Note: Descriptions are shown in the official language in which they were submitted.
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
CARDIOVASCULAR SIGNAL ACQUISITION, FUSION, AND NOISE MITIGATION
CROSS REFERENCE TO RELATED APPLICATIONS
[0001] This application is a continuation-in-part of co-pending U.S.
Application No.
15/743,154, filed on January 9, 2018, which is a National State Entry of
International
Application No. PCT/CA2015/051120, filed on November 2, 2015, which claims the
benefit of
priority to U.S. Provisional Application No. 62/191,318, filed on July 10,
2015, all of which are
incorporated by reference herein in their entirety. This application is also
related to U.S. Patent
Application No. 16/163,349 (Atty. Docket No. 35193-41271/US), filed on an even
date herewith,
and titled "CARDIOVASCULAR SIGNAL ACQUISITION, FUSION, AND NOISE
MITIGATION," and is also related to U.S. Patent Application No. 16/163,354
(Atty. Docket No.
35193-41272/US), filed on an even date herewith, and titled "CARDIOVASCULAR
HEALTH
MONITORING DEVICE," the contents of both are hereby incorporated by reference.
BACKGROUND
[0002] This disclosure relates generally to user cardiovascular disease
monitoring, and
more specifically to acquiring biometric signals relevant to cardiovascular
health, fusing signals,
and mitigating noise in the signal(s).
[0003] About 1 of 3 U.S. adults (over 70 million people) have high blood
pressure, but
only approximately half of these individuals their high blood pressure under
control. High blood
pressure is often called a "silent killer" because it typically produces no
warning signs or
symptoms, but is associated with increased risk factors for more serious
conditions, such as heart
disease and stroke. Frequent monitoring of blood pressure and other biometric
parameters
relevant to cardiovascular health can enable early detection of abnormal or
deteriorating
1
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
cardiovascular health states; however, currently available home-use devices
(e.g., pneumatic
cuffs) are not user-friendly, are uncomfortable, are difficult to use, and are
not designed to
promote regular use, in relation to adherence to a health-monitoring regimen.
Even further,
devices for consumer use are limited in the types of signals they can acquire
and effectively
process to generate composite features relevant to different cardiovascular
health states.
SUMMARY
[0004] A device including an array of electrodes generates one or more
electrical signals
from a user, extracts one or more noise signals, and generates one or more de-
noised electrical
signals upon processing the electrical signal(s) with the noise signal(s). The
array of electrodes is
coupled to a surface of the device, where the device also includes force
sensors in mechanical
communication with the surface for detecting user weight and other forces. The
device can be
configured to generate electrical signals from different subportions of the
array of electrodes and
to extract noise signals from different subportions of the array of
electrodes, where the
subportion(s) for electrical signal generation may or may not overlap with the
subportion(s) of
electrodes for noise signal extraction.
[0005] Collectively, the electrical signal(s) and the force-associated
signal(s) generated by
sensors of the device are processed by a computing subsystem with electronics
and architecture
configured for sensor fusion and extraction of composite features indicative
of cardiovascular
health states. In one or more embodiments, the device generates
electrocardiogram (ECG)
signals, impedance plethysmogram (IPG) signals, ballistocardiogram (BCG)
signals, and weight
measurements through an interface with feet of a user. Computing subsystem
components fuse
the ECG, IPG, and BCG data to efficiently generate analyses of cardiovascular
health of the user,
in relation to various parameters related to temporal components of cardiac
phases, force and
volume-associated parameters, and other relevant parameters. The parameters
are regularly
collected and analyzed to monitor user cardiovascular health and trigger
preventative health
interventions.
2
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
BRIEF DESCRIPTION OF THE DRAWINGS
[0006] FIG. 1A depicts a schematic of a system for cardiovascular signal
acquisition,
fusion, and noise mitigation, in accordance with one or more embodiments.
[0007] FIG. 1B depicts a plan view of components of the system shown in
FIG. 1A.
[0008] FIG. 2 depicts a plan view of components and vectors associated with
electrical
signal acquisition and noise acquisition, in accordance with one or more
embodiments;
[0009] FIG. 3A depicts system component configuration of a first embodiment
of the
system shown in FIG. 2.
[0010] FIG. 3B depicts a system component configuration of a second
embodiment of the
system shown in FIG. 2.
[0011] FIG. 3C depicts a system component configuration of a third
embodiment of the
system shown in FIG. 2.
[0012] FIG. 3D depicts a system component configuration of a fourth
embodiment of the
system shown in FIG. 2.
[0013] FIG. 4 depicts a plan view of a system component configuration, in
accordance with
one or more embodiments.
[0014] FIG. 5A depicts a flowchart of a method for cardiovascular signal
acquisition and
noise mitigation, in accordance with one or more embodiments.
[0015] FIG. 5B depicts a flowchart of a first embodiment of the method
shown in FIG. 5A.
[0016] FIG. 5C depicts a flowchart of a second embodiment of the method
shown in FIG.
5A.
[0017] FIG. 5D depicts a flowchart of a third embodiment of the method
shown in FIG.
5A.
[0018] FIG. 5E depicts a schematic flow of the embodiment of the method
shown in FIG.
5D.
[0019] FIG. 5F depicts a flowchart of a variation of the method for
cardiovascular signal
acquisition and noise mitigation shown in FIG. 5A.
3
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
[0020] FIG. 6 depicts a flowchart of a method for electrical and mechanical
cardiovascular
signal acquisition processing, in accordance with one or more embodiments.
[0021] FIG. 7A depicts a flow diagram of cardiovascular health parameter
extraction
processes, in accordance with one or more embodiments.
[0022] FIG. 7B depicts a first portion of the flow diagram shown in FIG.
7A.
[0023] FIG. 7C depicts an expanded portion of the flow diagram shown in
FIG. 7B.
[0024] FIG. 7D depicts a second portion of the flow diagram shown in FIG.
7A.
[0025] FIG. 7E depicts a third portion of the flow diagram shown in FIG.
7A.
[0026] FIG. 7F depicts a fourth portion of the flow diagram shown in FIG.
7A.
[0027] FIG. 7G depicts a fifth portion of the flow diagram shown in FIG.
7A.
[0028] FIG. 7H depicts a sixth portion of the flow diagram shown in FIG.
7A.
[0029] FIG. 8 depicts a flowchart of a method for processing cardiovascular
health
parameters with a risk model, in accordance with one or more embodiments.
[0030] FIG. 9 depicts a flowchart of longitudinal monitoring of
cardiovascular health of a
user, in accordance with one or more embodiments.
[0031] The figures depict various embodiments for purposes of illustration
only. One
skilled in the art will readily recognize from the following discussion that
alternative
embodiments of the structures and methods illustrated herein may be employed
without
departing from the principles described herein.
DETAILED DESCRIPTION
1. System for cardiovascular signal acquisition, fusion, and noise
mitigation
[0032] FIG. 1A depicts a schematic of a system 100 for cardiovascular
signal acquisition,
fusion, and noise mitigation, in accordance with one or more embodiments. FIG.
1B depicts a
plan view of components of the system 100 shown in FIG. 1A. The system
includes a substrate
110, an array of electrodes 120 coupled to the surface and including a left
subportion 122 and a
right subportion 128, one or more force sensors 130 in mechanical
communication with the
4
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
substrate 110, and electronics subsystem 140 including channels 144 and 146
for generation of
electrical signals from the array of electrodes 120, and for relaying and/or
pre-processing signals
from the force sensor(s) 130. The electronics subsystem 140 also includes
components of a
computing subsystem 150 and transmission hardware 149 for data communication
with other
components of the computing subsystem 150, where the computing subsystem 150
includes
architecture for generating de-noised signals and for fusion of electrical and
mechanical signal
data to extract features relevant to analyzing cardiovascular health. The
system 100 thus provides
structures, subsystem interfaces, and operation modes for signal acquisition
and processing,
including operations associated with methods described in more detail in
Section 2 below.
[0033] The system 100 functions to simultaneously acquire electrical and
mechanical
signals associated with cardiovascular health, and implement signal processing
methods to
mitigate noise induced by changes in position of the user during signal
acquisition, ambient
sources, and other sources. The system 100 also includes architecture for
receiving different
types of electrical and mechanical signals through interfaces with the feet of
a user, comparing
signals across different vectors defined by device sensor positions, and
extracting health-relevant
signal components and noise components based upon the comparison(s). In
particular, the system
100 is configured for routine assessment of hemodynamic parameters, including
systolic time
intervals, other temporal parameters (e.g., diastolic time intervals), and
other parameters, with
design considerations that promote regular use of the system.
1.1 System ¨ Substrate and Electrodes
[0034] As shown in FIGS. 1A and 1B, the system includes a substrate 110
that functions
to facilitate electrical signal transmission toward the array of electrodes
120 coupled to the
substrate 110, and to mechanically support the user's weight in relation to
weight measurements
and other force-associated signal generation functionality of sensor described
in more detail
below. The substrate 110 can additionally function to enable display (e.g.,
with integrated
display elements, with transparent materials, with translucent materials,
etc.) of information to
the user. The information can include information derived from analyses of
signals generated by
the system, instructions to the user, user verification information, or other
types of information.
[0035] In morphology, the substrate 110 includes a broad surface that,
during use, provides
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
an interface to the feet of the user for electrical and mechanical signal
generation. The broad
surface of the substrate 110 is planar, but can alternatively include recessed
and/or protruding
regions defined at the broad surface. Recessed and/or protruding regions of
the broad surface can
be configured to guide placement of the feet of the user and can include
features that are
complimentary to the soles of the user's feet.
[0036] The substrate 110 has a rectangular footprint when the broad surface
is projected
onto a horizontal plane, where the rectangular footprint has rounded edges.
The substrate 110 can
alternatively have any other suitable footprint. In dimensions, the substrate
110 can have a width
from 10-50 centimeters, a length from 10-50 centimeters, and a thickness from
0.2-2 centimeters;
however, the substrate 110 can alternatively have any other suitable
dimensions.
[0037] In material composition, the substrate 110 includes at least one
region that is
composed of glass, where the glass can be processed (e.g., tempered, etc.) to
have desired
properties in terms of mechanical properties, electrical properties, optical
properties, or other
properties described in more detail below. The substrate 110 can additionally
or alternatively be
composed of, or include regions that are composed of one or more of: a
polymeric material (e.g.,
plastic), a metallic material, a ceramic material, and a natural material
(e.g., wood, fiber, etc.).
