
`
`l
`
`1
`
`0:00:00 1:00:00 2:00:00 3:00:00 4:00:00 5:00:00 6:00:00 7:00:00
`
`Fig.4B. 1‘
`
`LIGHT/REM
`
`O
`
`|
`
`l
`
`|
`
`|
`
`|
`
`1
`
`l
`
`0:00:00 1:00:00 2:00:00 3:00:00 4:00:00 5:00:00 6:00:00 7:00:00
`
`Fig.4C. 1'
`
`DEEP
`
`O"
`
`|
`
`I
`
`|
`
`l
`
`|
`
`|
`
`0:00:00 1:00:00 2:00:00 3:00:00 4:00:00 5:00:00 6:00:00 7:00:00
`
`|
`
`Fig.4D. 1:
`
`WAKE-DEEP
`
`'1
`
`1
`
`l
`
`l
`
`|
`
`|
`
`1
`
`1
`
`0:00:00 1:00:00 2:00:00 3:00:00 4:00:00 5:00:00 6:00:00 7:00:00
`
`Fig.4E.W‘
`
`HYPNOGRAM
`
`4‘
`
`l
`
`l
`
`l
`
`l
`
`l
`
`l
`
`|
`
`0:00:00 1:00:00 2:00:00 3:00:00 4:00:00 5:00:00 6:00:00 7:00:00
`
`4 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 4 0f 11
`
`5,999,846
`
`Fig.5.
`
`MAKE SLEEP RECORDINGS /
`WITH STANDARD AND NOVEL
`EEG POSITIONS
`
`I
`
`HAND SCORE THE STANDARD
`3‘
`EEG CHANNEL USING 3
`INDEPENDENT SCORES TO /
`GENERATE STAGE
`CLASSIFICATIONS FOR EACH
`EPOCH
`
`IDENTIFY CONSENSUS
`SCORED EPOCHS
`
`32
`/
`
`I
`
`RANDOIvILY SEPARATE
`33
`CONSENSUS EPOCHS INTO
`TRAINING & TEST SETS WITH /
`BALANCED NUMBERS OF
`EACH STAGE TYPE IN EACH
`SET
`
`I
`
`PERFORM PREPROCESSING 34
`OF STANDARD AND NOVEL
`EEG CHANNELS
`
`I
`
`/-35
`
`USING EEG FROM TRAINING
`EPOCHS, TRAIN NETWORK 'A'
`ON THE STANDARD EEG'S
`PREPROCESSOR OUTPUTS
`USING HUMAN SCORES AS
`LABELS
`
`I
`
`/-36
`
`USING EEG FROM TRAINING
`EPOCHS, TRAIN NETWORK '8'
`ON THE NOVEL EEG'S
`PREPROCESSOR OUTPUTS
`USING HUMAN SCORES AS
`LABELS
`
`/-37
`
`ANALYSE ALL THE STANDARD
`EEG USING NETWORK 'A' TO
`GENERATE MULTIPLE
`PROBABILITY OUTPUTS
`EVERY SECOND
`
`I
`
`I
`
`/38
`
`ANALYSE ALL THE NOVEL
`EEG USING NETWORK '8' TO
`GENERATE MULTIPLE
`PROBABILITY OUTPUTS
`EVERY SECOND
`
`I
`
`I
`
`5 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 5 0f 11
`
`5,999,846
`
`I
`
`OVER EACH EPOCH,
`MULTIPLY ONE SECOND
`PROBABILITIES FOR EACH
`OUTPUT FROM NETWORK 'A'
`AND IDENTIFY THE MOST
`LIKELY SLEEP STAGE
`CLASSIFICATION FOR EACH
`EPOCH
`
`F|g.5 (Cont).
