U.S. patent application number 13/249812 was filed with the patent office on 2012-04-05 for fetal ecg monitoring.
This patent application is currently assigned to Tufts Medical Center, Inc.. Invention is credited to Gari D. Clifford, Jay Ward, Adam J. Wolfberg.
Application Number | 20120083676 13/249812 |
Document ID | / |
Family ID | 41164560 |
Filed Date | 2012-04-05 |
United States Patent
Application |
20120083676 |
Kind Code |
A1 |
Wolfberg; Adam J. ; et
al. |
April 5, 2012 |
FETAL ECG MONITORING
Abstract
A method for fetal monitoring includes acquiring electrical
signals from a set of electrodes, for example, a set of surface
electrodes applied to a maternal abdominal region. The electrical
signals are analyzed, including by performing a morphological
analysis of fetal electrocardiogram signals. A clinical indicator
is then determined from a result of performing the morphological
analysis.
Inventors: |
Wolfberg; Adam J.; (Newton,
MA) ; Clifford; Gari D.; (Boston, MA) ; Ward;
Jay; (Stratham, NH) |
Assignee: |
Tufts Medical Center, Inc.
Boston
MA
E-Trolz, Inc.
North Andover
MA
Massachusetts Institute of Technology
Cambridge
MA
|
Family ID: |
41164560 |
Appl. No.: |
13/249812 |
Filed: |
September 30, 2011 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
13113293 |
May 23, 2011 |
|
|
|
13249812 |
|
|
|
|
12424046 |
Apr 15, 2009 |
7949389 |
|
|
13113293 |
|
|
|
|
61100807 |
Sep 29, 2008 |
|
|
|
61045055 |
Apr 15, 2008 |
|
|
|
Current U.S.
Class: |
600/301 ;
600/382 |
Current CPC
Class: |
A61B 5/4362 20130101;
A61B 5/0245 20130101; A61B 5/02411 20130101; A61B 5/344 20210101;
A61B 5/412 20130101 |
Class at
Publication: |
600/301 ;
600/382 |
International
Class: |
A61B 5/0205 20060101
A61B005/0205; A61B 5/00 20060101 A61B005/00; A61B 5/0478 20060101
A61B005/0478; A61B 5/103 20060101 A61B005/103; A61B 5/04 20060101
A61B005/04; A61B 5/0448 20060101 A61B005/0448 |
Claims
1-28. (canceled)
29. A fetal monitor, comprising: a garment adapted for wearing
around a torso of a pregnant female; and a plurality of electrodes
associated with the garment such that, when the garment is disposed
around the torso, at least one physiological parameter of the fetus
is detected regardless of the position of the fetus within the
pregnant female.
30. The fetal monitor of claim 29 wherein the physiological
parameter is fetal heart rate, fetal brain activity, fetal body
position, or a combination thereof
31. The fetal monitor of claim 29 wherein the plurality of
electrodes comprise one or more dry electrodes.
32. The fetal monitor of claim 29 wherein the plurality of
electrodes comprise one or more gel-adhesive electrodes.
33. The fetal monitor of claim 29 wherein the plurality of
electrodes comprise at least one dry electrode and at least one
gel-adhesive electrode.
34. The fetal monitor of claim 29 further comprising electronics
that receives a collecting electrode signal from at least one of
the plurality of electrodes, receives a reference electrode signal
from at least another one of the plurality of electrodes, and
produces a lead signal based on the received collecting and
reference electrode signals.
35. The fetal monitor of claim 34 wherein the lead signal produced
by the electronics comprises a voltage differential between the
received collecting and reference electrode signals.
36. The fetal monitor of claim 34 wherein the electronics comprises
an amplifier, each of the received collecting and reference
electrode signals being an input to the amplifier and the lead
signal being an output of the amplifier.
37. The fetal monitor of claim 34 wherein the electronics receives
a maternal reference electrode signal from at least one of the
plurality of electrodes.
38. The fetal monitor of claim 34 wherein each of the at least one
of the plurality of electrodes producing the collecting electrode
signal is referred to as a collecting electrode and each of the at
least one of the plurality of electrodes producing the reference
electrode signal is referred to as a reference electrode, and
wherein at least one of the collecting electrodes and at least one
of the reference electrodes are disposed on different sides of the
torso when the garment is worn around the torso of the pregnant
female.
39. The fetal monitor of claim 38 wherein one or more of the
collecting electrodes are disposed within an abdominal region of
the torso and one or more of the reference electrodes are disposed
within a lumbar region of the torso.
40. The fetal monitor of claim 38 wherein one or more of the
collecting electrodes are disposed within a lumbar region of the
torso and one or more of the reference electrodes are disposed
within an abdominal region of torso.
41. The fetal monitor of claim 38 wherein one or more of the
collecting electrodes are disposed on a left side of the torso and
one or more of the reference electrodes are disposed on a right
side of the torso.
42. The fetal monitor of claim 38 wherein one or more of the
collecting electrodes are disposed on a right side of the torso and
one or more of the reference electrodes are disposed on a left side
of the torso.
43. The fetal monitor of claim 38 wherein one or more of the
collecting electrodes are disposed on a left side of the torso, and
one or more of the reference electrodes are disposed within an
abdominal region of the torso, the lumbar region of the torso, or a
combination thereof.
44. The fetal monitor of claim 38 wherein one or more of the
collecting electrodes are disposed on a right side of the torso,
and one or more of the reference electrodes are disposed within an
abdominal region of the torso, the lumbar region of the torso, or a
combination thereof.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application claims the benefit of U.S. Provisional
Application No. 61/045,055, titled "Fetal Monitoring System," filed
Apr. 15, 2008, and U.S. Provisional Application No. 61/100,807,
titled "Fetal ECG Monitoring," filed Sep. 29, 2008. The contents of
the above applications are incorporated herein by references.