The substrate 110 can thus be composed of a single material or can be a
composite material to
provide suitable physical properties.
[0038] In relation to mechanical properties, the material(s) of the
substrate 110 can have a
compressive strength, a shear strength, a tensile strength, a strength in
bending, an elastic
modulus, a hardness, a derivative of the above mechanical properties and/or
other properties that
enable structural support of the user and/or other system elements in various
operation modes
associated with use of the system 110.
[0039] In relation to electrical properties, the material(s) of the
substrate 110 can have a
conductivity, resistivity, a derivative of the above electrical properties
and/or other properties
that enable electrical signal transmission from the user's body to electrodes
of the system 100
described in more detail below. One or more surfaces of the substrate 110 can
be processed to
have desired electrical properties. For instance, the broad surface configured
to interface with
feet of the user can be surface treated with a conductive material (e.g.,
indium tin oxide) with a
desired pattern in relation to signal transduction through the system and/or
the body of the user.
6
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
The bulk material(s) of the substrate 110 can alternatively be selected to
have desired electrical
properties. As such, the substrate 110 can be an electrically conductive
substrate. Additionally or
alternatively, one or more portions of the substrate and/or elements coupled
to the substrate can
be capacitively coupled to the electrodes described below, for instance,
through an insulating
layer, where in these embodiments, the electrode(s) include a combination of a
conductive
material covered by an insulating material (and the user's feet are
capacitively coupled to the
conductive material through the insulating layer). As such, the substrate can
include electrically
conductive regions, but portions of the system contacting a user are
insulating. In relation to
optical properties, the material(s) of the substrate 110 can have a
transparency or translucency
suitable of conveying information to the user by way of an electronic display
coupled to,
positioned next to, or otherwise optically integrated with the substrate 110
in another manner.
The material(s) of the substrate can also be fabricated to manipulate (e.g.,
reflect, scatter, guide,
shape, etc.) light.
[0040] As shown in FIGS. 1A and 1B, the system 100 also includes an array
of electrodes
120 coupled to the surface and including a left subportion 122 and a right
subportion 128. One or
more electrodes of the left subportion 122 cooperate with one or more
electrodes of the right
subportion 128 to generate electrical signals from which parameters relevant
to cardiovascular
health can be generated, as described in more detail below. One or more
electrodes of the left
subportion 122 can also provide noise signals that the computing subsystem 150
can use to de-
noise the electrical signals. Similarly, one or more electrodes of the right
subportion 128 can also
provide noise signals that the computing subsystem 150 can use to de-noise the
electrical signals.
Thus, the arrangement of the array of electrodes 120 in space relative to the
substrate 110 can
allow the system to improve signal-to-noise (SNR) ratio with signal processing
methods, where
noise is associated with noise from ambient sources (e.g., 60 Hz mains), noise
from motion of a
user using the system 100, noise from poor or changing foot contact, and/or
any other noise
source. Noise sources and methods for signal de-noising are further described
in Section 2
below.
[0041] As shown in FIG. 1A, when the user interacts with the array of
electrodes 120 by
contacting the substrate 110 with his/her feet, the system 100 forms an
electrical circuit through
the user's body. The electrical circuit shown in FIG. 1A is defined through an
inferior portion of
the user's body, and passes through a left foot region, through a left leg
region, across the sagittal
7
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
plane of the user, through a right leg region, and through a right foot
region.
[0042] The electrodes of the array of electrodes 120 can be composed of a
conductive
material (e.g., conductive polymer, metal, etc.).
[0043] The array of electrodes 120 generate electrocardiogram (ECG) signals
during use.
The array of electrodes 120 can additionally generate impedance
plethysmography (IPG) signals
during use. The array of electrodes 120 can additionally generate other
bioelectrical signals upon
interacting with the user's body during use of the system 100.
[0044] The electrodes are arranged in a 2D array. The 2D array can be a
rectangular array,
where the rectangular array can have equal numbers of electrodes along its
width and height. The
size of the array of electrodes 120, in terms of number of electrodes,
distribution of electrodes in
space, and spacing between electrodes, can be configured based on
morphological constraints
governed by the substrate 120, other system aspects, or other design
considerations. In
alternative embodiments, however, the electrodes of the array of electrodes
120 can be arranged
in a polygonal array, ellipsoidal array, or in any other suitable manner
(e.g., an amorphous
array). The electrodes of the array of electrodes 120 can be arranged at
central regions of the
broad surface of the substrate 110 and/or at peripheral regions of the broad
surface of the
substrate 110.
[0045] The left subportion 122 is electrically isolated from the right
subportion 128 to
avoid bridging of electrodes of the left subportion 122 with electrodes of the
right subportion
128. Electrical isolation can be provided by patterning of electrically
conductive regions at the
broad surface of the substrate 110, use of insulating materials coupled to the
substrate 110, or in
another manner.
[0046] In the embodiment shown in FIG. 2, the array of electrodes 120
includes a left
anterior electrode 221, a right anterior electrode 222, a left posterior
electrode 223, and a right
posterior electrode 224. The left anterior electrode 221 and the left
posterior electrode 223 are
embodiments of the left subportion 122 of the array of electrodes 120, and the
right anterior
electrode 222 and the right posterior electrode 224 are embodiments of the
right subportion 128
of the array of electrodes 120 described in relation to FIGS. 1A and 1B above.
The left anterior
electrode 221 and the right anterior electrode 222 are associated with a first
electrical signal
channel 245 of the electronics subsystem described below, and the left
posterior electrode 223
8
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
and the right posterior electrode 224 are associated with a second electrical
signal channel 246 of
the electronics subsystem described below, where each of the first and the
second electrical
signal channels is associated with a circuit across an inferior sagittal plane
of the user's body
during use of the system. The left anterior electrode 221 and the left
posterior electrode 223 can
be used to generate a first noise signal associated with a first noise channel
247, and the right
anterior electrode 222 and the right posterior electrode 224 can be used to
generate a second
noise signal associated with a second noise channel 248, where methods of de-
noising are
described in more detail in Section 2 below.
[0047] FIG. 3A depicts system component configuration of a first embodiment
of the
system shown in FIG. 2, where the first embodiment of the system is configured
as a 2-channel
system for generation of two channels of ECG signals. The first embodiment
includes a left
anterior electrode 321, a right anterior electrode 322, a left posterior
electrode 323, and a right
posterior electrode 324, where the left and the right anterior electrodes 321,
322 are associated
with a first ECG channel 345a and the left and the right posterior electrodes
323, 324 are
associated with a second ECG channel 346a. Methods of signal and noise
extraction in the 2-
channel configuration are described in more detail in Section 2 below.
[0048] FIG. 3B depicts a system component configuration of a second
embodiment of the
system shown in FIG. 2, where the second embodiment of the system is
configured as a 3-
channel system for generation of channel of ECG signals and two channels of
noise signals. The
first embodiment includes a left posterior electrode 323 and a right posterior
electrode 324,
where the left and the right posterior electrodes 323, 324 are associated with
an ECG channel
346b, the left posterior electrode 323 is associated with a first noise
channel 347b, and the right
posterior electrode 324 is associated with a second noise channel 348b. The
first and the second
noise channels 347b, 348b can be coupled to a summation circuit 349b for noise
signal
aggregation and processing. Methods of signal and noise extraction in the 3-
channel
configuration are described in more detail in Section 2 below.
[0049] FIG. 3C depicts a system component configuration of a third
embodiment of the
system shown in FIG. 2, where the third embodiment of the system is configured
as a 3-channel
system for generation of channel of ECG signals and two channels of noise
signals. The first
embodiment includes a left anterior electrode 321 and a right anterior
electrode 322, where the
9
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
left and the right anterior electrodes 321, 322 are associated with an ECG
channel 345c, the left
anterior electrode 321 is associated with a first noise channel 347c, and the
right anterior
electrode 322 is associated with a second noise channel 348c. The first and
the second noise
channels 347c, 348c can be coupled to a summation circuit 349c. Methods of
signal and noise
extraction in the 3-channel configuration are described in more detail in
Section 2 below.
[0050] FIG. 3D depicts a system component configuration of a fourth
embodiment of the
system shown in FIG. 2, where the fourth embodiment of the system is
configured as a 4-channel
system for generation of two channels of ECG signals and two channels of noise
signals. The
first embodiment includes a left anterior electrode 321, a right anterior
electrode 322, a left
posterior electrode 323 and a right posterior electrode 324, where the left
and the right anterior
electrodes 321, 322 are associated with a first ECG channel 345d, the left and
the right posterior
electrodes 323, 324 are associated with a second ECG channel 346d, the left
anterior and
posterior electrodes 321, 323 are associated with a first noise channel 347d,
and the right anterior
and posterior electrodes 322, 324 are associated with a second noise channel
348d. The first and
the second noise channels 347d, 348d can be coupled to a summation circuit
349d. Methods of
signal and noise extraction in the 4-channel configuration are described in
more detail in Section
2 below.
[0051] In a variation related to FIGS 3A-3D, the system is configured as a
3-channel (or 4-
channel) system for generation of different channels of noise and one channel
(or two channels)
of ECG signals. In more detail, the system includes a left anterior electrode
321, a right anterior
electrode 322, a left posterior electrode 323 and a right posterior electrode
324, where both the
left anterior and posterior electrodes 321, 323 are used to derive a noise
source, and/or both the
right anterior and posterior electrodes 322, 324 are used to derive a noise
source. The left and the
right posterior electrodes 323, 324 are associated with a first ECG channel
and/or the left and
right anterior electrodes 321, 322 are associated with a second ECG channel.
FIG. 4 depicts a
plan view of a system component configuration, in accordance with one or more
embodiments.
The embodiment shown in FIG. 4 is configured as a multichannel system for
generation of
multiple channels of ECG signals and/or multiple channels of noise signals.
The array of
electrodes 420 is arranged as an anterior subportion 421, a posterior
subportion 422, a left
subportion 423, and a right subportion 424, where groupings of the anterior
subportion 421 are
associated with one or more ECG channels 445, groupings of the posterior
subportion 422 are
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
associated with one or more ECG channels 446, groupings of the left subportion
423 are
associated with one or more noise channels 447, and groupings of the right
subportion 424 are
associated with one or more noise channels 448.