`/39
`
`I
`
`/40
`
`OVER EACH EPOCH,
`MULTIPLY ONE SECOND
`PROBABILITIES FOR EACH
`OUTPUT FROM NETWORK 'B'
`AND IDENTIFY THE MOST
`LIKELY SLEEP STAGE
`CLASSIFICATION FOR EACH
`EPOCH
`
`T
`
`v
`
`COMPARE EACH
`HUMAN'S SCORES OF
`EVERY EPOCH IN ALL
`RECORDINGS WITH THE 41
`OTHER HUMANS /
`SCORES AND WITH
`EACH OF NETWORKS 'A'
`AND'B' IN TURN TO
`ENSURE THAT BOTH
`NETWORKS PERFORM
`ACCEPTABLY
`COMPARED WITH
`INTER-HUMAN SCORING
`
`I
`
`I
`
`COMPARE THE HUMAN
`CONSENSUS SCORES
`OF EPOCHS FROM THE
`42
`TEST SET WITH THE
`STAGE ASSIGNMENTS /
`GENERATED FROM
`NETWORKS 'A' AND '8' IN
`TURN TO ENSURE
`NETWORK 'B' ,USING
`NOVEL EEG, PERFORMS
`AS WELL AS NETWORK
`'A', USING STANDARD
`EEG
`
`6 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 6 0f 11
`
`5,999,846
`
`F|g.6A.
`
`1.0-
`
`'
`
`0.0-
`
`0:00:00
`
`-
`
`1:00:00
`
`F|g.6B.
`
`1.0-
`
`‘
`
`.
`
`.
`
`3:00:00
`2:00:00
`HOURS
`
`.
`
`4:00:00
`
`.
`
`5:00:00
`
`I
`
`8-.
`
`'.
`
`0.0M
`0:00:00
`
`.
`1 :00:00
`
`.
`.
`3:00:00
`2:00:00
`HOURS
`
`.
`4:00:00
`
`,
`5:00:00
`
`F|g.6C.
`
`1.0-
`
`0.0 1
`
`0:00:00
`
`‘
`
`.
`
`1 :00:00
`
`I"
`
`-
`
`'
`
`'
`
`-
`
`3:00:00
`2:00:00
`HOURS
`
`-
`
`'
`
`~
`
`.
`
`4:00:00
`
`5:00:00
`
`7 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 7 0f 11
`
`5,999,846
`
`Flg . 6D.
`
`1.0
`
`'
`
`'1 .0 "
`
`0:00:00
`
`I
`
`1 :00:00
`
`Flg .6 E.
`W 1
`
`R
`
`1 -
`
`2 _
`3
`
`i
`
`'1
`
`I
`
`I
`
`3:00:00
`2:00:00
`HOURS
`
`1
`
`4:00:00
`
`|
`
`5:00:00
`
`I
`
`.
`
`0:00:00
`
`1 :00:00
`
`3:00:00
`2:00:00
`HOURS
`
`4:00:00
`
`5:00:00
`
`Hg .6 F.
`
`W _
`
`M _
`
`R _
`
`1 -
`
`2 _
`
`3 _
`
`4 -
`
`0:00:00
`
`.
`
`1 :00:00
`
`.
`
`.
`
`3:00:00
`2:00:00
`HOURS
`
`.
`
`4:00:00
`
`,
`
`5:00:00
`
`8 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 8 0f 11
`
`5,999,846
`
`Fig.7.
`50
`GEE?
`
`CONTINUE
`RECORDING ?
`
`60\
`
`GENERATE & STORE
`SLEEP STATISTICS
`FROM EPOCH
`CLASSIFICATIONS
`
`62
`
`52
`ACQUIRE EEG DATA //
`SAMPLES
`
`5s
`
`COMPLETED
`SECOND ACQUIRED
`?
`
`NO
`
`RUN PREPROCESSING
`FILTER OVER CURRENT /
`SECONDS'S EEG AND
`GENERATE COEFFICIENTS
`
`I
`
`55
`FEED SECOND'S
`COEFFICIENT INTO NEURAL /
`NETWORK TO GENERATE 3
`PROBABILITY VALUES
`
`EPOCH'S
`WORTH OF DATA
`PROCESSED ?
`
`NO
`
`57
`FOR EACH PROBABILITY
`OUTPUT, MULTIPLY /
`TOGETHER THE
`PROBABILITIES FROM EACH
`SECOND OVER THE EPOCH
`
`I
`
`_
`
`_
`
`-
`
`9 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 9 0f 11
`
`5,999,846
`
`'
`
`Fig.7 (Cont).
