Optimal PEEP selection in Mechanical Ventilation using EIT · Optimal PEEP selection in Mechanical...
Transcript of Optimal PEEP selection in Mechanical Ventilation using EIT · Optimal PEEP selection in Mechanical...
-
Optimal PEEP selection in Mechanical Ventilation usingEIT
Ravi B. Bhanabhai
Carleton University - 2009/11 M.A.Sc
January 20, 2012
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 1 / 28
-
Outline
1 IntroductionThe ProblemHow to solve the problem?
2 ContributionsIP CalculationFuzzy Logic System
3 ResultsSigmoid vs. LinearLinear vs. VisualOptimal PEEP
4 References
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 2 / 28
-
Introduction
Introduction
This is a presentation outlining the work done within RaviBhanabhai M.A.Sc thesis.
Purpose: Investigate the use of Electrical Impedance Tomography(EIT) within mechanical ventilation.
Mathematical Tools:1 Linear and Non-Linear curve fitting techniques2 Fuzzy Logic.
Contribtions:1 Summarize scholarly papers on ALI.2 Inflection Point (IP) location on EIT and pressure data.3 Creation of Fuzzy Logic System using IP.
Novel Aspects:1 Use of short recruitment maneuever ( 2min)2 Regional Inflection Points used3 Use of Inflection Points within an automated classification system
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 3 / 28
-
Introduction
Introduction
This is a presentation outlining the work done within RaviBhanabhai M.A.Sc thesis.
Purpose: Investigate the use of Electrical Impedance Tomography(EIT) within mechanical ventilation.
Mathematical Tools:1 Linear and Non-Linear curve fitting techniques2 Fuzzy Logic.
Contribtions:1 Summarize scholarly papers on ALI.2 Inflection Point (IP) location on EIT and pressure data.3 Creation of Fuzzy Logic System using IP.
Novel Aspects:1 Use of short recruitment maneuever ( 2min)2 Regional Inflection Points used3 Use of Inflection Points within an automated classification system
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 3 / 28
-
Introduction
Introduction
This is a presentation outlining the work done within RaviBhanabhai M.A.Sc thesis.
Purpose: Investigate the use of Electrical Impedance Tomography(EIT) within mechanical ventilation.
Mathematical Tools:1 Linear and Non-Linear curve fitting techniques2 Fuzzy Logic.
Contribtions:1 Summarize scholarly papers on ALI.2 Inflection Point (IP) location on EIT and pressure data.3 Creation of Fuzzy Logic System using IP.
Novel Aspects:1 Use of short recruitment maneuever ( 2min)2 Regional Inflection Points used3 Use of Inflection Points within an automated classification system
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 3 / 28
-
Introduction
Introduction
This is a presentation outlining the work done within RaviBhanabhai M.A.Sc thesis.
Purpose: Investigate the use of Electrical Impedance Tomography(EIT) within mechanical ventilation.
Mathematical Tools:1 Linear and Non-Linear curve fitting techniques2 Fuzzy Logic.
Contribtions:1 Summarize scholarly papers on ALI.2 Inflection Point (IP) location on EIT and pressure data.3 Creation of Fuzzy Logic System using IP.
Novel Aspects:1 Use of short recruitment maneuever ( 2min)2 Regional Inflection Points used3 Use of Inflection Points within an automated classification system
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 3 / 28
-
Introduction The Problem
ALI & VILI
Respiratory Failure
OxygenationFailure
(hypoxemia)
VentilatoryFailure
(hypercapnia)
+ more oxygenrelated conditions
Acute Lung Injury(ALI)
Ventilator InduceLung Injury
(VILI)* Cyclic opening and closing* overdistensionPEEP
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 4 / 28
-
Introduction How to solve the problem?
Respiratory Function Models
Pao =V
C+ V R + V I Pmus (1)
Interested in VC only.
To remove other components this thesis data did two things:1 Slow Constant Flow2 Antheysia
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 5 / 28
-
Introduction How to solve the problem?
Respiratory Function Models
Pao =V
C+ V R + V I Pmus (1)
Interested in VC only.
To remove other components this thesis data did two things:1 Slow Constant Flow2 Antheysia
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 5 / 28
-
Introduction How to solve the problem?
Respiratory Function Models
Pao =V
C+ V R + V I Pmus (1)
Interested in VC only.
