ECG Imaging of Ventricular Activity Multi-Parametric ... · l e.g. a linear model n y: dependent variable=output x: independent variable=input. 12.12.2017 Multi-parametric Measurements
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ECG Imaging of Ventricular Activity in Clinical ApplicationsWalther Schulze
INSTITUTE OF BIOMEDICAL ENGINEERING
www.ibt.kit.eduKIT – University of the State of Baden-Württemberg and National Research Center of the Helmholtz Association
Multi-Parametric Measurements and Data Analysis in Electrophysiology
Olaf Dössel
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
The Team of Olaf Dössel at IBT Karlsruhe
2
Tobias Oesterlein
Axel Loewe
Lukas Baron
Stefan Pollnow
Michael Kircher
Yannick Lutz
Nicolas Pilia
Tobias Gerach Ekaterina Kovacheva
Laura UngerSteffen Schuler
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Multi-Parametric Measurements in Life Sciences
Multi-parametric measurements in flow cytometry Multi-parametric analysis of immunoassay results Cell microscopy with various biomarkers
Multi-parametric MRI imaging Multi-channel ECG data, blood pressure, oxygenation,… Voltage clamp protocol measurements for ion channel modelling … and hundreds more
3
Uncover hidden values/characteristics/parameters of
cells/tissue/patients from a variety of measured data
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Outline
Multi-parametric measurements in life sciences Basics of multivariate data analysis Example 1: Classification of Complex Fractionated Atrial Signals Example 2: Finding the best ion-channel model parameters for genetic diseases Quantification of uncertainty
Personalization of patient models by merging imaging, biosignals, biomarkers and genetics
4
Uncover hidden values/characteristics/parameters of
cells/tissue/patients from a variety of measured data
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Multivariate Data Analysis Clustering, Classification, Modelling: Find a Quantitative Input-Output Relation
Clustering: Find an appropriate type and number of classes, so that each subject belongs to just one class.
Classification: find the right class for the each subject put the cells of cell cytometry into the right box (cell sorter) predict whether a drug will have positive or negative effect on a patient is this melanoma benign or malign? is this ECG signal a normal beat, a ventricular extrasystole or an atrial extrasystole goodness of classification: specificity and selectivity
Uncover relations in the data to better understand - a qualitative approach
Modelling: find a quantitative (thus mathematical) input-output relation identify the appropriate „input“ data, that can be measured, find the best model, that can describe all relations in the measured data, goodness of prediction, boundaries of confidence, statistical significance of model parameters
5
modelling
classification
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Our Objective in Medicine: Predict what will happen, if…
What will happen, if we do not do anything? risk of sudden ventricular fibrillation >> implant a defibrillator?
What will happen, if we do A or B (decision points in guidelines) is drug A better than drug B or the other way around?
What will happen, if we do more or less of A the best dosage of a drug? the best field and dosage of radiation therapy?
6
Can we trust the result? What will happen, if our input data have an error? What will happen, if the parameters of our model were erroneous? What will happen, if we go beyond the boundaries of the model? What will happen, if we did not take important other aspects into account?
So we use computer assisted data analysis. But….
classificationmodelling
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
define an error function solve the problem of „optimization“
Is it a convex error function? Or are there any local minima? How to find the best free parameters of the algorithm?
Downhill-Simplex, wikipedia
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Optimization of the Optimization Parameters
Example: find the location of a ventricular extrasystole from multichannel ECG using support vector regression Parameters of the Support Vector Regression: C and s
s is the bandwidth of the Gaussian Kernel C takes care of the tradeoff between flatness and error
Within the training dataset the ground truth is known. Try out all combinations of and C and s and take the best.
13
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Software Packages
MATLAB
14
… and many more
Facebook Open Source
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Example 1: Classification of
Complex Fractionated Atrial Signals CFAEs
15
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Classification of Complex Fractionated Atrial Electrograms (CFAEs)
Many patients suffer from Atrial Flutter or Atrial Fibrillation. An RF-Ablation could solve the problem. Where are the areas in the atria, that drive the arrhythmia?
16
Zeit (s)
0 1 2 3 4 5
C3
C2
C1
C0
some electrograms measured in the atria
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Descriptors (Features) of CFAEs
17
Segmentation
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1CFAE 5
f(x)
f(1) (x
)
f(x)
f‘(x)
Phase Space
Harmonics
CF1CF2 = DF
Frequenz
Frequency Domain
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.60
50
100
150
200
250
300CFAE 5
Signalamplitude
Anz
ahl
Signalamplitude
Anz
ahl
Kurtosis =E{(X � µ)4}
�
4=
µ4
µ
22
Histogram
...