`'1
`FROM MULTIPLIED
`PROBABILITIES, DETERMINE 58
`THE OPTIMAL SLEEP /
`STAGE TO APPLY TO THE
`EPOCH
`
`59
`I
`STORE RESULT /
`
`VIGILANCE
`MONITOR ?
`
`NO
`
`VIGILANCE
`LEVEL TOO
`LOW ?
`
`NO
`
`65 \
`GENERATE/STORE
`WARNING
`
`fee
`
`CANCEL
`WARNING
`
`10 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 10 0f 11
`
`5,999,846
`
`'
`
`CENTRAL EEG: WAKE
`x.
`
`'
`
`.XXX
`
`XXXXXXXXXX
`
`MASTOID EEG: WAKE
`XXXXXXXX.
`x x x x x .
`
`XX.
`.XXX
`
`XX
`
`.XXXX
`
`XXXXXX...
`
`XXXXXXX..
`
`XXX
`
`XXX
`
`XXX
`
`XXX
`
`XX
`
`XX
`
`C ..X
`
`m X
`
`MXXX E
`E .XX C
`m; EU.
`MXXXX. HM
`ix. m”
`RXXXX MS
`EXXX MG
`
`G H B 8
`mg .ml9 F F
`
`XXXX.
`
`.XX
`XXX
`
`. X
`
`XX
`
`.XXXXXXX
`
`XXXXXXXX
`
`XXX
`
`XX.
`
`.X
`
`M ..
`
`
`W. .XXXXXXX .. X
`S. .XXXXXXX X
`
`$1 M
`m .. $8
`
`D .. mm
`
`G. .XXX. E2X E .. warm.
`.XXXXX mm X
`
`.XXXXX
`.XXXXXX
`
`.XX
`
`XX
`
`XX
`
`XX
`
`.XX.
`
`.XX .
`
`.XXXXX
`
`.XX
`
`.XX
`
`11 of 20
`
`

`
`U.S. Patent
`
`Dec. 7, 1999
`
`Sheet 11 0f 11
`
`5,999,846
`
`Fig.9A.
`AWAKE
`WWWWWWWWWW
`ALPHA ACTIVITY
`BETA ACTIVITY
`
`Fig . 9B.
`STAGE I SLEEP
`
`THETA ACTIVITY
`
`Fig . 9C.
`STAGE 2 SLEEP
`
`K COMPLEX
`
`SPINDLE
`
`STAGE 3 SLEEP
`
`I SECONDS
`012345
`
`DELTA ACTIVITY
`
`Fig.9 E.
`
`STAGE 4 SLEEP
`
`DELTA ACTIVITY
`
`REM SLEEP
`
`Fig .9 F.
`MWWWWWWWMW
`
`TH ETA ACTIVITY
`
`BETA ACTIVITY
`
`12 of 20
`
`

`
`1
`PHYSIOLOGICAL MONITORING
`
`FIELD OF THE INVENTION
`
`The present invention relates to improvements in physi
`ological monitoring, in particular sleep or vigilance moni
`toring.
`
`DESCRIPTION OF THE PRIOR ART
`
`The method currently employed World-Wide for scoring
`sleep recordings is described in Rechtschaffen and Kales
`(1968), “A Manual of Standardized Technology, Techniques
`and Scoring System for Sleep Stages of Human Subjects”.
`Scoring requires the folloWing signals to be recorded;
`electroencephalogram (EEG)—from a position near the
`top of the head,
`tWo eye channels (electro-oculogram (EOG))—from elec
`trodes near the outer canthus of each eye, and
`chin muscle tone (electromyogram (EMG))—from a pair
`of electrodes under the chin.
`Sleep scoring breaks the recording into epochs of typi
`cally 20, 30 or 40 seconds duration. Each epoch has a sleep
`stage classi?cation applied to it. The six recognised classi
`?cations are: Stage Wake; Stage REM (Rapid Eye
`Movement); Stages 1, 2, 3 and 4. The classi?cation of each
`epoch ?rst requires the identi?cation of particular features in
`the EEG and EOG, and measurement of the amplitude of the
`EMG relative to the background EMG level. The features
`are identi?ed using frequency and amplitude criteria. Such
`a recording technique and method of scoring is knoWn as
`polysomnography.