1.2 System ¨ Other sensors
[0052] As shown in FIGS. 1A and 1B, embodiments of the system also include
one or
more force sensors 130 in mechanical communication with the surface 110, where
the force
sensors can generate signals that are indicative of weight of the user (e.g.,
as the user steps onto
the substrate 110) and/or can detect forces and changes in forces that are
indicative of other
physiologically-relevant parameters. The force sensors 130, for instance, can
generate
ballistocardiogram (BCG) signals from forces generated by cardiovascular
physiological
behavior, which are detected and fused with other signal data according to
methods described
below.
[0053] Embodiments of the system can additionally or alternatively include
one or more
electrodes coupled to a right leg drive (RLD) electrode, where such a
configuration generates a
signal that is derived, at least in part, from a common mode portion of at
least one of the ECG
signals applied back to the body of the user, during use. Such a configuration
operates to enable
removal of common mode interference and can bias the ECG signals to within an
input voltage
range of respective signal amplifiers. The RLD signal can be derived from a
single ECG signal
or a combination of multiple ECG signals. In a configuration without an RLD
electrode and
associated circuitry, a signal input is AC-coupled and biased at mid-supply
voltage to bias the
ECG signals to within an input voltage range of respective signal amplifiers.
[0054] Embodiments of the system can additionally or alternatively include
other sensors
and/or biometric sensors for sensing aspects of the user, the user's
physiology, and/or the
environment of the user. Other sensors can include audio sensors (e.g.,
microphones),
motion/orientation sensors (e.g., accelerometers, gyroscopes, inertial
measurement units, etc.),
respiration sensors (e.g., plethysmography sensors), cardiovascular sensors
(e.g., electrical
signal-based cardiovascular sensors, radar-based cardiovascular sensors, force-
based
cardiovascular sensors, etc.), temperature sensors for monitoring
environmental temperature
(e.g., ambient temperature) and/or body temperature of the user, moistures
sensors (e.g., for
11
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
detecting environmental moisture), optical sensors (e.g., for optically
detecting blood flow
through user body tissue, optical sensors for detecting contact with the
user), capacitive touch
sensors, other electrophysiology sensors (e.g., skin conductance sensors),
and/or any other
suitable sensors.
1.3 System ¨ Electronics and Computing Subsystem
[0055] As shown in FIGS. 1A and 1B, embodiments of the system also include
an
electronics subsystem 140 including channels 144 and 146 for generation of
electrical signals
from the array of electrodes 120, and for relaying and/or pre-processing
signals from the force
sensor(s) 130, where channel configurations are described in more detail above
in relation to
configurations of the array of electrodes in different embodiments.
[0056] The electronics subsystem 140 includes components for receiving,
conditioning,
and relaying signals generated by the array of electrodes 120 and/or the force
sensor(s) 130. For
instance, electrical signals detected by the system from the feet of a user
are on the order of 10-
100 times smaller than the electrical signals collected by traditional methods
(e.g., through the
chest, hands, or upper extremity limbs), which significantly decreases signal-
to-noise ratio.
Therefore, the electronics subsystem 140 can include conditioning components,
such as a high-
resolution A/D converter and/or one or more filters. The electronics subsystem
140 can also
include components that provide power and/or manages power provision to one or
more other
system components. For instance, the electronics subsystem 140 can include a
battery (e.g.,
rechargeable battery, non-rechargeable battery) electrically coupled to a
power management
system that maintains desired circuit voltages and/or current draw appropriate
for different
system components. Power-associated components of the electronics subsystem
140 can be
retained within a housing of the system, where the electronics subsystem 140
can be electrically
and/or physically coupled to one or more of the substrate 110, the array of
electrodes 120, and
the force sensor(s) through the housing.
[0057] The electronics subsystem 140 also includes components of a
computing subsystem
150 and can also include data transmission hardware 149 for data communication
with other
components of the computing subsystem 150 that are remote from device
components that the
user physically interacts with. Remote computing components can be implemented
at other
12
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
networked computers, remote servers, in the cloud, and/or in another computing
platform. The
transmission hardware 149 can include receive and/or transmit components for
handling data
transfer between electronics of the device that the user physically interacts
with and remote
computing components through a network. Furthermore, the transmission hardware
149 can
provide a wired and/or wireless (e.g., WiFi, Bluetooth LE, etc.) interface
with the network or
other remote computing subsystem components.
[0058] In relation to methods described in Section 2 below, the computing
subsystem 150
can also include a non-transitory computer-readable storage medium containing
computer
program code for implementing one or more portions of the method(s) described
below. For
instance, the computing subsystem 150 can include program code and
architecture for generating
an electrocardiogram (ECG) signal from a left subportion 122 and a right
subportion 128 of the
array of electrodes 120, generating a first noise signal from the left
subportion 122 and/or a
second noise signal from the right subportion 128 of the array of electrodes,
and generating a de-
noised ECG signal upon processing the ECG signal with the first and/or the
second noise signals.
[0059] The computing subsystem 150 can also include architecture for
storing instructions
in non-transitory computer readable media for controlling operation states of
electrodes and/or
sensors, monitoring states of components coupled to the computing subsystem
150, storing data
in memory, coordinating data transfer (e.g., in relation to the transmission
hardware 149), and/or
performing any other suitable computing function of the system. The computing
component
160a can additionally or alternatively include signal conditioning elements
(e.g., amplifiers,
filters, analog-to-digital converters, digital-to-analog converters, etc.) for
processing signal
outputs of electrodes and sensors of the system 100.
2. Method ¨ Extracting and De-noising Electrical Signals
[0060] FIG. 5A depicts a flowchart of a method 500 for cardiovascular
signal acquisition
and noise mitigation, in accordance with one or more embodiments. As shown in
FIG. 5A, the
array of electrodes generates one or more ECG signals from the left and the
right subportions of
the array 510. Then, the computing subsystem (e.g., components of the
electronics subsystem
and/or the computing subsystem described above) extracts one or more noise
signals 540 and
generates one or more de-noised ECG signals 570 upon processing the one or
more ECG signals
13
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
to isolate components of the noise signal(s) from the ECG signals.
[0061] The method 500 functions to acquire electrical signals associated
with
cardiovascular health in a non-standard manner and with a system designed to
promote routine
usage by a user, and also functions to implement sensor distributions in space
to mitigate noise
induced by ambient sources, user motion (e.g., feet motion), and other
sources. The method can
include receiving electrical signals through the feet of a user, comparing
signals across different
vectors defined by device sensor positions, and extracting health-relevant
signal components and
noise components based upon noise-isolation and extraction methods. As such,
the method 500
significantly increases signal-to-noise ratios for electrical signals acquired
through feet of the
user. While applications of the method 500 for de-noising ECG signals are
described, the method
500 can additionally or alternatively be used to denoise other electrical
signals (e.g., IPG signals,
other passive electrical signals, other active electrical signals).
[0062] The method 500 can be implemented by one or more portions of the
system
embodiment(s) described above, where anterior, posterior, left, and right
electrode portions of an
array of electrodes can provide source signals that are processed to generate
de-noised signals of
interest. As configured by the structure of embodiments of the system
described above, noise and
artifacts present in outputs from anterior electrodes is largely uncorrelated
with noise and
artifacts present in outputs from posterior electrodes. Furthermore, as
configured by the structure
of embodiments of the system described above, anterior and posterior
subportions of electrodes
output both signal and noise components, while left and right subportions of
electrodes output
only noise components because they are not positioned across the body and
heart. In more detail,
noise outputs from left and right subportions of electrodes produce noise
signal components in
varying proportions and combinations relative to noise signal components from
anterior and
posterior subportions of electrodes.
2.1 Method ¨ Noise contributions
[0063] Electrical signals detected by the system from the feet of a user
are on the order of
10-100 times smaller than the electrical signals collected by traditional
methods (e.g., through
the chest, hands, or upper extremity limbs), which significantly decreases
signal-to-noise ratio.
As such, noise factors can have a much larger effect on signal acquisition
and/or processing as
14
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
compared to traditional methods for signal acquisition in relation to
cardiovascular health. In
relation to noise contributions to the ECG signals (or other electrical
signals) generated using the
array of electrodes, noise can come from ambient sources (e.g., 60 Hz mains,
50 Hz mains,
thermal fluctuations in the environment, industrial noise, etc.). Noise can
also come from motion
of the user while interfacing with the array of electrodes. For instance, in
relation to the weighing
scale form factor of the device described above, motion of the user's body
and/or or feet, such as
swaying motions while measuring body weight, curling of the feet, shifting of
the feet, motions
to maintain balance, poor contact between foot regions and the electrode(s),
and/or other motions
can induce significant noise that impacts the SNR of the desired signal(s).
Such motions can
induce electromyography (EMG) artifacts in electrical signals due to
generation of electrical
signals from muscular contraction and/or relaxation behavior. Such motions can
additionally or
alternatively induce force-associated artifacts that can interfere with force
associated
measurements of the system.
[0064] Methods for isolating and extracting noise induced by these and
other sources are
described below in relation to FIGS. 5B-5F, where noise can be extracted using
blind source
separation techniques (e.g., using independent component analysis), using
adaptive filtering
operations, using sensor channel windowing operations, using nonparametric
spectral estimation
processes, and/or using other operations that isolate desired signals and
undesired noise signal
components from source signals that include both desired components and noise.
2.2 Method ¨ Signal De-noising Using Blind Source Separation Techniques
[0065] FIG. 5B depicts a flowchart of a first embodiment of the method
shown in FIG. 5A,
where independent ECG signal sources can be estimated from source signals that
have noise.
The method 500b of FIG. 5B can be implemented using a two-channel array
configuration (e.g.,
the configuration shown in FIG. 3A), using a three-channel array configuration
(e.g., the
configuration shown in FIG. 3B or 3C), using a four-channel array
configuration (e.g., the
configuration shown in FIG. 3A), or using a configuration with more than four
channels. The
computing subsystem, in cooperation with the array of electrodes, thus
generates 510 one or
more ECG signals including an anterior ECG signal 520 and/or a posterior ECG
signal 530, and
extracts 540 a noise signal including a left noise signal and/or a right noise
signal 560, with
separation 570 of de-noised components from noise components based on
electrode array
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
configuration.