To remove other components this thesis data did two things:1 Slow Constant Flow2 Antheysia
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 5 / 28
-
Introduction How to solve the problem?
Pressure-Volume Curves
Used to help guide ventilation strategies by locating points ofcompliance change.
Points are Lower Inflection Point (LIP) and Upper Inflection Point(UIP)
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
(a) Linear Fit of PI data
20 10 0 10 20 300
0.2
0.4
0.6
0.8
1Time Alignment
global EITPressure
(b) Pressure andEIT data overlap
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 6 / 28
-
Introduction How to solve the problem?
Pressure-Volume Curves
Used to help guide ventilation strategies by locating points ofcompliance change.Points are Lower Inflection Point (LIP) and Upper Inflection Point(UIP)
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
(c) Linear Fit of PI data
20 10 0 10 20 300
0.2
0.4
0.6
0.8
1Time Alignment
global EITPressure
(d) Pressure andEIT data overlap
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 6 / 28
-
Introduction How to solve the problem?
Pressure-Volume Curves
Used to help guide ventilation strategies by locating points ofcompliance change.Points are Lower Inflection Point (LIP) and Upper Inflection Point(UIP)
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
(e) Linear Fit of PI data
20 10 0 10 20 300
0.2
0.4
0.6
0.8
1Time Alignment
global EITPressure
(f) Pressure andEIT data overlap
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 6 / 28
-
Introduction How to solve the problem?
Data used
Data used:
26 patients
low constant flow maneuver (4 L/min)
start 0 mbar 35 mbar / 2L
20 40 60 80 100 120 140 160 180 2000
5
10
15
20P
ress
ure
[mba
r]
Pressure Maneuver
20 40 60 80 100 120 140 160 180 2000
500
1000
1500
Vol
ume
[ml]
Volume Maneuver
20 40 60 80 100 120 140 160 180 200
20
0
20
40
Flo
w [L
/min
]
Flow Maneuver
Time [sec]
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 7 / 28
-
Introduction How to solve the problem?
Electrical Impedance Tomography (EIT)
EIT is real-time impedance tomography, it can be used to accuratlymeasure air distrubtion within the thorax.
(g) Start of Inflation (h) Max Pressure (i) End of Deflation
Figure: Example reconstruction using the GREIT methods of a healthy lungpatient (patient 7).
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 8 / 28
-
Introduction How to solve the problem?
Electrical Impedance Tomography (EIT)
EIT is real-time impedance tomography, it can be used to accuratlymeasure air distrubtion within the thorax.
(a) Start of Inflation (b) Max Pressure (c) End of Deflation
Figure: Example reconstruction using the GREIT methods of a healthy lungpatient (patient 7).
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 8 / 28
-
Contributions
Contributions
Automated IP calculation
Rule-base Fuzzy Logic Classifier
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 9 / 28
-
Contributions
Contributions
Automated IP calculation
Rule-base Fuzzy Logic Classifier
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 9 / 28
-
Contributions IP Calculation
IP Calculation
Three Types of IP location methods were used:
1 Sigmoid method
2 Visual heuristics
3 3-piece linear spline method
Multiple methods were implemented for comparison reasons.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 10 / 28
-
Contributions IP Calculation
IP Calculation
Three Types of IP location methods were used:
1 Sigmoid method
2 Visual heuristics
3 3-piece linear spline method
Multiple methods were implemented for comparison reasons.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 10 / 28
-
Contributions IP Calculation
IP Calculation
Three Types of IP location methods were used:
1 Sigmoid method
2 Visual heuristics
3 3-piece linear spline method
Multiple methods were implemented for comparison reasons.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 10 / 28
-
Contributions IP Calculation
IP Calculation
Three Types of IP location methods were used:
1 Sigmoid method
2 Visual heuristics
3 3-piece linear spline method
Multiple methods were implemented for comparison reasons.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 10 / 28
-
Contributions IP Calculation
IP Calculation
Three Types of IP location methods were used:
1 Sigmoid method
2 Visual heuristics
3 3-piece linear spline method
Multiple methods were implemented for comparison reasons.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 10 / 28
-
Contributions IP Calculation
Sigmoid Method
0 5 10 15 20 25 30 350
200
400
600
800
1000
1200
pressure mbar
volu
me
m
l
Sigmoid Method
a= 12ml
b= 1173 ml
Plip =c - 2d
=11.1Plip =c +2d
=22.9
c= 17 cm H20
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 11 / 28
-
Contributions IP Calculation
Visual Heuristics
Clinicians used this method to locate Inflection Points from global PVcurves.