BACKGROUND
[0002] This specification relates to fetal ECG (fECG)
monitoring.
[0003] Electrocardiogram (ECG) monitoring has been widely used on
adult patients for detecting medical conditions, for example,
abnormities associated with the heart. Signals representing a
patient's cardiac activities can be collected through a set of skin
surface electrodes distributed over the patient's body, for
example, attached to the patient's chest and limbs.
[0004] Monitoring of fetal ECG can be difficult due to the
co-existence of maternal and fetal signals in raw signals acquired
from a patient, as well as the relatively low fetal signal level
relative to the maternal signal and other noise sources. Some
conventional approaches to collecting fetal ECG signals include
placing a wire electrode onto the fetal scalp. Although the fetal
scalp electrode may provide a relatively clean fetal signal, this
procedure can only be performed under limited clinical
circumstances (e.g., when a patient is in labor, has ruptured
amniotic membranes, and has a dilated cervix) and thus may not be
suitable for the vast majority of pregnant and laboring patients.
The placement of the fetal scalp electrode may also present certain
risks to fetal safety, as rare cases of fetal scalp abscess and
newborn death have been reported.
SUMMARY
[0005] In one aspect, in general, a fetal monitoring system
includes a data acquisition system for acquiring signals including
signals representing surface measurements of cardiac activity. A
signal analyzer is coupled to the data acquisition system and is
configured to analyze the acquired signals to generate an output
having at least an clinical indicator characterizing a clinical
condition. The signal analyzer includes a signal processor for
extracting fetal electrocardiogram signals from the acquired
electrical signals, and a clinical condition detector for
performing a morphological analysis of the extracted fetal
electrocardiogram signals, and based on a result of the
morphological analysis, determining the clinical indicator. An
output system is provided for presenting a representation of the
clinical indicator.
[0006] Embodiments of this aspect may include one or more of the
following features.
[0007] The output system includes a display unit for generating a
visual representation of the output of the signal analyzer. The
display unit includes, for example, a computer screen and/or a
handheld device. A wireless transmitter may be provided for
transmitting the output of the signal analyzer to the handheld
device.
[0008] The data acquisition system includes an electrode array
having at least a plurality of electrodes attachable to a maternal
abdominal region. The electrode array may further include a second
plurality of electrodes attachable to a maternal lumbar region, and
potentially a third plurality of electrodes attachable to a
maternal side region. The pluralities of electrodes are positioned
in a pre-determined arrangement on a garment.
[0009] The signal analyzer further includes a heart rate detector
for determining a fetal heart rate from the acquired signals. The
output system is further configured for presenting a representation
of the fetal heart rate determined by the signal analyzer. The
heart rate detector may be further configured for determining a
degree of irregularity in the fetal heart rate.
[0010] The output system is further configured for presenting a
waveform representation of the fetal electrocardiogram signals.
[0011] The clinical condition detector is further configured for
determining a measure of morphological variation in the extracted
fetal electrocardiogram signals. The measure of morphological
variation includes an entropy of a sequence of segment
classifications.
[0012] The clinical indicator determined by the clinical condition
detector includes an indicator of a fetal condition. The clinical
indicator determined by the clinical condition detector may include
an indicator of at least one of chorioamnionitis, preeclampsia,
inflammation, and infection.
[0013] A signal selection unit is coupled to the signal analyzer
for selectively rejecting one or more of the acquired signals based
on a quality of the acquired signals.
[0014] In another aspect, in general, a fetal monitoring system
includes a data acquisition system for acquiring signals including
signals representing surface measurements of cardiac activity. A
signal analyzer is coupled to the data acquisition system and is
configured to analyze the acquired signals, including: obtaining
information characterizing a fetal orientation according to a
cardiac dipole model; and determining the fetal orientation based
on the obtained information. An output system is provided for
presenting a representation of the fetal orientation determined by
the signal analyzer.
[0015] In another aspect, in general, a method for fetal monitoring
includes acquiring electrical signals from a set of electrodes.
These electrodes include a set of electrodes applied to a maternal
abdominal region. The electrical signals are analyzed, including by
performing a morphological analysis of fetal electrocardiogram
signals. A clinical indicator is then determined from a result of
performing the morphological analysis.
[0016] Aspects can include one or more of the following.
[0017] Performing the morphological analysis includes determining a
quantitative measure of morphological variation. For example,
determining the measure of morphological variation includes
characterizing segments of signals determined from the acquired
electrical signals according to a group of classes, and determining
a measure of variation in sequences of segment classifications. The
quantitative measure of morphological variation may include an
entropy of a sequence of segment classifications.
[0018] Determining a clinical indicator includes determining an
indicator of a fetal condition.
[0019] Determining a clinical indicator includes determining an
indicator of an inflammation condition.
[0020] Determining a clinical indicator includes determining an
indicator of at least one of chorioamnionitis, preeclampsia,
inflammation, and infection.
[0021] In another aspect, in general, a method for fetal monitoring
includes acquiring electrical signals from a plurality of
electrodes. These electrodes include a plurality of electrodes
applied to a maternal abdominal region. The electrical signals are
analyzed, including obtaining information characterizing a fetal
orientation, for example, according to a cardiac dipole model. The
fetal orientation, including, for example, fetal movement and fetal
position, is then determined based on the obtained information.
