Fault Detection System for a Stent-GraftEndoleakage Monitor

Cristina Oliveira ∗ and José Machado da Silva ‡INESC TEC (former INESC Porto), Faculdade de Engenharia,

Universidade do PortoRua Dr Roberto Frias, 378, 4200-465 Porto, Portugal

∗ deb11016@fe.up.pt, ‡ jms@fe.up.pt

Abstract—A new inductive coupling based wireless system isbeing developed to monitor the condition status of aortic stent-grafts. It relies on the measure of the stent-graft’s outer pressureusing a capacitive sensor placed in a LC resonant circuit. Thework presented herein addresses the testing of the LC circuit todiagnose whether observed pressure deviations are due to defectsoccurring in the sensor’s inductor and capacitor or to an actualdegradation of the stent-graft.

Index Terms—fault detection, stent-graft, monitoring;

I. INTRODUCTION

A. Motivation

An abdominal aortic aneurysm (AAA) is a very dangerousarterial problem caused by a bulging area on the aorta. Theavailable treatments are the reconstruction of the aorta (opensurgery) or the endovascular aneurysm repair (EVAR). Thetreatment using a stent-graft to correct the problem (EVAR)is a less invasive procedure but requires an expensive surveil-lance protocol using computer tomography (CT) and magneticresonance imaging (MRI) scans almost on a year basis due tothe complications that can occur after the stent-graft placing,namely the endoleaks [1]. In order to detect the endoleaks awireless telemetry system is being developed for the intersacpressure monitoring after the endovascular repair.

B. Remote Monitoring System

The monitoring principle is based on the detection ofpressure variations within the aneurysm sac by means of acluster of capacitive pressure sensors attached to the stent-graft (figure 1). Each sensor comprises an LC resonant circuit,whose oscillation frequency is sensitive to pressure variations.An external reader delivers energy and detects sensor’s reso-nance frequency through an inductive-coupling link. The useof a group of sensors placed in the same stent-graft maximizesthe system sensitivity to leakages, which is an advantagecomparing to the system presented in [2] that uses only onesensor. The communication with the sensor cluster uses the12.5 MHz to 20.0 MHz frequency band, specifically allocatedfor use in medical applications [3].

The pressure sensor is based on two square plate electrodes(diaphragm) separated by a dielectric (air at a pressure P0),where changes on the outside pressure (Pout) will deformthe square plates and consequently will generate a capacitivechange. The sensing capacitor is attached to an inductor

Figure 1: Sensor cluster in an abdominal aorta stent-graft.

forming together an LC resonant circuit, whose oscillationfrequency is sensitive to pressure variations. An externalreader delivers energy and detects the corresponding resonancefrequency through an inductive-coupling link. Altogether, theexternal stimulating inductor (primary) and the embodiedinductor (secondary), establish an inductive communicationlink that can be modelled by a transformer.

The flexible capacitive sensor fabrication process usesaligned carbon nanotubes (ACNTs) embedded in a flexiblepolymer matrix. Given the specification of the application,the device must be foldable, extremely flexible and charac-terized by a small profile. The ACNTs embedded in a flexi-ble substrate of polydimethylsiloxane (PDMS), a transparent,nontoxic and biocompatible silicone elastomer, are used tofabricate the elements of the capacitive measurement system(an inductor and the capacitive electrodes)(figure 2).

Figure 2: PDMS membrane with embedded ACNTs.

2012 IEEE 18th International Mixed-Signal, Sensors, and Systems Test Workshop

978-0-7695-4726-8/12 $26.00 © 2012 IEEE

DOI 10.1109/IMS3TW.2012.14

17

The pressure sensor is composed of three thin layers, wherethe inductor and the electrodes are defined by the top andbottom layers (each membrane has 1% CNTs volume fraction),and the dielectric (air) is in between (figure 3) [4].

Figure 3: Schematic of the pressure square (sidelength = 2a) sensora) 3D view and b) section cut B-B [4].

C. Fault Occurrences

An important aspect to be considered concerns the fre-quency shift due to the sensor’s inductor bending after place-ment in the stent. Consequently the corresponding inductancecan change [5] due to bending. This deformation may havea direct impact on variations of the sensor’s oscillation fre-quency, producing uncertainty in the pressure detection. Theapproach being used here relies on the direct analysis of LC’sresonant frequency. This allows decreasing the sensitivity toangle misalignments between the two inductors. However,interpreting the captured oscillation frequency cannot be madein a simple single observation. The pressure in the stent-graft varies periodically with the arterial blood pressure. Inconsequence, what one actually captures is a distribution ofresonant frequencies, modulated, on one hand, by the cardiacbeating (60 to 100 beats per minute for a healthy heart) andthe range of pressures it generates. On the other hand, thefrequency is also affected by the operating condition status ofthe pressure sensor. Therefore, in case an abnormal resonantfrequency distribution is observed how can one distinguisha defective stent-graft from a defective sensor? The workpresented here addresses this problem in order to improve thereliability of the proposed endoleakage monitor.

