U.S. patent application number 17/694789 was filed with the patent office on 2022-08-25 for detection of asynchrony.
The applicant listed for this patent is ResMed Pty Ltd. Invention is credited to Nigel Hamilton LOVELL, Qestra Camille MULQUEENY, Stephen James REDMOND.
Application Number | 20220265944 17/694789 |
Document ID | / |
Family ID | 1000006330434 |
Filed Date | 2022-08-25 |
United States Patent
Application |
20220265944 |
Kind Code |
A1 |
MULQUEENY; Qestra Camille ;
et al. |
August 25, 2022 |
DETECTION OF ASYNCHRONY
Abstract
A controller or processor(s) implements detection of respiratory
related conditions, such as asynchrony, associated with use of a
respiratory treatment apparatus or ventilator. Based on data
derived from sensor signals associated with the respiratory
treatment, the detector may evaluate a feature set of detection
values to determine whether or not an asynchrony occurs in a breath
of the patient's respiratory cycle such as by comparing the values
against a set of thresholds. Different events may also be
identified based on the particular feature set and threshold(s)
involved in the detection processing. Automated determination of
feature sets may also be implemented to design different asynchrony
event classifiers. The methodologies may be implemented by
computers or by respiratory treatment apparatus. The detection of
such asynchrony events can then also serve as part of control logic
for automated adjustments to the control parameters of the
respiratory treatment generated by the respiratory treatment
apparatus.
Inventors: |
MULQUEENY; Qestra Camille;
(Maroubra, AU) ; REDMOND; Stephen James; (South
Coogee, AU) ; LOVELL; Nigel Hamilton; (Coogee,
AU) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
ResMed Pty Ltd |
Bella Vista |
|
AU |
|
|
Family ID: |
1000006330434 |
Appl. No.: |
17/694789 |
Filed: |
March 15, 2022 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15172543 |
Jun 3, 2016 |
11291784 |
|
|
17694789 |
|
|
|
|
13264012 |
Oct 12, 2011 |
9392964 |
|
|
PCT/AU2010/000457 |
Apr 22, 2010 |
|
|
|
15172543 |
|
|
|
|
61171606 |
Apr 22, 2009 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
A61M 2205/52 20130101;
A61B 5/091 20130101; A61M 2230/65 20130101; A61B 5/7264 20130101;
A61B 5/087 20130101; A61B 5/085 20130101; A61M 2016/0036 20130101;
A61M 2230/005 20130101; A61M 16/06 20130101; A61M 2205/502
20130101; A61M 16/026 20170801; A61M 2016/0027 20130101; A61M
2016/003 20130101; A61M 16/0875 20130101; A61M 16/04 20130101; A61M
16/0066 20130101; A61M 16/0003 20140204; A61M 2230/42 20130101;
A61M 2230/46 20130101; A61M 2210/1014 20130101 |
International
Class: |
A61M 16/00 20060101
A61M016/00; A61B 5/085 20060101 A61B005/085; A61B 5/091 20060101
A61B005/091; A61B 5/087 20060101 A61B005/087; A61M 16/04 20060101
A61M016/04; A61M 16/06 20060101 A61M016/06; A61M 16/08 20060101
A61M016/08 |
Claims
1. A respiratory treatment apparatus for estimating respiratory
resistance and compliance based on measures of flow and pressure
comprising: one or more sensors to generate signals representative
of flow and pressure; and a processor, coupled with the one or more
sensors, the processor being configured to control: detecting a
portion of expiration of a breathing cycle from data of the flow
signal; and calculating a resistance value and compliance value
with flow, pressure and volume measures that correspond to the
detected portion of expiration.
2. The respiratory treatment apparatus of claim 1 wherein the
detected portion of expiration begins at an expiration cycle and
ends when an expired tidal volume for the expiration cycle exceeds
a limit in a range of about 85 to 90 percent.
3. The respiratory treatment apparatus of claim 2 wherein the
calculating comprises a multiple linear regression process with
data representing the flow, pressure and volume measures.
4. The respiratory treatment apparatus of claim 3 wherein the
resistance value and compliance value are calculated in a
breath-by-breath process.
5. The respiratory treatment apparatus of claim 2 wherein the
processor is further configured to control an assessment of
accuracy of the calculated resistance and compliance values.
6. The respiratory treatment apparatus of claim 5 wherein the
assessment of accuracy comprises calculating a coefficient of
determination and comparing it to a threshold.
7. The respiratory treatment apparatus of claim 1 wherein the
processor is further configured to control determining a PEEP
control parameter based on a plurality of compliance values
determined by the calculating of the processor.
8. The respiratory treatment apparatus of claim 7 wherein the
processor controls a repeated change to a preset PEEP control
parameter during which a plurality of pressure and flow measures
are determined, and wherein the plurality of compliance values are
determined based on the plurality of pressure and flow
measures.
9. The respiratory treatment apparatus of claim 8 wherein the
determining of the PEEP control parameter comprises detecting an
inflection point from data representing the plurality of compliance
values.
10. The respiratory treatment apparatus of claim 1 wherein the
processor is further configured to control determining a maximum
pressure support limit based on a plurality of compliance values
determined by the calculating of the processor.
11. The respiratory treatment apparatus of claim 10 wherein the
processor controls a repeated change to a preset PEEP control
parameter during which a plurality of pressure and flow measures
are determined, and wherein the plurality of compliance values are
determined based on the plurality of pressure and flow
measures.
12. The respiratory treatment apparatus of claim 8 wherein the
determining of the maximum pressure support limit comprises
detecting an inflection point from data representing the plurality
of compliance values.
13. The respiratory treatment apparatus of claim 4 further
comprising a flow generator, coupled with the processor, to
generate a flow of breathable gas at pressures above atmospheric to
a patient interface based on control signals from the
processor.
14. A method for estimating respiratory resistance and compliance
based on measures of flow and pressure comprising: generating, with
one or more sensors, signals representative of flow and pressure;
and detecting, in a processor, a portion of expiration of a
breathing cycle from data of the flow signal; and calculating, in
the processor, a resistance value and compliance value with flow,
pressure and volume measures that correspond to the detected
portion of expiration.
15. The method of claim 14 wherein the detected portion of
expiration begins at an expiration cycle and ends when an expired
tidal volume for the expiration cycle exceeds a limit in a range of
about 85 to 90 percent.
16. The method of claim 15 wherein the calculating comprises a
multiple linear regression process with data representing the flow,
pressure and volume measures.
17. The method of claim 16 wherein the resistance value and
compliance value are calculated in a breath-by-breath process.
18. The method of claim 15 wherein the processor is further
configured to control an assessment of accuracy of the calculated
resistance and compliance values.
19. The method of claim 18 wherein the assessment of accuracy
comprises calculating a coefficient of determination and comparing
it to a threshold.
20. The method of claim 14 further comprising, determining, in the
processor a PEEP control parameter based on a plurality of
compliance values determined by the calculating of the
processor.
21. The method of claim 20, further comprising, repeatedly,
changing, in the processor, a preset PEEP control parameter during
which a plurality of pressure and flow measures are determined, and
wherein the plurality of compliance values are determined based on
the plurality of pressure and flow measures.
22. The method of claim 21 wherein the determining of the PEEP
control parameter comprises detecting an inflection point from data
representing the plurality of compliance values.
23. The method of claim 14, further comprising, determining, in the
processor, a maximum pressure support limit based on a plurality of
compliance values determined by the calculating of the resistance
value and compliance value.
24. The method of claim 23 further comprising repeatedly changing a
preset PEEP control parameter during which a plurality of pressure
and flow measures are determined, and wherein the plurality of
compliance values are determined based on the plurality of pressure
and flow measures.
25. The method of claim 21 wherein the determining of the maximum
pressure support limit comprises detecting an inflection point from
data representing the plurality of compliance values.
26. The method of claim 17 further comprising controlling, with the
processor, a flow generator to generate a flow of breathable gas at
pressures above atmospheric to a patient interface.
27. The method of claim 14 further comprising controlling, with the
processor, an operation of a flow generator based on the calculated
resistance value and compliance value.
Description
CROSS REFERENCE TO RELATED APPLICATIONS
[0001] The present application is a continuation of U.S. patent
application Ser. No. 15/172,543, filed on Jun. 3, 2016, which is a
continuation of U.S. patent application Ser. No. 13/264,012, filed
on Oct. 12, 2011, now U.S. Pat. No. 9,392,964, which is a national
phase entry under 35 U.S.C. .sctn. 371 of International Application
No. PCT/AU2010/000457, filed Apr. 22, 2010, published in English,
which claims priority from U.S. Provisional Patent Application No.
61/171,606, filed Apr. 22, 2009, all of which are incorporated
herein by reference.
FIELD OF THE TECHNOLOGY
[0002] The present technology relates to methods and apparatus for
the detection of respiratory treatment related conditions or events
such as the conditions indicative of asynchrony.
BACKGROUND OF THE TECHNOLOGY
[0003] Ensuring quality interaction between a patient and their
respiratory treatment apparatus can be significant. For example,
quality interaction between a patient and their ventilator is
critical to minimize the work of breathing of the patient.
Increased work of breathing caused by patient-ventilator asynchrony
is associated with negative patient outcomes and has been shown to
be highly prevalent; studies have shown up to 47% of ventilator
delivered breaths may be asynchronous.
[0004] With proper supervision and adjustment of ventilator
settings, asynchrony may be manually mitigated by the clinician.
However, monitoring asynchrony is a challenge. Knowledge of its
existence may require high resolution and real-time display of
airway pressure delivered by the ventilator, the patient's airflow
and a measure of inspiratory activity, such as diaphragmatic EMG or
esophageal pressure measured invasively via a balloon catheter.
Furthermore, expert visual interpretation of these signals can be
critical to diagnose the quality of patient-ventilator
interaction.
[0005] It may be desirable to develop further methods and devices
for automating detection of asynchrony that may improve respiratory
treatment apparatus.
SUMMARY OF THE TECHNOLOGY
[0006] Aspects of the present technology involve methods and
apparatus for a detection of asynchrony.
[0007] Still further aspects of the technology involve methods and
apparatus for development of devices for detection of
asynchrony.
[0008] In one such example embodiment, an automated processing
method is provided for detection of asynchrony of a respiratory
treatment apparatus that controls a delivery of a synchronized
respiratory treatment. The method may include determining detection
values of a feature set with data derived from signals of at least
one sensor coupled with the respiratory treatment apparatus. The
method may further include comparing with a processor the feature
set of detection values with a set of stored data thresholds. The
method may also include determining with the processor an
occurrence of an asynchrony event between the respiratory treatment
apparatus and a patient respiratory cycle based on the
comparison.
[0009] In another embodiment, an apparatus for detection of
asynchrony between a synchronized respiratory treatment and a
respiratory cycle is provided. The apparatus may include a memory
containing data representing a feature set of detection values
derived from signals of at least one sensor coupled with the
respiratory treatment apparatus. The apparatus may also include a
processor to access the data. The processor may be configured to
control a comparison of the data of the feature set of detection
values with a set of stored data thresholds of the memory. It may
also be configured to control a determination of an occurrence of
an asynchrony event between the respiratory treatment apparatus and
a patient respiratory cycle based on the comparison. In addition,
it may be configured to control a generation of an output
representing the occurrence of the asynchrony event.
[0010] In some embodiments, the feature set of detection values may
include any one or more of a respiratory rate based feature, a
respiratory volume based feature, a respiratory mechanics based
feature, an expiratory flow morphology based feature.
[0011] In some embodiments, the apparatus or method may also
determine or identify a plurality of distinct asynchrony events
based on the comparison. The asynchrony events may be any one or
more of an expiratory ineffective effort event, a post-triggering
effort event, a double triggering event, an autotriggering event, a
late triggering event, an early cycling event, a late cycling
event, and/or an inspiratory ineffective effort event.
