embs2007_torres

8/6/2019 EMBS2007_Torres

1/24

Application of the Empirical Mode Decomposition method to the Analysis of

Respiratory Mechanomyographic Signals ([email protected])

APPLICATION OF THE EMPIRICAL MODEDECOMPOSITION METHOD TO THE ANALYSIS OF

RESPIRATORY MECHANOMYOGRAPHIC SIGNALS

Abel Torres1, J. A. Fiz2, B. Gldiz3, J. Gea4, J. Morera2and R. Jan1

1 Dept. ESAII, Centre de Recerca en Enginyeria Biomdica, UPC, Barcelona, Espaa.2 Dept. of Pneumology, Hospital Germans Trias i Pujol, Badalona, Espaa

3

Dept. of Pneumology, Hospital Cruces, Baracaldo, Espaa4 Dept. of Pneumology, Hospital del Mar, Barcelona, Espaa

E-mail: [email protected]

29th IEEE EMBS Annual International Conference, August 23-26, 2007, Lyon, France


2/24



I. INTRODUCTION

During muscular contraction takes place a transverse movement produced by the

lateral expansion of the active muscle fibers. This movement can be separated in

two parts, according to the movement type: MOV: A movement of great amplitude, due to the overall lateral movement of

the muscle, produced by the sum of the variation in diameter of all the active

muscle fibers during the contraction.

VIB: A movement of small amplitude that consist in oscillations or vibrations,

produced by the repetitive and asynchronous contraction of the muscle fibers

(mechanomyogram: MMG)

20 mm

MUSCLE SURFACE

PIEZOELECTRIC

ELEMENTSEISMIC

MASS

10

mm

ACCELEROMETERS (ACC)

10-15 mm

MUSCLE SURFACE

ELECTRET

MICROPHONE

AIR

CHAMBER

2mm

5 mm

AIR COUPLED

MICROPHONES

2

9mm

28 mm

PIEZOELECTRIC

ELEMENT

SEISMI

C MASS

MUSCLE SURFACE

TEE TIP

PIEZOELECTRIC

CONTACT SENSORS (PCS)

Both types of movement can be acquired by means different kind of surface

sensors:


3/24



I. INTRODUCTION

LF component (0-5 Hz) large

movement of the thoracic cage

-0.02

-0.01

0.01

0.02

0 f >5 Hz

-0.02

-0.01

0

0.01

0.02

f


4/24



II. OBJECTIVES

A critical point in MMG studies is the selection of the cut-off frequency that

must separate the low frequency (MOV) and high frequency (VIB-MMG)

components of the muscle movement signal The selection of the cut-off frequency of the filter

is today still controversial (DC, 2 Hz, 5 Hz, 10 Hz, )

Fourier-Wavelet decomposition methods: fixed basis functions (do not

necessarily match the varying nature of MMG signals).

Empirical Mode Decomposition (EMD): adaptively decomposes the signal

into oscillating components.

The main objective of this study is introduce the EMD technique to

decompose vibratory MMG signals in order to remove respiration andother low frequency artifacts and to relate the resulting high frequency

component with the force developed by the respiratory muscles


5/24



III. MATERIAL & METHODS

Population of study:

- Two traqueostomized mongrel dogs (15-20 kg)

- Animals awake and in all fours position during the study

Respiratory tests:

- Spontaneous ventilations against a progressive

resistive inspiratory load

A. Signal acquisition


6/24




Experimental Setup

Mechano-

myographic

signal

(MMG)

Inspiratory

Pressure

(Pins )

18 mm

5 mm

Kistler 8302A Capacitive

Accelerometer

0 2 4 6 8 10 12 14 16 18 20

t (s)

0

2

4

6

8

10

Inspiratory Pressure (Pins )

(cmH2O)

-0.04-0.02

0

0.02

0.04(V)

Mechanomyographic signal (MMG)

12 bit A/D system, 4 kHz sampling frequency, decimation 100 Hz

A. Signal acquisition


7/24



III. MATERIAL & METHODSA. Signal acquisition

5 10 15 20 25

0

5

10

15

20

25

30

1 2 3 4 5 6

Inspiratory Pressure Signal

P(cmH2O)

t(s)

5 10 15 20 25-0.05

0

0.05

1 2 3 4 5 6

Diaphragmatic MMG signal

MMG(V)

t(s)


8/24



III. MATERIAL & METHODSB. Empirical Mode Decomposition (EMD)

The aim of the EMD is to decompose the signal into a sum of oscillating

components or Intrinsic Mode Functions (IMFs).