II. FAULT DETECTION

As this detection process provides an indirect pressuremeasurement, the occurrence of defects in the pressure sensormay mask the detection of anomalies in the stent-graft. Theextreme cases in the LC network are an open-circuit due to aconnection break or a short-circuit. Since the pressure is beingsensed by the two plates capacitor, this is the most sensitiveelement of the LC network and the most likely to cause faults

in the pressure measurements. Possible capacitor defects arelisted in table I. The work presented herein addresses theimplementation of additional measurements and operationswhich allow diagnosing the occurrence of defects in the LCsensor.

Table I: List of possible sensor defects and their influence on thepressure measurements.

Capacitor Defects Effects on MeasurementsStuck capacitor Leads to a constant resonant frequency

measurementReduction of capacitor’snominal measurementrange

Allows detecting pressure deviations butin a narrow range, these measurementscould still be taken as admissible

Large deviation of capaci-tor’s nominal value

Could lead to a false defective stent-graftdetection (e.g. a leaking stent-graft)

Collapsed capacitor Shows no oscillation frequencyBending of the structure Deviation of the inductance and capaci-

tor valuesAging of the structure Increase of inductor resistance

The aortic blood pressure (ABP) waveform (figure 4)conveys information about the cardiovascular system suchas heart rate, systolic, diastolic and mean arterial pressures.Moreover, it provides information on possible complications inthe endovascular stent-graft. Reading the stent-graft pressure(LC resonant frequency) at a 100-200 Hz sampling frequencyallows reconstructing the ABP signal for future features ex-traction.

When the sensor is calibrated and ready to be placedaround the stent-graft all the circuit elements (Ls, Rs and Cs)values are known. However, these components’ performancemay change overtime due to bending, wear and friction. Incases where the measured ABP is outside the nominal range,distinguishing that these values are caused by the pressureapplied on the stent and not due to changes in the sensorcomponents’ values are of utmost importance.

Figure 4: Aortic blood pressure waveform.

The set of resonant frequencies captured during a number ofABP cycles allows building an histogram of the captured val-ues distribution. The shape of this histogram is a preliminaryindicator of the occurrence of anomalies.

A. System Description

To distinguish between each possible situation, one canresort to the measure of the power transmission from thesensor to the reader circuit, as well as from the impedance seenat node vo(ω). These additional measurements combined withthe measured oscillating frequency enable the determinationof the circuit components values.

18

Figure 5: Circuit used to measure coupling power and impedance.

Using the circuit presented in figure 5 one can performa frequency sweep to obtain the characteristics of inputimpedance (ZL) and power (P ) as functions of frequency –power is proportional to the product of vo(ω) and vi(ω) andZL = vo

vinRi. With m frequency sweeps one can obtain the

inductive coupling power spectrum (figure 6) which allow usto obtain the quality factor (Q = 1

Rs

√Ls

Cs) of the resonant

circuit.

P =1

m

m∑j=0

vo(ω)i(ω) ≡ 1

m

m∑j=0

vo(ω)vi(ω)

nRi(1)

Transmitted Power

Frequency (Hz)

Mea

n P

ow

er (

dB

W)

Figure 6: Mean transmitted power.

The real and imaginary parts of the impedance obtainedfrom the frequency sweep (figure 7) are given by equations 2.

Once power and impedance measurements have been per-formed, the preliminary values for the coupling factor k andRs, using fo and Q equations and the nominal Ls and Cs

values as a first guess ("seed") values, can be estimated. TheLs and Cs values are then estimated by means of a fittingprocess of the measured impedance real and imaginary partsto equations 2. This process is then iterated until a specifiedestimation error (difference between measured and estimatedresonant frequencies) is verified. The following pseudo-codesummarizes this approach.

Introduce the estimated valuesstart_points =

Impedance ZL

Imp

edan

ce (

Oh

m)

Frequency (Hz)

Figure 7: Real and imaginary parts of impedance ZL.