[0066] In one embodiment of FIG. 5B, the computing subsystem, in
cooperation with
electronics components that receive source signals from the array of
electrodes, processes input
signals from input channels and produces output signals where independent
desired signal
components are separated from undesired noise components. In this embodiment,
the computing
system includes architecture in code for performing an independent component
analysis (ICA)
operation the separates multivariate signals from the input channels into
subcomponents
associated with desired signals and noise.
[0067] As noted above in relation to FIGS. 3A-3D, the input channels can be
an anterior
ECG channel coupled to left and right anterior electrodes and a posterior ECG
channel coupled
to left and right posterior electrodes (as in the 2-channel configuration of
FIG. 3A), where the
output of the ICA operation recreates a new anterior ECG signal and a new
posterior ECG signal
with noise components separated out. Alternatively, the input channels can be
a posterior ECG
channel coupled to left and right posterior electrodes and two noise channels
coupled to left
electrodes and right electrodes, respectively (as in the 3-channel
configuration of FIG. 3B),
where the output of the ICA operation recreates a new posterior ECG signal and
isolated noise
signals associated with the two noise channels. Alternatively, the input
channels can be an
anterior ECG channel coupled to left and right anterior electrodes and two
noise channels
coupled to left electrodes and right electrodes, respectively (as in the 3-
channel configuration of
FIG. 3C), where the output of the ICA operation recreates a new anterior ECG
signal and
isolated noise signals associated with the two noise channels. Alternatively,
the input channels
can be an anterior ECG channel coupled to left and right anterior electrodes,
a posterior ECG
channel coupled to left and right posterior electrodes, and two noise channels
coupled to left
electrodes and right electrodes, respectively (as in the 4-channel
configuration of FIG. 3D),
where the output of the ICA operation recreates a new anterior ECG signal, a
new posterior ECG
signal, and isolated noise signals associated with the two noise channels.
[0068] The ICA operation implemented by the computing subsystem separates
the
independent signal and noise components by increasing the statistical
independence of the
estimated signal and noise components, with a parallel or deflational ICA
algorithm. The ICA
operation can be based on maximization of non-Gaussianity (e.g., as motivated
by central limit
16
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
theory, considering kurtosis, considering negentropy), or minimization of
mutual information
(e.g., considering maximum entropy, based on a divergence factor). The ICA
operation can be
based on a non-linear function or a linear function that transforms the
multivariate input into
resolved independent components. In alternative embodiments, another blind
source separation
operation, such as principal components analysis, singular value
decomposition, dependent
component analysis, matrix factorization, coding and decoding, stationary
subspace analysis, or
another operation can be used to resolve signal and noise components.
2.3 Method ¨ Signal De-noising Using Adaptive Filtering Techniques
[0069] FIG. 5C depicts a flowchart of a second embodiment of the method
shown in FIG.
5A, where the second embodiment implements an adaptive filtering operation.
The method 500c
of FIG. 5C can be implemented using a four-channel array configuration (e.g.,
the configuration
shown in FIG. 3D) or with another array/channel configuration. In performing
the method 500c,
the computing subsystem, in cooperation with the array of electrodes,
generates 510 one or more
ECG signals including an anterior ECG signal 520 and a posterior ECG signal
530, and extracts
540 a noise signal including a left noise signal and/or a right noise signal
560, with separation
570 of de-noised components from noise components based on an adaptive
filtering operation.
[0070] The adaptive filtering operation can use an affine projection
algorithm with the
filter equation y(k) = XT(k)*w(k), where y is the filtered signal, X is the
filter input matrix that is
a function of x, where x is a vector of adaptive filter parameters, w is a
function for adaptation of
adaptive parameters, and k is a time index. The input signals to the affine
projection operation
are the anterior and posterior ECG signals and the summation of signals from
two noise
channels.
[0071] The adaptive filtering operation can alternatively use a recursive
least squares
algorithm or a least mean squares algorithm with the filter equation y(k) =
xT(k)*w(k), where y
is the filtered signal, x is a vector of adaptive filter parameters, w is a
function for adaptation of
adaptive parameters, and k is a time index. The input signals to the least
squares operation(s) are
the anterior and posterior ECG signals and the summation of signals from two
noise channels.
[0072] Alternative embodiments of the adaptive filtering operation can use
a generalized
17
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
normalized gradient descent algorithm, a least mean fourth algorithm, or
another suitable
adaptive filtering algorithm.
2.4 Method ¨ Signal De-noising Using Sensor Selection Techniques
[0073] FIG. 5D depicts a flowchart of a third embodiment of the method
shown in FIG.
5A, and FIG. 5E depicts a schematic flow of the embodiment of the method shown
in FIG. 5D.
In performing the method 500d, the computing subsystem, in cooperation with
the array of
electrodes, generates 510 one or more ECG signals including an anterior ECG
signal 520 and a
posterior ECG signal 530, and generates 570 a de-noised signal upon segmenting
the anterior
ECG signal 591a and the posterior ECG signal 591b, performs a quality
assessment operation
592 based on analysis of noise present in each segment of the anterior ECG
signal and the
posterior ECG signal, and stitches segments that pass the quality assessment
operation 593 to
form a composite de-noised ECG signal. In relation to segmentation, the
computing subsystem
can segment signals into windows with a desired window length (e.g., 1 second,
less than one
second, more than 1 second), where the windows can be non-overlapping or
overlapping. In the
quality assessment operation, the criteria for selection of the signal window
to include in the
composite de-noised ECG signal can be based on a SNR criterion or a variance-
associated
criterion. In more detail, for each matching window across the anterior ECG
signal and the
posterior ECG signal, the computing subsystem can determine which window has
less noise
based on the SNR or the variance-associated criterion, and pass the "winning"
window onward
for stitching to generate the composite de-noised ECG signal.
2.5 Method ¨ Signal De-noising Using Nonparametric Spectral Estimation
Techniques
[0074] FIG. 5F depicts a flowchart of a variation of the method for
cardiovascular signal
acquisition and noise mitigation shown in FIG. 5A. In performing the method
500f, the
computing subsystem, in cooperation with the array of electrodes, performs a
non-parametric
spectral estimation process by generating 510 one or more ECG signals from
left and right
subportions of an array of electrodes, and generating 580 a de-noised signal
upon decomposing
the ECG signal(s) 581 and reconstructing the ECG signal(s) 582 to extract a de-
noised signal
component and a noise component. In performing the method 500f, the computing
subsystem
18
CA 03116846 2021-04-16
WO 2020/081471
PCT/US2019/056160
can extract the de-noised ECG signals by using the quasi-periodic nature of
the ECG signal(s). In
a specific example of 500f, the computing subsystem embeds an input ECG signal
(e.g., an
anterior ECG signal, a posterior ECG signal) into a Hankel matrix having a
desired length (e.g.,
of 100 samples, of less than 100 samples, of more than 100 samples) based on
accuracy
considerations, where longer matrices produce greater accuracy, but are
computationally
expensive. The computing subsystem then uses a singular value decomposition
operation with
the Hankel matrix to decompose the input signal. Then, the computing subsystem
reconstructs
the ECG signal by splitting the output from the decomposition operation into
two groups
including a first group for the ECG signal component and a second group for
the noise signal
component. In the specific example, the reconstructed time series is formed
using diagonal
averaging. However, alternative variations of the method 500f can implement
another spectral
estimation architecture having another decomposition and/or reconstruction
algorithm.
3.
Method ¨ Generation of and Fusion of Multiple Signals for Cardiovascular
Health
Monitoring
[0075] FIG. 6 depicts a flowchart of a method 600 for electrical and
mechanical
cardiovascular signal acquisition processing, in accordance with one or more
embodiments. As
shown in FIG. 6, responsive to contacting the feet of the user, the array of
electrodes generates
610 one or more electrical signals (e.g., ECG signals, IPG signals).
Responsive to contacting the
feet of the user, the set of force sensors also generate 615 one or more force-
derived signals (e.g.,
BGC signals, a weight signal). Then, the computing subsystem (e.g., components
of the
electronics subsystem and/or the computing subsystem described above)
generates 620 values of
a set of cardiovascular health parameters, where generating values of a set of
cardiovascular
health parameters can include generating 625 values of systolic temporal
parameters, several of
which are described below. The computing subsystem then processes 640 values
of the set of
cardiovascular health parameters with a cardiovascular risk model, and returns
650 an output of
the cardiovascular risk model.
[0076] The computing system can also provide 660 a prediction of the
cardiovascular
health state of the user, derived from the output, to an entity associated
with the user. The entity
can be another computing entity that provides further analysis of the
prediction in relation to
automated interventions for actively improving an undesired cardiovascular
health state, or
19
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
maintaining a desired cardiovascular health state. Automated interventions can
be provided
through medical devices (e.g., electrical stimulation devices, medication
eluting devices,
medication dispensing devices, etc.) in communication with the computing
subsystem, such that
the computing subsystem can also generate and/or provide instructions for
controlling operation
states of the medical device(s) for automated interventions. The entity can
additionally or
alternatively be a non-computing entity, such as a practitioner, emergency
personnel, caretaker,
family member, friend, or other acquaintance of the user.
[0077] The method 600 functions to process and fuse parameters derived from
electrical
signal(s) and force-derived signal(s) that are collected simultaneously or
contemporaneously as a
user steps onto a sensing surface, in order to extract values of
cardiovascular health parameters.
The parameter values can then be used to determine, in real time, a
cardiovascular health state of
the user. In one or more embodiments, systems associated with the method 600
generate ECG
signals (e.g., such as in manners described above), IPG signals, BCG signals,
and weight
measurements through an interface with feet of a user. The method 600 and
associated system
components are configured such that the parameter values are regularly
collected in a non-
disruptive/non-invasive manner, and can be analyzed to monitor user
cardiovascular health to
trigger interventions at critical times, if needed.
[0078] The method 600 can be implemented by one or more portions of the
system
embodiment(s) described above, where portions of an array of electrodes (e.g.,
anterior,
posterior, left, and right subportions of the array of electrodes) can provide
electrical signals that
are processed in different channels (e.g., ECG channels, IPG channels) and one
or more force
sensors can provide force-derived signals. The signals are then conditioned
with electronics
subsystem components and processed by computing subsystem to provide processed
outputs that
can be used to maintain or improve user health.