The IMFs are obtained adaptively directly from the signal

(not required a priori known basis)

An IMF is defined as a function that satisfies two conditions:

(1) the number of extrema and zero crossings must be either

equal or differ at most by one

(2) the mean value of the envelope defined by the local maximaand the envelope defined by the local minima must be zero (or

close to zero).


9/24




Given a signalx(t), the algorithm of the EMD can be summarized as follows:

10 20 30 40 50 60 70 80 90 100 110 120

-2

-1

0

1

2

Original signal: x(t)=d0(t)

(http://perso.ens-lyon.fr/patrick.flandrin/emd.html)


10/24





10 20 30 40 50 60 70 80 90 100 110 120

-2

-1

0

1

2


1. Identify all local maxima and minima ofd0(t)=x(t).



11/24

Application of the Empirical Mode Decomposition method to the Analysis ofRespiratory Mechanomyographic Signals ([email protected])



10 20 30 40 50 60 70 80 90 100 110 120

-2

-1

0

1

2


2. Obtain the upper and lower envelopes eu(t) and el(t)

1. Identify all local maxima and minima ofd0(t)=x(t)



12/24




10 20 30 40 50 60 70 80 90 100 110 120

-2

-1

0

1

2




3. Compute the mean of the envelopes: m(t)=(eu(t)+el(t))/2



13/24




10 20 30 40 50 60 70 80 90 100 110 120

-2

-1

0

1

2


4. Extract the detail d1(t)= d0(t)-m(t)

10 20 30 40 50 60 70 80 90 100 110 120

-1.5

-1

-0.5

0

0.5

1

1.5

residue


No IMF(extrema

zeros 1)


14/24


B. Empirical Mode Decomposition (EMD)



d1(t)

10 20 30 40 50 60 70 80 90 100 110 120

-1.5

-1

-0.5

0

0.51

1.5

5. Iterate 1-4 until dk(t) can be considered an IMF

10 20 30 40 50 60 70 80 90 100 110 120

-1

-0.5

0

0.5

1

residue

No IMF(extrema

zeros 1)


15/24





d2(t)

10 20 30 40 50 60 70 80 90 100 110 120

-1.5

-1

-0.5

0

0.5

1

1.5


10 20 30 40 50 60 70 80 90 100 110 120

-1

-0.5

0

0.5

1

residue

No IMF(extrema

zeros 1)


16/24





d8(t)


10 20 30 40 50 60 70 80 90 100 110 120

-1

-0.5

0

0.5

1

residue

10 20 30 40 50 60 70 80 90 100 110 120

-1

0

0.5

1

IMF OK:

IMF1=d8(t)


17/24







3. Compute the mean of the envelopes: m(t)=(eu(t)+el(t))/2

10 20 30 40 50 60 70 80 90 100 110 120

-1

-0.5

0

0.5

1

residue

4. Extract the detail d1(t)= d0(t)-m(t)

6. Iterate 1-5 in order to obtain all the IMFs of the signal



18/24



-0.010

0.01 1 2 3 4 5 6

IMF1

-0.02

0

0.02 1 2 3 4 5 6

IMF2

-0.010

0.011 2 3 4 5 6

IMF3

-0.01

0

0.01 1 2 3 4 5 6

IMF4

-0.020

0.021 2 3 4 5 6

IMF5

B. Empirical Mode Decomposition

-0.05

0

0.05 1 2 3 4 5 6

Signal

0 10 15 20 25 t(s)

MMG

PSD

0 25 50f(Hz)5

-0.04-0.02

00.020.04 1 2 3 4 5 6

Res.