[0022] In another aspect, in general, a method for fetal monitoring
includes acquiring electrical signals from a plurality of
electrodes. These electrodes include a plurality of electrodes
applied to a maternal abdominal region. The electrical signals are
analyzed, including obtaining information characterizing a muscle
movement associated with uterine contraction. A characteristic of
the uterine contraction (e.g., a frequency or a strength the
contraction) is then determined based on the obtained
information.
[0023] In other aspects, in general, a medical apparatus is
configured to acquiring signals from a plurality of electrodes and
perform steps of the methods identified above.
[0024] In another aspect, in general, software stored on a computer
readable medium includes instructions for causing a computing
system to receive data representing signals from a plurality of
electrodes, and perform steps of the methods described above.
[0025] Some embodiments may have one or more of the following
advantages.
[0026] In some embodiments, morphologic entropy in fetal ECG
signals is used as a risk metric for early detection of
inflammation and neuronal injury during pregnancy, for example, due
to conditions such as intrauterine infection that are associated
with an increased risk of cerebral palsy and sepsis in newborns.
Early detection of inflammation may allow for interventions that
can reduce the risk of adverse new born outcome.
[0027] In some examples, morphologic entropy of the fetal ECG
signal is measured using an unsupervised algorithm to first
partition heart beats into different classes of activity based on
their morphology, and then to compute the entropy of the symbolic
sequence obtained by replacing each beat in the original signal
with a label corresponding to its morphologic class. When evaluated
on fetal ECG recordings, morphologic entropy shows a statistically
significant correlation (e.g., a substantially linear association)
with the level of certain biochemical marker (e.g., interleukin-8)
in umbilical cord serum. This may provide a noninvasive means to
detect inflammation and neuronal injury before the onset of
permanent disability, thereby facilitating clinical
intervention.
[0028] Other features and advantages are apparent from the
following description, and from the claims.
DESCRIPTION OF DRAWINGS
[0029] FIG. 1 is a block diagram of one embodiment of a fetal
monitoring system.
[0030] FIG. 2 is a block diagram of one embodiment of the ECG
analyzer of FIG. 1.
[0031] FIGS. 3A-3C illustrate fetal position changes during
pregnancy.
[0032] FIG. 4 shows an example of data display of the fetal
monitoring system of FIG. 1.
[0033] FIG. 5 shows one example of electrode configuration.
[0034] FIG. 6 shows ECG waveforms collected using the electrode
configuration of FIG. 5.
[0035] FIG. 7 shows another example of electrode configuration.
[0036] FIG. 8A shows a waveform of fetal-maternal mixture.
[0037] FIG. 8B shows a waveform of fetal ECG extracted from the
fetal-maternal mixture of FIG. 8A.
[0038] FIGS. 9A-9C show three exemplary classes of ECG waveforms,
respectively.
[0039] FIG. 9D shows the occurrence of different classes of ECG
waveforms in one patient with respect to time.
[0040] FIG. 10A illustrates the distribution of heart rate
variability among fever and normal populations.
[0041] FIG. 10B illustrates the distribution of ECG entropy among
fever and normal populations.
[0042] FIG. 11C illustrates a correlation between ECG entropy and
IL-8 level.
DETAILED DESCRIPTION
Overview
[0043] Referring to FIG. 1, in some embodiments, a fetal monitoring
system 100 is configured to identify characteristics of fetal ECG
(fECG) signals collected from a patient 110 and based on these
characteristics to detect events of clinical significance,
including, for example, predicting impending fetal injury caused by
inflammatory, hypoxic, or ischemic insults.
[0044] Very generally, the fetal monitoring system 100 includes an
ECG monitor 120 that obtains and analyzes fetal ECG signals to
generate data of clinical relevance. In some embodiments, the ECG
monitor 120 makes use of morphological information in the fECG
signal in addition to or instead of solely determining heart rate
information. Data generated by the ECG monitor 120 can be presented
to physicians in a variety of forms, for example, as printed on
paper charts, shown on a display unit 160 (e.g., a computer
screen), and transmitted via wireless signals to a handheld device
170 (e.g., a smart phone or PDA).
[0045] In this example, the ECG monitor 120 includes a data
acquisition system 130, a channel selection module 140 (optional),
and an ECG analyzer 150.
[0046] The data acquisition system 130 collects electrical signals,
for example, electric potentials in the form of fetal-maternal
mixtures, through a set of electrodes 132. These electrodes 132
include a set of electrodes distributed over the maternal abdomen,
lower back, and/or sides, from which one or more leads are formed
to generate electrical signals.
[0047] In this description, a lead is generally defined in
association with a combination (e.g., a pair) of electrodes, which
can be associated with an imaginary line in the body along which
electrical signals are measured. A lead records the electrical
signals produced by the heart (e.g., in the form of a voltage
differential) from the corresponding combination of electrodes
placed at specific points on the patient's body. Two different
leads may use one or more common electrodes and therefore the
number of leads in an ECG system is not necessarily in direct
proportion to the number of electrodes placed on the patient's
body. In some examples, the electrodes 132 are placed relatively
far away from the maternal heart to reduce the influence of
maternal signal in the fetal-maternal mixtures. In some other
examples, the electrodes 132 may also include one or more
electrodes placed on the maternal chest near the heart from which a
maternal reference lead can be determined. The arrangement of the
electrodes on the patient's body and the definition of lead pattern
are selected depending on the particular implementation, as is
discussed later is this document.