Multiple methods exist:
1 Find location where PV curve has linear compliance2 Pressure where rapid increase in compliance occurs3 Place two line. 1) Along low compliance. 2) Along High compliance.
Locate Intersection.
This thesis:1 5 participants2 Fit in linear manner to get closest to all the data points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 12 / 28
-
Contributions IP Calculation
Visual Heuristics
Clinicians used this method to locate Inflection Points from global PVcurves.
Multiple methods exist:1 Find location where PV curve has linear compliance
2 Pressure where rapid increase in compliance occurs3 Place two line. 1) Along low compliance. 2) Along High compliance.
Locate Intersection.
This thesis:1 5 participants2 Fit in linear manner to get closest to all the data points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 12 / 28
-
Contributions IP Calculation
Visual Heuristics
Clinicians used this method to locate Inflection Points from global PVcurves.
Multiple methods exist:1 Find location where PV curve has linear compliance2 Pressure where rapid increase in compliance occurs
3 Place two line. 1) Along low compliance. 2) Along High compliance.Locate Intersection.
This thesis:1 5 participants2 Fit in linear manner to get closest to all the data points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 12 / 28
-
Contributions IP Calculation
Visual Heuristics
Clinicians used this method to locate Inflection Points from global PVcurves.
Multiple methods exist:1 Find location where PV curve has linear compliance2 Pressure where rapid increase in compliance occurs3 Place two line. 1) Along low compliance. 2) Along High compliance.
Locate Intersection.
This thesis:1 5 participants2 Fit in linear manner to get closest to all the data points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 12 / 28
-
Contributions IP Calculation
Visual Heuristics
Clinicians used this method to locate Inflection Points from global PVcurves.
Multiple methods exist:1 Find location where PV curve has linear compliance2 Pressure where rapid increase in compliance occurs3 Place two line. 1) Along low compliance. 2) Along High compliance.
Locate Intersection.
This thesis:1 5 participants2 Fit in linear manner to get closest to all the data points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 12 / 28
-
Contributions IP Calculation
Visual Heuristic
0 5 10 15 20 25 30 350.4
0.2
0
0.2
0.4
0.6
0.8
1
Pressure (mbar)
EIT
Con
duct
ivity
Trial3/ 4 Deflation
One chance only, be carefull
0 5 10 15 20 25 30 350.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Pressure (mbar)
EIT
Con
duct
ivity
Trial1/ 4 Deflation
One chance only, be carefull
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 13 / 28
-
Contributions IP Calculation
3-piece Linear Spline Method
Similar to visual methods
Fits to 3 lines with Inflection Points being located at intersection
This Thesis: No Constraints on fitting other then minimization ofcriteria
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 14 / 28
-
Contributions IP Calculation
3-piece Linear Spline Method
Similar to visual methods
Fits to 3 lines with Inflection Points being located at intersection
This Thesis: No Constraints on fitting other then minimization ofcriteria
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 14 / 28
-
Contributions IP Calculation
3-piece Linear Spline Method
Similar to visual methods
Fits to 3 lines with Inflection Points being located at intersection
This Thesis: No Constraints on fitting other then minimization ofcriteria
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 14 / 28
-
Contributions IP Calculation
3-piece Linear Spline Method
Similar to visual methods
Fits to 3 lines with Inflection Points being located at intersection
This Thesis: No Constraints on fitting other then minimization ofcriteria
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 352
0
2
4
6
Pressure (mbar)
EIT
Con
duct
ivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 14 / 28
-
Contributions Fuzzy Logic System
Introduction
The Fuzzy System is designed into 4 sections:
1 Location of IP
2 Fuzzification
3 Premise Calculation (Application of IF-THEN)
4 Defuzzification and Optimization
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 15 / 28
-
Contributions Fuzzy Logic System
Introduction
The Fuzzy System is designed into 4 sections:
1 Location of IP
2 Fuzzification
