This article was downloaded by: [Eindhoven Technical University]On: 19 October 2014, At: 23:06Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,UK

Instrumentation Science &TechnologyPublication details, including instructions forauthors and subscription information:http://www.tandfonline.com/loi/list20

Determination of FemoralArtery Occlusion UsingPrincipal Component Analysisof Doppler SignalsSadık Kara a & Prof. Semra Kemaloğlu b

a Erciyes University, Department of ElectronicsEngineering , Kayseri, Turkeyb Erciyes University, Department of BiomedicalDevices Technology , Kayseri, TurkeyPublished online: 16 Aug 2006.

To cite this article: Sadık Kara & Prof. Semra Kemaloğlu (2005) Determination ofFemoral Artery Occlusion Using Principal Component Analysis of Doppler Signals,Instrumentation Science & Technology, 33:3, 329-338, DOI: 10.1081/CI-200056144

To link to this article: http://dx.doi.org/10.1081/CI-200056144

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all theinformation (the “Content”) contained in the publications on our platform.However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness,or suitability for any purpose of the Content. Any opinions and viewsexpressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of theContent should not be relied upon and should be independently verified withprimary sources of information. Taylor and Francis shall not be liable for anylosses, actions, claims, proceedings, demands, costs, expenses, damages,and other liabilities whatsoever or howsoever caused arising directly or

indirectly in connection with, in relation to or arising out of the use of theContent.

This article may be used for research, teaching, and private study purposes.Any substantial or systematic reproduction, redistribution, reselling, loan,sub-licensing, systematic supply, or distribution in any form to anyone isexpressly forbidden. Terms & Conditions of access and use can be found athttp://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

Determination of Femoral Artery OcclusionUsing Principal Component Analysis of

Doppler Signals

Sadık Kara

Erciyes University, Department of Electronics Engineering,

Kayseri, Turkey

Semra Kemaloglu

Erciyes University, Department of Biomedical Devices Technology,

Kayseri, Turkey

Abstract: The aim of this study is to scrutinize the ability of principal component

analysis (PCA) over power spectral densities (PSD) for common femoral artery

blood flow study. Doppler femoral artery signals of patients with occluded arteries and

of healthy subjects were recorded. Then, power spectral densities of these signals were

obtained using the Welch method. To clearly determine the difference between the

groups of occluded patients and healthy subjects, PCA was implemented with

patients and healthy matrices derived from PSD. The basic differences between

the healthy and occluded patients were acquired with 1st principal component. The

use of PCA of physiological waveforms is presented as a powerful method likely to

be incorporated into future medical signal processing.

Keywords: Femoral artery occlusion, Power spectral density, Principal component

analysis

INTRODUCTION

Doppler ultrasound is widely used as a non-invasive method for the assess-

ment of blood flow, both in the central and peripheral circulation. It may be

Address correspondence to Prof. Sadık Kara, Erciyes University, Department of

Electronics Engineering, Biomedical Eng. Group, 38039 Kayseri, Turkey. E-mail:

kara@erciyes.edu.tr

Instrumentation Science and Technology, 33: 329–338, 2005

Copyright # Taylor & Francis, Inc.

ISSN 1073-9149 print/1525-6030 online

DOI: 10.1081/CI-200056144

329

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

used to estimate blood flow, to image regions of blood flow, and to locate sites

of arterial disease, as well as flow characteristics and resistance of common

femoral arteries.[1–4]

Doppler systems are based on the principle that ultrasound, emitted by an

ultrasonic transducer, is returned partially towards the transducer by the

moving targets, thereby inducing a shift in frequency proportional to the

emitted frequency and the velocity along the ultrasound beam. The results

of the studies in the literature have shown that Doppler ultrasound evaluation

can give reliable information on both systolic and diastolic blood velocities of

arteries and have supported that Doppler ultrasound is useful in screening

certain hemodynamic alterations in arteries.[1–4]

In recent years, color Doppler ultrasonography (CDU) has found increa-

sing use in the assessment of vascular disease in lower limb circulation,

displaying high sensitivities when compared to arteriography.[5–7] CDU is

relatively cheap, quick, non-invasive, and safe, whereas angiography carries

significant risks of morbidity and mortality.[5]

Doppler ultrasound is usually the first-line investigation of lower limb

arterial disease and, thus, identifies patients with treatable peripheral vascular

disease.