[0048] The signals collected by the data acquisition system 130 are
transmitted to an ECG analyzer 150 that first digitizes raw ECG
signals (e.g., at a sampling rate of 1,000 Hz and a resolution of
16 bits) for subsequent processing and analysis. In some examples,
the raw signals are transmitted over multiple independent channels,
for example, each channel for a different lead. In this example, a
channel selection module 140 applies a channel selection algorithm
that can discard certain channels of "weak" (low quality) signals
to allow only "strong" (high quality) signals to be passed to the
ECG analyzer 150. Some of the discarded channels contain primarily
noise, for example, due to fetal position change or poor electrode
conductivity (e.g., caused by the non-conductive gel used in an
earlier ultrasound procedure). These channels are preferably
rejected as the noise characteristics may not be amendable to the
type of filtering technique designed for the system. Further
discussion of the channel selection algorithm is provided in a
later section.
[0049] Referring to FIG. 2, to obtain data of clinical significance
from raw ECG signals, some embodiments of an ECG analyzer 250
include a pre-processor 251 that applies one or more filtering
techniques (as will be discussed later) to generate processed ECG
signals, for example, in the form of "clean" fetal ECG waveforms or
metrics (i.e., parameters) of fetal-maternal ECG models. These
processed signals are used by one or more analyzing modules, as
described below.
1.1 Clinical Condition Detector
[0050] One example of a type of an analyzing module is a clinical
condition detector 252. Very generally, the clinical condition
detector 252 includes a feature extractor 253 for extracting
characteristics of the ECG signals, such as heart rate variability,
ECG morphology, and morphology classification and entropy, to
assist clinical evaluation. These characteristics are then provided
to a clinical condition evaluator 254, which identifies specific
ECG patterns that are correlated with events of clinical
significance. For example, the clinical condition evaluator 254 may
use a clinical model 255 to correlate electrophysiological
behaviors (e.g., ECG patterns) of the fetus and/or the mother with
statistical behaviors in large populations associated
well-established medical conditions, such as chorioamnionitis,
histopathologic chorioamnionitis, and clinical neonatal infection.
The resulting correlation is used to determine the susceptibility
of the patient (mother and/or fetus) to such conditions. Depending
on the particular implementation, the clinical condition evaluator
254 may have separate modules (e.g., a chorioamnionitis evaluator,
an intrapartum fever evaluator), with each module providing a
measure of a degree of the presence of a particular aspect of fetal
and/or maternal distress. Physicians may receive the outputs of the
individual modules in confidence scores, for example, presented on
a scale of 0 to 10 with "0" indicating no (or least) distress and
"10" indicating the highest level of distress. The individual
scores can also be combined to form an evaluation of overall fetal
distress level indicating the general health condition of the
fetus.
[0051] In some embodiments, the clinical condition evaluator 254
performs an automated diagnosis to identify medical conditions
(e.g., using expert systems and/or human intervention) and/or to
provide recommendation for follow-up procedures. In some examples,
other clinical data (such as pathologic evaluations of serum
samples from the umbilical cord) are collected from the patient in
pregnancy or during labor and are used by the clinical condition
evaluator 254 in conjunction with the identified ECG
characteristics to help further determine the likelihood of
impending fetal/neonatal injuries (such as brain injuries, cerebral
palsy, and death).
[0052] Using the feature extractor 253, high quality fetal ECG data
can be obtained from the patient under a variety of clinical
conditions (e.g., pregnant or in-labor). The characteristics of the
ECG data can be well preserved to enable clinical analysis that is
otherwise unavailable using conventional techniques.
Implementations of the feature extractor 253 and examples of
clinical condition evaluator 254 are described in greater detail at
a later section.
1.2 Fetal Orientation Detector
[0053] A second example of an analyzing module is a fetal
orientation detector 256 that provides an estimate of fetal
position within the mother.
[0054] Referring to FIGS. 3A-3C, fetal position may change during
various stages of pregnancy and the pre-labor position can affect
the way by which the mother will deliver and whether certain
cautionary steps need to be taken. In some applications, it is
desirable to generate an estimate of fetal position as an output of
the monitoring system, for example, providing a clinician with a
continuous output.
[0055] In some examples, such a position estimate is determined as
part of a multiple dipole modeling approach for extracting the fECG
signal from the raw signals that include both fetal and maternal
signals, in which estimated orientation of the dipole of the fetal
heart provides an estimate of the orientation of the fetus relative
to the mother's body.
[0056] In some examples, the fetal position is used as part of the
feature extraction procedure, or as part of the clinical evaluation
procedure. For example, signal acquisition in certain fetal
positions may result in characteristically distinct signals, for
example, that exhibit higher signal-to-noise characteristics. In
some examples, automated clinical determinations are made as a
function of the fetal position, for example, being performed only
in certain fetal positions. An example of such a fetal position is
a fetus with its back to the maternal abdominal wall, which may
result in particularly high quality signals due to the short
distance between the fetal heart and the surface electrodes. In
some examples, the estimated fetal position is used to select
electrodes in the channel selection module 140. In some examples,
the estimated fetal position is used to determine signal and/or
model characteristics related to various electrodes, for example,
to determine signal transmission characteristics between the signal
source (e.g., fetal heart) and the electrodes.
[0057] Other examples of analyzing modules implemented in the ECG
analyzer 250 include a heart rate tracker 258, a fetal ECG waveform
extractor (not shown), and possibly other modules that associate
user-determined statistics with clinical analysis. The heart rate
tracker 258 may provide a continuous output of fetal heart beat
over time and automatically identify the occurrence of heart rate
acceleration, deceleration, and certain types of irregularity that
can be early manifestation of serious medical conditions such as
cardiac arrhythmia.