=[k=1, Rs=50, Cs=10e-12, Ls=10e-6]

WHILE (abs(fo_meas-fo_calc))>100Rs_iter = 1/Q_meas*sqrt(Ls/Cs)Rs = Rs_iterFUNCTION curvefitting(ZL_real, ZL_imag)

estimate Ls_iterestimate Cs_iter

Ls=Ls_iterCs=Cs_iterfo_calc=1/2*pi*sqrt(Ls*Cs*(1-k^2))

END

This data modelling assumes capacitor’s value Cs remainsconstant during the measurements but in a real time measure-ment it varies with the pressure inside the aneurysm sac. Asequence of operations must be performed in order to acquirethe needed data (transmitted power and impedance) duringa time interval corresponding to a known pressure when thecapacitor has a low variability (figure 8).

The telemetry system comprises an electrocardiogram(ECG) equipment to record a real time electrocardiogram andthe fault detection circuit to measure the transmitted powerand impedance. The ECG acquisition has two purposes, theassessment of the patient’s cardiac condition and to providethe fault detection triggering signal. The ECG is a widelyused exam in patients with diseases associated with cardiacfunction, and is an important factor in the determination ofrisk of open surgery and EVAR [6].

To ensure that the transmitted power and impedance mea-surements are carried out during a period when the capacitanceis almost constant, the measurements must start in the onsetof the systolic pressure and stop when the blood pressuredrops again. The systole occurs around 0.2 seconds afterthe contraction of the ventricles (QRS complex), so if theECG is recorded in real time it is possible to determine theQRS occurrence and start the fault detection measurementsduring the systole. Then, the recorded transmitted power and

19

R(ZL) = Rp +ω4RsL

2mC

2s

1 + ω2(R2sC

2s − 2LsCs) + ω4L2

sC2s

I(ZL) = ω(Lp − Lm) +ωLm + ω3(R2

sLmC2s + L2

mCs − 2LsLmCs) + ω5C2s (L2

sLm − L2mLs)

1 + ω2(R2sC

2s − 2LsCs) + ω4L2

sC2s

(2)

impedance are used for the data modelling to extract thenominal values of the sensor’s components. With this approachthe small variability of the capacitor sensor is assured andthe signal acquisition is carried-out during the most criticalsituation when the pressure is maximum. The fault detectionsystem enables the assessment of sensor’s components valuesand, depending on the fault severity, it could help recalibratingthe reading system so the sensor’s readings are used insteadof its dismissal.

B. Simulation Results

If the measured oscillating frequency is higher or lowerthan the expected one, it is possible to estimate the LCstructure components values to make sure the frequency shiftwas caused by the pressure sensor and not by an eventualchange of the inductor value. This situation was simulatedusing Agilent ADS to obtain the corresponding transmittedpower and impedance curves for Rs = 50 Ω, k=0.1 and thesensor’s capacitance Cs and inductance Ls were varied toevaluate the fitting algorithm performance. The initial guesseswere Rs = 50 Ω, Ls = 10 µH, Cs = 10 pF and k=0.1. Firstthe Rs, k and the Cs values were kept constant and equal tothe fitting algorithm initial guesses, and the inductance Ls wasvaried between 7 and 16 µH with a 1 µH step. The estimationerrors for each inductance value are shown in figure 9. For allthe inductance values the estimation value was below the 0.8% and the lowest estimation error occured for a Ls equal tothe initial guess value has expected.

Figure 9: Inductance estimation errors.

A similar process was used for the capacitance estimationerrors, with the Rs, Ls and k kept constant during thesimulations, and with the capacitance Cs varied between 7pF and 16 pF with a 1 pF step. The estimation errors for eachinductance value are shown in figure 10. Again the estimationerrors were very low indicating that the fitting algorithm canadapt to different input impedance and power curves and

output a sensor component nominal value very close to itsactual value.

Figure 10: Capacitance estimation errors.

C. Experimental Results

An experimental set-up was used to evaluate the concept. Itcomprises a PCB circuit with a planar rectangular inductor, aresistor and a capacitor emulating the LC network. AnotherPCB mounted inductance acts as the primary winding. Asignal generator provides a frequency sweeping of a sinewave,being a GPIB controlled data acquisition system used tocapture node voltages vi and vo (figure 5). ECG and ABPsignals from the MGH/MF Waveform Database [7] are usedto synchronize data capture triggering. The Pan-Tompkinsalgorithm is used to filter the ECG signal and detect the QRScomplexes [8].

After a few seconds of calibration the system is readyto start the frequency sweeping. The measurements must bedone in 0.2 seconds (duration of systole) and the system’sresolution, i.e. the minimum Cs and Ls detectable deviations,is dependent on the number of points (frequencies) that canbe measured within this period. A MatLab script is used tocontrol the GPIB connected devices, the detection of QRScomplex in the ECG, to calculate the primary impedance andtransmitted power after the captured raw data, and to estimateCs, Ls and Rs using a data fitting based algorithm.