3 Premise Calculation (Application of IF-THEN)
4 Defuzzification and Optimization
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 15 / 28
-
Contributions Fuzzy Logic System
Introduction
The Fuzzy System is designed into 4 sections:
1 Location of IP
2 Fuzzification
3 Premise Calculation (Application of IF-THEN)
4 Defuzzification and Optimization
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 15 / 28
-
Contributions Fuzzy Logic System
Introduction
The Fuzzy System is designed into 4 sections:
1 Location of IP
2 Fuzzification
3 Premise Calculation (Application of IF-THEN)
4 Defuzzification and Optimization
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 15 / 28
-
Contributions Fuzzy Logic System
Introduction
0 5 10 15 20 25 30 35 401
0
1
2
3
Pressure (mbar)
EITConductivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 35 401
0
1
2
3
Pressure (mbar)
EITConductivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 35 400.5
0
0.5
1
1.5
2
Pressure (mbar)
EITConductivity
Linear Fit Inflation
OriginalSigmoid FittedInflection Points
0 5 10 15 20 25 30 35 401
0
1
2
Pressure (mbar)
EITConductivity
Linear Fit Deflation
OriginalSigmoid FittedInflection Points
5 10 15 20 25 30 350
0.2
0.4
0.6
0.8
1
Pressure
Mem
bershipValue
Pressure based Membership Graph Inflation
BelowIn BetweenAbove
5 10 15 20 25 30 350
0.2
0.4
0.6
0.8
1
Pressure
Mem
bershipValue
Pressure based Membership Graph Deflation
BelowIn BetweenAbove
5 10 15 20 25 30 350
0.2
0.4
0.6
0.8
1
Pressure
Mem
bershipValue
Pressure based Membership Graph Inflation
BelowIn BetweenAbove
5 10 15 20 25 30 350
0.2
0.4
0.6
0.8
1
Pressure
Mem
bershipValue
Pressure based Membership Graph Deflation
BelowIn BetweenAbove
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
0.2
0.4
0.6
0.8
1
EIT
Mem
bershipValue
LIP UIP
EIT based Membership Graph Inflation
BelowIn BetweenAbove
0.5 0 0.5 1 1.5
0
0.2
0.4
0.6
0.8
1
EIT
Mem
bershipValue
LIP UIP
EIT based Membership Graph Deflation
BelowIn BetweenAbove
0 0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
EIT
Mem
bershipValue
LIP UIP
EIT based Membership Graph Inflation
BelowIn BetweenAbove
0 0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
EIT
Mem
bershipValue
LIP UIP
EIT based Membership Graph Deflation
BelowIn BetweenAbove
5 10 15 20 25 30 35
0
0.2
0.4
0.6
0.8
1
Pressure
Magnitude
Pressure based Optimal Pressure Graph Inflation
Good StatesBad States
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
0.2
0.4
0.6
0.8
1
EIT
Magnitude
EIT based Optimal Pressure Graph Inflation
Good StatesBad States
5 10 15 20 25 30 35
0
0.2
0.4
0.6
0.8
1
Pressure
Magnitude
Pressure based Optimal Pressure Graph Deflation
Good StatesBad States
0.5 0 0.5 1 1.5
0
0.2
0.4
0.6
0.8
1
EIT
Magnitude
EIT based Optimal Pressure Graph Deflation
Good StatesBad States
5 10 15 20 25 30 35
0
0.2
0.4
0.6
0.8
1
Pressure
Magnitude
Pressure based Optimal Pressure Graph Inflation
Good StatesBad States
0 0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
EIT
Magnitude
EIT based Optimal Pressure Graph Inflation
Good StatesBad States
5 10 15 20 25 30 35
0
0.2
0.4
0.6
0.8
1
Pressure
Magnitude
Pressure based Optimal Pressure Graph Deflation
Good StatesBad States
0 0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
EIT
Magnitude
EIT based Optimal Pressure Graph Deflation
Good StatesBad States
Average over all pixelsCalculate Optimal PEEP
IF P ressure/E I T THEN StateBelow CollpasedI n Between NormalAbove Overdistended
Inflection Points Fuzzification Premise Calculation OptimizationRule Base
0 50 100 150 200 250 300 350 400 4500
50
100
150
200
250
300Good vs Bad states Pressure based system
Pressure Index
Mantiu
de
Optimal PEEP Optimal PEEP
0 50 100 150 200 250 300 350 4000
50
100
150
200
250
300Good vs Bad states Pressure based system
Pressure Index
Mantiu
de
Optimal PEEP
0 50 100 150 200 250 300 350 400 4500
50
100
150
200
250
300Good vs Bad states EIT based system
Pressure Index
Mantiu
de
Optimal PEEP
0 50 100 150 200 250 300 350 4000
50
100
150
200
250
300
350Good vs Bad states EIT based system
Pressure Index
Mantiu
de
Optimal PEEP
Pressure - Inflation
Pressure - Deflation
EIT - Inflation
EIT - Deflation
Inference and Defuzzification
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 16 / 28
-
Contributions Fuzzy Logic System
IP and Fuzzification
IP were taken from the 3-piece linear optimization portion and used in thecreation of the fuzzification graphs.