0 500 1000 1500 2000 25000
500
1000
1500
2000
t [samples]
NLEO [(mV)
2 ]
0 500 1000 1500 2000−150
−100
−50
0
50
100
t [samples]
Signal [mV]
NLEO
Signal
NLEO
1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2 3.4
CFAE5
CFAE4
Common CFAEs
Zeit in Sekunden
1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2 3.4
CFAE5
CFAE4
Common CFAEs
Zeit in Sekunden
Time Domain
1120 1130 1140 1150 1160 1170
−2
−1
0
1
2
3
4
5
6
x 10−5
Zeit / ms
0 5 10 15 20 25−1
−0.5
0
0.5
1
1.5coif4
Coiflet 4
Wavelet Features
CFAE 0
CFAE 1
CFAE 2
CFAE 3
{
{
C0
C1
C2
C3
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Classification using a Fuzzy Decision Tree
18
The fuzzy decision tree delivers a probability of class membership
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Benchmarking Models of Human Atrial Electrophysiology
22
Ito IKur IKr IKs IK1IKACh If
INa ICaL IbCa IbNa IbCl ICl(Ca)
NCX
NKA
PMCA[Cl2-]i
[K+]i
[Na+]i
[Ca2+]i
Common to allmodelsGrandi et al. only
Maleckar et al. only
Koivumäki et al. only
SR
IKp B
C
-100-80-60-40-20
0 20 40 60
0 50 100 150 200 250 300 350 400
volta
ge (m
V)
time (ms)
Courtemanche et al.Nygren et al.
Maleckar et al.Koivumäki et al.
Grandi et al.
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0 100 200 300 400 500 600
[Ca2+
] i ( µ
M)
time (ms)
D
A
extracleft
dyad
IleakItr
Irel
Idi
IrelIup
Itr IrelIup
Nygren et al.Maleckar et al. Koivumäki et al.
Courtemanche et al.
subintra
extra
ICaLICaL
intra
Iup
ICaL
extraintra
Iup
intra
extrasub
cleft
ICaL
Irel Itr
Grandi et al.
Wilhelms, M.; Hettmann, H.; Maleckar, M. M. C.; Koivumäki, J. T.; Dössel, O.; Seemann, G., Benchmarking electrophysiological models of human atrial myocytes, Frontiers in Physiology, 3, 1-16, 2013
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Voltage Clamp Protocol for Characterization of Chanellopathies
Adapting the IKr channel due to hERG mutations
23
I Kr (µA
)
t (s)
-10-8-6-4-2 0 2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
N588K400 ms 200 ms15 ms
-80 mV -80 mV
+50 mV
400 ms-120 mV
-70 mV
-10-8-6-4-2 0 2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
WT K897T
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
simuliertgemessen xI K
r (µA
)
t (s)
WT
simulatedmeasured
L532P
COO–NH3+
S3 S4 S5 S6S2S1
Voltage clamp protocolLoewe, A.; Wilhelms, M.; Fischer, F.; Scholz, E. P.; Dössel, O.; Seemann, G., Arrhythmic potency of human ether-a-go-go-related gene mutations L532P and N588K in a computational model of human atrial myocytes, Europace, 16, 435-443, 2014
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
The parameters of the Ion Channel IKr
24
I
Kr
=g
Kr
x
r
(Vm
�E
K
)
1+ exp
⇣V
m
+g
Kr1g
Kr2
⌘
axr
= Xr
a1V
m
+Xr
a2
1� exp
⇣V
m
+Xr
a2Xr
a3
⌘
bxr
= 7.3898e�5V
m
+Xr
b1
exp
⇣V
m
+Xr
b1Xr
b2
⌘�1
txr
= [Xr
KQ10(axr
+bxr
)]�1
x
r• =
1+ exp
✓�V
m
+Xr
m1
Xr
m2
◆��1
E
K
=RT
F
log
[K+]o
[K+]i
Find 12 model parameters, so that all voltage clamp measurements can be reproduced.
Particle Swarm Algorithm
Lutz, 2008
3.1. Mathematical Modeling of Cardiac Cells 17
Ib,Ca Ib,Na INa
ICa,L
INaCa Ip,Ca Ca2+
Ca2+
3Na+
2K+
3Na+ INaK Ito IKur IKr IKs
IK1
Iup Itr
Irel
NSR JSR
Cytosol
Extracellular Space
Ileak
Fig. 3.2. Schematic of the atrial cell modeled by Courtemanche et al..