[0012] In an example embodiment, the feature set of detection
values for detecting an inspiratory ineffective effort event may
include (a) power of a piecewise bilinear approximation of a
remainder of expiration after a location of a maximum expiratory
flow, (b) a distance between a maximum and minimum values of a
moving average expiratory flow, (c) an integral of a rectified and
de-trended moving average of expiratory flow, (d) an inspiratory
time constant (e) and a fraction of said distance and a peak
expiratory flow.
[0013] In some embodiments, the processor may be or be part of a
controller of the respiratory treatment apparatus, which may be a
ventilator, and the sensor may be a flow sensor. The controller may
then be configured to detect a respiratory cycle with the signal of
the flow sensor and to generate flow generator control signals for
producing the respiratory treatment.
[0014] In some embodiments of the method or apparatus, the feature
set are selected by a selecting process such that the feature set
comprises a subset of a superset of features. Such a selecting
process may include evaluating, with a processor, values of the
superset for known asynchronous events occurring in data of a
plurality of breaths established with a plurality of respiratory
treatment apparatus. The evaluating process may involve calculating
posterior-probabilities with values of the superset by
Parzen-window estimation, wherein groups of values of the superset
are selected by iteratively including and removing values. The
evaluation may also include detecting asynchrony events with the
selected subset of features and comparing the events detected with
known events data to test or assess the performance of the
subset.
[0015] In some embodiments, the apparatus may include a patient
interface to carry a flow of breathable gas to a patient. It may
also include a flow generator coupled with the patient interface to
generate a flow of the breathable gas through the patient
interface. The apparatus may also include a transducer, such as a
pressure transducer or differential pressure transducer. The
transducer may provide a signal indicative of patient flow through
the patient interface. The processor may serve as a controller of
the flow generator so that the controller is configured to detect a
respiratory cycle with the signal of the flow sensor and to
generate flow generator control signals for producing the
respiratory treatment.
[0016] In some embodiments, a system is provided for detection of
asynchrony between a synchronized respiratory treatment and a
respiratory cycle. The system may include a processing means for
processing data derived from one or more pressure transducer
signals from use of respiratory treatment apparatus. The processing
means may be configured for (a) comparing a feature set of
detection values calculated from the data with a set of stored data
thresholds, and (b) determining an occurrence of an asynchrony
event between the respiratory treatment apparatus and a patient
respiratory cycle based on the comparison. The system may also
include classification means for selecting the feature set such
that the feature set comprises a subset of a superset of features,
wherein the selecting comprises evaluating values of the superset
for known asynchronous events occurring in data of a plurality of
breaths established with a plurality of respiratory treatment
apparatus. The processing means of the system may also be
configured for calculating the detection values of the feature set
with the data derived from signals of at least one sensor.
[0017] In some embodiments of such a system, the feature set of
detection values may be two or more of features of a group of
features consisting of (a) a respiratory rate based feature; (b) a
respiratory volume based feature; (b) a respiratory mechanics based
feature; (d) an expiratory flow morphology based feature.
[0018] In further embodiments of such a system, the occurrence of
the asynchrony event may be an asynchrony event of a group of
asynchrony events consisting of (a) an expiratory ineffective
effort event, (b) a post-triggering effort event, (c) a double
triggering event, (d) an autotriggering event (e) a late triggering
event, (f) an early cycling event, and (g) a late cycling
event.
[0019] The system may also include an interface means to carry a
flow of breathable gas, a flow generation means, coupled with the
interface means, for generating the breathable gas, and a flow
sensor for generating the pressure transducer signals. In such a
system, the processing means may be further configured for
controlling the flow generation means to provide a synchronized
respiratory treatment, such as respiratory support ventilation.
[0020] In some embodiments, the methodologies may be contained on
information bearing medium such as a memory. For example, an
information-bearing medium may have processor-readable information
thereon. The processor-readable information may control an
apparatus for detection of asynchrony between a synchronized
respiratory treatment and a respiratory cycle. The
processor-readable information may then include control
instructions to access detection values of a feature set with data
derived from signals of at least one sensor coupled with the
respiratory treatment apparatus, to compare the feature set of
detection values with a set of stored data thresholds, and to
determine an occurrence of an asynchrony event between the
respiratory treatment apparatus and a patient respiratory cycle
based on the comparison.
[0021] In some embodiments of the present technology, a respiratory
treatment apparatus estimates respiratory resistance and compliance
based on measures of flow and pressure. The apparatus may include
one or more sensors to generate signals representative of flow and
pressure. The apparatus may also include a processor, coupled with
the one or more sensors. The processor may be configured to control
detecting a portion of expiration of a breathing cycle from data of
the flow signal. It may also be configured to control calculating a
resistance value and compliance value with flow, pressure and
volume measures that correspond to the detected portion of
expiration.
[0022] In some embodiments, the detected portion of expiration may
a portion that begins at an expiration cycle and ends when an
expired tidal volume for the expiration cycle exceeds a limit in a
range of about 85 to 90 percent. The compliance and resistance
values may optionally be calculated by a multiple linear regression
process with data representing the flow, pressure and volume
measures. Moreover, this may be calculated in a breath by breath
process. The processor may be further configured to control an
assessment of accuracy of the determined resistance and compliance
values. The assessment of accuracy may involve calculating a
coefficient of determination and comparing it to a threshold.
[0023] In some embodiment, the processor of the respiratory
treatment apparatus may be further configured to control
determining a PEEP control parameter based on a plurality of
compliance values determined by the processor. In such a process,
the processor may control a repeated change, such as a step
reduction, to a preset PEEP control parameter during which a
plurality of pressure and flow measures are determined. The
plurality of compliance values may then be determined based on the
plurality of pressure and flow measures. This determination may
involve detecting an inflection point from data representing the
plurality of compliance values.
[0024] In some embodiments, the processor of the respiratory
treatment apparatus may also be configured to control determining a
maximum pressure support limit based on a plurality of determined
compliance values. In such a process, the processor may controls a
repeated change, such as a stepped reduction, to a preset PEEP
control parameter during which a plurality of pressure and flow
measures are determined. The processor then may determine the
plurality of compliance values based on the plurality of pressure
and flow measures. Optionally, the determining of the maximum
pressure support limit may involve detecting an inflection point
from data representing the plurality of compliance values. In some
such embodiments, the respiratory treatment apparatus may also
include a flow generator, coupled with the processor, to generate a
flow of breathable gas at pressures above atmospheric to a patient
interface based on control signals from the processor.
[0025] Further embodiments and features of the technology will be
apparent from the following detailed disclosure, claims and
drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
[0026] The present technology is illustrated by way of example, and
not by way of limitation, in the figures of the accompanying
drawings, in which like reference numerals refer to similar
elements including:
[0027] FIG. 1 is an block diagram of an example asynchrony detector
of the present technology;
[0028] FIG. 2 is an flow chart with an example methodology that may
be implemented in some embodiments of the asynchrony detector of
FIG. 1;
[0029] FIG. 3 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to an expiratory
ineffective effort asynchrony event that may be detected with the
present asynchrony detection technology;
[0030] FIG. 4 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to a post-triggering
effort asynchrony event that may be detected with the present
asynchrony detection technology;
[0031] FIG. 5 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to a double triggering
asynchrony event that may be detected with the present asynchrony
detection technology;
[0032] FIG. 6 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to an autotriggering
asynchrony event that may be detected with the present asynchrony
detection technology;
[0033] FIG. 7 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to a late triggering
asynchrony event that may be detected with the present asynchrony
detection technology;
[0034] FIG. 8 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to an early cycling
asynchrony event that may be detected with the present asynchrony
detection technology;
[0035] FIG. 9 is a graph illustrating flow, pressure and
transdiaphragmatic pressure corresponding to a late cycling
asynchrony event and an inspiratory ineffective effort asynchrony
event that may be detected with the present asynchrony detection
technology;
[0036] FIG. 10 is an diagram for a state machine to detect
respiration related states in some embodiments of the
technology;
[0037] FIG. 11 is an example respiratory treatment apparatus with
an asynchrony detection controller;
[0038] FIG. 12 is a block diagram illustrating suitable components
that may be implemented as a processing apparatus or controller for
embodiments of the present technology;
[0039] FIG. 13 is a screenshot of an example user interface
suitable for manual classification of asynchrony events;
[0040] FIG. 14 includes two graphs illustrating steps of an
automated process or analysis of expiratory flow and an output
signal with an indicator of an expiratory ineffective effort
asynchrony event from which features may be determined;
[0041] FIG. 15 includes two graphs illustrating further steps of an
automated process or analysis of expiratory flow and an output
signal with an indicator of an expiratory ineffective effort
asynchrony event from which features may be determined;
[0042] FIG. 16 is a graph illustrating additional aspects of an
automated analysis in the determination of some features suitable
for detecting expiratory ineffective effort asynchrony events;
[0043] FIG. 17 is a graph illustrating further aspects of an
automated analysis in the determination of some features suitable
for detecting expiratory ineffective effort asynchrony events as
well as illustrating some of the features;
[0044] FIG. 18 is a graph illustrating aspects of an automated
analysis in the determination of some features suitable for
detecting double triggering asynchrony events as well as
illustrating some of the features; and
[0045] FIG. 19 is a graph illustrating example titration methods
based on determined respiratory mechanics values of the present
technology.
DETAILED DESCRIPTION
[0046] The present technology involves automated detection of
asynchrony events related to the delivery of respiratory treatment.
Asynchrony involves situations when the actual state of the
patient's respiration is not in phase with an automated respiratory
treatment that is intended to be synchronized with the patient's
respiration. As previously mentioned, asynchrony can occur in
patient-ventilator interaction. The current technology may be
employed to develop or implement asynchrony detectors. Such
detectors may serve to determine whether or not one or more
asynchrony events has or is occurring. The detectors may also be
implemented for identification or classification of one or more
different types of asynchrony events.
[0047] An asynchrony detection embodiment is illustrated in FIG. 1.
Typically, asynchrony detector 100, such as a programmed processor,
may access data or signals from one or more sensors associated with
a respiratory treatment apparatus that provides a synchronized
respiratory treatment. For example, the data may be based on flow
f(t) and/or pressure p(t) signals taken from a flow sensor and/or
pressure sensor of such an apparatus. Based on this data the
detector 100 may then detect or identify asynchrony events.
[0048] For example, in some embodiments the detector may implement
a methodology illustrated in the flow chart of FIG. 2. At 220, a
processor may access detection values of a feature set with data
derived from signals of at least one sensor coupled with the
respiratory treatment apparatus. The processor may then evaluate
the detection values. For example, at 222, the processor may then
compare the feature set of detection values with a set of stored
data thresholds. At 224, the processor may determine an occurrence
of an asynchrony event between the respiratory treatment apparatus
and a patient respiratory cycle based on the evaluation or
comparison.
[0049] Information of the detected events may be recorded in the
memory of a device and/or output to a display apparatus. Similarly,
it may be transmitted (e.g., via wired or wireless communication)
for review or analysis with other processing apparatus. The stored
or recorded data representing information of the events may, for
example, include an identification of event type, time of
occurrence, count of all asynchrony events for a period of use,
count of each type of detected event for a period of use, etc.
Other data of the sensors of the apparatus may be associated with
the asynchrony event(s) and may also be stored for further analysis
in conjunction with the asynchrony event information. For example,
the flow and/or pressure signals data for each patient breathing
cycle identified as asynchronous may also be stored.
[0050] In some embodiments, asynchrony events may be scored and
output by the processor. The score may be used as a performance
index for a particular respiratory treatment apparatus or treatment
session with the apparatus. For example, the score may be
represented as a ratio of a number of asynchrony events per time
period (e.g., a treatment period) or per ventilator breaths
delivered.