[0058] Note that the pre-processor 250 may provide signals to
various analyzing modules in different forms. In other words, the
input data to the clinical condition detector 252 is not
necessarily the same data provided to the orientation detector 256
or the heart rate tracker 258. Depending on the particular
implementation, some analyzing modules may accept data representing
"clean" fetal ECG waveforms, whereas others may accept data
representing metrics of predefined fetal-maternal ECG models.
[0059] FIG. 4 shows one example of a data display by which the
outputs of various analyzing modules are presented to physicians,
for example, on a computer screen or a handheld device. This
display includes multiple regions that respectively show, for
example, a fetal ECG waveform along with observed fetal heart rate,
a fetal orientation pointer, an overall fetal distress index, an
entropy index, and possibly other indices. In some examples,
changes in fetal position since the most recent examination (or
over the entire course of pregnancy) are also presented, for
example, by loading prior position data from a patient database. In
some examples, each index has a predefined "alert" level (e.g., a
score of 6 out of 10) beyond which special attention (e.g.,
follow-up procedures) is indicated. In some examples, the
monitoring system 100 also allows physicians to view detailed data,
for example, the statistics upon which a particular index value is
determined, when there is a need.
Electrode Configuration
[0060] Depending on the particular implementation, ECG signals can
be collected using invasive and/or non-invasive approaches with the
electrodes 132 placed in a variety of arrangements. The following
description provides two examples of electrode configurations
suitable for use with the monitoring system 100 of FIG. 1.
2.1 Example I
[0061] Referring to FIG. 5, a first electrode configuration of some
embodiments of the data acquisition system 130 is shown. In this
example, the configuration is capable of simultaneously collecting
fetal scalp electrode ECG data ("gold-standard" fetal data),
maternal ECG data ("gold-standard" maternal data), and combined
data (fetal-maternal mixture) from the maternal abdomen. Fetal ECG
data can be isolated from the combined data using the gold standard
maternal data and can be further compared with the gold standard
fetal data.
[0062] In this example, ECG signals are obtained using 32 adhesive
electrodes, including: 3 maternal chest electrodes (producing a
robust maternal gold standard reference), 28 abdominal and back
electrodes (producing an over-complete set of maternal/fetal
mixtures), and a fetal single scalp electrode inserted using an
intra-uterine probe. The single intra-uterine probe, although not
employed without indication, can be optionally used on a
significant number of patients (e.g., in-labor patients). This
probe can provide a strong, low-noise, fetal ECG signal, and hence
a "gold standard" with which to compare the extracted fetal ECG
from the abdominal probes. The three chest electrodes provide a
strong maternal ECG representation with no (or negligible) fetal
contamination. Using the chest and scalp electrodes, the quality of
both the maternal removal and the fetal extraction can be
evaluated. Depending on implementation, these electrodes can either
be dry electrodes (e.g., Orbital Research, Cleveland, Ohio) or
commercial gel adhesive electrodes (e.g., Red Dot, 3M, St. Paul,
Minn.). In some examples, the electrodes are mounted onto the
maternal body using a mesh (or garment), which can stabilize
electrodes and improve electrode-skin contact during
examination.
[0063] FIG. 6 shows exemplary ECG waveforms detected using the
above described data acquisition system. These waveforms include
fetal ECG, fetal-maternal ECG, and maternal ECG obtained
respectively from fetal scalp electrodes, abdominal electrodes, and
chest electrodes.
2.2 Example II
[0064] Referring to FIG. 7, a second electrode configuration of
some embodiments of the data acquisition system 130 is shown. Here,
a set of dry electrodes (e.g., 32) are mounted on a convenient
elastic monitoring garment that is strapped around the maternal
abdomen to allow the electrodes to be distributed in a
predetermined arrangement over the abdomen, the back, and on the
sides of the patient. No fetal scalp electrode is necessary with
this configuration. This configuration provides a non-invasive
means to monitor fECG signals yet still capable of providing a
sufficient set of useful fECG signals regardless of the fetal
status.
[0065] In some embodiments, the electrode arrangement and the lead
pattern by which electrical signals are collected can use
conventional standards developed on adult patients. One example of
such a conventional standard makes use of a well-established
12-lead pattern, with each lead recording the electrical activity
of the adult heart from a different perspective. The signal of each
lead can correlate with a different anatomical area of the heart,
for example, to help identify acute coronary ischemia or injury.
Fetal ECG signals are contained in some or all of the lead signals
and may be extracted using various data extraction and filtering
methods (as will be described later). In some cases, the isolation
of fetal signals from fetal-maternal mixtures can be difficult as
the conventional standards were developed based on adult models
without accounting for the influence of fetal presence and the
resulting fetal-maternal mixtures can be either poorly
characterized or contain very low fetal components relative to the
predominant maternal signals.
[0066] In some other embodiments, the electrode arrangement and the
lead pattern use a design that suits the particular need of fetal
ECG monitoring. One example of the design is shown in FIG. 7, which
illustrates the placement of some electrodes in a side view, a back
view, and a sectional view of the patient body. In this example,
the entire set of electrodes forms at least of a group of
cross-body leads each of which generates electrical signals along
an imaginary line across the body, for example, from back to front,
or from left side to right side. Some of these leads are each
formed by a respective pair of electrodes, one being referred to as
a collecting/positive electrode (e.g., E1) and the other being
referred to as a reference/negative electrode (e.g., R1). The
corresponding lead signal (e.g., L1) is obtained, for example,
using a biomedical instrumentation amplifier that forms an
amplified signal representing a voltage differential between the
collecting electrode and the reference electrode. For some of these
leads, the reference electrode is placed at the opposite side of
the body to which the collecting electrode is attached. For
example, some of the collecting electrodes are placed in the
abdominal region while the corresponding reference electrode(s) are
placed in the lumbar region. Similarly, some of the collecting
electrodes can be placed in the left side of the body while the
corresponding reference electrode(s) are placed in the right side
of the body.