0 25 50f(Hz)0 10 15 20 25 t(s)5

Lower IMFs: Fast oscillation modes Higuer IMFs: Slow oscillation modes

III M M


19/24



IMF1,IMF2, ,

IMF5,Res.,

um123

PINS

Data analysis

RMS and Shannon Entropy (H)

1 2 3 4 5 60.002

0.004

0.006

0.008

RMS

Maximum and Mean Inspiratory Pressure

0

10

20

30

PINS max

1 2 3 4 5 6

1 2 3 4 5 6

Identification of respiratory cycles, and detection of initial and final time of diaphragm

muscle contraction was made by means the PINS signal:

n. cycle

n. cycle

III M M


20/24



Data analysis

- Pearson correlation coefficient (r):

1 1 1

2 2

2 2

1 1 1 1

N N N

i i i i

i i i

N N N N

i i i i

i i i i

N x y x y

r

N x x N y y

= = =

= = = =

=

Relationship between maximum and mean PINS (PM and Pm) developed, and the

RMS and H obtained in the first five Intrinsic Mode Functions (IMF1,IMF2, ,IMF5), the residue (Res.), and the sum of the first three IMFs (IMF1+IMF2+IMF3)

were evaluated by means the Pearson correlation coefficient:

Maximum and Mean PINS:x

RMS and H:y0 5 10 150

0.02

0.04

0.06

0.08

RMS

PM

IV R


21/24


IV. RESULTS

CORRELATION COEFFICIENTS BETWEEN PINS AND MMGSIGNAL PARAMETERS

Correlation coefficients are higher with lower IMFs

Correlation coefficients are higher with the entropy parameter

The vibratory activity of the diaphragm can be reconstructed as the sum of

the first three IMFs

IMF Dog 1 Dog 2

Pm

PM

Pm

PM

RMS H RMS H RMS H RMS H

IMF1 0.60 0.63 0.63 0.64 0.75 0.71 0.69 0.66

IMF2 0.44 0.68 0.45 0.70 0.67 0.61 0.63 0.61

IMF3 0.17 0.43 0.17 0.41 0.67 0.65 0.63 0.63

IMF4 0.07 0.43 0.08 0.40 0.68 0.61 0.64 0.53

IMF5 0.28 0.47 0.28 0.36 0.60 0.57 0.54 0.48

Residue 0.21 0.20 0.19 0.17 0.16 0.37 0.12 0.29

Sum1,2,3 0.54 0.65 0.56 0.65 0.73 0.77 0.70 0.74

IV R S S


22/24


IV. RESULTS

0 5 10 150

0.02

0.04

0.06

0.08

0 5 10 15 200

0.01

0.02

0.03

0.04

0.05

0.06

0.07

20 40 60 800

10

20

0

0.35

0.7

0

0.35

0.7

20 40 60 800

5

10

15H H

H H

Pm Pm

n. cycle n. cycle

Pm Pm

0 5 10 150

0.02

0.04

0.06

0.08r=0.77

0 5 10 15 200

0.01

0.02

0.03

0.04

0.05

0.06

0.07r=0.65

Pm Pm

r=0.96 r=0.99

Relationship between Pm and IMF1+IMF2+IMF3:

IV DISCUSSION AND CONCLUSIONS


23/24


IV. DISCUSSION AND CONCLUSIONS

- The study of MMG signals acquired during dynamic contractions has the dilemma of

separation of the LF and HF components. The EMD method has the advantage

that automatically estimates the vibratory components of the MMG signal in thefirst 2-3 components of the EMD decomposition, and these components are highly

related with the force developed by the respiratory muscles.

- The EMD method could probably identify other types of vibration or movements that could not be related

with the respiratory activity. Nevertheless, this supposition should be validated with further experiments in

which it can be caused the appearance of this type of artifacts

IV DISCUSSION AND CONCLUSIONS


24/24

Application of the Empirical Mode Decomposition method to the Analysis ofRespiratory Mechanomyographic Signals ( b l t @ d )

IV. DISCUSSION AND CONCLUSIONS

-

Further investigation should be carried out to develop a more automatic criterion to interpret each IMF with its physiological

meaning, but, in general, the HF vibratory activity of the muscle is concentrated in the lower IMFs while the respiratory

movement LF component is concentrated in the higher IMFs of the EMD decomposition.

- The presented EMD method is an interesting technique to study the

LF and HF components of MMG signals, because it takes into

consideration the nonlinear and nonstationary nature of the MMGsignals.

embs2007_torres

Documents