[0051] The processor may also be configured to analyze the scored
events to implement or suggest changes to the treatment control of
the respiratory treatment apparatus. For example, as discussed in
more detail herein, detected asynchrony events may serve as part of
the processing control logic to automatically change certain
control parameters of a pressure treatment delivered by a
respiratory treatment apparatus. Similarly, detected asynchrony
events may serve as part of the processing control logic to
automatically output messages to suggest that manual changes to
certain control parameters of a pressure treatment delivered by a
respiratory treatment apparatus be changed.
Example Detectable Asynchrony Events
[0052] The asynchrony detector 100 in some embodiments may be
configured to detect any one or more of the different types of
asynchrony events graphically illustrated in FIGS. 3-9. As further
discussed herein, the illustrated events may be considered
expiratory ineffective effort events, post-triggering effort
events, double triggering events, autotriggering events, late
triggering events, early cycling events, late cycling events and/or
inspiratory ineffective effort events.
[0053] FIG. 3. is a graphic illustration of an example expiratory
ineffective effort event. The graph has a plot of a flow signal,
airway pressure ("Pao") signal and transdiaphragmatic pressure
("Pdi") signal during a normal breath in section A and a breath
with an expiratory ineffective effort event in section B. A
positive swing in the Pdi indicates inspiratory effort by the
patient. As illustrated, the breath of section B has a predominant
perturbation on the flow signal during mid-expiration. This
asynchrony event may also be characterized as a positive Pdi tidal
swing during expiration not followed by inspiratory support from a
respiratory treatment apparatus or ventilator.
[0054] FIG. 4. is a graphic illustration of an example
post-triggering effort event. The graph has a plot of a flow
signal, airway pressure Pao signal and transdiaphragmatic pressure
Pdi signal during a normal breath in section A and a breath with an
post-triggering effort event in section B. The graph illustrates a
positive Pdi tidal swing beginning after the respiratory treatment
apparatus or ventilator has triggered and occurring while the
apparatus is delivering inspiratory support that results in an
increase in flow. The onset of inspiratory support in section B in
this example is machine-triggered.
[0055] FIG. 5. is a graphic illustration of an example double
triggering event. The graph plots a flow signal, airway pressure
Pao signal and transdiaphragmatic pressure Pdi signal during two
normal breaths in sections A and C and a breath with a double
triggering event in section B. Two mechanical breaths over a single
Pdi tidal swing are delivered. The first cycle is patient-triggered
and separated from the second by a very short expiratory time that
may be defined as being less than about one-half of the mean
inspiratory time (over 5 mins).
[0056] FIG. 6. is a graphic illustration of an example
autotriggering event. The graph plots a flow signal, airway
pressure Pao signal and transdiaphragmatic pressure Pdi signal
during two normal breaths in section A and C and a sequence of
autotriggering events in section B. The event may involve the
delivery of inspiratory support by the respiratory treatment
apparatus or ventilator without a concomitant swing in Pdi that
would otherwise indicate patient effort. This can be taken to
indicate that the device delivered a breath that was not triggered
by the patient.
[0057] FIG. 7. is a graphic illustration of an example late
triggering event. The graph plots a flow signal, airway pressure
Pao signal and transdiaphragmatic pressure Pdi signal during a
normal breath in section B and a breath with an late triggering
event in section A. The graph shows a positive tidal swing in Pdi
during expiration that is not followed by inspiratory pressure
support from the device until more than 150 ms has elapsed. The
increase in Pdi indicates that the start of patient effort was more
than 150 ms after the device delivered pressure support.
[0058] FIG. 8. is a graphic illustration of an example early
cycling event. The graph plots a flow signal, airway pressure Pao
signal and transdiaphragmatic pressure Pdi signal during a normal
breath in section A and a breath with the early cycling event in
section B. A cessation or termination of inspiratory support by the
device (e.g., advancing to the device's expiratory state) prior to
the end of the patient's Pdi tidal swing can indicate that the
patient's effort is prolonged beyond the device's mechanical
breath.
[0059] FIG. 9. is a graphic illustration of an example late cycling
event. The graph plots a flow signal, airway pressure Pao signal
and transdiaphragmatic pressure Pdi signal during a breath with the
late cycling event. The event may be characterized as a
prolongation of the device's inspiratory time or inspiratory state
that is beyond the end of the patient's inspiratory effort. For
example, it may be considered to exist when the time T.sub.b is
greater than half of the time T.sub.a (e.g., T.sub.b>T.sub.a/2)
where: [0060] T.sub.b is the prolongation of the device inspiratory
time beyond the end of patient effort; and [0061] T.sub.a is the
duration of patient inspiratory effort.
[0062] FIG. 9 also illustrates an inspiratory ineffective effort
event. This event is shown occurring on the time scale between two
and three seconds. The graph of this time period shows the
transdiaphragmatic pressure Pdi signal increasing and decreasing
without a related or corresponding change in the airway pressure
Pao signal.
Example Features Sets
[0063] As previously mentioned, the detector 100 in some
embodiments may be configured to detect the aforementioned
different types of asynchrony events based on feature sets of
detection values or variables, for example, by detecting a certain
pattern or patterns with detection values of one or more feature
sets. These sets may include one or more features calculated or
determined from the sensor signals that detect information about
patient respiration and/or information about the respiratory
treatment apparatus. For example, embodiments of the technology may
be implemented to utilize, calculate values for or determine the
features based on the measurement signals as illustrated in the
graphs of FIGS. 3-9. Nevertheless, in still further embodiments,
the features may be based on non-invasive measurement signals of a
pressure sensor and/or flow sensor rather than an invasive
transdiaphragmatic pressure sensor.
[0064] Example features of the set of detection values may include,
respiratory rate based features, volume based features, respiratory
mechanics based features, expiratory flow morphology based
features, power based power based features, etc., derivable with
pressure and/or flow data. For example, as discussed in more detail
herein, such detection variables may be any one or more of the
following detection variables: RR breath, RR ratio, IE ratio,
Vol.sub.I, Vol.sub.E, breathLeak, Vol ratio, PEEPii, R.sub.i,
C.sub.i, Tau.sub.i, PEEPie, Re, Ce, Tau.sub.e, InterpMinima,
InterpMinima-2/3, PW linear approx power, PW vol deviation, PW dist
flow maxmin, PW distance ratio. Such features may be calculated on
a breath by breath basis as detected from the signals. These
features are summarized below and are contained in the following
Table II. Additional features are discussed in more detail
below.
TABLE-US-00001 TABLE II Feature List 1 RR breath Respiratory Rate
based 2 RR ratio features 3 IE ratio 4 Vol.sub.I Volume based
features 5 Vol.sub.E 6 breathLeak 7 Vol ratio 8 PEEPi.sub.i
Respiratory Mechanics 9 R.sub.i based features 10 C.sub.i 11
Tau.sub.i 12 PEEPi.sub.e 13 R.sub.e 14 C.sub.e 15 Tau.sub.e 16
InterpMinima Expiratory Flow 17 InterpMinima-2/3 Morphology based
18 PW linear approx power features 19 PW vol deviation 20 PW dist
flow maxmin 21 PW distance ratio
[0065] Respiratory rate based features may include the following:
[0066] RR.sub.breath=1/T.sub.breath. This respiratory rate feature
may be determined on a per breath basis and values thereof can be
calculated as the inverse of the period. [0067] RR
ratio=RR.sub.breath/RR.sub.mean. This feature based on the
respiratory rate may be determined as the ratio of the respiratory
rate of the breath to the mean respiratory rate calculated over a
period of time such as 5 minutes. [0068] IE
ratio=T.sub.i/T.sub.breath. This feature based on the respiratory
rate may be determined as the ratio of inspiratory time for the
breath to the period of the breath.
[0069] Volume based features for each breath may include the
following: [0070] Vol.sub.I--This determined feature may be
considered an absolute volume inhaled in a breath. It may be
determined as the integral of the flow during inspiration. [0071]
Vol.sub.E--This determined feature may be considered an absolute
volume exhaled in a breath. It may be determined as the integral of
the flow during expiration. breathLeak=Vol.sub.I-Vol.sub.E. This
determined feature may be considered the volume leaked between
inspiration and expiration. It may be determined as the difference
between the absolute volume inhaled and the absolute volume
exhaled. [0072] Vol ratio=Vol.sub.I/Vol.sub.E. This determined
feature may be considered a ratio of inspiratory and expiratory
volumes of a breath. It may be calculated as the absolute volume
inhaled divided by the absolute volume exhaled.
[0073] Respiratory Mechanics based features for each breath may
also be determined. For example, mechanics parameters can be
calculated for both inspiratory and expiratory phases (denoted by
subscript i and e respectively) of each breath by a multiple linear
regression fit of pressure, flow and volume data to a first order
single compartment lung model as follows:
Pressure = Flow * R + Volume / C + PEEPtot ##EQU00001##
[0074] Where: [0075] PEEPtot is the sum of the applied pressure
during expiration and the patient's intrinsic PEEP; [0076] R is the
resistance of the respiratory system; and [0077] C is the
compliance of the respiratory system. [0078] Tau=R*C. The patient's
time constant (Tau), a respiratory mechanics based feature, may
also be determined as the resistance times the compliance.
[0079] Expiratory flow morphology based features may also be
determined. Asynchronies may be characterized by irregularities or
perturbations on the pressure and flow signals. To identify these
during expiration, significant deviations of the flow curve may be
derived from an approximated "normal" expiratory shape. A segmented
moving average filter with time constant, such as about 0.1
seconds, can be applied to a flow signal separately for each
respiratory phase. For each period of expiration, the location of
the maximum expiratory flow occurring in the first portion (e.g.,
25%) of expiration was obtained. Additional expiratory flow
morphology based features may include the following:
[0080] interpMinima--For the remainder of expiration beyond the
first portion, local minima may be determined at discrete
intervals, such as at intervals of about 150 milliseconds, for
example. Interpolation from these minima for the duration of
expiration can be determined. The result may be used to de-trend
the moving average by subtraction, and the power (RMS) can be
obtained for the resultant signal and normalised across the mode
value for the patient.
[0081] interpMinima-2/3--This feature may be determined with the
de-trended interpolated minima signal above. From the interpolated
minima signal, the power (RMS) can be calculated for the last two
thirds of expiration, and normalised across the mode value for the
patient.
[0082] PW linear approx power--This feature may be considered a
piecewise bilinear approximation for the remainder of expiration
after the location of maximum expiratory flow. The second linear
component can be used to de-trend the moving average by
subtraction. The power (RMS) can then be determined for the
resultant signal, and normalised across the mode value for the
patient.
[0083] PW vol deviation--This feature may be considered a piecewise
volume deviation. It may be calculated as the integral of the
rectified and de-trended moving average. The de-trending may be
determined by subtraction of the piecewise linear approximation. It
can be normalised across the mode value for each patient.
[0084] PW dist flow maxmin--This feature may be considered a
piecewise distance flow maximum minimum. It may be determined with
indices of the minimum and maximum of the de-trended moving average
(via subtraction of the piecewise linear approximation). The
minimum and maximum were then used to locate the corresponding real
flow values in the moving average expiratory flow signal. The
distance between these two real flow values may then be calculated
in the Y direction.
[0085] PW distance ratio--This feature may be considered a
piecewise distance ratio. It may be determined as a ratio of the PW
dist flow maxmin. It may be calculated as a fraction of the
distance between the maximum flow point (calculated previously) and
the peak expiratory flow.
[0086] Additional features for features sets are discussed in
further embodiments below.