`A set of rules is then applied to the features to obtain the
`classi?cation for each epoch.
`Examples of conventional EEG traces Which have been
`assigned to the sleep stages mentioned above are shoWn in
`FIGS. 9(a)—(}‘). FIGS. 9(a)—(f) shoW the folloWing stages;
`FIG. 9(a)-aWake;
`FIG. 9(b)-stage 1;
`FIG. 9(c)-stage 2;
`FIG. 9(d)-stage 3;
`FIG. 9(e)-stage 4;
`FIG. 9(f)-REM.
`Once each epoch has been assigned a classi?cation clean
`up rules are applied that can reclassify certain epochs
`according to their context.
`The classi?cations of each epoch for the entire night’s
`recording can be plotted against time. This is a hypnogram.
`Summary statistics can be derived from the hypnogram
`that alloW objective measures of the quality of sleep to be
`made.
`
`SUMMARY OF THE INVENTION
`
`According to a ?rst aspect of the present invention there
`is provided an insomnia monitor comprising
`(1) one or more electrodes for obtaining an electrical
`signal from a subject over a period of epochs, the
`electrical signal being related to the sleep stage type
`being experienced by the subject;
`(2) a processor adapted to analyZe the electrical signal and
`assign a sleep stage type to each epoch to generate a
`hypnogram;
`(3) means for analysing the hypnogram to generate a
`summary index of sleep quality over the period of
`epochs; and
`
`10
`
`15
`
`25
`
`35
`
`45
`
`55
`
`65
`
`5,999,846
`
`2
`(4) means responsive to the means for analysing the
`hypnogram to display the summary index of sleep
`quality.
`The present invention provides a device Which can be self
`contained, portable and cheap. The device generates and
`displays a summary index Which provides a simple objective
`indicator of the degree of insomnia suffered by the subject.
`Insomnia can manifest itself in many forms, and therefore
`different sleep summary indices may be generated and
`displayed. For instance a subject may experience a simple
`lack of sleep. In this case a loW Sleep Ef?ciency Index
`(Which is the ratio of the time asleep to the time in bed) Will
`provide the required indication and is generated and dis
`played. Alternatively the subject may have a high Sleep
`Ef?ciency Index but may sleep “badly”. For instance the
`subject may experience irregular sleep cycles (eg alternating
`long/short periods of REM sleep). Therefore an alternative
`or additional summary index may comprise an indication of
`the periodicity of the sleep/Wake continuum. For instance
`the index may be derived from the periodicity, variance of
`frequency etc. of one or more categories of sleep.
`Typically the device is Worn by a subject during the night,
`during Which time it continually acquires and analyses the
`electrical signal Which is typically an EEG signal from the
`subject’s scalp. When the recording is terminated in the
`morning it provides one or more simple indices of sleep
`quality Which indicate hoW Well the subject slept. Values
`beloW predetermined thresholds indicate that the subject
`should refer to either a general practitioner or a sleep
`laboratory for further investigation.
`The device is primarily intended for use by general
`practitioners for use as a screening tool for subjects Who
`claim to be insomniacs and for members of the public Who
`Wish to monitor the quality of their oWn sleep.
`Providing a summary index reduces the time taken by the
`physician to make a decision on Whether additional treat
`ment is required, and it does not need particular skills,
`making it more suitable for GPs to use. Preferably the
`summary index comprises a Yes/No value indicating
`Whether or not the subject suffers from some form of
`insomnia.
`The ?rst aspect of the present invention also extends to a
`method of sleep monitoring, the method comprising:
`(1) obtaining an electrical signal from a subject over a
`period of epochs, the electrical signal being related to
`the sleep stage type being experienced by the subject;
`(2) analysing the electrical signal and assigning a sleep
`stage type to each epoch to generate a hypnogram;
`(3) analysing the hypnogram to generate a summary index
`of sleep quality over the period of epochs; and
`(4) displaying the summary index of sleep quality.