Membership Input 1 (mbar) Input 2 (mbar) Input 3 (mbar) Input 4 (mbar)
Below min(p) min(p) -2+LIP LIPIn Between -2+LIP LIP UIP 2+UIPAbove UIP 2+UIP max(p) max(p)
Table: Details on creating the trapezoidal based fuzzy membership functions
0
0.25
0.5
0.75
1
a b c d
Figure: Trapezoidal Fuzzy Membership graph
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 17 / 28
-
Contributions Fuzzy Logic System
IP and Fuzzification
IP were taken from the 3-piece linear optimization portion and used in thecreation of the fuzzification graphs.
Membership Input 1 (mbar) Input 2 (mbar) Input 3 (mbar) Input 4 (mbar)
Below min(p) min(p) -2+LIP LIPIn Between -2+LIP LIP UIP 2+UIPAbove UIP 2+UIP max(p) max(p)
Table: Details on creating the trapezoidal based fuzzy membership functions
0
0.25
0.5
0.75
1
a b c d
Figure: Trapezoidal Fuzzy Membership graph
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 17 / 28
-
Contributions Fuzzy Logic System
IP and Fuzzification
IP were taken from the 3-piece linear optimization portion and used in thecreation of the fuzzification graphs.
Membership Input 1 (mbar) Input 2 (mbar) Input 3 (mbar) Input 4 (mbar)
Below min(p) min(p) -2+LIP LIPIn Between -2+LIP LIP UIP 2+UIPAbove UIP 2+UIP max(p) max(p)
Table: Details on creating the trapezoidal based fuzzy membership functions
0
0.25
0.5
0.75
1
a b c d
Figure: Trapezoidal Fuzzy Membership graph
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 17 / 28
-
Contributions Fuzzy Logic System
Inference and Defuzzification
The Inference is conducted using the Rule base. With key relations toprevious papers:
1 Pressure below the LIP is considered collapsed
2 Pressure above the UIP is considered overdistended
Defuzzification was done by breaking the output states into twoclassifications:
1 Good = Normal states
2 Bad = Collpased + Overdistended states
Upon averaging over lung region MAX value between the difference ofGood and Bad states is performed to locate the PEEP. Fuzzy Logic Schematic
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 18 / 28
-
Contributions Fuzzy Logic System
Inference and Defuzzification
The Inference is conducted using the Rule base. With key relations toprevious papers:
1 Pressure below the LIP is considered collapsed
2 Pressure above the UIP is considered overdistended
Defuzzification was done by breaking the output states into twoclassifications:
1 Good = Normal states
2 Bad = Collpased + Overdistended states
Upon averaging over lung region MAX value between the difference ofGood and Bad states is performed to locate the PEEP. Fuzzy Logic Schematic
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 18 / 28
-
Contributions Fuzzy Logic System
Inference and Defuzzification
The Inference is conducted using the Rule base. With key relations toprevious papers:
1 Pressure below the LIP is considered collapsed
2 Pressure above the UIP is considered overdistended
Defuzzification was done by breaking the output states into twoclassifications:
1 Good = Normal states
2 Bad = Collpased + Overdistended states
Upon averaging over lung region MAX value between the difference ofGood and Bad states is performed to locate the PEEP. Fuzzy Logic Schematic
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 18 / 28
-
Contributions Fuzzy Logic System
Inference and Defuzzification
The Inference is conducted using the Rule base. With key relations toprevious papers:
1 Pressure below the LIP is considered collapsed
2 Pressure above the UIP is considered overdistended
Defuzzification was done by breaking the output states into twoclassifications:
1 Good = Normal states
2 Bad = Collpased + Overdistended states
Upon averaging over lung region MAX value between the difference ofGood and Bad states is performed to locate the PEEP. Fuzzy Logic Schematic
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 18 / 28
-
Contributions Fuzzy Logic System
Inference and Defuzzification
The Inference is conducted using the Rule base. With key relations toprevious papers:
1 Pressure below the LIP is considered collapsed
2 Pressure above the UIP is considered overdistended
Defuzzification was done by breaking the output states into twoclassifications:
1 Good = Normal states
2 Bad = Collpased + Overdistended states
Upon averaging over lung region MAX value between the difference ofGood and Bad states is performed to locate the PEEP. Fuzzy Logic Schematic
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 18 / 28
-
Results Sigmoid vs. Linear
Sigmoid vs. Linear
0
5
10
15
20x 103
LIPInf UIPInf LIPDef UIPDef
Linear Method
% n
ot fo
und
0
0.2
0.4
0.6
0.8
1
LIPInf UIPInf LIPDef UIPDef
Sigmoid Method
% n
ot fo
und
Figure: How frequent each sigmoid and linear method are not able to find IP.