Feature Set Selection
[0087] Sets of the features such as the respiratory rate based
features, volume based features, respiratory mechanics based
features and expiratory flow morphology based features, for
implementation by the detector 100 may be chosen at design time
based on the indicative nature of each feature relative to a
particular asynchrony event. Thus, empirical analysis of the
features in association with known breath asynchrony data may be
utilized to derive feature sets indicative of asynchrony and a
pattern of threshold values of the features that may be suitable
for use in detectors described herein. This empirical analysis may
be based on data from multiple patients. For example, a classifier
model, such as a Parzen Window Estimator, may be implemented to
identify or select suitable feature sets. The following is a
description of an example implementation of such a classifier.
Classifier Model: Parzen Window Estimation
[0088] A Parzen window estimator may be implemented by a processing
apparatus to evaluate feature subsets indicative of asynchrony
events by analysis of groups of features such as a superset of some
or all of the above features in conjunction with the known breath
asynchrony data. For example, after the above features are
determined or extracted from the sensor signals, each breath may
have an associated set of features, x. This feature vector will
have up to X features depending on the desired application. In such
a classifier, if .omega..sub.i signifies the i.sub.th class, a
Bayes' rule methodology may be implemented to find the class for a
given test feature vector that will maximize the posterior
probability as follows:
P .function. ( .omega. i | x ) = P .function. ( .omega. i ) .times.
p .function. ( x | c .times. .omega. i ) p .function. ( x ) ( 1 )
##EQU00002##
[0089] Posterior probability for each category may be calculated
via Parzen-window estimation. An advantage of this nonparametric
method is that it can be used without the assumption that the forms
of the underlying densities are known. To estimate the density p(x)
at x, first a sequence of regions R.sub.1, . . . R.sub.n is created
across the feature space. Each region R.sub.n is a d-dimensional
hypercube. If h.sub.n is the length of the side of the hypercube,
its volume is given by:
V n = h n d ( 2 ) ##EQU00003##
[0090] It follows that the number of samples falling into the
hypercube k.sub.n can be defined by the following window
function:
k n = i = 1 n .phi. .function. ( x - x i h n ) ( 3 )
##EQU00004##
[0091] Where .PHI.(u) is a unit hypercube having the value one
inside and the value zero outside the unit hypercube centred at
origin. Combing (2) and (3), the Parzen-window density estimate
using n training samples and the window function can be defined
by:
p n ( x ) = 1 n .times. i = 1 n 1 V n .times. .phi. .function. ( x
- x i h n ) ##EQU00005##
[0092] A Gaussian window function may be used with radius equal to
the median of the Euclidean distances between feature vectors.
[0093] Due to the possibility of the estimator being applied to
breath data that is labelled with multiple classes, the classifier
can be operated as a dichotomizer such that each class can be
treated separately and sequentially.
Feature Subset Selection
[0094] The processing apparatus can be implemented with a form of a
sequential selection algorithm, such as the Sequential Forward
Floating Search (SFSS) algorithm, to serve as a subset selector
that can identify the feature subset that maximizes a
classification performance criterion. For example, an accuracy
criterion may be calculated as the percentage of breaths correctly
classified with each subset based on the known classification.
Initially, such an algorithm can make several (e.g., 3) passes with
the ordinary sequential forward selection (SFS), such that three
features are selected. Each of these evaluation passes adds the
particular feature that contributes the greatest improvement to
performance over and above the performance threshold of the already
selected features. To avoid "nesting" of features that are not part
of the optimal feature set, features can be removed one at a time
and the sequence of the remaining features can be re-shuffled. The
classifier performance may then be checked for any improvement. New
combinations of features are only chosen subsequent to an
improvement in performance. Following such a feature trimming
phase, the SFS algorithm can then run to add a new feature. This
cycle of feature addition and trimming iterates until performance
of a particular feature subset has been maximized.
Example Feature Sets
[0095] For example, one embodiment of a detector 100 designed by
such a process can be implemented with a subset of features for
detecting expiratory ineffective effort asynchrony. Such a subset
may include PW linear approx power, PW dist flow maxmin, PW vol
deviation, Tau.sub.i and PW distance ratio. Values for these
features on a breath-by-breath basis may be calculated from the
respiratory apparatus sensor data and compared to a corresponding
set of thresholds to detect the existence of the expiratory
ineffective effort asynchrony event in one or more breaths.
Additional feature set examples are discussed in more detail
herein.
Experimental Testing
[0096] Based on the aforementioned methodologies, 5627 breaths from
23 patients in a pulmonary ward were examined. The patients were
undergoing both conventional and non-invasive ventilation (NIV).
Patients were undergoing NIV after recovering from acute
respiratory failure or conventional ventilation in the advanced
stage of weaning. For each subject between 10-20 minutes of
respiratory data was obtained. The following traces were obtained.
Flow at the airway opening was measured with a heated
pneumotachograph and a differential pressure transducer placed
between the mask and the Y-piece of the ventilator or at the
Y-piece. Airway pressure was measured from a side port between the
pneumotachograph and the face mask or the endotracheal tube.
Esophageal and gastric pressures were measured with a
balloon-catheter system. To this end, an esophageal balloon was
positioned at the lower third of the esophagus, filled with 0.5 ml
of air and a gastric balloon filled with 1 ml of air. The proper
position of the balloon was verified using the occlusion test.
Transdiaphragmatic pressure (Pdi) was calculated as the difference
between gastric (Pga) and esophageal (Pes) pressures.
Breath Segmentation Algorithm
[0097] A segmentation algorithm was applied to the data with the
processing apparatus to automatically demark each breath instance.
As illustrated in FIG. 10, the programmed state-machine included
four states and analyzed the flow and volume signals (shown as
"Flow" and "Volume" conditions of FIG. 10) to identify the points
in time at the beginning and end of each inspiration. This
identification corresponded to the state machine entering the
"Confirmed Inspiration" state 1010 and the "Confirmed Expiration"
state 1020 shown in FIG. 10 based on the flow and volume
conditions. The chosen flow and volume thresholds expressed in the
conditions of the state diagram logic are .alpha.=0.05 L/s and
.beta.=150 mL. Other methods of breath demarcation might also be
implemented.
[0098] As illustrated in FIG. 10, and starting in the Confirmed
Expiration state 1020, the programmed processor waits for the flow
to be greater than a positive value near zero (a) before it
advances into the Provisional Inspiration state 1030. Then, only
when it detects sufficiently accumulated volume (.beta.) to
indicate that the signal is established in inspiration beyond noise
fluctuations close to zero, does the processor advance to the
Confirmed Inspiration state 1010. A similar processing logic
applies to the transition from the Confirmed Inspiration state 1010
through Provisional Expiration 1040 and back to the Confirmed
Expiration state. By monitoring these locations in time relative to
the data of the flow signal, times are identified at the beginning
and end of each inspiration, corresponding to when the state
machine enters the Confirmed Inspiration and the Confirmed
Expiration states.
Visual Asynchrony Scoring
[0099] Visual scoring of asynchronies and correction of breath
demarcation was carried out by a single physician specializing in
mechanical ventilation using a custom designed graphical user
interface (Matlab). Alternately, an automated computer system may
be utilized to calculate the scoring of the asynchronies.
[0100] An example graphic user interface (GUI) for manual event
scoring is illustrated in FIG. 13. The interface permits manual
scoring by easy visual assessment of each and every breath
individually and within the context of the breath series. The
interface allows categorization according to its display and
navigation through the signals. Data files with the data of the
signals may be managed with buttons in the top right corner of the
interface. On loading a data file, the program of the interface
parses data headers as variable names and stores the column data
accordingly. Files with saved data may include a reserved set of
workspace variables reserved as follows:
[0101] CompFlow: [0102] Patient flow data compensated for offsets
and leak.
[0103] Pao: [0104] Pressure at the airway
[0105] Pdi: [0106] Transdiaphragmatic pressure
[0107] Ppl: [0108] Pleural pressure
[0109] Pab: [0110] Abdominal pressure
[0111] Inspirations: [0112] Array of indices demarking the
beginning of inspiration for each breath
[0113] Expirations: [0114] Array of indices demarking the beginning
of expiration (accordingly, the end of inspiration) for each
breath
[0115] AsynchronyClass: [0116] Array of labels designated to each
breath indicating the types of asynchrony present if any
[0117] Fs: [0118] Sampling rate of the data, default at 100 Hz
[0119] Filename: [0120] Unique file identifier
[0121] In the example interface, five plots are used to display the
time series data which from top-to-bottom, including the patient
flow, pressure at the airway, transdiaphragmatic pressure, pleural
pressure, and abdominal pressure. A control button may be included
to allow inversion of the flow if necessary.
[0122] The time scale of the data may be adjusted by the button
groups in the bottom left hand corner of the panel. For example,
shortening of the scale may halves the range of seconds displayed
and conversely lengthening the scale may double the range.
Navigation through the data is achieved by flipping forward or
backwards screen by screen according to the time scale set via up
and down arrow keys of a keyboard of the computer programmed with
the interface. Markers can be shown on the data traces indicating
the beginning of both inspiration and expiration for every breath
as demarked by an automated breath detection algorithm as discussed
in more detail herein. One breath can be selected and the
corresponding portions of each trace are then highlighted at any
one time. A breath can be selected either by using the left and
right arrow keys to move sequentially forwards and backwards, or
entering numeric information (breath number or time at which the
breath occurs) into the first two user input boxes in the bottom
right hand corner of the panel. The denoted start of inspiration
for a breath can be adjusted by entering an offset in milliseconds
in the last user input box.
[0123] The bottom right panel of the interface shows a more
detailed graph of a single breath, where the patient flow, pressure
at the airway and transdiaphragmatic pressure are overlaid on
common graphing axes. The pressure scale is given on the left axis
and the flow on the right. This allows the scorer to view timing
detail of the signals with greater resolution.
[0124] Finally, the upper right panel functions to receive and
store user input regarding the categorization of the breaths with
checkboxes. More than one category may be checked at a time. The
categories available to choose from and their definition criteria
may include:
[0125] Normal Trigger: [0126] The triggering of the ventilator is
reasonably coordinated with respect to the beginning of the
patient's effort.
[0127] Normal Inspiration: [0128] The interaction between the
ventilator and patient during inspiration is harmonious.
[0129] Normal Cycle: [0130] The cycling of the ventilator is
reasonably coordinated with respect to the termination of the
patient's effort.
[0131] Normal Expiration: [0132] The interaction between the
ventilator and patient during expiration is harmonious.
[0133] Inspiratory Ineffective Effort: [0134] A positive Pdi tidal
swing beginning after the ventilator has triggered and is
delivering inspiratory support, resulting in an increasing swing in
the flow trace.
[0135] Expiratory Ineffective Effort: [0136] A positive Pdi tidal
swing occurred during expiration, but was not followed by
inspiratory support from the ventilator.
[0137] Multiple Ineffective Efforts: [0138] Two or more positive
Pdi tidal swings occurring during expiration, where none are
followed by inspiratory support from the ventilator.
[0139] Late Cycling: [0140] A positive tidal swing occurs in Pdi
during expiration that is not followed by inspiratory pressure
support from the ventilator until more than 150 ms has elapsed. The
increase in Pdi indicates that the start of patient effort was more
than 150 ms after the device delivered pressure support.
[0141] Late Trigger: [0142] A prolongation of the ventilator's
inspiratory time beyond the end of the patient's inspiratory
effort. For example, where Ti.sub.excess is the prolongation of the
ventilator inspiratory time beyond the end of patient effort; and
Ti.sub.pat is the duration of patient inspiratory effort, a late
trigger may be considered to exist when the time Ti.sub.excess is
greater than half of the time Ti.sub.pat,
Ti.sub.excess>Ti.sub.pat/2).
[0143] Early Cycling: [0144] A termination of inspiratory support
by the ventilator prior to the end of the patient's Pdi tidal
swing, indicating that the patient's effort is prolonged beyond the
device's mechanical breath.