`Using the same physiological parameters and using a
`similar process to that described above but employing an
`alternative set of rules to those of Rechtschaffen and Kales
`it is possible to construct a “Wakeogram” that indicates the
`degree of Wakefulness of the subject before they fall asleep,
`as Well as their depth of sleep once they are asleep. From the
`Wakeogram it is possible to derive a measure of the vigi
`lance of the subject.
`In accordance With a second aspect of the present inven
`tion there is provided a vigilance monitor comprising
`(1) one or more electrodes for obtaining an electrical
`signal from a subject over a period of epochs, the
`electrical signal being related to the Wakefulness stage
`type being experienced by the subject;
`(2) a processor adapted to analyZe the electrical signal and
`assign a Wakefulness stage type to each epoch to
`generate a Wakeogram;
`
`13 of 20
`
`

`
`3
`(3) means for monitoring the output of the Wakeogram to
`determine whether the output of the Wakeogram meets
`predetermined criteria; and
`(4) means responsive to the means for monitoring the
`Wakeogram to generate a message when the output of
`the Wakeogram meets the predetermined criteria.
`The second aspect of the invention provides a vigilance
`monitor which allows people in safety-critical jobs to have
`their vigilance directly monitored. Vigilance analysis
`requires segmentation and classi?cation of the electrical
`signal during wakefulness as well as during sleep. A suitable
`scoring technique classi?es wakefulness into several differ
`ent categories, each representing a lower state of alertness or
`vigilance. The scorers may use an arbitrary epoch length of
`typically 20 seconds, but other epochs could be chosen (as
`per sleep). Just as for sleep, scoring is based on visual
`methods.
`The electrical signal or Wakeogram may be stored for
`later analysis or monitoring. In this case the electrical signal
`or Wakeogram may be analysed at a later date to determine
`whether a predetermined level of vigilance has been main
`tained over the period of epochs. In this case the vigilance
`monitor typically comprises means for analysing the Wakeo
`gram to generate a summary indeX of vigilance quality over
`the period of epochs. The summary indeX may be stored for
`later output, or may be displayed by the vigilance monitor.
`Preferably however the Wakeogram is analysed and the
`message is generated during the period of epochs. In this
`case, the message gives a real-time continuous indication of
`the vigilance of the subject.
`When the output of the Wakeogram meets the predeter
`mined criteria, which typically have been determined in
`advance as representing a lowered level of vigilance, the
`device generates a message of some kind. This might be an
`audible, visual or electronic message and may be used to
`alarm the subject.
`The second aspect of the present invention also eXtends to
`a method of vigilance monitoring, the method comprising:
`(1) obtaining an electrical signal from a subject over a
`period of epochs, the electrical signal being related to
`the wakefulness stage type being experienced by the
`subject;
`(2) analysing the electrical signal and assigning a wake
`fulness stage type to each epoch to generate a Wakeo
`gram;
`(3) monitoring the Wakeogram to determine whether the
`output of the Wakeogram meets predetermined criteria;
`and
`(4) generating a message when the output of the Wakeo
`gram meets the predetermined criteria.
`The following comments apply both to the insomnia
`monitor according to the ?rst aspect of the present invention
`and to the vigilance monitor according to the second aspect
`of the present invention.
`The hypnogram or Wakeogram may be generated from a
`plurality of electrical signals from standard sites (eg EEG,
`EOG, EMG etc). Preferably however the device generates
`the hypnogram or Wakeogram from a single channel only
`(preferably an EEG channel). The use of only a single EEG
`channel reduces the cost of ampli?cation circuitry, and fewer
`electrodes are required to perform a recording than tradi
`tional polysomnography recordings.
`The device typically comprises a portable, battery
`powered, self contained unit.
`The summary indeX or message is typically generated
`within the device. The indeX or message can then be
`
`10
`
`15
`
`25
`
`35
`
`45
`
`55
`
`65
`
`5,999,846
`
`4
`represented by displaying it on a digital display and/or by
`sounding an audible alarm and/or by storing it within the
`device for later review. Alternatively the indeX or message
`may be transmitted to another device (eg. by transmitting it
`to a computer via a serial interface).