ion concentrations are not included in the model nor are processes of intracellular Ca
Loewe, A.; Wilhelms, M.; Schmid, J.; Krause, M. J.; Fischer, F.; Thomas, D.; Scholz, E. P.; Dössel, O.; Seemann, G., Parameter estimation of ion current formulations requires hybrid optimization approach to be both accurate and reliable, Frontiers in Bioengineering and Biotechnology, 3, 209, 2016
Institute of Biomedical EngineeringAxel Loewe - In Silico Assessment of Dynamic Drug Effects on Human Atrial Patho-Electrophysiology07.05.2015
Predict the Arrhythmogeneity: Effect on rotor inducability and lifetime
26
-2 +2 +10 +350 +6 +20
0 0.2
0.4
0.6
0.8
1
Zeit (s)
Time (s)0 521 43
Control
N588K-1:1
L532P-1:1
4 cm
Stimulation time with respect to end of refractory phase
Loewe et al., Europace, 2014
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Quantification of Uncertainty
27
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Evaluate the Quality of a Classification Problem
Train the method with data A, evaluate with data B Measure sensitivity, selectivity, ….
If there are only few data, where the ground truth is known? cross validation
confusion matrix for binary classification problems
Statistical significance? Is the p-value below the limit?
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Quantification of Uncertainty for Mathematical Modelling
29
propagation of measurement errors of the input data
influence of errors of model parameters on the result interindividual variability of model parameters errors due to application with variables outside of the recommended boundaries
errors due to using the wrong model, errors due to neglecting important variables.
easy model - easy solution non-linear model - not so easy…
sensitivity analysis - time consuming…
???
???
???
latin hypercube sampling
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Quantification of Uncertainty
s1
s2
s1 and s2 : vector in state space
prediction of the mathematical model
uncertainty of the model
uncertainty of the initial
state
uncertainty of prediction
s3
s3 uncertainty by omitting one dimension in state space
30
A good mathematical model of uncertainties is better than intuition.
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Personalization of patient models by merging imaging, biosignals, biomarkers and genetics.
31
Uncover hidden values/characteristics/parameters of
cells/tissue/patients from a variety of measured data
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Atrial Electrogram
32
[PentaRay Mapping Catheter, Biosense]
[Constellation, Boston Scientific] and FIRMap, Topera
2.4. CATHETER ABLATION OF ATRIAL FIBRILLATION 21
(a) A typical 4-pole ablation catheter with equidistantspaced electrodes. The tip electrode at the distalend of the catheter is 4 mm, all other electrodes are2 mm. The inter-electrode spacing is 2 mm.
(b) A multi-polar circular mapping catheter. The elec-trodes have a width of 1 mm and the inter-electrodespacing is 1 mm and 4 mm, respectively.
Fig. 2.13: Two examples of common catheters used during catheter ablation. The straight catheter on theleft is used for ablation as well as for analyzing EGMs. The circular catheter on the right is atypical analyzing catheter that gives spatial information about the tachycardia.
Therefore, tip temperature and impedance are constantly monitored during the ab-lation process. To avoid local temperature peaks at the electrode tip, irrigated tipelectrodes are used [65]. To ensure not only superficial but also transmural lesions,ablation currents are typically applied for 60 s to set an ablation point.
(a) The ablation catheter is pushed via the inferiorvena cava into the right atrium.
(b) Crossing the intra-atrial septum, irregular tis-sue can be destroyed by use of electrical en-ergy.
Fig. 2.14: Radio-frequency catheter ablation. Displayed is a typical ablation catheter with four electrodes.
[Orbiter, Boston Scientific] and: Lasso, Biosense
[Inquiry AFocusII, St. Jude Medical]
[Rhythmia Orion, Boston Scientific]
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Steps to Personalized Modeling of Atrial Tachycardia
33
70 year old female, persistent atrial flutter, cycle length ≈460ms Macro-Reentry passing roof and posterior wall Critical substrate: slow conducting region at central anterior wall
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Steps to Personalized Modeling of Atrial Tachycardia
34
individual geometry of the atria imported LAT pattern imported to MATLAB
Manual initialization: Lines of block, refractory tissue, trigger
CV estimated using FaMaS
Resulting excitation in agreement with clinical data:
Region of slow conduction correctly identified Cycle length corresponds to clinical data
… opens the door to computer assisted personalized therapy optimization.
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
The Integrated Heart Model
35
0 1 2 3 4 5 6 7 8
B
A
�A
Time [s]0 1 2 3 4 5 6 7 8
B
A
�V
Time [s]
Institute of Biomedical Engineering12.12.2017 Multi-parametric Measurements and Data Analysis in Electrophysiology
Algorithms and Software for Medical Applications
… give a soft advice ("second opinion“)
36
… give a strong advice (follow the guide lines!)
… are part of a closed circle
… will need a CE mark „medical device“ and an FDA approval
open questions: There are too many input / output options to test them all! How can we validate a computer model? How can we validate a „machine learning“ algorithm that ist changing every day? Do we have to approve the „learning dataset“?