[0145] Auto-trigger: [0146] The delivery of inspiratory support by
the respiratory treatment apparatus or ventilator without a
concomitant swing in Pdi that would otherwise indicate patient
effort. This can be taken to indicate that the device delivered a
breath that was not triggered by the patient.
[0147] Double Trigger: [0148] Two mechanical breaths over a single
Pdi tidal swing are delivered. The first cycle is patient-triggered
and separated from the second by a very short expiratory time that
may be defined as being less than about one-half of the mean
inspiratory time of previous normal breaths.
[0149] Cough: [0150] The patient is coughing indicated by large
negative spike in expiration with abdominal effort.
[0151] Airway Obstruction: [0152] The upper airways are largely
obstructed, as indicated either by flattened or no flow concomitant
with patient effort, or briefly during a swallowing manoeuvre
occurring predominantly in expiration.
[0153] Timed Breath: [0154] A mandatory breath triggered by the
ventilator.
[0155] Discard: [0156] The data is corrupted and not suitable for
use.
[0157] Don't Know: [0158] The event/activity is ambiguous and needs
further review.
[0159] The physician designated to carry out the visual scoring was
blinded to the clinical data of the patients and was not involved
in their care.
[0160] The beginning of inspiration was globally adjusted for each
patient by 200 ms (.+-.50 ms) to align with the deflection of Pdi
corresponding to the initiation of patient effort. Subsequently,
each breath was labeled as either normal or asynchronous by the
categories and definitions of the asynchronous events previously
identified including, expiratory ineffective effort events,
post-triggering effort events, double triggering events,
autotriggering events, late triggering events, early cycling
events, late cycling events and inspiratory ineffective effort
events.
[0161] The total of 5627 breaths were analyzed and labeled as
summarized in Table II below.
TABLE-US-00002 TABLE II Summary of Visual Classifications % of
Category Count Total Normal 2960 52.63 Expiratory Ineffective
Effort 567 10.08 Inspiratory Ineffective Effort 93 1.65 Double
Trigger 48 0.85 Auto-trigger 77 1.37 Late Trigger 811 14.42 Early
Cycling 527 9.37 Late Cycling 646 11.49 Other events 300 5.33
Unknown 276 4.91 Physical breaths 5624 100 Total classifications
6305 112.11
Feature Selection
[0162] As a goal of this experiment was to automate the detection
of asynchronies via non-invasive sensors, a superset of detection
features were derived using only the flow and pressure signals
rather than with the transdiaphragmatic pressure signal. Thus, a
programmed processing apparatus derived features for each breath
based on data from the flow and pressure signals. The derived
features were those summarized previously in Table I above.
Parzen Window Estimator and Feature Subset Selector
[0163] The feature data was then applied to a processing apparatus
with a programmed parzen window estimator and feature subset
selector as previously described. In testing, the data was analyzed
to develop a feature set for an ineffective efforts asynchrony
event classifier. Additional, feature sets for the remaining
asynchrony events may be similarly developed.
Cross-Fold Validator
[0164] Cross-fold validation was applied to assess the results of
the developed asynchrony classifier. To this end the superset of
features from all but one patient formed the training data for the
development of the detector/classifier. The classifier with the
subset of features was then tested on the corresponding feature
data of the excluded patient. This process was iterated 23 times
for all patients, whereby one patient was excluded from the
training data and reserved to test the automatic asynchrony
detection system in each iteration. In other words, the classifier
that was automatically developed with the data from the larger
group of patients was then used to automatically detect the
asynchrony events from the data of the excluded patient.
Accuracies, sensitivities, specificities, positive and negative
predictive values and Cohen's kappa coefficient were then averaged
across all 23 passes for overall estimate of performance of the
detector/classifier.
Testing Results
[0165] Table III below contains the positive predictivity, negative
predictivity, sensitivity and specificity results for the
detections made by the automated asynchrony detector using a
developed feature set for detecting ineffective effort asynchrony
events. The performance of the detector was compared to the
manually annotated data for 5627 breaths. Also shown is the number
of records, N, for which each parameter that was calculable. The
feature subset for the example detector selected in order were: 1.
PW linear approx power; 2. PW dist flow maxmin; 3. PW vol
deviation; 4. Tau.sub.i; and 5. PW distance ratio.
TABLE-US-00003 TABLE III Performance results for Ineffective
Efforts Classification Measure Mean Positive Predictivity 84.03%
Negative Predictivity 95.32% Sensitivity 58.74% Specificity 98.69%
Overall Accuracy 94.49%
Experimental Testing Assessments
[0166] A non-invasive method for detecting major forms of
asynchrony in patient-ventilator interaction has been designed and
compared to a human expert-based classification system using a
database of 23 subjects.
[0167] In the case of classifying expiratory ineffective efforts,
overall accuracy and specificity of the detector/classifier is
high, whereas the sensitivity is moderate. Possible causes for the
lower sensitivity relate to the fact that timed pressure support
breaths were common amongst the data. In such modes containing
these breaths, it is intentional for the ventilator to ignore
normal efforts exerted by the patient, preferring to dictate the
timing in a metronomic fashion. As a result, normal size efforts
may influence the flow signal producing normal positive swings that
are not accompanied by pressure support. These efforts, while still
classed as ineffective, are not the pathological variety,
influenced largely by intrinsic PEEP and hyperinflation, which
preoccupy the main concern of clinicians for patients on pressure
support ventilation. Pathologic ineffective efforts occur with
smaller and negative value flow swings relative to their timed mode
induced counterparts. As such, the combined feature vectors
intersect a greater proportion of other breath classes. Relabeling
of this class as two subsets may improve the sensitivity of the
classifier.
[0168] Another factor influencing the sensitivity of the classifier
is the similar morphology between expiratory efforts and early
cycling: both are categorized by a single perturbation on the flow
signal during expiration (and correspondingly on the pressure),
however are distinguished merely by distance in time they occur
from the start of expiration. This distance is subject dependent,
and may be ambiguous even by expert visual inspection.
[0169] Factors limiting classifier specificity include artifacts
such as coughs, swallowing, and abdominal effort, as well as
incorrect breath labeling brought about by human error and
ambiguity where the Pdi signal is noisy, exhibits large drifts, or
where esophageal spasms occur.
[0170] It is interesting to note that as well as morphological
features, the inspiratory time constant Tau.sub.i was selected as a
distinguishing feature by the SFSS algorithm. It is interesting to
relate this physiological parameter to the mechanism of ineffective
efforts, which occur predominantly in obstructive patients with
hyperinflation. We could expect then, that patients with high
inspiratory time constant exhibit a greater number of ineffective
efforts.
[0171] In sum, using a pattern classification approach to
automating the detection of asynchronies is feasible and may be of
clinical utility in assessing the quality of patient ventilator
interaction.
[0172] Example Expiratory Ineffective Efforts Related Feature
Sets
[0173] In some embodiments of the detector, feature sets may be
specially devised for one or more particular asynchrony events. For
example, often asynchronies are characterized by irregularities or
perturbations on the pressure and flow signals. Specifically with
respect to expiratory ineffective effort events (eIEs), these
irregularities occur during expiration. As the signal-to-noise
ratio of these irregularities is generally higher on the flow
signal than on the pressure signal, flow may be used as the main
signal for deriving the feature set. To this end, a segmented
moving average filter with time constant of about 0.1 seconds may
be applied to the flow signal separately for each respiratory
phase. In order to identify irregularities during expiration,
significant deviations of the flow curve from a more `normal`
predicted trajectory of the breath's expiratory shape may be
derived. We call this predicted trajectory through expiration,
.THETA.. The predicted trajectory may be used to detrend the
filtered flow signal, {dot over (V)}.sub.f, by subtraction as
follows:
.PHI. = V . f - .THETA. ##EQU00006##
such that the resultant signal, .PHI., emphasized any irregular
deviations that may represent asynchronies. Features were derived
from further operations on .PHI..
[0174] The predicted trajectory through expiration signal .THETA.
may be derived by one of two example methods as follows. Initially
for both methods, the location of the maximum expiratory flow
occurring in the first 25% of expiration was obtained. Up until
this location, .THETA. was equivalent to the filtered flow.
Thereafter, the remainder of .THETA. may be calculated by one of
the following methods:
[0175] (1) Interpolated minima method: [0176] Local minima may be
obtained at intervals of 500 ms from the filtered flow signal.
.THETA..sub.int can be derived by interpolation from these minima
between the maximum expiratory flow and the end of expiration. FIG.
14 illustrates the method applied to an expiratory sample
containing an expiratory ineffective effort. In the figure, the
filtered expiratory flow data, {dot over (V)}.sub.f contains an
expiratory ineffective effort. The derivation of the predicted
normal trajectory, .THETA..sub.int, entails finding local minima at
500 ms intervals of the data occurring after the peak expiratory
flow, and interpolating between the minima to derive a smoother
version of the original. The signal .PHI..sub.int, which is
obtained by subtracting .THETA..sub.int from {dot over (V)}.sub.f,
highlights or indicates the ineffective effort.
[0177] (2) Piecewise bilinear fit method: [0178] A piecewise
bilinear approximation may be performed for the remainder of
expiratory flow data after the location of maximum expiratory flow.
This procedure may involve least squares fitting of the data to two
linear sections, shown as a-15 and b-15, which can be continuously
linked by an elbow or breakpoint at location k. In order to
determine where the breakpoint needed to be so as to offer the best
and most appropriate fit, values of k may be tested sequentially
from the first sample to that equal to a third of the length of the
data. The test condition can be the root-mean-square of the
differences between the data and the fitted values, and the k that
returned the minimum condition value can be chosen. The location of
k may be restricted to the first third of expiration because a
typical expiratory profile changes its curvature most distinctly
within this nominal period. Furthermore, ineffective efforts tended
to occur most commonly during the latter two thirds of expiration.
For the purposes of emphasizing the ineffective effort in the
calculation of .PHI..sub.pw, a linear fit that had higher variance
around the region pertaining to the ineffective effort feature was
desirable. If no restriction was put on the placement of k, it may
be the case that for expirations containing ineffective efforts,
the best k would locate within the ineffective effort morphology
itself, and thus render the procedure inadequate for augmenting it.
The signal .THETA..sub.pw may be derived by concatenating the
filtered flow up to its first intersection with the second linear
component, b-15, and the remaining portion of b-15 beyond this
intersection. FIG. 15 shows the method applied to the same sample
shown in FIG. 14. Thus, in FIG. 15, the filtered expiratory flow
data, {dot over (V)}.sub.f contains an expiratory ineffective
effort. The derivation of the predicted normal trajectory,
.THETA..sub.pw may entail fitting a piecewise bilinear function
(sections a-15 and b-15 joined at breakpoint k) to the data after
the location of the maximum expiratory peak, and concatenating the
second linear component to the filtered flow at their intersection
(indicated by the upwards arrow 1550). The signal .PHI..sub.pw,
which may be obtained by subtracting .THETA..sub.pw from {dot over
(V)}.sub.f, highlights or indicates the ineffective effort.
[0179] Fiducial points may be located on .THETA..sub.pw with the
data of the signal to derive the feature set using these traces.
Examples are illustrated in FIG. 16, which shows the detrended
flow, .THETA..sub.int, marked with fiducial points shown on FIG. 16
as d-16, e-16, f-16 and g-16. The points may include the following:
[0180] d-16: [d.sub.x,d.sub.y], the zero-crossing with positive
slope occurring immediately prior to the maximum value; [0181]
e-16: [e.sub.x,e.sub.y], the maximum value; [0182] f-16:
[f.sub.x,f.sub.y], the zero-crossing with negative slope occurring
immediately after the maximum value, that also occurs prior to the
minimum; and [0183] g-16: [g.sub.x,g.sub.y], the minimum value.