`The insomnia or vigilance monitor may store the original
`EEG signal and analyZe the stored signal at the end of a
`period of sleep or at the end of a period of vigilance
`monitoring. Typically however the insomnia monitor or
`vigilance generates the hypnogram or Wakeogram “on-the
`?y” and further comprises a memory adapted to store the
`hypnogram or Wakeogram. This minimises the cost of
`memory—12 hours of EEG would require typically 4MB of
`non-volatile storage. If the analysis is performed on-line
`only the results need to be stored, which can be done more
`cheaply, in a few kB.
`According to a third aspect of the invention there is
`provided a method of sleep or vigilance monitoring, the
`method comprising obtaining an EEG signal from the mas
`toid site behind a subject’s ear, and performing sleep or
`vigilance analysis on the EEG signal.
`The mastoid site provides a novel site for monitoring
`electrical activity to monitor sleep or vigilance quality. The
`mastoid sites lie below the hairline. This allows disposable,
`stick-on electrodes to be used instead of the glued-on
`electrodes normally required for sleep studies. The latter
`require acetone based glues and solvents to be used, and
`require trained personnel to ?t them.
`The mastoid signal cannot necessarily be interpreted by
`humans but offers advantages in terms of hook-up time,
`convenience, comfort, aesthetics, and lowered skill require
`ments for application.
`Typically the method comprises obtaining a differential
`signal between the two mastoid sites.
`Preferably the sleep or vigilance analysis (such as poly
`somnography analysis) is carried out on the mastoid EEG
`signal alone. Typically the analysis is carried out by a neural
`network.
`Preferably the ?rst and/or second and third aspects of the
`invention are combined, ie. the electrical signal is obtained
`in step (1) from the mastoid site.
`The sleep-wake continuum can be fully described in terms
`of a ?nite number of continuous processes; for insomnia
`monitoring these are Wakefulness, Dreaming/Light Sleep
`and Deep Sleep. These correspond to the human-scored
`stages of Wake, REM/Stage 1 and Stage 4. Wakefulness can
`be further partitioned into different degrees of Wakefulness
`or Vigilance, such as Active Wake, Quiet Wake, Wake with
`high alpha content, Wake with high theta content. What is
`required is a means of tracking the time course of the EEG
`as it moves between these processes.
`Typically the processor of the ?rst or second aspect of the
`invention and the means for performing sleep or vigilance
`analysis according to the third aspect of the invention
`comprises a neural network such as a multilayer perceptron
`(MLP). The requirements outlined above are ideally
`matched to the functional capabilities of an MLP. This
`neural network can be trained to perform polysomnography
`analysis on a single EEG channel (instead of 1 EEG, 2 EOG
`and 1 EMG). It also can be trained to analyZe an uncon
`ventional EEG signal, such as the mastoid signal.
`Since conventional neural networks are static pattern
`classi?ers, the EEG signal must be segmented into “iframes”
`during which the signal properties can be deemed to be
`stationary. The EEG is usually considered to be quasi
`stationary over intervals of the order of one second, as this
`is the characteristic time of key transient features such as
`
`14 of 20
`
`

`
`5,999,846
`
`5
`sleep spindles. The important information in the EEG is in
`the frequency domain. An auto-regressive
`model of the
`EEG signal provides adequate representation of the EEG
`during the sleep-Wake continuum and may be used as an
`input representation to train and test the neural netWork. If
`a 10 coef?cient model is used, for example, this gives a
`10-dimensional input vector to the neural netWork for every
`one-second segment of EEG. In the case of vigilance moni
`toring the optimal number of coef?cients is typically higher
`than 10, as the signal becomes more complex during Wake
`fulness so that a higher order model is needed to describe it
`fully.
`The signal from a novel electrode site such as the mastoid
`site cannot be easily analyZed using the standard human
`scoring methods discussed above.
