Outline
What is at stake ?
Test Bench
Application
Conclusion
Towards a suitable time-scale representation of cardio-respiratory signals through EMD algorithms: a simulation and validation tool
C. Franco1,2 , J. Fontecave-Jallon1 , N. Vuillerme2 and P-Y. Guméry1 (1) TIMC-IMAG UMR 5525 Laboratory, UJF Grenoble 1-CNRS PRETA, Experimental, Theoretical and Applied cardio-Respiratory Physiology, team (2) AGIM FRE 3405 Laboratory, CNRS-UJF-EPHE-UPMF AFIRM, Acquisition, Fusion of Information and netwoRk for Medicine, team Grenoble, France Wednesday, August 31st 2011
0 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Outline
Context & aims in physiology in data processing
The test bench Application & results : The cardio-respiratory model Cardio-respiratory interactions analysis Test of EMD algorithms performances
Perspectives
1 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Outline
Context & aims in physiology in data processing
The test bench Application & results : The cardio-respiratory model Cardio-respiratory interactions analysis Test of EMD algorithms performances
Perspectives
1 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Outline
Context & aims in physiology in data processing
The test bench Application & results : The cardio-respiratory model Cardio-respiratory interactions analysis Test of EMD algorithms performances
Perspectives
1 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Outline
Context & aims in physiology in data processing
The test bench Application & results : The cardio-respiratory model Cardio-respiratory interactions analysis Test of EMD algorithms performances
Perspectives
1 / 22
Outline
What is at stake ?
Context
Test Bench
Application
Conclusion
& aims
2 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
From a physiological point of view
Non-invasive monitoring of the stroke volume from volumetric measurements by means of respiratory inductive plethysmography. Cardiac data retrieval from a cardio-respiratory (CR) signal.
*http://visuresp.com/index.html
3 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
From a signal processing point of view Dealing with :
non-stationnarities from respiratory and cardiac signals : swallowing, apnea, vocalization, eort... interaction between the two present physiological sources Putting back the EMD limits of separation
Figure: Behavior of EMD algorithm
towards the separation of a two-tone signal. Adapted from Rilling & Flandrin, IEEE Signal Process. Lett., 2007 [8].
Lack of analytical foundation
,→ Data-driven approach
4 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
From a signal processing point of view Dealing with :
non-stationnarities from respiratory and cardiac signals : swallowing, apnea, vocalization, eort... interaction between the two present physiological sources Putting back the EMD limits of separation
Figure: Behavior of EMD algorithm
towards the separation of a two-tone signal. Adapted from Rilling & Flandrin, IEEE Signal Process. Lett., 2007 [8].
Lack of analytical foundation
,→ Data-driven approach
4 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
From a signal processing point of view Dealing with :
non-stationnarities from respiratory and cardiac signals : swallowing, apnea, vocalization, eort... interaction between the two present physiological sources Putting back the EMD limits of separation
Figure: Behavior of EMD algorithm
towards the separation of a two-tone signal. Adapted from Rilling & Flandrin, IEEE Signal Process. Lett., 2007 [8].
Lack of analytical foundation
,→ Data-driven approach
4 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
A simulation and validation tool
5 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Test bench
Generation of cardiac and respiratory signals 1. Simulation
Simulated res-piratory signal Simulated cardiac signal
Simulated cardio-respira -tory signal
Time-scale representation : EMD, CEEMD algorithms 2. Data processing
Simulated stroke volume
Reconstructed respiratory signal Reconstructed cardiac signal Estimated stroke volume
SV
SV
res
ref
Bland & Altman test to compare SV with SV
Ability of EMD, CEEMD algorithms to separate components 3a. Limits of Agreement 3b. Method comparison ref
res
6 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Application
7 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
1. Generation of simulated CR signals (Fontecave-Jallon et al., Phil. Trans. R. Soc. A, 2009) [4]
Cardiac compartment Heart volume
Vh (t ) = Ah (t ).vh (t , fh ) 1/fh
Rib cage
Amplitude modulation
A (t )
Mechanical
h
||
a (t ). h
coupling
1
c
A (t ).c2
v
Cardio-respiratory volume
Vth (t )
1/fA
Alveolar volume
VA (t ) = aA (t ).vA (t , fA )
Ventilatory compartment 8 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
2. Signal Processing Time-scale representations of the simulated CR signal
Figure: EMD Decomposition
Figure: CEEMD Decomposition 9 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
2. Signal Processing Time-scale representations of the simulated CR signal
Figure: EMD Decomposition
Figure: CEEMD Decomposition 9 / 22
Outline
What is at stake ?
Test Bench
2. CEEMD algorithm
Application
Conclusion
(Yeh et al. AADA, 2010) [9]
Signal to decompose
White noise 1
+
White noise 2
−
+
White noise n
...
−
−
+
+
−
Empirical Mode Decomposition
IMF1,1 IMF1,2
IMF2,1 IMF2,2
IMF3,1 IMF3,2
IMF4,1 IMF4,2
IMF...,1 IMF...,2
IMF...,1 IMF...,2
IMF1,m
IMF2,m
IMF3,m
IMF4,m
IMF...,m
IMF...,m
...
...
Means through the dierent scales
...
...
...
P2n
IMF1 = (
IMF = ( m
n−1,1 IMF2n,1 n−1,2 IMF2n,2 ... ...