3.1 Method ¨ Passive Electrical Signal Extraction
[0079] As shown in FIG. 6, responsive to contacting the feet of the user,
the array of
electrodes generates 610 one or more electrical signals.
[0080] Passive electrical signals, including the ECG signals described in
relation to the
method 500 above, can be generated. The passive electrical signals can thus
include periodic
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
signals generated by depolarization of the heart. The passive electrical
signals can also include
time varying components that include muscle activity information associated
with muscles of the
legs of the user and derived from electrical potentials produced by muscles
used to stand and/or
balance. The signals capturing leg muscle activity have a lower frequency due
to contact
impedance between the user's feet and the array of electrodes of the system.
The signals
capturing leg muscle activity are also modulated by changes in foot positon,
electrodermal
activity of the skin, activity of sweat glands in the skin, and can be
indicative of physiological
and/or psychological arousal (in relation to autonomic nervous system
activation).
3.2 Method ¨ Active Electrical Signal Extraction
[0081] The same array of electrodes used to generate passive electrical
signals can also
generate active electrical signals, such as the IPG signals noted above, when
the user steps onto
the surface(s) in electrical contact with the array of electrodes; however, in
alternative
embodiments, IPG signals (or other electrical signals) can be collected with
another set of
electrodes. Each active electrical signal has a periodic component attributed
to changes in
resistance of the lower extremities as blood volume and flow changes with each
heartbeat, and
the periodic component (i.e., the IPG signal), is extracted by the computing
subsystem and
associated electronics with bandpass filtering (e.g., with a 0.5-30Hz
frequency band). Each
active electrical signal also has a static or slow varying DC component that
is representative of
body impedance, and this DC component is indicative of water content in the
body. The
computing subsystem and associated electronics extract values of parameters
from the DC
component, where the parameters include one or more of: fluid status,
extracellular and
intracellular water content, body composition, body fat, and edema status. The
periodic
components and the DC components are derived at multiple frequencies by the
computing
subsystem, as described in more detail below, to extract additional
information. For instance, a
higher frequency signal (-64 kHz) can pass through more of the cell membranes
in the body and
thus represents overall body water content. A lower frequency signal (-8kHz)
less easily passes
through cell membranes and represents extracellular water content. Thus, the
computing
subsystem can process signals at different frequencies to extract values of
parameters related to
total body water (TBW), extracellular water (ECW), and intracellular water
(ICW) content.
[0082] In relation to active electrical signal generation, the system can
provide a
21
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
stimulation current that travels from one foot and through one leg of the
user, and then through
the other leg and the other foot of the user. The flow and presence of blood
and other body fluids
in the user's body presents a varying resistance to the stimulation current,
where the resistance
varies with fluid in a respective body region (e.g., a leg region) at any
given time. The
stimulation current encounters this change in resistivity which produces a
detectable voltage
change. In relation to a detectable voltage change, an active electrical
waveform thus has
characteristic peaks representative of the maximum and minimum fluid volume
(e.g., blood
volume) in a body region of the user associated with the stimulation current.
[0083] A subportion of electrodes used to generate the active signals can
be configured to
apply a stimulation current to the feet of the user through conductive aspects
of the substrate
described above. The stimulation current can be a small current (e.g., a
current below 500uA, a
current below lmA, a current below 5 mA, a current below 10mA, etc.). The
stimulation current
can also be a variable current with a regular waveform (e.g., sinusoidal
waveform, non-
sinusoidal waveform, square waveform, sawtooth waveform, etc.) or a non-
regular waveform.
However, the stimulation current can be non-variable, with known
characteristics that can be
used to assess body-region impedance. In one embodiment, the stimulation
current is a current of
approximately 500 uA having a frequency of 8-64 kHz.
[0084] In a configuration using paired electrodes associated with left and
right sides of a
device (and left and right sides of the body of the user), a first pair of
electrodes can be used to
apply the stimulation current, and a second pair of electrodes can be used to
detect the active
electrical signal(s). In relation to the device configuration shown in FIG. 2
above, the anterior
electrodes can be used to apply the stimulation current, and the posterior
electrodes can be used
to detect the IPG signal(s). Alternatively, the posterior electrodes can be
used to apply the
stimulation current, and the anterior electrodes can be used to detect the IPG
signal(s).
Alternatively, with configured timing electronics architecture, the
stimulation current can be
applied with a pair of electrodes followed by detection of the IPG signal with
the same pair of
electrodes. Furthermore, in relation to timing electronics architecture, the
electronics subsystem
can implement timing operation modes for detection of the passive electrical
signal(s) through
the array of electrodes used to detect the active electrical signals.
22
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
3.3 Method ¨ Force-derived signal extraction and other signals
[0085] As shown in FIG. 6, responsive to contacting the feet of the user,
the set of force
sensors also generate 615 one or more force-derived signals (e.g., BGC
signals, a weight signal).
The set of force sensors and associated components of the electronics
subsystem (e.g., analog
circuitry) have a signal-to-noise ratio (SNR) and resolution sufficient for
ballistocardiography,
where the system detects small forces produced by physiological operation of
the user's
cardiovascular system (e.g., such as small perturbations of the body as the
heart beats). Such
forces can be associated with ejection of blood from the heart into the aorta
(e.g., corresponding
to a J-wave of a BCG signal), and travel of blood through the ascending and
descending portions
of the aorta to other portions of the user's body. BCG signals are extracted
by the computing
subsystem through bandpass filtering (e.g., of 0.5-50Hz). A low frequency or
DC component of
the forced-derived signal is derived by the computing subsystem through
lowpass filtering (e.g.,
with a cutoff frequency of 5Hz), and characterizes motion of the user on the
substrate as well as
weight of the user standing on the device. In more detail, body weight can be
extracted through
summation of signals from all force sensors of the system, and motion can be
extracted through
lowpass filtering each force sensor independently and extracting center of
pressure information.
[0086] During signal generation, additional sensors coupled to the
electronics subsystem
can also generate additional signals associated with the environment of the
user. Such signals can
include temperature signals and/or moisture signals, which can inform or
affect other electrical
signal measurements or force signal measurements. Additional sensors that can
be implemented
are described in more detail above.
3.4 Method ¨ Cardiovascular and other physiological health parameter
extraction
[0087] FIG. 7A depicts a flow diagram of cardiovascular health parameter
extraction
processes, in accordance with one or more embodiments. As shown in FIG. 7A,
the system
generates one or more of (or one or more instances of, depending on sensor
multiplicity and
configuration): an ECG signal, an IPG signal, a BCG signal, a weight signal, a
temperature
signal, and a moisture signal. The electronics subsystem and/or computing
subsystem then passes
respective signals through different operation flows in order to extract
values of parameters
relevant to cardiovascular health, as described in more detail below in
relation to FIGS. 7B-7G.
23
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
Parameter values are then processed with a cardiovascular risk model in order
to generate
predictions of cardiovascular health state of the user, where the predictions
can be used to trigger
appropriate interventions to support the health of the user. The flows shown
in FIGS. 7A-7G can
be repeated regularly (e.g., multiple times a day, daily, weekly, etc.)
whenever the user uses the
device, where regular use is promoted by configuring elements of the system in
a weighing scale
form factor that can contemporaneously measure signals beyond weight signals.
Regular
measurements can thus provide rich data to longitudinally analyze
cardiovascular health of the
user.
[0088] The sensors of the system can also generate values of other
physiological health
parameters including galvanic skin potential, foot contact to electrodes, and
foot-to-foot
electromyography signals from passive electrical potentials; body water
content (ECW, ICW,
and TBW), body composition, and fluid status from active electrical signals;
and body weight,
center of pressure, and motion from force-derived signals. These parameters
are used by the
system to provide additional clinical context in a wide range of patient and
user populations can
are used by the system to detect noise and motion in the system, for noise
mitigation and artifact
removal, as described above and below.
3.4.1 Method ¨ Ensemble Averaging
[0089] FIG. 7B depicts a first portion of the flow diagram shown in FIG.
7A, which
corresponds to an embodiment of a portion 620 of the method shown in FIG. 6.
As shown in
FIGS. 7A and 7B, the electronics subsystem, with associated computing
architecture, can pass
each of the ECG signal(s), the IPG signal(s) and the BCG signal(s) through an
interpolation
operation 721 and a set of filtering operations 722. In an embodiment, prior
to interpolation and
filtering, active and passive electrical signals are measured at 250Hz
sampling rate using a 24-bit
delta sigma analog to digital converter (ADC) of circuitry of the system. In
an embodiment, prior
to interpolation and filtering, force-derived signals from each of the set of
force sensors are
sequentially sampled at 1 kHz using a 24-bit delta sigma ADC of circuitry of
the system, where a
higher sampling rate is associated with an increased number of force sensors.
[0090] As executed by the computing subsystem, in an embodiment, the
interpolation
operation can include interpolation of signals to lkHz in order to increase
temporal resolution of
24
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
the passive and active electrical signals, as well as force-derived signals.
However, interpolation
can be implemented by the computing subsystem with another suitable frequency
of
interpolation.
[0091] The filtering operations can include a bandpass filtering operation,
as described
above, and/or other filtering operations. The filtering operations can vary
across different
electrical signals and/or force-derived signals, and can include digital
finite impulse response
(FIR) techniques and/or infinite impulse response (IIR) techniques). In an
embodiment, the
filtering operations include a bandpass filter of 0.1-100Hz for passive
electrical signals
associated with ECG signals. In an embodiment, the filtering operations
include a bandpass filter
of 5-100Hz for passive electrical signals associated with leg muscle-derived
electrical potentials.
In an embodiment, the filtering operations include a bandpass filter of 0.5-
30Hz for active
electrical signals associated with IPG signals. In an embodiment, the
filtering operations include
a bandpass filter of 0.5-50Hz for force-derived signals associated with BCG
signals. However, in
variations, other frequency ranges can be used for different signal types, in
different bandpass
filtering operations.
[0092] Additionally or alternatively, the filtering operations can include
a fourth-order
high-pass filter operation followed by a low-pass filter operation. For each
signal type, the high
pass filter can include a cutoff frequency associated with higher-order
derivatives of each signal
type in order to preserver higher-order derivative features of the signal,
where the cutoff
frequencies can differ across signal type. However, the cutoff frequencies or
frequency ranges
can alternatively overlap. Similarly, the low pass filter can include a cutoff
frequency associated
with each signal type, where the cutoff frequencies differ across signal type.