Mean Std Median
LIP - Inflation 1.47 3.02 1.50UIP - Inflation -6.80 2.54 -6.82LIP - Deflation 4.07 1.84 4.07UIP - Deflation -2.37 2.24 -2.78
Table: Difference between Sigmoid and Linear Method
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 19 / 28
-
Results Sigmoid vs. Linear
Sigmoid vs. Linear
0
5
10
15
20x 103
LIPInf UIPInf LIPDef UIPDef
Linear Method
% n
ot fo
und
0
0.2
0.4
0.6
0.8
1
LIPInf UIPInf LIPDef UIPDef
Sigmoid Method
% n
ot fo
und
Figure: How frequent each sigmoid and linear method are not able to find IP.
Mean Std Median
LIP - Inflation 1.47 3.02 1.50UIP - Inflation -6.80 2.54 -6.82LIP - Deflation 4.07 1.84 4.07UIP - Deflation -2.37 2.24 -2.78
Table: Difference between Sigmoid and Linear Method
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 19 / 28
-
Results Linear vs. Visual
Linear vs. Visual Heuristics
Difference = Linear IP Visual Heuristic IP
0.6247mbar for LIP 0.4662mbar for UIPbest average = 0.016mbar worst average = 1.507mbarProvides insight to accuracy of linear method
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 20 / 28
-
Results Linear vs. Visual
Linear vs. Visual Heuristics
Difference = Linear IP Visual Heuristic IP0.6247mbar for LIP 0.4662mbar for UIP
best average = 0.016mbar worst average = 1.507mbarProvides insight to accuracy of linear method
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 20 / 28
-
Results Linear vs. Visual
Linear vs. Visual Heuristics
Difference = Linear IP Visual Heuristic IP0.6247mbar for LIP 0.4662mbar for UIPbest average = 0.016mbar worst average = 1.507mbar
Provides insight to accuracy of linear method
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 20 / 28
-
Results Linear vs. Visual
Linear vs. Visual Heuristics
Difference = Linear IP Visual Heuristic IP0.6247mbar for LIP 0.4662mbar for UIPbest average = 0.016mbar worst average = 1.507mbarProvides insight to accuracy of linear method
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 20 / 28
-
Results Optimal PEEP
Hetergenaity
Figure: Progressive change of lung state with pressure
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 21 / 28
-
Results Optimal PEEP
LIP+2 vs FLS
0 5 10 15 20 25 30 350.5
0
0.5
1
1.5
2
Pressure [mbar]
EIT
Linear Fit graph with LIP+2 mbar PEEP
LIP+2cm = 9.4837 mbar
0 5 10 15 20 25 30 35
1
0.5
0
0.5
1
Pressure [mbar]
Mem
bers
hip
Optimal Selection graph with Optimal PEEP
FLS = 17.2 mbar
PILinear FitLower IPUpper IP
BadGoodDifference
(a) Patient 12
10 0 10 20 30 400.5
0
0.5
1
1.5
Pressure [mbar]
EIT
Linear Fit graph with LIP+2 mbar PEEP
LIP+2cm = 2.1 mbar
0 10 20 30 40
1
0.5
0
0.5
1
Pressure [mbar]
Mem
bers
hip
Optimal Selection graph with Optimal PEEP
FLS = 16.8 mbar
PILinear FitLower IPUpper IP
BadGoodDifference
(b) Patient 16
0 5 10 15 201
0
1
2
3
4
5
Pressure [mbar]
EIT
Linear Fit graph with LIP+2 mbar PEEP
LIP+2cm = 8.5263 mbar
PILinear FitLower IPUpper IP
0 5 10 15 20
1
0.5
0
0.5
1
Pressure [mbar]
Mem
bers
hip
Optimal Selection graph with Optimal PEEP
FLS = 8.6 mbar
BadGoodDifference
(c) Patient 8
0 5 10 15 20 251
0
1
2
3
4
Pressure [mbar]
EIT
Linear Fit graph with LIP+2 mbar PEEP
LIP+2cm = 9.628 mbar
0 5 10 15 20 25
1
0.5
0
0.5
1
Pressure [mbar]
Mem
bers
hip
Optimal Selection graph with Optimal PEEP
FLS = 10.3 mbar
PILinear FitLower IPUpper IP