[0067] Using such a lead pattern, some of the collected signals can
exhibit a stronger fetal component and/or contain less noise
compared with lead signals collected using conventional adult
standards. Depending on the particular implementation, each lead
does not necessarily use a different reference electrode. In other
words, some leads may be formed using collecting electrodes at
various positions in the abdominal region against a single
reference electrode in the lumbar region. In some examples, the
reference electrodes and the collecting electrodes can be
electrodes of different characteristics (for example, made from
different materials, having different sizes, and/or exhibiting
different levels of signal sensitivity) and be attached to the body
using different attachment mechanisms (e.g., dry vs. wet). In some
examples, the set of electrodes is coupled to a lead
reconfiguration module that can dynamically adjust electrode
paring, lead selection, and/or garment positioning based on
feedback signals provided by the ECG analyzer 150 to account for,
for example, fetal position changes, loss of electrode contact, and
other events that may cause abrupt changes in certain electrode or
lead signals.
Channel Selection
[0068] In the exemplary electrode configurations shown in FIGS. 5
and 7, one reason to record a large number of abdominal and back
signals described above is that the fetal ECG tends to manifest in
only a subset of these leads, yet the actual combination is
dependent on the state of the fetus, the time through pregnancy,
the degree of electrical contact, and the location and orientation
of the fetus or fetuses. Therefore, the channel selection module
140 is configured to adaptively select channels of "strong" (high
quality) signals and discards channels of "weak" signals. As some
of the abdominal signals will contain primarily noise, preferably,
these channels are discarded from processing.
[0069] One technique used by the channel selection unit 140 to
select channels of useful signals is based on fusing multiple
signal quality indices (SQI) derived from multiple ECG leads. In
some examples, physiological SQIs are obtained by analyzing the
statistical characteristics of each channel and their relationships
to each other. For instance, by computing spectral coherence,
statistical departures from Gaussianity, and the performance of
differently-sensitive event detectors, this technique allows the
automatic location of channels that contain useful signal, and
discarding of those that contain primarily noise. Furthermore, a
sliding scale of quality is available to enable the selection of
different channels for different applications. Further discussion
of this technique is provided by Li et al., in "Robust Heart Rate
Estimation from Multiple Asynchronous Noisy Sources Using Signal
Quality Indices and a Kalman Filter," published in Physiological
Measurement 29 (2008) 15-32, the disclosure of which is
incorporated herein by reference.
4 Extraction of Fetal Signals from Fetal-Maternal Mixtures
[0070] Some techniques to extract waveforms of fetal ECG signals
from the fetal-maternal mixtures include signal processing and
filtering techniques such as adaptive filtering (AF), nonlinear
projective filtering (NLPF), neural networks, independent component
analysis (ICA) and joint time-frequency analysis (JTFA). One
limitation of these techniques lies in their dependencies on the
signal-to-noise ratio (SNR) of the data and sensitivity to the
frequent artifacts that manifest during FECG acquisition. Each
technique may either perform an "in-band" filtering (removing
frequency signals that are present in the fetal signal) or produce
a phase distortion in the signal that has an unknown affect on the
fECG morphology. These issues may result in significant changes in
the clinical parameters one wishes to extract from the fECG.
[0071] Another issue in fetal ECG recording and analysis deals with
signal distortions that result from the transmission of the fetal
signal trough the mother's abdomen. To reach the surface
electrodes, fECG signals pass through multiple layers of media
(e.g., the vernix caseosa) each of which may have very different
electric properties and some may cause significant attenuation the
fetal ECG signals collected from surface electrodes. Since the
effective frequency range of the ECG is below 1-2 KHz and
considering the distance between the body surface electrodes and
the cardiac sources, the propagation medium of the maternal body
may be considered as a linear instantaneous medium. The body
surface recordings are hence a linear instantaneous projection of
the cardiac sources and artifacts onto the axes of the recording
electrode pairs. It is however known that the electrical impedance
of the body volume conductor changes with respiration. Therefore
despite its linearity, the propagation medium is time-varying and
the body surface recordings are rather non-stationary.
[0072] One method to address the issue of fetal ECG distortion due
to transmission through media of varying dielectric constants is to
use a model of the fetal cardiac source to constrain the filtering
and feature extraction process. One technique, for example, applies
a three-dimensional dynamic model to represent the electrical
activity of the heart. More specifically, this model is based on a
single dipole model of the heart and is later related to the body
surface potentials through a linear model which accounts for the
temporal movements and rotations of the cardiac dipole, together
with a realistic ECG noise model. Details of this technique are
further described by Sameni et al., in "Multichannel ECG and Noise
Modeling: Application to Maternal and Fetal ECG Signals," published
in EURASIP Journal on Advances in Signal Processing, Volume 2007,
Article ID 43407, the disclosure of which is incorporated herein by
reference.
[0073] FIG. 8A illustrates a typical mixture of maternal and fetal
ECG. The maternal beats appear as negative spikes (HR=90 bpm), and
the fetal beats appear as the smaller, positive spikes (HR=138
bpm). Both the fetal and maternal peak heights appear to be
modulated by some low-frequency component (including, e.g.,
respiration). A fetus will "practice" respiration prior to birth,
and this can lead to changes in intra-thoracic pressure.