[0184] These fiducial points may be mapped to the data of the
filtered flow, {dot over (V)}.sub.f, contributing to its complete
fiducial point set. This is illustrated in FIG. 17, which shows the
filtered flow, {dot over (V)}.sub.f, marked with fiducial points
labeled as c-17, d'-17, e'-17, f-17 and g'-17. These points may be
considered as follows: [0185] c-17: [c.sub.x,c.sub.y], the peak
expiratory flow on the filtered flow, {dot over (V)}.sub.f [0186]
d'-17: [d.sub.x,d'.sub.y], the onset of the ineffective effort
[0187] e'-17: [e.sub.x,e'y], the peak value of the ineffective
effort [0188] g'-17: [g.sub.x,g'.sub.y], the completion of the
ineffective effort
[0189] One or more of the following features may then be derived
from .PHI..sub.pw and .PHI..sub.int, and where possible are
illustrated in FIGS. 16 and 17.
[0190] power .PHI..sub.int: [0191] This feature may be considered
the power (RMS) of .PHI..sub.int, given by
[0191] .PHI. r .times. m .times. s = .times. .PHI. i 2 n
##EQU00007## [0192] where i is the sample number, and n is the
number of samples in the signal;
[0193] power 2/3 .PHI..sub.int: [0194] This feature may be
considered the power (RMS) of the last two thirds of
.PHI..sub.int;
[0195] power .PHI..sub.pw: [0196] This feature may be considered
the power (RMS) of .PHI..sub.pw;
[0197] integral .PHI..sub.pw: [0198] This feature may be considered
the integral of .PHI..sub.pw after rectification equivalent;
[0199] gradient-a: [0200] This feature may be considered the
gradient of the first linearly fitted section, a-15;
[0201] gradient-b: [0202] This feature may be considered the
gradient of the second linearly fitted section, b-15;
[0203] amplitude p-t .PHI..sub.pw; [0204] This feature may be
considered the vertical distance between the maximum and minimum
values in .PHI..sub.pw, e.sub.y-g.sub.y;
[0205] duration p-t .PHI..sub.pw: [0206] This feature may be
considered the horizontal distance between the maximum and minimum
values in .PHI..sub.pw, g.sub.x-e.sub.x;
[0207] volIE: [0208] This feature may be considered the volume of
air moved during the ineffective effort. This may be calculated by
adding the positive area under the curve
d-16.fwdarw.e-16.fwdarw.f-16 and the absolute value of the area
enclosed by d-16.fwdarw.f-16.fwdarw.g-16.fwdarw.d-16 or, the
integral of the arc d'-17.fwdarw.e'-17.fwdarw.g'-17 with respect to
the line d'-17.fwdarw.g'-17;
[0209] zxUpCount: [0210] This feature may be considered the number
of zero crossings in .PHI..sub.pw with positive slope; maxFlowIE:
[0211] This feature may be considered the value of the maximum flow
reached in the ineffective effort, e'.sub.y.
[0212] heightIE: [0213] This feature may be considered the vertical
distance between the values in {dot over (V)}.sub.f corresponding
to the indices e'.sub.y and g'.sub.y;
[0214] durIE: [0215] This feature may be considered the duration of
the ineffective effort, g.sub.x-d.sub.x;
[0216] rangeP-P: [0217] This feature may be considered the vertical
distance between the maximum expiratory peak in expiration and the
local maxima in the ineffective effort, e'.sub.y-c.sub.y;
[0218] durP-P: [0219] This feature may be considered the horizontal
distance between the maximum expiratory peak in expiration and the
local maxima in the ineffective effort, e.sub.x-c.sub.x;
[0220] ampRatio: [0221] This feature may be considered the ratio of
heightIE to rangeP-P;
[0222] volRatio: [0223] This feature may be considered the ratio of
the volume of air moved during the ineffective effort, volIE, and
the total expired volume, volE.
[0224] durRatio: [0225] This feature may be considered the ratio of
the ineffective effort duration, durIE, to the length of
expiration, T.sub.e;
[0226] locationIE: [0227] This feature may be considered the
location of the ineffective effort in expiration calculated as the
ratio of e.sub.x to the length of expiration, T.sub.e;
[0228] A feature set or subset of the above expiratory flow
morphological based features may be used in both original and
normalized forms. Normalization may be performed such that values
may be divided by the mode value for each patient. In testing as
discussed above, it may be assumed that as the majority of breaths
for a patient are `normal` (without asynchronies), the mode or most
frequently occurring values of a given feature would represent
those breaths with normal patterns. The normalization process is
intended to reduce inter-subject variability. Once the feature set
is derived for each patient, the range of values for each feature
and corresponding histograms may be calculated. Feature values may
be assigned a bin number based on a bin width equal to about 5% of
the range. The feature mode value can then be calculated as the
highest density bin number multiplied by the bin width, added to
the minimum feature value. Thus, as with other features previously
discussed, in some embodiments the feature mode value relative to
each feature determined from multiple patients may then be
implemented as a threshold for detection of asynchrony events when
compared to a particular feature calculated by a detector. However,
in some embodiments, such a feature mode value relative to each
feature may be specifically determined for past data of one
particular patient and utilized for future detection with that
particular patient.
[0229] As will be apparent from the aforementioned discussions, the
detection of respiratory events, such as the different types of
asynchrony events described herein, may be based on one or more
features derived from signal morphologies (e.g., flow and/or
pressure signal morphologies). This determination of features for
events may also involve the identification of fiducial points,
distances related thereto, ratios or other relationships between
them. The automated identification of the fiducial points may
include various operations such as finding zero crossings, local
maxima and minima of the originally acquired physiological signals
or any intermediate signals. Intermediate signals may be derived
from original signals (such as flow or pressure), by mathematically
fitting the data by MLR, or other functions such as linear,
quadratic, exponential, sinusoidal, trapezoidal, or triangular, or
any piecewise combination of the above. The configuration of
piecewise functions may include implementation of an optimization
routine that seeks to locate breakpoints with a minimum error of
fit. Intermediate signals, furthermore, can be derived from
operations on one or more other intermediate signals (e.g.
subtraction, superposition, desampling, interpolation). In such
processes, the fiducial points may be one or more of the following
examples: (a) the inflection point on a pressure signal prior to
the supra plateau pressure; (b) the inflection point indicating
maximum change in slope on a flow signal after the inspiratory peak
and before the of end inspiration; (c) an inflection point
indication maximum change in negative slope on the pressure at end
expiration and immediately prior to triggering; (d) the minimum
pressure immediately prior to triggering; (e) the inflection point
indicating maximum change in positive slope on the flow signal
during inspiration until the inspiratory peak; and/or (f)
breakpoints between piecewise functions fitted to data.
[0230] Example Double Triggering Related Feature Sets
[0231] Characteristic to double triggering is rapid retriggering to
form a `rabbit-ears` pattern on the pressure signal. This pattern
may be taken into account for purposes of devising feature sets for
use by a suitable detector. In this regard, the flow signal will
commonly follow the `rabbit-ears` pattern, with a jagged profile
revealing a retardation of inspiration that may either remain
positive throughout the breath, or dip briefly into expiration
(`false expiration`) and return to inspiration concomitant with the
re-trigger (see FIG. 5). It is desirable that the breath framer
recognize this pattern as pertaining to a single inspiration.
However, if the flow becomes negative between pressurizations and
the corresponding volume is greater than about 150 mL, it is
possible that two breaths are segmented.
[0232] The following morphological features may be derived, based
on these patterns, some of which are illustrated in FIG. 18. The
figure shows flow and pressure signals of a breath with double
triggering. Significant peaks and troughs are circumscribed, and
the features of `peaksCount`, `falseExp` and `timeElapsed`
described below are illustrated.
[0233] peaksCount: [0234] This feature may be considered the number
of significant maxima separated by significant minima more than
about 5 cmH2O away. The simulated real-time peak-trough finding
algorithm is described as follows. If f(x) represents a signal
where x is a datum, the following action is taken on each x: [0235]
1. If f(x)> Tmax, let Tmax=f(x). Tmax is the current transitory
maximum which acts as a holding variable. [0236] 2. If
f(x)<Tmin, let Tmin=f(x). Tmin is the current transitory minimum
which acts as a holding variable. [0237] (Steps (1) and (2) may
oscillate around noise. Until a significant difference between Tmin
and Tmax is evident, Lmax and Lmin, the local maximum and minimum
values, are not confirmed.) [0238] 3. If Tmax-Tmin>5 cmH2O, set
Lmax=Tmax and Lmin=Tmin. [0239] 4. If we have started to descending
from a peak, and f(x) is less than 75% of the range between Lmax
and Lmin, set the descending flag, nullify the ascending flag,
reset the transitory variables, and confirm storage of the new set
of Lmax and Lmin values. [0240] 5. If we have started to ascend
from a trough, and f(x) is greater than the midpoint between the
most recent Lmax and Lmin, set the ascending flag, nullify the
descending flag, set Lmin=Tmin, and Tmax=f(x). [0241] 6. Iterate
next datum . . .
[0242] timeElapsed: [0243] This feature may be considered the time
elapsed between significant maxima.
[0244] falseExpVol:
[0245] This feature may be considered the total expiratory volume
occurring during inspiration. It is considered that this will most
often be greater than 0 mL but less than 150 mL for double
triggers, comprising a false expiration.
[0246] expRatio: [0247] This feature may be considered the ratio of
the current expiration time to the mean expiratory time calculated
over the preceding two minutes. It may occur infrequently that the
false expiration volume may be greater than 150 mL and the breath
framer will erroneously split the breath. In this instance, the
scorer was instructed to categorise the first breath as a double
trigger. In this case, it is expected that the false expiration
volume will be small relative to the average expired volume,
contributing to the hyperplane separating the normal and double
trigger breaths.
[0248] Thus, one or more of these double triggering related
features may be calculated or determined by an automated detection
processor for a particular patient and be assessed against suitable
threshold(s) to determine whether a double triggering event has
occurred.
Respiratory Mechanics Based Feature Determination
[0249] As previously described, feature sets or subsets may include
respiratory mechanics based features. In some embodiments, these
features may be determined automatically by a processor of a
detector 100. In this regard, an automated determination of
resistance R and compliance C values for the feature sets as
previously described may be calculated by a multiple linear
regression method.
[0250] The basis for such an algorithm for processing by the
detector can rely on the application of multiple linear regression
(MLR) of numerical data to a model of the patient-ventilator
system. Multiple linear regression extends simple linear regression
and is used to describe the relationship between a single response
variable with a set of two or more explanatory variables. The
relationship is linear and can be written in its basic form as
Y i = .beta. 0 + .beta. 1 .times. x i , 1 + .beta. 2 .times. x i ,
2 + + .beta. k .times. x i , k + .epsilon. i ##EQU00008##
[0251] where the random errors .sub.i,i=1, 2, . . . , n, are
normally distributed random variables with zero mean and constant
variance .sigma..sup.2.
[0252] The patient-ventilator system can be modeled as a single
compartment such that the total driving pressure is the sum of the
elastic and resistive properties of the system. This can be
described by a first order differential equation
P t .times. o .times. t = R .times. V ( t ) + V .function. ( t ) C
+ P 0 ##EQU00009##
where P.sub.tot is the driving pressure, {dot over (V)} is the flow
through the airways, V is the volume displaced, R is the airway
resistance, C is the compliance of the respiratory system, and
P.sub.0 is the pressure at end-expiration which is the sum of the
applied external PEEP and internal PEEP of the patient. By defining
the driving pressure as the single response variable and flow and
volume as the explanatory variables a processor can apply MLR to
the measured data to determine the parameters R and C.