IMF2 IMF2
n−1,m IMF2n,m
IMF2
1 IMFi ,1 )/(2n)
i=
P2n
IMF2 = (
...
=1 IMF ,2 )/(2n) ... i
P2n
i=
i
1 IMFi ,m )/(2n)
Aim : Overcoming the Mode-Mixing phenomenon (Abdulhay IEEE EMBC, 2009) [1]
et al.
Means : Redistributing the scales uniformly 10 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
3a. Limits of agreement Cardiac volume reconstruction
Figure: Comparison of the cardiac volume (plain) reconstructed from EMD and CEEMD resp. with the one of reference (dotted) Discrepancies appear with the EMD approach 11 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
3a. Limits of agreement Stroke volume reconstruction
Stroke volume : volume of blood ejected with each beat Amplitude of the cardiac volume at each cardiac cycle
,→
Stroke volume agreement
Bland & Altman statistical test : 92 measures over 120s Limits of agreement with 95% condence interval : up to ±30% (Critchley & Critchley, J Clin. Monitor. Comp., 1999) [3]
Figure: Bland-Altman plot (Bland &
Altman, Lancet, 1986) [2] : comparison of SV with SV estimated from CEMMD algorithm SV . ref
CEEMD
12 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
3b. Method comparison
Bland & Altman statistical test for 143 simulations Limits of agreement depending on amplitude a = aA /ah and frequency f = fA /fh ratios :
Better achievement of CEEMD approach in the separation of the cardiac and the respiratory components. 13 / 22
Outline
What is at stake ?
Test Bench
Application
Conclusion
Final thoughts Limitations
Simulation : Improvement of both the respiratory and cardiac models Signal processing : Unsupervised selection of the cardiac IMFs Future prospects
Clarify the result discrepancies observed between EMD and CEEMD algorithms (resolution, ...) (Fontecave-Jallon et al., GRETSI, 2011) [5] Validation on real data by coupling plethysmography and cardiac impedancemetry (Fontecave-Jallon et al., P'Health, 2011) [6]
14 / 22
Application
Acknowledgement
This research was granted by the company IDS SA (Montceau les Mines, France) through a CIFRE process ran by ANRT and initiated by the French Ministry of Higher Education and Research.
15 / 22
Application
References I [1]
E. Abdulhay, PY Gumery, J. Fontecave, and P. Baconnier.
[2]
J. M. Bland and D. G. Altman.
[3]
L. A. H. Critchley and J. A. J. H. Critchley.
[4]
J. Fontecave Jallon, E. Abdulhay, P. Calabrese, P. Baconnier, and PY. Gumery.
Cardiogenic oscillations extraction in inductive plethysmography : Ensemble empirical mode decomposition. In Engineering in Medicine and Biology Society, 2009. EMBC 2009. Annual International Conference of the IEEE, volume 1, pages 22402243. Conf Proc IEEE Eng Med Biol Soc, Sept. 2009. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet, 1(8476) :307310, Feb 1986. A meta-analysis of studies using bias and precision statistics to compare cardiac output measurement techniques. Journal of Clinical Monitoring and Computing, 15(2) :8591, February 1999. A model of mechanical interactions between heart and lungs.
Philosophical Transactions of the Royal Society A : Mathematical,Physical and Engineering
, 367(1908) :47414757, 2009.
Sciences
[5]
J. Fontecave-Jallon, E. Abdulhay, and PY. Guméry.
[6]
J. Fontecave-Jallon, PY. Guméry, P. Calabrese, R. Briot, and P. Baconnier.
Empirical mode decomposition for the investigation of cardio-respiratory interactions within volumetric signals : a simulated approach. In GRETSI'11, 2011. A wearable technology revisited for cardio-respiratory functional exploration : Stroke volume estimation from respiratory inductive plethysmography. In pHealth, Lyon, France, July 2011 2011.
16 / 22
Application
References II
[7]
G. Rilling and P. Flandrin.
[8]
G. Rilling, P. Flandrin, P. Gonalves, and J.M. Lilly.
[9]
J.R. Yeh, J.s. Shieh, and N.e. Huang.
One or two frequencies ? the empirical mode decomposition answers. Signal Processing, IEEE Transactions on, 56(1) :8595, Jan. 2008. Bivariate empirical mode decomposition. Signal Processing Letters, IEEE, 14(12) :936939, Dec. 2007. Complementary ensemble empirical mode decomposition : a novel noise enhanced data analysis method. Advances in Adaptative Data Analysis, 2(2) :135156, 2010.
Contact :
[email protected]
17 / 22
Application
EMD ability to separate two nonlinear components
Figure: One or two frequencies ? Rilling's work Conclusion
EMD achievement depends on amplitude and frequency ratios from the present components[7]. 18 / 22
Application
Application to real data
19 / 22
Application
Application to real data
IMF3 to IMF6 were selected and summed up to reconstruct the cardiac volume. 20 / 22
Application
Stroke volume estimation
Figure
: Comparison of the cardiac signal reconstructed from RIP with the one from ECG
Figure : For each cardiac cycle, stroke volume is estimated as the amplitude of the cardiac signal from RIP
21 / 22
Application
Results for 24 healthy subjects [6] RIP versus cardiac impedance
Figure
: Correlation between stroke volumes extracted from cardiac RIP and impedance signals
Figure
: Bland & Altman plot to compare RIP and impedance measurements
Limitations :Scale mixing persist
22 / 22