The accuracy of the interpretation of any changes present in the shape of

the common femoral arterial (CFA) waveform becomes all the more vital,

either in confidently excluding the presence of significant proximal disease

(and, thus, ensuring that the patient does not undergo an unnecessary

angiogram with its concomitant risks)[8] or in identifying and assessing any

proximal disease and, thus, indicating further investigation and/or treatment.

Spectral analysis and signal decomposition continue to find wide use in a

multitude of biomedical signal processing applications. Spectral analysis of the

Doppler signals produces information concerning the blood flow in the arteries.

The maximum frequency envelope from the Doppler waveforms obtained

from the common femoral artery retrospectively are analyzed using a mathe-

matical feature extraction technique, i.e., principal component analysis (PCA).

The results were compared with the arteriographic findings and showed that

PCA represents a significant improvement in diagnostic accuracy when

compared with other techniques.[9]

Guo et al. compared time-frequency distributions of femoral artery

Doppler signals using various methods. The results showed that the Bessel

distribution performed the best, but the Choi-Williams distribution and auto-

regressive modeling are also techniques which can generate good time-

frequency distributions of Doppler signals.[10]

Eiberg et al. studied common femoral Doppler waveforms and demon-

strated that only three simple waveform characteristics belonging to common

femoral arteries were necessary in order to characterize the waveform as

normal or abnormal and to predict upstream aorto-iliac disease.[11]

Wright and Gough applied artificial neural networks (ANNs) to the

problem of the diagnosis of aorto-iliac arterial disease on the basis of the

S. Kara and S. Kemaloglu330

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

profile of the common femoral artery (CFA) Doppler flow velocity waveform.

Thus, they demonstrated the ability of an ANN to identify the severity of

aorto-iliac disease from the CFA waveform.[12]

Smith et al. compared network classification results with a Bayesian

classifier following a principal component analysis of the waveforms.[13]

Several authors have reported good results using, during the past decades,

the diagnostic performance of ultrasound complex analysis of the common

femoral waveform.[14,15] However, none of these methods have gained wide-

spread use, mainly because of their complexity and need for additional

equipment.

Nevertheless, subjective visual examination of the CFA waveform profile

has a potential for inter- and intra-observer variability. To overcome this,

several methods of numerical analysis have been explored. Of these, the

most straightforward are the waveform pro-file indices, such as the pulsatility

index (PI)[16] and the Pourcelot or resistance index (RI).[17] More sophisticated

methods have also been developed, such as the Laplace transform[18] and

principal components analysis (PCA).[19] But, neither the simple nor the

more complex analytical techniques have yielded an acceptable diagnostic

accuracy to make them commonplace in the vascular clinic.

The aim of this study was to apply and evaluate a principal component

analysis method to power spectral density acquired with Welch’s averaged,

modified periodogram of femoral artery Doppler signals, because more

reliable method was revealed to diagnose femoral artery congestion.

EXPERIMENTAL

Subjects

In this study, common femoral artery Doppler signals were obtained from 54

subjects. The group consisted of 22 females and 32 males, with ages ranging

from 20 to 65 years and mean age 36.5 years (standard deviation, SD ¼ 9.1).

Color Doppler Ultrasound was used during examinations and sonograms were

taken under study. According to examination results, 30 of 54 subjects

suffered from femoral artery occlusion and the rest were healthy subjects.

The group having femoral artery occlusion consisted of 12 females and 18

males, with a mean age 37.5 years (SD ¼ 7.8, range 22–65) and the

healthy subjects consisted of 10 females and 14 males with a mean age 34.5

years (SD ¼ 8.0, range 20–60).