However, the cutoff
frequencies or frequency ranges can alternatively overlap. In still other
variations, the filtering
operations can be applied to non-fourth order derivatives of the signal(s).
Furthermore, the
filter(s) can be applied to inbound signals in any other suitable order.
[0093] As indicated above, electrical and force-derived signals are sampled
simultaneously
when the user contacts the substrate with his/her feet, in order to facilitate
extraction of
cardiovascular health parameters that are reliant on phase relationships and
accurate time
synchronization between signals. As such, the system configuration enables
automatic signal
synchronization and accounts for misalignments due to filtering and other
signal processing
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
operations. However, in alternative embodiments, the system can collect
different signals with
non-simultaneous sampling, and implement signal registration and alignment
techniques to
extract cardiovascular health parameters that are reliant on phase
relationships.
[0094] In relation to FIGS. 7A and 7B, the electronics subsystem, with
associated
computing architecture, can use an extracted feature of one signal type as
references to ensemble
average other signal types, with ensemble averaging techniques gated off a
specific signal type.
In embodiments described, the active and passive electrical signals, as well
as dynamic force-
derived signals, are small in magnitude and contaminated by noise and
artifact, which motivates
use of ensemble averaging. In embodiments described, the IPG signals have the
highest signal-
to-noise ratio (SNR), so characteristic feature(s) of the IPG signals are used
to ensemble average
each of the signals.
[0095] In one embodiment, as shown in FIGS. 7A, 7B, and 7C, the computing
subsystem
can detect a set of peaks from the IPG signal, as the IPG signal is generated,
and use the
extracted peaks to generate 723 an ensemble averaged waveform for each of the
ECG signal(s),
the IPG signal(s), and/or the BCG signal(s). In more detail, the computing
subsystem generates a
first derivative of the IPG signal, smooths the first derivative of the IPG
signal with a moving
average filter, and squares the output of the smoothing operation. The maximum
peak of the IPG
derivative signal is used as a gating feature for the ensemble averages of the
processed electrical
and force-derived signals. Each incoming peak of the IPG signal is then used
as a temporal
marker to collect and store a window (e.g., a window of 500-1000ms) on each
side of each
temporal marker for each electrical and force-derived signal. As additional
peaks are detected,
windows of signals about each peak are summated and averaged to create
ensemble averages of
each signal type over a measurement period (e.g., associated with a session of
a user standing on
the substrate of the device). In this embodiment, the resulting ensemble
averages result in
approximately one heart beat cycle of information for each electrical signal
and force-derived
signal type. In various embodiments, however, the number of peaks over which
an ensemble
average is calculated can be adjusted by the computing subsystem to reduce
noise, and in one
embodiment, each ensemble average is generated over 20 heart beats. Averaging
over a number
(N) peaks is associated with a reduction in noise by a factor of the square
root of N. In
performing the ensemble averaging process 723, the computing subsystem can
remove sources
of noise that are non-periodic. For instance, an electromyography component
(associated with
26
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
lower limb muscle activation) of a passive electrical signal can overwhelm an
ECG component
of the passive electrical signal, and ensemble averaging using the IPG signal
can remove non-
periodic noise associated with the electromyography component.
[0096] In other embodiments, other features can be used to create the
ensemble averages.
For example, the computing subsystem can generate ensemble averages of signals
based upon
other IPG signal features (e.g., other maximum or minima), features of higher
order derivatives
of the IPG signal, and features of other transformations of the IPG signal. In
still other
embodiments, the computing subsystem can implement other non-IPG signals as
the gating
source(s) for ensemble averages. In one such embodiment, a BCG signal having
sufficient
quality can be processed by the computing subsystem to detect characteristic
features (e.g., of an
I-wave, of a J-wave) for use in ensemble averaging. Additionally or
alternatively, in another
embodiment, an ECG signal having sufficient quality can be processed by the
computing
subsystem to detect characteristic features (e.g., of a QRS peak) for use in
ensemble averaging.
The gating feature(s) can be constant across all users, or can be changed
automatically and
adaptively selected by the computing subsystem based characterization of
quality of each signal
type for each user.
3.4.2 Method ¨ Noise Mitigation in Relation to Ensemble Averaging
[0097] Furthermore, in performing the ensemble averaging operation 723, the
computing
subsystem can use a weighted window process, whereby a variance-associated
parameter (e.g.,
local variance, standard deviation) can be used to assign a weight to each
signal window as it is
processed to generate the ensemble average, where the weight decreases for a
noisier signal
window.
[0098] In related processes, in relation to noise mitigation using the
ensemble averaging
process, the computing subsystem blocks gating features from being further
used in an ensemble
averaging process, thereby blocking ensembling for windows of signals
associated with high
levels of noise or other artifacts. The computing subsystem can trigger
blocking of gating
features based upon comparison to a threshold noise condition. The computing
subsystem can
additionally or alternatively trigger blocking of gating features based upon
another parameter
value (e.g., center of pressure from force sensor-derived data, as a measure
of motion).
27
CA 03116846 2021-04-16
WO 2020/081471
PCT/US2019/056160
Threshold conditions for gating feature blocking can be constant for each
measurement session,
or can be adapted to each signal type. The computing subsystem also implements
threshold
condition comparisons in a manner that does not filter out features of
interest (e.g., such as QRS
complexes of ECG signals).
3.4.3 Method ¨ Cardiovascular Parameter Extraction
[0099] Once
signals have been measured and pre-processed, characteristic features and
relationships between the signals are extracted by the computing subsystem to
determine
cardiovascular health states and/or other physiological states of the user(s).
As described below,
features of each of the IPG, ECG, and BCG signals can be extracted and co-
processed to
generate values of features correlated with cardiovascular health parameters.
[00100] FIG.
7D depicts a second portion of the flow diagram shown in FIG. 7A, which
corresponds to an embodiment of a portion 620 of the method shown in FIG. 6.
As shown in
FIGS. 7A and 7D, the computing subsystem can identify an R-peak of an ECG
signal or
averaged ECG waveform and a peak of an I-wave of the BCG signal or averaged
BCG
waveform. The computing subsystem can then use the positions of the R-peak and
the peak of
the I-wave to extract 724 a pre-ejection period (PEP) for the user. The R-peak
is a peak of the
QRS complex corresponding to depolarization of the right and left ventricles
of the heart, and
captured in the ECG signal. The computing subsystem can use a wavelet analysis
to identify the
R-peak in the signal. The wavelet analysis can include a discrete wavelet
transform to enhance
the R-peak(s) in the ECG signal, followed by a peak finding process to find
the time point
associated with the R-peak. The I-peak can be a good proxy for the end of PEP,
given that the I-
wave represents a post-ejection of blood from the aorta, and the computing
subsystem can use a
peak finding process to locate the time point corresponding to a peak of the I-
wave. The
computing subsystem can then apply a correction operation to extract a more
exact end of the
PEP period, where the correction operation can be based upon modeling against
a reference
device that outputs a true value of PEP. Alternatively, the PEP and/or
relative time points
associated with the PEP can be estimated with a correction factor using other
measured
parameters, such as the pulse rate and/or pulse transit time. The correction
operation can be
universally applied to or alternatively customized to signals from different
users (e.g., during
different measurement systems). However, other features can be good proxies
for locating an end
28
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
of the PEP period (e.g., a B-point of an IPG-derived signal). The PEP
characterizes a time
between electrical depolarization of the heart and ejection of blood into the
ascending aorta,
which is related to a length of time the heart is contracting and reflects
cardiac contractility.
[00101] FIG. 7E depicts a third portion of the flow diagram shown in FIG.
7A, which
corresponds to an embodiment of a portion 620 of the method shown in FIG. 6.
In an
embodiment, the computing subsystem generates a first derivative of the
average IPG waveform
and applies a lowpass filter (e.g., of 15 Hz) to the first derivative. In an
embodiment, the
computing subsystem also generates a second derivative of the average IPG
waveform and
applies a lowpass filter (e.g., of 21 Hz) to the second derivative. In an
embodiment, the
computing subsystem also generates a third derivative of the average IPG
waveform and applies
a lowpass filter (e.g., of 21 Hz) to the third derivative. As shown in FIGS.
7A and 7E, the
computing subsystem can identify a B-point of an IPG signal or averaged IPG
waveform and an
X-point from at least one of a first derivative, a second derivative, and a
third derivative of the
IPG signal or averaged IPG waveform. In particular, identification of one or
more of the B-point
and the X-point can be obscured by noise or atypical signal morphology. As
such, one or more of
the other signals (e.g., ECG-derived signals, BCG-derived signals, IPG-derived
signals) can be
used to correctly identify B and/or X-points. For instance, the R-peak of an
ECG-derived signal
can be used to define a physiological window in which the B-point is expected
to be found. As
such, other signals can be used to generate physiologically relevant time
windows where other
signal features are expected to be found, in order to improve localization of
such features. The
computing subsystem can then determine 725 the left ventricular ejection time
(LVET) for the
user from the time distance between the B-point and the X-point, where the
LVET is a time
period of blood flow across the aortic valve, as influenced by the heart rate
(HR) of the user, the
pre-load on the aortic valve, the afterload on the aortic valve, and
contractile state. In more
detail, in determining the LVET for the user, the computing subsystem can
generate a second
derivative of the averaged IPG waveform (or IPG signal) and identify a first
minimum
immediately preceding a maximum change in impedance in the averaged IPG
waveform (or IPG
signal), where the time point associated with the first minimum corresponds to
the B-point. The
computing subsystem can also identify an absolute minimum of the second
derivative of the
averaged IPG waveform (or IPG signal), where the time point associated with
the absolute
minimum corresponds to the X-point. Then, the computing subsystem can
determine left
29
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
ventricular ejection time (LVET) from positions of the first minimum and the
absolute minimum.