`According to a fourth aspect of the present invention there
`is provided a method of training and testing a ?rst neural
`netWork for use in a physiological monitor, the method
`comprising;
`(1) obtaining a ?rst set of physiological signals from a
`subject, each member of the set being obtained over a
`period of epochs on a subject;
`(2) obtaining a second set of physiological signals from
`the subject, each member of the set being obtained over
`the same epochs as a respective member of the ?rst set
`of signals, and having a correlation With the respective
`member of the ?rst set of signals;
`(3) assigning a class to each epoch by analysing the set of
`?rst signals by a knoWn method;
`(4) separating each set of signals into a set of training
`signals and a set of test signals;
`(5) training a second neural netWork by inputting the
`training set of ?rst signals and using the classes
`assigned to each epoch as training labels;
`(6) training the ?rst neural netWork by inputting the
`training set of second signals and using the classes
`assigned to respective epochs of the ?rst set of signals
`as the training labels; and
`(7) monitoring the performance of the ?rst neural netWork
`by comparing the class assigned to each epoch by the
`?rst and second netWorks When input With the second
`and ?rst set of test signals.
`The fourth aspect of the present invention provides a
`method of training a neural netWork to act on input data (ie
`the second signals) Which cannot be analyZed in a conven
`tional Way. Provided that there is a correlation (Which may
`be non-linear) betWeen the tWo sets of signals, the ?rst
`neural netWork can be trained using the second signals as
`input data, but using labels Which are obtained from the ?rst
`signals in a conventional Way.
`For example, the signals may be derived from different
`physiological measurements; for example, the ?rst set of
`signals may comprise signals derived from the EEG of a
`subject, and the second set of signals may relate to the blood
`pressure of the subject over simultaneous epochs.
`Preferably hoWever the method is used to train and test the
`neural netWork of a sleep or vigilance monitor Which is used
`to monitor an electrical signal (ie. the second electrical
`signal) Which cannot be analyZed using standard polysom
`nography analysis. The second netWork is trained using
`labels obtained from a ?rst set of signals from standard sites
`(typically EEG from the scalp, tWo EOG signals and an
`EMG signal) Which can be analyZed using standard poly
`somnography analysis. Typically the second electrical signal
`is an EEG signal obtained from an electrode site Which
`cannot necessarily be scored by humans (such as the mastoid
`site).
`
`10
`
`15
`
`3O
`
`35
`
`45
`
`55
`
`6
`The technique of training a neural netWork to obtain the
`same output from indirectly related but different input
`signals (in this case, of EEG taken from different sites) can
`equally be applied to other analysis systems including an
`off-line sleep or vigilance analysis system that retrospec
`tively analyses stored data.
`According to a ?fth aspect of the present invention there
`is provided a method of assigning a class to an epoch of a
`physiological signal obtained from a subject as a set of
`samples, the method comprising
`(1) estimating the probability of each of a plurality of
`stage types for each sample;
`(2) cumulatively multiplying the probabilities for each
`sample With the probabilities of a previous sample;
`(3) determining Which stage type has the highest prob
`ability When all samples in the epoch have been cumu
`latively multiplied; and
`(4) assigning that stage type to the epoch.
`Typically the physiological signal comprises an EEG
`signal. The signal may be obtained from the mastoid site.
`Preferably the methods of the fourth and/or ?fth aspects
`of the invention are employed in the production and/or
`operation of an insomnia monitor according to the ?rst
`aspect of the invention or a vigilance monitor according to
`the second aspect of the invention. In addition, the methods
`of the fourth and/or ?fth aspects of the present invention
`may be combined With a method according to the third
`aspect of the present invention.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`Embodiments of all aspects of the present invention Will
`noW be described With reference to the accompanying
`drawings, in Which:
`FIG. 1 is a block diagram of an embodiment of an
`insomnia monitor or vigilance monitor according to the ?rst
`and/or second and ?fth aspects of the present invention;
`FIG. 2 illustrates an example of the multi-layer perceptron
`(MLP) neural netWork used by the monitor;
`FIG. 3 illustrates the sigmoidal non-linearity used in each
`of the MLP’s hidden and output units;
`FIG. 4 is an example of the neural networks outputs;
`FIG. 5 is a block diagram illustrating an embodiment of
`the method of training and testing the neural netWork
`according to the fourth aspect of the present invention;
`FIGS. 6(a) to 6(c) illustrate second-by-second output
`probabilities from a three-class MLP for (a) Wakefulness,
`(b) Light Sleep/REM, (c) Deep Sleep;
`FIG. 6(a) illustrates the pseudo-hypnogram obtained by
`subtracting (c) from (a);
`FIG. 6(e) illustrates a hypnogram obtained by multiplying
`output probabilities according to the fourth aspect of the
`invention from a six-class MLP over 30-second epochs;
`FIG. 6(}‘) illustrates the corresponding human/scored hyp
`nogram;
`FIG. 7 illustrates the insomnia or vigilance monitor main
`processing loop;
`FIG. 8 illustrates the Kohonen maps from independent
`analysis made on data recorded simultaneously from stan
`dard and novel EEG electrode sites; and, FIGS. 9(a)—9(}‘)
`illustrate conventional EEG traces for the various stages of
`sleep.