[0074] FIG. 8B illustrates the same signal after maternal
subtraction using a model-based Kalman Filter tracking method
described above. Note that the respiratory-modulation of the
R-peaks and other features of the fECG are preserved in the
waveform. These subtle features are essential in performing
accurate feature analysis, such as R-peak location (e.g., for heart
rate variability evaluation of sepsis), ST-elevation analysis
(e.g., for ischemia) and QT interval analysis (for pro-arrhythmic
indications).
[0075] Using these "clean" fetal ECG waveforms, the feature
extractor 253 of FIG. 2 is able to identify characteristics of the
waveforms that are associated with clinically relevant activities.
Examples of ECG characteristics include heart rate variability, ECG
morphology, and entropy. For instance, fECG signals may be grouped
into different morphological classes, and each class may be further
divided based on subtle morphological characteristics, based on
which patterns of clinical relevance may be identified. Techniques
of feature extraction are described in greater detail below in the
following sections.
[0076] In some examples, the feature extractor 253 does not need
the "clean" fetal ECG waveforms in order to obtain features of
interest. For instance, the pre-processor 251 may process the raw
ECG data to obtain metrics of ECG models or symbolization of ECG
classification, based on which the feature extractor 150 may
extract interesting features.
5. Feature Extraction and Clinical Analysis
5.1 Heart Rate Variability Analysis
[0077] Heart rate variability (HRV) can be an important
quantitative marker of cardiovascular regulation by the autonomic
nervous system. Heart rate is generated by the intrinsic rhythm of
the sinoatrial node in the heart, but constant input from the
brainstem through a feedback loop in the autonomic nervous system
closely modulates this rate. At rest, variation in heart rate
arises predominantly from vagal tone governed by the vagus nerve
nuclei. However, this variation is affected by the interaction
between vagal and sympathetic activity, as well as by central
respiratory and motor centers and peripheral oscillations in blood
pressure and respiration.
[0078] In many clinical settings, evaluation of HRV is based on the
subjective interpretation of this variable by clinicians using
paper printouts that plot the fetal heart rate as a function of
time. In some embodiments, heart beat may be detected by
cross-correlating the cardiac signal with a reference heart beat
trace from data recorded using the fetal ECG. The height of the
cross-correlation peak (if it is not normalized) provides a measure
of the strength of the signal and its similarity to the reference.
The position of the peak provided an accurate measure of the exact
time the beat occurred. These measures provided a way to reject
signal that is not a fetal beat as well as to measure accurately
the time between beats (the fetal heart rate). This approach
provides data that can be used for analyses based on rate and
HRV.
[0079] The cross-correlation can be used to locate fetal heart
beats in the data, which can then be "windowed" out into a series
of individual heart beats. The data is then subjected to a
multivariate statistical analysis, and the results are used to
group beats according to variations in the ensemble of heart beats.
These data can be later used for the analysis of waveform
morphology.
5.2 Morphological Analysis
[0080] In some embodiments, the feature extractor 253 performs
morphological analysis on the fECG signal. One approach to
analyzing fetal ECG morphology uses clustering and symbolic
analysis of ECG signals to discover medically relevant patterns.
Very generally, ECG signals are classified into groupings that are
morphologically similar according to a signal waveform similarity
measure. In some examples, successive segments of the fECG waveform
are formed with one segment per beat, and min-max clustering is
then used to form the groupings according to pair-wise distance
between the waveform segments. In some embodiments, the pair-wise
distance between segments uses a dynamic time-warping (DTW)
measure. In other examples, each segment is modeled using a
parametric model (e.g., using a sum of displaced Gaussian
components) and the distance between segments is based on a
distance between the model parameters of the segments. The
characteristics of the identified groups are used to determine a
measure of morphological variation. In some examples, the segments
of the fECG are labeled, for example, with discrete labels from an
alphabet of symbols (e.g., 5 arbitrary labels). Then a statistical
measure is determined from the sequence of labels, for example, in
a sliding window of the signal.
[0081] One measure of morphological variation is an entropy of a
sample distribution of the labels. In some examples, the entropy of
a finite state model of the sequence is used. In some examples, the
segments are not necessarily deterministically labeled (relying on
a probability measure for beats in each hidden class), and the
entropy of a underlying (e.g., hidden) sequence of segment classes
is computed, thereby avoiding a need to first determine an accurate
series of class labels, which may require a "clean" estimate of the
fECG signal. Some aspects of these approaches are described by Syed
et al., in "Clustering and Symbolic Analysis of Cardiovascular
Signals: Discovery and Visualization of Medically Relevant Patterns
in Long-Term Data Using Limited Prior Knowledge," published in
EURASIP Journal on Advances in Signal Processing, Volume 2007,
Article ID 67938, the disclosure of which is incorporated herein by
reference.
[0082] Unlike the techniques incorporated into ECG monitors and ICU
monitoring devices that compare observed phenomena to standardized
patterns representing pathophysiological conditions (ventricular
tachycardia or ST-depression, for example), some entropy-based
approaches of the types described above do not necessarily assume a
priori information about the ECG morphology. Each morphological
class is represented by a symbol, and various patterns of symbols
in sequence may have clinical significance. This analytic approach
is suited for the fetal ECG data collected in the present system
100, because with the exception of ST-segment analysis, there are
no formal systems for fetal ECG evaluation. Independence from a
priori information can be useful in fetal applications where the
information may not be available, or may be highly variable based
on factors such as fetal age.