Measurement of Common Femoral Artery Doppler Signals

Flow signals from common femoral arteries were recorded via a color Doppler

Ultrasound unit (Toshiba PowerVision 6000, Osaka, Japan). A linear

Determination of Femoral Artery Occlusion Using PCA 331

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

ultrasound probe of 10MHz was used to transmit pulsed ultrasound signals to

arterials. The subjects were lying supine and breathing naturally during the

data sampling. In all recordings, the insonation angle and the presetting of

the ultrasound were kept constant. The ultrasonic transducer was applied,

on a horizontal plane, to the inguinal using a water-soluble gel as a

coupling agent. Care was taken not to apply pressure to the inguinal in

order to avoid artifacts. The insonation angle was adjusted, via electronic

steering methods, and manually, in order to keep a constant value of 45

degrees on a longitudinal view. The amplification gain was carefully set to

obtain a clean spectral output with minimal background noise on the

spectral display. The audio output of ultrasound unit was sampled at

44,100Hz and then sent to a PC via an input-output card.[20] Thus, the

system hardware was composed of Digital Doppler Ultrasound unit that can

operate in the pulsed mode, a linear ultrasound probe, an input-output card,

and a personal computer (PC) which was used for storage, display, and

spectral analysis of the acquired data (Figure 1).

Welch Method for Spectral Analysis

The FFT based Welch method is defined as a classical (nonparametric)

method. In the Welch method, signals are divided into overlapping

segments; each data segment is windowed. Thus, the segments can be

defined as

xmðnÞ ¼ xðnþ mDÞ n ¼ 0; 1; . . . ; L� 1 m ¼ 0; 1; . . . ; K � 1 ð1Þ

where D is the size of the segments and D = L. According to this segmen-

tation, A second modification is made, which is to window data segments prior

to computing the periodogram.[21] As a result

PðmÞPERð f Þ ¼

1

LU

XL�1

n¼0

xmðnÞwðnÞ expð�j2pfnÞ

�����

�����

2

ð2Þ

Figure 1. Block diagram of the system hardware used to acquire Doppler data.

S. Kara and S. Kemaloglu332

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

where

U ¼1

M

XM�1

n¼0

w2ðnÞ ð3Þ

and the Welch spectrum estimate is the average of these modified periodo-

grams,

PW ð f Þ ¼1

K

XK�1

m¼0

PðmÞW ð f Þ ð4Þ

In this study, X is divided with 50% overlap, each section is windowed

with a Hamming window, and modified periodograms are computed and

averaged.

Principal Component Analysis for Doppler Spectral Waveforms

Principal Component Analysis (PCA) is a way of identifying patterns in data,

and expressing the data in such a way as to highlight their similarities and their

differences. Since patterns among data can be hard to find for high dimen-

sional data, where the luxury of graphical representation is not available,

PCA is a powerful tool for analyzing the data. The other main advantage of

PCA is that, once you have found these patterns in the data, you can

compress the data, ie., by reducing the number of dimensions, without

much loss of information.[22]

In this paper, the use of PCA for the characterization and interpretation of

physiological signals was explained. Principal components are calculated

using eigenvectors and eigenvalues of covariance matrices or correlation

matrix.

CovðXijÞ ¼1

n� 1

Xn

k¼1

ðXik � �XiÞðXjk � �X jÞ ð5Þ

n: total number of signals (30 for patients, 24 for healthy), i ¼ 1, 2, . . . ,m,j ¼ 1, 2, . . . ,m (m signal dimension). Xi is the average signal of the

population.

First, three principal components were computed by the solution of

Cwp ¼ lpwp p ¼ 1; 2; 3: ð6Þ

C, covariance matrix, wp, pth principal component (eigenvector), and lp is

the corresponding eigenvalue. The lp’s are positive values which are

proportional to the fraction of the total variance accounted for by each

component wp which has the important property of forming an orthogonal set.

Determination of Femoral Artery Occlusion Using PCA 333

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

The coefficients of the principal components of qth signal are then

given by

aqp ¼Xm

i¼1

xqiwip p ¼ 1; 2; 3: q ¼ 1; 2; . . . ; n: ð7Þ

Data matrices were formed from the power spectral densities of femoral

artery Doppler signals belonging to patients and healthy subjects, separately.

PCA was applied to these matrices. Matlab 6.5 was used for signal processing

and power spectral density procedures, and SAS 9.1.2 data analysis software

was used for the PCA.