BadGoodDifference
(d) Patient 17
Figure: Global PI curve with LIP, UIP, and the LIP+2 mbar pressure and theFuzzy optimal selection with according FLS based pressure.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 22 / 28
-
References I
Adler, A., Amato, M., Arnold, J., Bayford, R., Bodenstein, M., Bohm,S., Brown, B., Frerichs, I., Stenqvist, O., Weiler, N., and Wolf, G.(2012).Whither lung EIT: where are we, where do we want to go, and whatdo we need to get there?Submitted for publication to Journal of Applied Physiology.
EIDORS (2011).EIDORS: Electrical impedance tomography and diffuse opticaltomography reconstruction software.http://eidors3d.sourceforge.net/.
Graham, B. M. (2007).Enhancements in Electrical Impedance Tomography (EIT) ImageReconstruction for 3D Lung Imaging.PhD thesis, Carleton University.
http://eidors3d.sourceforge.net/
-
References
References II
Grychtol, B., Wolf, G. K., Adler, A., and Arnold, J. H. (2010).Towards lung EIT image segmentation: automatic classification oflung tissue state from analysis of EIT monitored recruitmentmanoeuvres.Physiological Measurement, 31(8):S31.
Grychtol, B., Wolf, G. K., and Arnold, J. H. (2009).Differences in regional pulmonary pressure impedance curves beforeand after lung injury assessed with a novel algorithm.Physiological Measurement, 30(6):S137.
Holder, D. (2004).Electrical Impedance Tomography: Methods, History, Applications.Institute of Physics Publishing.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 24 / 28
-
References
References III
Luepschen, H., Meier, T., Grossherr, M., Leibecke, T., Karsten, J.,and Leonhardt, S. (2007).Protective ventilation using electrical impedance tomography.Physiological Measurement, 28(7):S247.
Mendel, J. M. (2001).Uncertain rule-based fuzzy logic system: introduction and newdirections.Prentice-Hall PTR, 1st edition.
Pulletz, S., Adler, A., Kott, M., Elke, B., Hawelczyk, B., Schadler, D.,Zick, G., Weiler, N., and Frerichs, I. (2011).Regional lung opening and closing pressures in patients with acutelung injury.Journal of Critical Care, 0000(0000):0000.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 25 / 28
-
References
References IV
Victorino, J. A., Borges, J. B., Okamoto, V. N., Matos, G. F. J.,Tucci, M. R., Caramez, M. P. R., Tanaka, H., Sipmann, F. S., Santos,D. C. B., Barbas, C. S. V., Carvalho, C. R. R., and Amato, M. B. P.(2004).Imbalances in regional lung ventilation: A validation study onelectrical impedance tomography.Am. J. Respir. Crit. Care Med., 169(7):791800.
Wolf, G. K., Grychtol, B., Frerichs, I., Zurakowski, D., and Arnold,J. H. (2010).Regional lung volume changes during high-frequency oscillatoryventilation.11(5):610615.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 26 / 28
-
References
References V
Wu, D. and Mendel, J. M. (2011).Linguistic summarization using IFTHEN rules and interval type-2fuzzy sets.Fuzzy Systems, IEEE Transactions on, 19(1):136151.
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 27 / 28
-
References
ending
Fuzzy Logic Schematic
Ravi B. Bhanabhai (Carleton U) Safe Keeping Ventilation Patients 01/20/2012 28 / 28
IntroductionThe ProblemHow to solve the problem?
ContributionsIP CalculationFuzzy Logic System
ResultsSigmoid vs. LinearLinear vs. VisualOptimal PEEP
References