`
`65
`
`EMBODIMENT
`Ablock diagram of a typical implementation of the device
`according to the invention is described beloW, Which refers
`to FIG. 1.
`
`15 of 20
`
`

`
`10
`
`15
`
`25
`
`7
`The device is a small, self contained portable unit that can
`continually acquire and analyZe EEG signals for at least 12
`hours. Results are held in non-volatile data memory 7 for
`later display on LCD display 11 or doWn-loading via an
`isolated RS232 link 12.
`PoWer is provided by internal primary or rechargeable
`batteries and a regulated poWer supply 9.
`Signals are acquired from electrodes 1 mounted on the
`subject’s head via sWitching circuitry 2 and input ampli?er
`3. The input ampli?er has an analogue bandWidth of at least
`0.5HZ—30HZ and is of a high gain, loW noise instrumentation
`design. The signal is input to loW-pass ?lter 21 to reduce
`unWanted aliasing components before analogue to digital
`conversion. The signal is regularly held by sample and hold
`circuit 22 and converted to digital format by analogue to
`digital converter 4. The resultant quantised data samples are
`transferred to loW poWer microcontroller 5 for processing.
`The sampling rate is typically 128HZ and the quantisation of
`the analogue to digital converter 4 is typically 12 bits, Which
`provides suf?cient dynamic range not to require a gain
`control on the input ampli?er 3.
`When recording one channel of EEG from the mastoid
`site, three electrodes are typically necessary; tWo of them
`(1a,1b) comprise the differential inputs to input ampli?er 3
`(the recording is of one part of the body With respect to
`another; in this case the tWo parts are the tWo mastoid sites),
`and the third is an “indifferent” lead (not shoWn) Whose sole
`function is to alloW input ampli?er return currents to ?oW.
`The indifferent lead can be attached to any part of the
`subject’s body. It is possible to produce an ampli?er Without
`an indifferent lead, but the performance is not as satisfactory.
`Due to the loW amplitude of the signals (in the 0—200 pV
`range), the fact that the subject is fairly mobile, and the
`uncontrolled environment in Which the equipment is
`operating, the number of potential sources of artefact con
`tamination of the signal are substantial; a poorly designed
`system Will even detect the passage of the subject through
`the earth’s magnetic ?eld.
`Before signal acquisition begins, the impedances of the
`electrodes on the subject’s head are measured by causing
`impedance measurement circuitry 10 to drive a signal of
`knoWn amplitude and source impedance via the sWitching
`circuitry 2 through each of the electrodes 1a, 1b in turn onto
`the subject’s scalp. The resultant signal is measured by the
`microcontroller by the process described above and from it
`the impedances of each of the electrodes ?tted to the
`subject’s scalp are calculated in turn. A Warning message is
`displayed on the LCD display 11 if the impedance of either
`of the connections to the subject’s head is unacceptably
`high.
`During data acquisition, the device continually acquires
`EEG signals from the subject’s head for analysis. The
`microcontroller 5 analyses the quantised values and from
`them generates results that are stored in the non-volatile data
`memory 7.
`The programme for the microcontroller is held in pro
`gramme memory 6.
`Real-time clock 8 Which can be read from and Written to
`by the microcontroller 5 alloWs the results to be stored
`relative to the time of day.
`Watchdog 20 resets the microcontroller 5 if the micro
`controller fails to Write to it periodically. If the microcon
`troller 5 is reset it Will identify Whether it Was in record
`before the reset Was received and if so, go back into record
`so that a minimal amo