[0253] The noise in the system is predominantly random produced by
turbulent flow and the ventilator turbine. The latter also produces
deterministic and cyclostationary components that may influence the
outcome of the model. However within the frequency range of
interest (e.g., <10 Hz) they are considered to be relatively
insignificant. Thus the error in the model, , is a good
approximation for most of the system noise.
[0254] Patient muscle effort, however, is an example of a
non-random noise source that may have significant implications for
the accuracy of the model. In a spontaneously breathing patient
using a ventilator the driving pressure at any time is generated by
both the ventilator (P.sub.v) and the patient's respiratory muscles
(P.sub.mus):
P t .times. o .times. t = P v + P m .times. u .times. s
##EQU00010##
[0255] Because P.sub.mus may not be measured directly without using
an esophageal balloon catheter, it is difficult to accurately
determine the mechanical parameters non-invasively while the
patient's muscles are activated. Predominantly this is a concern
during inspiration when the diaphragm and accessory muscles
contract. Fitting flow and volume data to ventilator pressure data
during inspiration without consideration of muscle activity would
underestimate resistance and overestimate compliance:
P v = R .times. V ( t ) + V .function. ( t ) C + P 0
##EQU00011##
[0256] This approach, which may be implemented by a processor in an
MLR process, may allow the exclusion of influences from patient
effort that would otherwise impede the accuracy of R and C.
[0257] Another example of non-random system noise is cardiogenic
oscillation appearing at heart-rate bandwidth (1-2 Hz). Typically,
cardiogenic oscillation produces low amplitude signals relative to
the ventilated patient pressure and flow, such that it has limited
effect on the model.
[0258] For these processes, differentiation between inspiration and
expiration may be based on the state machine described in more
detail herein that includes four states: Provisional inspiration
(PI), Confirmed Inspiration (CI), Provisional Expiration (PE),
Confirmed Expiration (CE). Thus, Expiratory data may be considered
as that occurring from the start of a Provisional Expiration
preceding a Confirmed Expiration, and up to but not including the
start of the next Provisional inspiration preceding a Confirmed
Inspiration.
[0259] Data used for the regression in each breath may include the
beginning of expiration up to the point when approximately 85-95%
(or more preferably 90%) of the tidal volume had been expired.
Thus, in some embodiments only a portion (e.g., an expiratory
portion) of a breathing cycle (e.g., the pressure, flow and volume
measures attributable or corresponding with the expiratory
portion), may be evaluated from an initial time of expiration up to
a time when a desired limit (e.g., a range of 85-90 percent) of
expired tidal volume is reached. A reason for this may be to
optimally capture the dynamic behavior in the data. The intention
of fitting the data to this model is to describe the dynamics of
the passive system subjected to the pressure source. Thus, the
calculations may exclude data that could be associated with patient
respiratory muscle activity. Where slow flow occurs at
end-expiration accompanying minimal change in pressure, the
increased signal to noise ratio has the potential to introduce
inaccuracy in the parameter estimation, and so this portion of the
data may be omitted from the computation.
[0260] The desire for a good fit is to achieve an expiratory breath
without expiratory effort or other spurious artefacts, such that
the flow waveform follows an exponential decay. Thus, it may be
desirable to preclude artefactual breaths from involvement with the
fitting algorithm. Thus, in some embodiments, an accuracy
assessment of the determination of resistance and compliance may be
implemented. One way to accomplish this may be to implement a
median filter, such as one with a width of about 15, such that the
median values of the fitted mechanics parameters from the last
number breaths (e.g., about 15) can be taken to represent the
expiratory breaths with more regular exponential decay. Also a
coefficient of determination R.sup.2 value from the statistical fit
may be used as a criterion to eliminate poorly fitted breaths. For
example, a threshold of about 0.8 may be used. Thus, in such an
example, a processor of an apparatus may determine accuracy, such
as by determining the coefficient of determination R.sup.2 value
for each MLR determination of compliance and resistance values and
comparing the value against with an accuracy threshold, so that
certain determinations of resistance and compliance may be
disregarded (e.g., determinations where the coefficient does not
equal or exceed the threshold).
[0261] While the aforementioned determination of resistance and
compliance values may be implemented as part of the asynchrony
detection methodologies discussed herein, they may also be
determined and recorded for other purposes. Thus, a processing
device or respiratory treatment apparatus of the present technology
may calculate these values and record this data over time, such as
on a breath-by-breath basis. The values may even be averaged over a
number of breaths or over a desired time period (e.g., an hour or a
treatment session) or a number of treatment sessions with a
respiratory treatment apparatus. The values may be output (e.g., to
another processing apparatus, a display and/or memory store) so
that trends in the resistance and compliance values may be
evaluated. In addition to the control implementations discussed
herein, the values may even serve as additional bases for making
automated adjustments to the control parameters of a respiratory
treatment apparatus. For example, they may be utilized for
automated or manual titration of pressure settings (e.g., pressure
support (PS) and/or PEEP) of a respiratory treatment device for
treatment of the patient. In this regard, they may serve as a basis
for an automated algorithm that makes determines, changes and/or
sets a control parameter of a positive end expiratory pressure
(PEEP) and/or pressure support of a respiratory treatment
apparatus.
[0262] In an example of such an embodiment, automated titration of
PEEP control parameters may be implemented with the aforementioned
estimation of respiratory mechanics, such as the compliance value.
Compliance represents the gradient of the pressure-volume (PV)
curve which is used commonly in clinical practice as a diagnostic
aide. The PV curve is sigmoid in shape where low compliance occurs
at the lower and upper extremities of the curve beyond the lower
and upper inflection points. Optimal ventilation strategy, in
particular the titration of PEEP and PS, would exploit the
mid-linear portion of the PV curve.
[0263] In an example embodiment of the automatic titration of PEEP,
a processor of a titration apparatus (e.g., an autotitrating
respiratory treatment apparatus) may be implemented with a PEEP
titration methodology as follows: [0264] 1. Without additionally
using pressure support PS, a PEEP control parameter of a
respiratory treatment apparatus is initialized at a higher value
(e.g., 10-15 cmH2O) to ensure maximal recruitment of alveoli.
[0265] 2. The PEEP parameter is then decremented (e.g.,
breath-by-breath), and the corresponding compliance value is
determined or monitored at each step with respect to its associated
PEEP parameter of the breath or breaths from which the compliance
is determined. [0266] 3. Using the resulting trend in compliance,
the inflection point that indicates a significant reduction is
located. [0267] 4. An appropriate PEEP parameter is chosen, for
example, as the pressure level one increment higher than the
pressure level corresponding to the inflection point. This
inflection point is illustrated in the graph of FIG. 19 at arrow
a-19.
[0268] In an example embodiment of an automatic titration of a
maximum limit for pressure support PS, a processor of a titration
apparatus (e.g., an autotitrating respiratory treatment apparatus)
may also be implemented with a methodology, which may optionally
follow the preceding PEEP titration methodology, as follows: [0269]
1. Without additionally using PS, the PEEP control parameter is
initialized at a higher value (e.g., 10-15 cmH2O) to ensure maximal
recruitment of alveoli. [0270] 2. The PEEP parameter is then
incremented (e.g., breath by breath) until a significant drop is
observed in compliance which is monitored at each step. [0271] 3.
An appropriate maximum PS limit is chosen, such as the pressure
level one increment lower than that corresponding to the inflection
point. This is also illustrated in FIG. 19 at inflection point
shown at arrow b-19.
Example Technological Embodiments
[0272] In some embodiments of the present technology, automated
methods are implemented to detect not only a general occurrence of
a ventilator-patient asynchrony event but different types of
asynchrony events such as two or more of the illustrated asynchrony
events of FIGS. 3-9. This detection of a plurality of distinct
asynchrony events may be implemented by automated processing of the
data from a flow signal and/or a pressure signal measured by
suitable sensors during delivery of ventilatory assistance by a
ventilator or other respiratory treatment device. For example, the
detection of the distinct events may be based on a
pre-determination of a set of detection variables including, for
example, respiratory rate based features, volume based features,
respiratory mechanics based features, expiratory flow morphology
based features, power based power based features, or other features
referred to herein etc., that are derivable with pressure and/or
flow data. For example, such detection variables may be any one or
more of the following example detection variables: RR breath, RR
ratio, IE ratio, Vol.sub.I, Vol.sub.E, breathLeak, Vol ratio,
PEEPi.sub.i, R.sub.i, C.sub.i, Tau.sub.i, PEEPi.sub.e, R.sub.e,
C.sub.e, Tau.sub.e, InterpMinima, InterpMinima-2/3, PW linear
approx power, PW vol deviation, PW dist flow maxmin, PW distance
ratio, power .PHI..sub.int, power 2/3 .PHI..sub.int, power
.PHI..sub.pw integral .PHI..sub.pw, gradient-a, gradient-b,
amplitude p-t .PHI..sub.pw, duration p-t .PHI..sub.pw, volIE,
zxUpCount, maxFlowIE, heightIE, durIE, rangeP-P, durP-P, ampRatio,
volRatio, durRatio, locationIE, peaksCount, timeElapsed,
falseExpVol, and/or expRatio.
[0273] The set of variables may be chosen according to an optional
feature subset selection process as previously described. The
chosen subset of detection variables may be compared to one or more
thresholds or threshold functions to detect the event conditions
with the variables. The thresholds for chosen variables may be
determined or optimized by a suitable automated classification
criteria according to empirical/clinician classification of
observed events such as those discussed herein. In addition, the
evaluation of variables and detection of the event conditions may
be further based on an automated respiratory phase detection (e.g.,
detection of inspiration or expiration phases, such as with an
implemented version of the state diagram shown above).
[0274] The automated processing to detect the different types of
asynchrony by evaluating the set of variables may be performed by a
processing apparatus (e.g., a programmed general purpose computer
having memory and one or more processors or a specific purpose
computer) with previously recorded flow and/or pressure data or by
a ventilator during treatment while such measures are being
determined. Thus, the methodologies may be implemented by one or
more processors with software or firmware and/or with application
specific integrated chips that embody the detection methodologies
illustrated herein. In some embodiments of the technology,
adjustments to respiratory treatment by a ventilator may be
automated according to the detected events. For example, a
controlled treatment response of a ventilator may be modified
according to which asynchrony type is detected such that one type
of event may generate a different automatic ventilator or
respiratory treatment response. Furthermore, the detected events
may be recorded or scored. Thus, such an apparatus may generate a
report of occurrence details of the different events such as how
many have occurred, when they have occurred etc. These details may
optionally be displayed in association with the values of the
detection variables or set of detection variables and other flow or
pressure data.
[0275] For example, in one embodiment, a plurality of distinct
asynchrony events or an asynchrony event may be detected by a
processing apparatus or respiratory treatment apparatus from data
representing a flow and/or pressure measurements taken during
respiratory treatment with a cycling respiratory treatment
apparatus. In one such embodiment, a set of detection variables may
include one or more of measures of PW linear approx power, PW dist
flow maxmin, PW vol deviation, Tau.sub.i and PW distance ratio
determined by a processor of the processing apparatus or
respiratory treatment apparatus. These variables may then be
compared to their associated predetermined thresholds by the
processing apparatus to detect whether or not an expiratory
ineffective effort asynchrony event has occurred or not. The
processing apparatus or respiratory treatment apparatus may further
detect or score other types of asynchrony events such as one or
more of Post-triggering Effort, Double Triggering, Autotriggering,
Late Triggering, Early Cycling and/or Late Cycling.
[0276] Optionally, in some embodiments, the detection of these
asynchrony events may be excluded (i.e., not scored) by the
apparatus upon automated detection of artifacts such as coughs,
swallowing, etc. The determination of such events is illustrated in
International Patent Publication No. WO 2006/079152, the entire
disclosure of which is incorporated herein by reference.