Although only the first three components have been computed, the first

component indicate that 97.9 and 98.2 percent of the total variance of X is

explained by this reduced set of components for patients and healthy

subjects, respectively, and therefore, only minimal information is lost by

ignoring higher order components.

RESULTS AND DISCUSSION

The demonstration of Doppler shift frequency magnitude on a time domain

belonging to femoral artery occlusion patients and healthy persons is given

in Figure 2. In this figure, the signals represented in time domain seem

similar to each other and there is no obvious difference between them.

Figure 3 shows power spectral density graphics of a femoral artery

occlusion patient (patient, no. 28) and healthy person (healthy, no. 19). In

the graphics of patients who have femoral artery occlusion; the blood

velocity falls sharply from maximum to minimum. Conversely, the graphics

Figure 2. Time domain graphics of common femoral artery Doppler signals.

(a) Healthy person (healthy, no. 19); (b) patient with occlusion (patient, no. 28).

S. Kara and S. Kemaloglu334

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

of healthy subjects have stepped fall characteristics and a tendency to form

a second peak.

To determine the difference between the groups of patient and healthy

subjects, PCA was implemented with patient and healthy matrices derived

from PSD. Afterwards, the principal components of Doppler ultrasound for

the femoral artery was analyzed. While the 1st principal component of

the data matrix acquired from patients represents 97.6% total variance, the

1st principal component of the healthy data matrix represents 98.2% total

variance.

Therefore, the 1st principal component clearly shows the basic differ-

ence between the healthy people and occluded patients. While the

Figure 3. Power Spectral Density graphics of common femoral artery Doppler sig-

nals. (a) Healthy person (healthy, no. 19); (b) patient with occlusion (patient, no. 28).

Figure 4. Healthy and patient subjects’ first two principal component graphics vs.

time axis. (a) 1st principal component; (b) 2nd principal component.

Determination of Femoral Artery Occlusion Using PCA 335

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

standard deviation of the 1st principal component of healthy subjects is

0.0149, the patient subjects have 0.0303 standard deviation in the 1st

principal component. First and second principal components are depicted

in Figure 4. Second and later principal components exhibited appearances

similar to each other for healthy and patient subjects.

The graphic of the 2nd principal component vs. 1st principal component

of healthy subjects and patient subjects is given in Figure 5. As a result,

the patient and healthy groups are separated clearly from each other.

CONCLUSION

In this paper, the application of PCA to the spectral waveform of Doppler

signals of femoral arteries belonging to patients with occlusion and healthy

subjects was presented. The capability of PCA over PSD was examined in

terms of distinguishing between patients with occlusion and healthy

subjects. The basic differences between the healthy and occluded patients

were observed with the 1st principal component. The use of PCA of the

physiological waveform is presented as a powerful method likely to be

incorporated into future medical signal processing.

Figure 5. Graphics of 2nd principal component vs. 1st principal component of

healthy subjects and patient subjects.

S. Kara and S. Kemaloglu336

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

ACKNOWLEDGMENT

This project was supported as Post-Graduate Education and Research Project

by Erciyes University (Project no. FBT-04-25).

REFERENCES

1. Evans, D.H.; McDicken, W.N.; Skidmore, R.; Woodcock, J.P. Doppler Ultra-sound: Physics, Instrumentation and Clinical Applications; Wiley: Chichester,1989.

2. Sigel, B. A brief history of Doppler ultrasound in the diagnosis of peripheralvascular disease, Ultrasound Med. Biol. 1998, 24 (2), 169–176.

3. Ubeyli, E.D.; Guler, I. Spectral analysis of internal carotid arterial Doppler signalsusing FFT AR MA and ARMA methods. Comput. Biol. Med. 2004, 34 (4),293–306.

4. Muller, M.; Ciccotti, P.; Reiche, W.T. Comparison of color-flow Dopplerscanning, power Doppler scanning, and frequency shift for assessment of carotidartery stenosis. J. Vasc. Surg. 2001, 34 (6), 1090–1095.

5. Chervu, A.; Moore, W.S. Carotid endarterectomy without arteriography: personalseries and review of the literature. Ann. Vasc. Surg. 1994, 8, 296–302.