[00102] In a related embodiment, the LVET can be determined from features
of BCG-
derived signals and/or IPG-derived signals. For instance, a BCG-derived signal
can be high-pass
filtered and/or derivatives of the BCG-derived signal can be calculated, such
that higher
frequency components of the signal are emphasized and extracted. The resulting
features can
represent vibrations of the user's body due to the aortic valve opening and
closing, and can be
used by the computing subsystem to determine temporal markers representative
of the opening
and closing of the valves. These temporal markers are then used, with or
without combination of
IPG-derived features, to calculate the LVET for a user. These features can
also be used with
ECG-derived features to calculate PEP. For instance, the computing subsystem
can process an R-
peak time point and a time point of an aortic valve opening feature of a BCG-
derived signal to
determine PEP.
[00103] As such, transformations on ensemble average signals can be used to
extract
features, where derivatives and higher order derivatives (e.g., second
derivatives, third
derivatives, fourth derivatives, etc.) of an averaged ensemble signal (e.g.,
averaged IPG signal)
can be used to extract features (e.g., peaks and valleys) associated with
different cardiovascular
time intervals. Furthermore, time intervals associated with transformations of
a signal can be
used to extract derivative features.
[00104] The computing subsystem can extract amplitude features from the
ensemble
averaged signals. In particular, because ensembling involves gating, small
changes in timing and
phases of the signals during a measurement session across different signal
types can result in a
reduction in feature amplitudes for signals that do not contain the gating
feature. Thus, the
computing subsystem can recover true amplitudes of features in each signal
type by realigning
each individual ensemble averaged waveform using its individual component
signals. In one
example, to realign a BCG signal, the J-wave of each component signal used to
generate the
ensemble averaged BCG signal can be used to realign the ensemble components.
Since the J-
wave location is known, a tighter window (e.g., window less than 500m5) can be
used to detect
local peaks associated with the J-wave location, and used to realign the
ensemble components.
Then, after realignment, the true amplitude of the J-wave components can be
extracted by the
computing subsystem.
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
[00105] FIG. 7F depicts a fourth portion of the flow diagram shown in FIG.
7A, which
corresponds to an embodiment of a portion 620 of the method shown in FIG. 6.
As shown in
FIGS. 7A and 7F, the computing subsystem determines a PEP/LVET ratio derived
from signal
fusion processes applied to the ECG signal and the IPG signal, where the PEP
and LVET for the
user can be determined as described in relation to FIGS. 7C and 7E above. The
PEP/LVET ratio
characterizes an index of left ventricular systolic performance (i.e.,
systolic time ratio, STR) that
is correlated with ejection fraction, which is a measurement of the fraction
of blood leaving the
heart of the user each time it contracts. In particular, a PEP/LVET ratio that
is above a threshold
value can be used by the computing subsystem to diagnose a patient with
systolic heart failure.
For example, a PEP/LVET ratio greater than 0.40 (or another threshold) can
indicate that a
patient has an ejection fraction less than 40% (or another value). The
computing subsystem can
also use the PEP/LVET ratio to phenotype patients. For instance, the PEP/LVET
ratio can be
used to discriminate between the two most common forms of heart failure (e.g.,
reduced ejection
fraction-associated heart failure and preserved ejection fraction-associated
heart failure).
[00106] FIG. 7G depicts a fifth portion of the flow diagram shown in FIG.
7A, which
corresponds to an embodiment of a portion 620 of the method shown in FIG. 6.
As shown in
FIGS. 7A and 7G, the computing subsystem identifies a J-wave position of the
averaged BCG
waveform (or BCG signal), where the J wave corresponds to a deflection in a
signal between a
QRS complex of a cardiac phase and an ST segment of a cardiac phase. The
computing
subsystem can then detect an arrival time of a pulse associated with the J-
wave at least at one of
the left and the right foot of the user, through the set of force sensors of
the system. Then, based
upon the time point corresponding to the J wave position, the arrival time of
the pulse, and a
height of the user, the computing subsystem can generate 727 a pulse transit
time (PTT) and/or
pulse wave velocity (PWV) for the user. In more detail, the computing
subsystem determines the
PTT and the PAT from multiple signals, where, in one embodiment, the PTT is
calculated using
an IPG-derived signal and a BCG-derived signal. In more detail, a peak of the
I-wave is used as a
first temporal maker, and a maximum value of the IPG-derived signal is used as
a second
temporal marker, where the distance between the first and the second temporal
markers is equal
to the PTT. However, other features of the IPG and/or BCG-derived signals can
be used to
determine PTT. For example, the J-wave peak and the X-point of the IPG-
derivative can be used
to determine the PTT, the B-point of the IPG-derivative and the J-wave of the
BCG-derived
31
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
signal can be used to determine PTT, the I-wave peak of the BCG-derived signal
and the X-point
of the IPG-derivative can be used to determine PTT, and/or the I-wave peak of
the BCG-derived
signal and the maximum peak of the IPG-derivative can be used to determine
PTT.
[00107] The PTT characterizes the time it takes for a pulse pressure
waveform to travel
along a portion of an arterial tree (e.g., from the aortic arch to a lower
torso region of the user),
and the PWV characterizes a speed of travel of the pulse pressure waveform.
The computing
subsystem can also locate an I-peak of the averaged BCG waveform, as shown in
FIG. 7G, by
implementing a peak finding algorithm in relation to the J-wave position. The
computing
subsystem can then use the position, amplitude, or other aspect of the I-peak
to derive systolic
temporal parameter values or other parameter values related to health risk.
[00108] FIG. 7A also depicts a portion of a method where the computing
subsystem fuses
signals of multiple types to extract one or more of: a mean arterial pressure,
systolic blood
pressure (SBP), and a diastolic blood pressure (DPB) for the user. In more
detail, the computing
subsystem identifies a pulse rate from at least one of the averaged ECG
waveform, the averaged
IPG waveform, the averaged BCG waveform, and the ensemble waveform. In an
embodiment,
the pulse rate can be determined from peaks of the IPG signal, and a heart
rate ensemble
averaged signal can be generated with a windowing operation, as described
above (e.g., with a
window of -1500 to 500 ms about respective peaks in the IPG signal). The
computing subsystem
then identifies a BCG amplitude from the averaged BCG waveform. Then, the
computing
subsystem transforms 728 the PEP (determined as described above), the PTT
(determined as
described above), the pulse rate, the BCG amplitude, and a user weight derived
from the weight
signal into a cardiac output (CO) value, a systemic vascular resistance (SVR)
value, and a central
venous pressure (CVP) value. Finally, with the CO, SVR, and CVP values, the
computing
subsystem determines 729 a mean arterial pressure (MAP) for the user from a
product of the
cardiac output (CO) value and the systemic vascular resistance (SVR) value
added to the central
venous pressure (CVP) value.
[00109] In relation to pulse rate, the computing subsystem can determine
pulse rate in real
time from any one or more of ECG-derived signals, IPG-derived signals, and BCG-
derived
signals. The computing subsystem can additionally or alternatively determine
pulse rate (i.e.,
average pulse rate determined over the course of a measurement session) from
one or more
32
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
averaged waveforms (i.e., averaged ensemble signals). For instance, if the
window for an
ensemble operation is extended (e.g., to approximately 2x or longer than the
period of an average
pulse), the computing subsystem captures multiple heart beats in a given
ensemble. The average
pulse rate can then be derived by detecting time points of instances of a
characteristic feature
(e.g., peak of an IPG-derived signal, R-peak of an ECG signal) across each
waveform period
used to generate a final ensemble, where the difference between the time
points is used to
calculate pulse rate. In this embodiment, the determined pulse rate only
encompasses beats that
were included in the determination of a respective ensemble averaged waveform,
and is robust in
relation to low-quality and/or low resolution signals. Furthermore, if certain
features are blocked
(by the filtering operations described) due to motion or other artifacts
associated with a
measurement, the features are automatically removed from consideration during
generation of an
ensemble averaged waveform and also pulse rate determination. Thus, the pulse
rate can be
robustly determined from generating an ensemble averaged waveform of one or
more of the
ECG signal, the IPG signal, and the BCG signal.
[00110] In relation to previously described parameters, the computing
subsystem, as shown
in FIG. 7A, also further generates 730 a pulse arrival time (PAT) for the user
from a summation
of the PEP (determined as described above) and the PTT (determined as
described above).
Additionally or alternatively, PAT can be determined as inferred from the ECG-
derived signals,
the BCG-derived signals, and the IPG-derived signals. For instance, the PAT
can be determined
by the computing subsystem based on the R-peak of the ECG-derived signal and a
maximum of
the first derivative of the IPG-derived signal.
[00111] Furthermore, in some embodiments, physiologically-relevant time
intervals (e.g.,
PEP and LVET) determined by the computing subsystem are influenced by pulse
rate. As such,
the computing subsystem can also correct these physiologically-relevant time
intervals based on
the pulse rate determination so that their physiological significance is
properly assessed (e.g., in
relation to generation of appropriate interventions). In one example, a
corrected LVET, can be
generated based on the formula LVET, = 1.5*HR + LVET, where HR is the pulse
rate. In one
example, a corrected PEP, can be generated based on the formula PEP, = 0.4*HR
+ PEP. The
corrected time intervals (e.g., PEP, LVETc) can be determined from real-time
ECG, BCG,
and/or IPG signals, and/or with generation of ensemble averaged waveforms (as
described
above), where corrected and uncorrected time intervals can be used as inputs
to predictive
33
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
models (e.g., predictive models of cardiovascular health risk described in
relation to FIG. 8
below). For instance, the PEP/LVET ratio can be determined using corrected
intervals (e.g.,
PEPc/LVEK), as described above.
[00112] FIG. 7H depicts a sixth portion of the flow diagram shown in FIG.
7A, which
corresponds to an embodiment of a portion 620 of the method shown in FIG. 6.
As shown in
FIGS. 7A and 7G, the computing subsystem modulates 731 one or more of the
averaged ECG
waveform, the averaged IPG waveform, and the averaged BCG waveform with input
temperature signals from the temperature sensor and/or moisture signals from
the moisture
sensor described in relation to the system above. As such, responsive to
contact the left foot and
the right foot of the user, the computing subsystem can generate a temperature
signal and a
humidity signal and modulate a value of at least one of the set of systolic
temporal parameters
based upon the temperature signal and the humidity signal. For instance, the
computing
subsystem can modulate operation due to device changes (e.g., changes in
electrode resistance
due to changes in humidity) and/or physiological changes of the user due to
excessive heat
and/or humidity.