[0083] In some examples, model-based filtering is applied to the
fECG signal, for example, prior to entropy-based analysis. For
example, Gaussian based modeling as described in Clifford et al.,
"Model-based filtering, compression and classification of ECG,"
International Journal of Bioelectromagnetism Vol. 7, No. 1, pp
158-161, 2005, and in U.S. Patent Publication 2007/0260151,"Method
and Device for Filtering, Segmenting, Compressing and Classifying
Oscillatory Signals," published Nov. 8, 2007, are used in
processing the fECG signals. These references are incorporated
herein by reference. In some examples, the classification based on
these techniques is used in determining entropy measures as
described in the Syed reference. For example, each class may be
characterized by a range of model parameters for that class (e.g.,
by partitioning the space of parameters values) or each class be
associated with a distribution of the model parameters for that
class.
Examples of Clinical Applications
[0084] In some embodiments, characteristics of ECG patterns are
associated with events of clinical activity. Some examples of such
clinical applications includes using an entropy measure of a fECG
signal as an indicator of an inflammation condition, or as an
indicator of a cause of an inflammation condition, for example, an
infection-based cause of inflammation.
[0085] In an experimental application of signal processing and
analysis techniques described above, the fECG waveforms of 30
recordings discovered a change in the morphology of the heart beat
that occurs prior to the development of chorioamnionitis.
[0086] FIGS. 9A-9C illustrate three classes of QRS complexes
classified from a 7-hour dataset collected from a woman who
developed chorioamnionitis during labor. FIG. 9D shows the
occurrence of each beat during 10-minute intervals timed with
respect to the onset of maternal fever of the same patient. Note
the consistent appearance of class 1 ECG signals one hour prior to
the development of fever.
[0087] Analyses of the fetal ECG waveforms also show that a measure
of entropy--the degree of disorder in the similarity of the
morphology of sequences the fetal heart beats--distinguishes those
fetuses subject to intra-amniotic infection from those without
exposure to infection.
[0088] FIGS. 10A and 10B illustrate respectively the HRV analysis
and entropy analysis of 30 fetal ECG datasets from women with
chorioamnionitis and women without infection. As shown in FIG. 10A,
the distribution of fetal HRV for fetuses subjected to
chorioamnionitis (e.g., exhibiting maternal fever symptom) is not
easily distinguishable from that of fetuses in an uninfected
intrauterine environment. In comparison, FIG. 10B shows that, when
the entropy of the fetal ECG signal is calculated for the same set
of fetal ECG data, fetuses subjected to chorioamnionitis are
bimodally distributed with respect to entropy, whereas fetuses in
an uninfected environment are essentially normally distributed. In
other words, an ECG waveform having a very low (e.g., 0) or very
high (e.g., 4) entropy indicates a higher probability of developing
chorioamnionitis. In some examples, the distributions of observed
entropy measures in two known classes of patients (e.g., condition
present versus normal) are used to form a likelihood ratio test to
classify a patient based on an observed entropy.
[0089] In some examples, different patterns of electrophysiological
behaviors can be correlated with medical conditions using specific
biochemical markers of such conditions, e.g., markers of
inflammation and brain injury measured from fetal umbilical cord
collected from the patient. Umbilical cord blood interleukin-6, for
example, is significantly elevated in fetuses that develop sepsis
compared with fetuses that do not develop sepsis. Cord blood levels
of IL-6 greater than 108.5 pg/ml are considered 95% sensitive and
100% specific for neonatal sepsis.
[0090] FIG. 11 shows an association between the morphologic entropy
of the fetal ECG and fetal umbilical cord serum interleukin-8
(IL-8) levels. Increasing levels of IL-8 are correlated (e.g.,
having a substantially linear relationship) with increasing
disorder in the fetal ECG morphology. One possible explanation of
this correlation is that an in-utero fetal inflammation/infection
is associated with quantitative changes in the fetal ECG,
reflecting altered electrophysiological signaling at the level of
the fetal brainstem, fetal myocardium, or both.
[0091] Another related application relates to using characteristics
of ECG signals to discriminate between different possible causes of
medical conditions. Various causes of diseases may induce changes
in ECG morphology through different mechanisms, which may in turn
lead to distinguishable patterns in ECG morphologies. For example,
infection, which is one explanation for inflammation, may induce a
morphological change in fetal ECG signals through brain stem and
myocardium level; while preeclampsia (pregnancy-induced
hypertension) is likely to affect the ECG morphologies through
mechanism of placental failure. The various presentations of ECG
morphologies can therefore be used as a basis for discriminating
between different causes of certain diseases.
[0092] In some embodiments, the feature extractor 253 performs
signal analysis that is not necessarily related to ECG signals. For
example, muscle signals are detected using the surface electrodes
or conventional pressure sensors for contractions, and timing and
intensity of uterine contractions are estimated. This approach has
an advantage of providing a single monitoring device being applied
to the mother, while providing multiple clinically-relevant
signals.
[0093] In some embodiments, the fetal monitoring system 100 may
incorporate functions of other medical diagnostic tools to enhance
fetal ECG detection and/or assist clinical evaluations. For
example, a maternal reference signal can be obtained using other
sensing modes, such as ultrasound, imaging, and blood pressure
sensing, to facilitate fetal ECG extraction. Also, histological and
pathological data of a patient can be assessed in conjunction with
ECG data to detect inflammation and neuronal injury before the
onset of permanent disability.
[0094] It is to be understood that the foregoing description is
intended to illustrate and not to limit the scope of the invention,
which is defined by the scope of the appended claims. Other
embodiments are within the scope of the following claims.
* * * * *