6. Davies, A.H.; Willcox, J.H.; Magee, I.; Currie, S.E.; Cole, S.E.A.; Murphy, P.Colour duplex in assessing the infrainguinal arteries in patients with claudication.Cardiovasc. Surg. 1995, 3 (2), 211–212.

7. Langsfeld, M.; Nepute, J.; Hershey, F.B. The use of deep duplex scanning topredict hemodynamically significant aortoiliac stenoses. J. Vasc. Surg. 1988, 7,395–399.

8. Aburahma, A.F.; Robinson, P.A.; Boland, J.P.; Umstot, R.K.; Clubb, E.A.;Grandia, R.A.; Kennard, W.; Bastug, D.F. Complications of arteriography in arecent series of 707 cases: factors affecting outcome. Ann. Vasc. Surg. 1993, 7,122–129.

9. Walton, L. An objective feature extraction technique applied to the Dopplerwaveforms from the groin: a prospective study. Ultrasound Med. Biol. 1983, 2,263–8.

10. Guo, Z. Comparison of time-frequency distrubution techniques for analysis ofsimulated doppler ultrasound signals of the femoral artery. IEEE Trans. Biomed.Eng. 1994, 41 (4), 332–342.

11. Eiberg, J.P. Screening for aortoiliac lesions by visual interpretation of thecommon femoral doppler waveform. Eur. J. Vasc. Endovasc. Surg. 2001, 22,331–336.

12. Isabel, A. Artificial neural network analysis of common femoral artery dopplershift signals: classification of proximal disease. Ultrasound Med. Biol. 1999,24 (5), 735–743.

13. Smith, J.H.; Graham, J.; Taylor, R.J. The application of an artificial neural networkto Doppler ultrasound waveforms for the classification of arterial disease. Int.J. Clin. Monit. Comput. 1996, 13, 85–91.

14. Van Asten, W.N.; Beijneveld, W.J.; Pieters, B.R. Assessment of aortoiliac obstruc-tive disease by Doppler spectrum analysis of blood flow velocities in the commonfemoral artery at rest and during reactive hyperemia. Surgery 1991, 109, 633–639.

Determination of Femoral Artery Occlusion Using PCA 337

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4

15. Currie, I.C.; Wilson, Y.G.; Baird, R.N.; Lamont, P.M. Post occlusive hyperaemicduplex scan: a new method of aortoiliac assessment. Brit. J. Surg. 1995, 82,1226–1229.

16. Gosling, R.G.; King, D.H. Continuous wave ultrasound as an alternative andcomplement to X-rays in vascular examination. In Cardiovascular Applicationsof Ultrasound; Reneman, R.S., Ed.; Amsterdam: North Holland 1974; 266–282.

17. Planiol, T.; Pourcelot, L. Doppler Effect Study of the Carotid Circulation. Ultra-sonics in Medicine; Elsevier: New York, 1973.

18. Baird, R.N.; Bird, D.R.; Clifford, P.C.; Lusby, R.J.; Skidmore, R.; Woodcock, J.P.Upstream stenosis: diagnosis by Doppler signals from the femoral artery. Arch.Surg. 1980, 115 (11), 1316–1322.

19. Macpherson, D.S.; Evans, D.H.; Bell, P.R.F. Common femoral artery Dopplerwaveforms: a comparison of three methods of objective analysis with directpressure measurements. Brit. J. Surg. 1984, 71, 46–49.

20. Arenson, J.W. Real-time Two-dimensional Blood Flow Imaging Using a DopplerUltrasound Array; McGraw-Hill: New Jersey, 1982.

21. Proakis, J.G.; Rader, C.M.; Fuyun, L.; Chrysostomos, L. Advanced Digital SignalProcessing; Macmillan: Princeton, NJ, 1992.

22. Lindsay, I.; Smith, A. A Tutorial on Principal Components Analysis; 26 February2002. http://kybele.psych.cornell.edu/�edelman/Psych-465-Spring-2003/PCA-tutorial.

Received October 25, 2004

Accepted November 15, 2004

Manuscript 1492

S. Kara and S. Kemaloglu338

Dow

nloa

ded

by [

Ein

dhov

en T

echn

ical

Uni

vers

ity]

at 2

3:06

19

Oct

ober

201

4