[0277] In some embodiments of the technology, an apparatus is
configured for automated detection of two or more distinct types of
asynchrony events from data representing flow and/or pressure
measurements taken during pressure treatment with a cycling
respiratory treatment device or ventilator.
[0278] In some embodiments of the technology, an apparatus is
configured for automated detection of an expiratory ineffective
effort asynchrony event based on data representing flow and/or
pressure measurements taken during pressure treatment with a
cycling respiratory treatment device or ventilator.
[0279] In some embodiments of the technology, an apparatus is
configured for automated detection of an expiratory ineffective
effort asynchrony event based on data representing flow and/or
pressure measurements taken during pressure treatment with a
cycling respiratory treatment device or ventilator, where the
detection involves an analysis of two or more calculated features
representing respiratory mechanics based features and/or expiratory
flow morphology based features, such as, PW linear approx power, PW
dist flow maxmin, PW vol deviation, Taut and PW distance ratio. For
example, the analysis may be a comparison of the calculated
features with a set of thresholds or threshold functions.
[0280] In some embodiments of the technology, an apparatus is
configured for automated detection of a post-triggering effort
asynchrony event based on data representing flow and/or pressure
measurements taken during pressure treatment with a cycling
respiratory treatment device or ventilator.
[0281] In some embodiments of the technology, an apparatus is
configured for automated detection of a double triggering
asynchrony event based on data representing flow and/or pressure
measurements taken during pressure treatment with a cycling
respiratory treatment device or ventilator.
[0282] In some embodiments of the technology, an apparatus is
configured for automated detection of an autotriggering asynchrony
event based on data representing flow and/or pressure measurements
taken during pressure treatment with a cycling respiratory
treatment device or ventilator.
[0283] In some embodiments of the technology, an apparatus is
configured for automated detection of a late triggering asynchrony
event from based on representing flow and/or pressure measurements
taken during pressure treatment with a cycling respiratory
treatment device or ventilator.
[0284] In some embodiments of the technology, an apparatus is
configured for automated detection of an early cycling asynchrony
event from based on representing flow and/or pressure measurements
taken during pressure treatment with a cycling respiratory
treatment device or ventilator.
[0285] In some embodiments of the technology, an apparatus is
configured for automated detection of a late cycling asynchrony
event based on data representing flow and/or pressure measurements
taken during pressure treatment with a cycling respiratory
treatment device or ventilator.
[0286] Any one, some or all of the detected asynchrony events may
serve as a metric for non-invasively and simply determining the
quality of patient-ventilator interaction based on processing of
flow or pressure measurements. Thus, other measurement devices or
equipment does not necessarily need to be utilized.
[0287] Although in some embodiments implemented thresholds may be
optimized from empirical or clinical observation of data from many
patients, asynchrony detection utilizing any of the previous
mentioned features may be optimized for a particular patient
through clinical observation of flow and/pressure data from the
particular patient and by utilizing automated feature selection and
automated classification of the selected features for that
particular patient by the methods previously mentioned. Thus, a
detection or treatment apparatus may be tuned or programmed for an
individual subject according to the results of an automated
selection of features and thresholds determination made by a
separate tuning apparatus or computer that is programmed to perform
the feature subset selection and threshold determination.
Optionally, these methodologies may also be integrated within the
programming of a ventilator or respiratory treatment device.
[0288] An apparatus with the implemented detection of asynchrony
events can serve as a tool to permit understanding of the
mechanisms of asynchrony by relating influence of physiological
parameters on particular asynchronies.
[0289] The apparatus may also serve as a tool for predicting
asynchronies in patient-ventilator use for particular subject.
Thus, prognosis for ventilator therapy for that subject may be
assessed with respect to compliance to therapy, hospitalization
stay, weaning success, mortality, onset of COPD exacerbation, sleep
quality etc.
[0290] The automated detection of one or more of the different
types of asynchrony events in a ventilator or cycling respiratory
treatment device can serve as a control feedback for
servo-titration of the device during patient treatment (e.g., smart
therapy algorithms).
[0291] Other embodiments and modifications thereof will be apparent
from consideration of all of the details of the present
disclosure.
Example Respiratory Treatment Apparatus Embodiment
[0292] In reference to FIG. 11, the present asynchrony detection
technology may be implemented with a respiratory treatment
apparatus 1102, such as a ventilator or other respiratory treatment
apparatus that synchronizes treatment operations with patient
respiration. Such an apparatus may include a flow generator such as
a servo-controlled blower 1104. The blower 1104 can typically
include an air inlet and impeller driven by a motor (not
shown).
[0293] The respiratory treatment apparatus 1102 will also typically
include a patient interface such as an air delivery conduit 1106
and a mask 1108 or endotracheal tube (not shown) to carry a flow of
air or breathable gas to and/or from a patient.
[0294] The apparatus 1102 also may include sensors 1110, such as a
pressure sensor and/or flow sensor. In such an embodiment, the
pressure sensor, such as a pressure transducer, may measure the
pressure generated by the blower 1104 and generate a pressure
signal p(t) indicative of the measurements of pressure, such as the
pressure at the airway. Similarly, the flow sensor generates a
signal representative of the patient's respiratory flow. For
example, flow proximate to the patient interface or a sense tube
(not shown) may be measured using a pneumotachograph and
differential pressure transducer or similar device such as one
employing a bundle of tubes or ducts to derive a flow signal f(t).
This flow signal may optionally be adjusted to compensate for
offset and leak. Other sensors may be utilized to generate data
indicative of flow or pressure for the purposes of the
methodologies of the apparatus. For example, in some embodiments it
may also include a transdiaphragmatic pressure sensor to generate a
signal representative of transdiaphragmatic pressure. Similarly, it
may also optionally include a sensor(s) suitable for generating a
signal(s) representative of Pleural pressure and/or abdominal
pressure.
[0295] Based on the sensor signals, such as the flow f(t) and/or
pressure p(t) signals, a controller 1112 may generate blower
control signals. For example, the controller may generate a desired
pressure set point and servo-control the speed of the blower to
meet the set point by comparing the set point with the measured
condition of the pressure sensor. Thus, the controller 1112 may
make controlled changes to the pressure delivered to the patient
interface by the blower 1104. Typically, such settings may be made
to synchronize a treatment with patient respiration or to support
the patient's respiration and may be made in conjunction with a
detection of a state of a patient respiration such as by analysis
of the flow signals in conjunction with control parameters such as
trigger and cycling thresholds. Optionally, changes to pressure may
be implemented by controlling an exhaust with a mechanical release
valve (not shown) to increase or decrease the exhaust while
maintaining a relatively constant blower speed. Similarly, based on
flow f(t) and/or pressure p(t) signals, a controller 1112 may
implement the present asynchrony event detection methodologies
described in more detail herein.
[0296] Thus, the controller 1112 may include one or more processors
programmed to implement particular methodologies or algorithms
described in more detail herein. To this end, the controller may
include integrated chips, a memory and/or other control
instruction, data or information storage medium. For example,
programmed instructions encompassing such a control methodology may
be coded on integrated chips in the memory of the device. Such
instructions may also or alternatively be loaded as software or
firmware using an appropriate data storage medium.
[0297] In some such embodiments, the controller may detect or score
asynchrony events and modify pressure control parameters for the
respiratory treatment based on the detection or scoring of such
asynchrony events. For example, if one or more of ineffective
inspiratory effort asynchrony events have been detected, the
controller may automatically reduce (e.g., step down) a triggering
threshold used by the controller to detect patient inspiration and
initiate control of an inspiratory pressure level. Further step
changes may also be made upon continued detection of the asynchrony
events. Optionally, in response to the detection of a number of
such asynchrony events, the controller or processor may generate an
output message for a clinician to indicate that a change to the
trigger threshold setting should be manually made or has been
automatically made. Similarly, detection of late triggering
asynchrony events may serve to control step changes or messages
concerning the trigger threshold of the respiratory treatment
apparatus. In such as case, the step change made or suggested in
response to a late triggering may be smaller than the step change
associated with the detection of ineffective efforts asynchrony
events.
[0298] By way of further example, detection of autotriggering
asynchrony events may serve to control one or more step changes or
messages to increase the trigger threshold of the respiratory
treatment apparatus. Through controlling step increases and
decreases in the triggering threshold based on the different
asynchrony events to reduce the counts of these events, the trigger
threshold may be servo controlled to an optimum setting.
[0299] In a similar fashion, adjustments to the respiratory
treatment apparatus' cycling threshold may be made or suggested by
the controller or processor upon detection of late cycling and/or
early cycling events to reduce these events and thereby optimize
the setting for the cycling threshold.
[0300] Other changes to the control parameters of the respiratory
treatment apparatus may also be made or suggested in accordance
with the detection of the asynchrony events.
[0301] Example Controller/Processing Apparatus Architecture
[0302] An example system architecture of a processing apparatus
(e.g., computer or controller 1112) is illustrated in the block
diagram of FIG. 12. The system architecture may serve as a
respiratory treatment apparatus controller with asynchrony
detection or more simply as a stand alone asynchrony detector such
as a special purpose computer. In the illustration, the detection
device 1206 may be implemented by a general purpose computer with
one or more programmable processors 1208. The device may also
include a display interface 1210 to output data from the detection
methodologies as previously discussed (e.g., asynchrony events and
information, etc.), results or graphs as described herein to a
display such as on a monitor or LCD panel. A user control/input
interface 1212, for example, for a keyboard, touch panel, control
buttons, mouse etc. may also be included for inputting data, or
otherwise activating or operating the methodologies described
herein. The device may also include a sensor or data interface
1214, such as a bus, for receiving/transmitting data such as
programming instructions, flow data, pressure data, asynchrony
data, feature set calculation algorithms, feature subset selection
algorithms, feature set values and other output or input of the
previously described methodologies.
[0303] The device also includes memory/data storage components 1220
containing control instructions and data of the aforementioned
methodologies and algorithms. At 1222, these may also include
stored processor control instructions for asynchrony event
detection as discussed in more detail herein. For example, this may
include feature set value calculation instructions and threshold or
threshold function comparison instructions. For example, at 1224,
they may include stored processor control instructions for flow and
pressure signals processing such as feature calculation and breath
segmentation. At 1226, they may also include processor control
instructions for device activation control such as respiratory
treatment control parameters, ventilation control parameters or
pressure treatment synchronization methodologies. This may also
include, for example, pressure control adjustment based on detected
asynchrony events, etc. Optionally or alternatively, at 1228, they
may also include processor control instructions for feature set
determination such as instructions for a Parzen window estimator,
SFSS selector, etc. Finally, they may also include stored data at
1230 for these methodologies such as flow data, pressure data,
breath segmentation data, features sets for detecting one or more
asynchrony events, features set values, thresholds, threshold
functions, detected asynchrony events and related information,
known asynchrony events data, feature subsets, features supersets,
reports and graphs, etc.
[0304] In some embodiments, some or all of the processor control
instructions and data for controlling the above described
methodologies may be contained in a computer readable recording
medium as software for use by a general purpose computer so that
the general purpose computer may serve as a specific purpose
computer according to any of the methodologies discussed herein
upon loading the software into the general purpose computer.
[0305] In the foregoing description and in the accompanying
drawings, specific terminology, equations and drawing symbols are
set forth to provide a thorough understanding of the present
technology. In some instances, the terminology and symbols may
imply specific details that are not required to practice the
technology. Moreover, although the technology herein has been
described with reference to particular embodiments, it is to be
understood that these embodiments are merely illustrative of the
principles and applications of the technology. It is therefore to
be understood that numerous modifications may be made to the
illustrative embodiments and that other arrangements may be devised
without departing from the spirit and scope of the technology. For
example, in some detector embodiments, in addition to flow and/or
pressure data, feature sets may be also be based on data from
transdiaphragmatic pressure signals.
* * * * *