[00113] Also in relation to the system described above, the computing
subsystem can
calculate body impedance, which is correlated with body water content, from
the electrical
signals generated. The computing subsystem can also determine balance of the
user as the user
steps onto the substrate, where the balance analysis can include one or more
of: movement in
multiple directions (e.g., lateral directions, anterior/posterior directions),
center of pressure,
postural sway, sway path, sway velocity, balance index, and any other suitable
components of
the user's balance.
[00114] Furthermore, in some embodiments, any derived parameters (e.g.,
MAP, SV, CO,
systolic time intervals, etc.) can be absolute measurements or relative
measurements (e.g.,
compared to a baseline or other reference measurement). Relative and/or
absolute measurements
can be calibrated against a reference device for improved accuracy. For
example, a derived
stroke volume model can be calibrated for a specific user by collecting data
from a reference
device (e.g., a device operating according to the Fick method, thermodilution
device, impedance
cardiography device, etc.) contemporaneously with collection of data from an
embodiment of the
system described above, in order to improve accuracy in the values of the
parameters determined
34
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
from the embodiment of the system described above.
4. Method ¨ Processing Cardiovascular Health Parameter Values with Risk
Model and
Generating Predictions of Cardiovascular Health State
[00115] FIG. 8 depicts a flowchart of a method for processing
cardiovascular health
parameters with a risk model, in accordance with one or more embodiments. As
shown in FIG. 8,
the computing subsystem generates 841 values of cardiovascular health
parameters during each
measurement session for a user, as described above. In embodiments, as
described above, the
computing subsystem generates time interval and amplitude-derived features.
These features are
used to build models of clinical parameters associated with cardiovascular
health risks. The
computing subsystem can thus transform 842 values of time interval and
amplitude-derived
features into clinically relevant parameters, including stroke volume, cardiac
output, blood
pressure, system vascular resistance, and other parameters. The clinical
parameters can then be
input into trained risk models configured for generating predictions of
cardiovascular health
states of the user(s), where cardiovascular health states can be related to
stable states, worsening
states (e.g., of various forms of heart disease), and/or indeterminate states.
In one example,
outputs of the cardiovascular risk model for a particular user can be
processed with a distance
analysis 843 or another analysis that compares parameters for a particular
user to outputs of the
model associated with cardiovascular health states. The computing subsystem
can then use the
distance analysis or another analysis to return a prediction 844 of the
cardiovascular health state
of the user. In an example, the prediction can indicate decompensation in a
heart failure patient
and the computing subsystem can use the prediction to drive remote
interventions (e.g., for
reduction of unnecessary hospitalizations).
[00116] Additionally, the computing subsystem can include architecture for
predicting and
generating models of disease phenotypes (e.g., disease phenotypes of heart
failure between
systolic and diastolic variants). In another example, the computing subsystem
can transform
stroke volume inputs, systemic vascular resistance inputs, and impedance
inputs into a
hypertension phenotype. Such phenotypes can be used by the computing subsystem
to order to
identify if a user is suffering from a fluid status issue or a blood volume
issue. As described in
relation to intervention provision below, phenotyping can subsequently be used
to more precisely
administer the therapy targeting at underlying mechanisms of undesired health
states.
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
[00117] In relation to generating parameter values and processing parameter
values with
models, the computing subsystem can process combinations of cardiovascular and
other
physiological parameters generated according to methods described above, in
order to generate
predictions. For instance, the computing subsystem can use weight and baseline
impedance
parameters to generate an index of fluid status in addition to generation of
outputs related to
cardiac status. The combination of fluid status and cardiac status information
can be used by the
computing subsystem to augment sensitivity and specificity for certain
conditions where, for
instance, fluid status changes (e.g., related to fluid retention) in
association with cardiac status
changes (e.g., related to deterioration in state) can indicate statuses (e.g.,
related to heart failure,
related to chronic obstructive pulmonary disease, related to chronic kidney
disease, etc.) with
increased sensitivity and specificity. Furthermore, combination of weight
information,
impedance information, and other cardiac data can be used by the computing
subsystem to
determine dry weight (i.e., the normal weight of a patient's body without any
fluid
accumulation). In more detail, the computing subsystem can determine dry
weight upon
assessing fluid status in combination with simultaneous measures of
hemodynamic performance
(MAP, CO, systolic time intervals, etc.). Dry weight assessment is important
in relation to
conditions (e.g., heart failure, kidney disease, etc.), where changes in fluid
can be measured
relative to a dry weight baseline. Furthermore, effective diuresis benefits
from knowledge of a
user's dry weight.
[00118] In another example, the computing subsystem can generate model
outputs based on
body weight to improve a user's cardiac status in an actionable feedback loop.
In more detail, if a
user is determined to have high blood pressure due to excessive body weight,
the computing
subsystem can generate an associated prediction and generate intervention
protocols (e.g., a
weight loss program, control instructions for an exercise regimen administered
by connected
exercise equipment, control instructions for a connected dispenser containing
weight loss
supplements, etc.) for the user. The interventions can also include tailored
modifications to
operation of the systems described above, where the system measures body
weight and cardiac
status for the user simultaneously and provides such information to the user
or another associated
entity to promote improvements to health statuses of the user. In more detail,
simultaneous
measurement of weight, in combination with height information (e.g., as input
by the user or
another entity, as determined in another manner) can be used by the computing
subsystem to
36
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
generate indices of cardiovascular function normalized to the user's body type
(e.g., in terms of
BMI, body surface area, or other derivative measures of body type). In an
example, stroke
volume and cardiac output can be calculated and transformed into a stroke
index and a cardiac
index, respectively, by dividing stroke volume and cardiac output by body
surface area (as
determined from height and weight using the Du Bois formula, using a Haycock
method, etc.). In
particular, cardiac index is a hemodynamic parameter that relates the cardiac
output (CO) from
the left ventricle in one minute to body surface area, and thus relates heart
performance to a size
of a user.
[00119] FIG. 9 depicts a flowchart of longitudinal monitoring of
cardiovascular health of a
user, in accordance with one or more embodiments. As shown in FIG. 9,
responsive to
contacting the feet of the user, the system can generate 910 passive and
active electrical signals
(e.g., ECG and IPG signals) from a set of electrodes, according to embodiments
described above.
Responsive to contacting the feet of the user, the system can also generate
915 force-derived
signals (e.g., weight signals and BCG signals), according to embodiments
described above. The
computing subsystem can then generate 920 a set of cardiovascular health
parameters with a
signal fusion operation according to embodiments derived above, where the
cardiovascular
health parameters are processed 940 by the computing subsystem with a risk
model. The
computing subsystem can the return outputs 950 of the cardiovascular risk
model at multiple
time points associated with different measurement sessions for the user. The
outputs associated
with different time points can be processed 951 with a longitudinal analysis,
in order to generate
insights into changes in the user's health condition over time. Longitudinal
analyses can be used
to promote interventions that are more tailored to the user's specific
condition. For instance, the
computing subsystem can generate instructions for automatic medication
adjustments for a user.
In one specific example, the computing subsystem's outputs can be used for
automatic titration
of diuretic dosing for a heart failure patient. In other examples, automatic
medication adjustment,
as determined using outputs of the computing subsystem, can be applied to
other chronic disease
conditions (e.g., hypertension).
4. Conclusion
[00120] The system and method(s) described can confer benefits and/or
technological
improvements, several of which are described herein. For example, the system
and method(s)
37
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
can produce fused or composite data that characterize complex physiological
behavior, which is
analyzed to provide insights into improving user health interventions. Such
data structures and
processing methods can be used to efficiently generate comparisons across a
large amount of
data from different sources, for a large number of users over time.
[00121] The system and method(s) can further employ non-typical use of
sensors. For
instance, the system and method(s) can employ sensor arrays including
different types of sensors
in a spatial and structural configuration that enables significant
improvements in increasing SNR
for extremely noise biometric signals taken from non-traditional body regions.
As such, the
system and method(s) can provide several technological improvements.
[00122] The foregoing description of the embodiments has been presented for
the purpose
of illustration; it is not intended to be exhaustive or to limit the patent
rights to the precise forms
disclosed. Persons skilled in the relevant art can appreciate that many
modifications and
variations are possible in light of the above disclosure.
[00123] Some portions of this description describe the embodiments in terms
of algorithms
and symbolic representations of operations on information. These algorithmic
descriptions and
representations are commonly used by those skilled in the data processing arts
to convey the
substance of their work effectively to others skilled in the art. These
operations, while described
functionally, computationally, or logically, are understood to be implemented
by computer
programs or equivalent electrical circuits, microcode, or the like.
Furthermore, it has also proven
convenient at times, to refer to these arrangements of operations as modules,
without loss of
generality. The described operations and their associated modules may be
embodied in software,
firmware, hardware, or any combinations thereof.
[00124] Any of the steps, operations, or processes described herein may be
performed or
implemented with one or more hardware or software modules, alone or in
combination with
other devices. In one embodiment, a software module is implemented with a
computer program
product comprising a computer-readable medium containing computer program
code, which can
be executed by a computer processor for performing any or all of the steps,
operations, or
processes described.
[00125] Embodiments may also relate to an apparatus for performing the
operations herein.
This apparatus may be specially constructed for the required purposes, and/or
it may comprise a
38
CA 03116846 2021-04-16
WO 2020/081471 PCT/US2019/056160
general-purpose computing device selectively activated or reconfigured by a
computer program
stored in the computer. Such a computer program may be stored in a non-
transitory, tangible
computer readable storage medium, or any type of media suitable for storing
electronic
instructions, which may be coupled to a computer system bus. Furthermore, any
computing
systems referred to in the specification may include a single processor or may
be architectures
employing multiple processor designs for increased computing capability.
[00126] Embodiments may also relate to a product that is produced by a
computing process
described herein. Such a product may comprise information resulting from a
computing process,
where the information is stored on a non-transitory, tangible computer
readable storage medium
and may include any embodiment of a computer program product or other data
combination
described herein.
[00127] Finally, the language used in the specification has been
principally selected for
readability and instructional purposes, and it may not have been selected to
delineate or
circumscribe the patent rights. It is therefore intended that the scope of the
patent rights be
limited not by this detailed description, but rather by any claims that issue
on an application
based hereon. Accordingly, the disclosure of the embodiments is intended to be
illustrative, but
not limiting, of the scope of the patent rights, which is set forth in the
following claims.
39