Table II shows results obtained for Rs=12 Ω, Ls=45.8µH , a coupling factor k around 0.1 (set by proper separationbetween inductors), and for two different Cs values. Theinitial guesses for the fitting algorithm were Rs = 12 Ω, Ls

= 45 µH, Cs = 10 pF and k=0.1.

The fault detection approach allows detecting a shift in thesensor’s capacitance Cs in spite of the observed error. Possiblereasons for the fitting algorithm difficulty to estimate a moreaccurate Cs value are the noise associated with the frequency

20

Figure 8: Fault detection block diagram.

Table II: Experimental results obtained for different capacitancevalues.

EstExp

Cs=3.68 pF Cs=3.9 pF

Ls (H) 4.5e-5 (1.7%) 4.5e-5 (1.7%)Cs (F) 3.12e-12 (15%) 3.35e-12 (14%)

sweeping process and the presence of the probes capacitancewhich affect the measured frequency and power. In order toimprove and increase the number of the measurements forthe fault detection system tests, a PCB board with adjustablecomponents (variable inductor, capacitor and resistor) is beingimplemented, and a de-embedding process is being studied.

Considering the case of a cluster of sensors one can takeadvantage of data fusion to improve diagnosis. In this case, ifa sensor provides a high pressure reading due to a endoleak orendotension episode the surrounding sensors must also showan increase on the measured pressure, but if only a sensorhas an abnormal reading (extremely high or low pressurecompared with the other sensors’ readings) then this sensoris faulty and its readings should be ignored.

III. CONCLUSIONS AND FUTURE WORK

A methodology for testing and diagnosing failures in ainductive-link based aortic stent-graft monitoring system hasbeen proposed. This monitoring is performed after the mea-surement of the aneurysm sac pressure resorting to the place-ment of LC cells in the stent-grafts fabrics, whose resonantfrequency is determined by the variations of the capacitivepressure sensor. Nevertheless, as this is an indirect pressuremeasurement based approach, detection uncertainty is proneto occur since variations can be caused either by endoleaks inthe stent-graft or due to a degradation of the LC network per-formance. The work presented here addresses a methodologyto diagnose deviations in the LC values after measurements ofthe detected signal power and inductive-coupling impedance.Preliminary experimental results show that both L and C

values can be estimated with good accuracy. A new prototypeis being implemented to further evaluate this fault detectionsystem.

ACKNOWLEDGMENTS

This work has been carried-out in the frameworkof projects EDAM-SenseCardioHealth and Eureka/CatreneTOETS CT302, with the support of programme MIT|Portugal,Fundação para a Ciência e a Tecnologia, grant MIT-Pt/EDAM-EMD/0007/2008 and grant contract SFRH/BD/81476/2011.

REFERENCES

[1] F. Springerand, R. W. Günther, and T. Schmitz-Rode, “Aneurysm SacPressure Measurement with Minimally Invasive Implantable PressureSensors: An Alternative to Current Surveillance Regimes after EVAR?”Cardiovasc Intervent Radiol, vol. 31, no. 3, pp. 460–467, 2008.

[2] T. Ohki, K. Ouriel, P. G. Silveira, B. Katzen, R. White, F. Criado, andE. Diethrich, “Initial results of wireless pressure sensing for endovascularaneurysm repair: The apex trial–acute pressure measurement to confirmaneurysm sac exclusion,” Journal of Vascular Surgery, vol. 45, no. 2, pp.236–242, 2007.

[3] “The European table of frequency allocations and utilizations in thefrequency range 9 KHz to 3000 GHz,” Electronic CommunicationsCommitee (ECC), Tech. Rep., 2009.

[4] A. T. Sepúlveda, A. J. Pontes, J. C. Viana, L. A. Rocha, I. C. T.Santos, F. Fachin, R. G. de Villoria, and B. L. Wardle, “Design of apressure sensor for monitoring of post-endovascular aneurysm repair,” inBIODEVICES, 2011, pp. 14–22.

[5] S. Leung and D. Lam, “Performance of printed polymer-based RFIDantenna on curvilinear surface,” IEEE Transactions on Electronics Pack-aging Manufacturing, vol. 30, no. 3, pp. 200–205, Jul. 2007.

[6] D. C. Brewster and et. al, “Guidelines for the treatment of abdominalaortic aneurysms: Report of a subcommittee of the joint council ofthe american association for vascular surgery and society for vascularsurgery,” Journal of Vascular Surgery, vol. 37, no. 5, pp. 1106–1117,May 2003.

[7] A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff, P. C.Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, and H. E.Stanley, “PhysioBank, PhysioToolkit, and PhysioNet: Components of anew research resource for complex physiologic signals,” Circulation, vol.101, no. 23, pp. e215–e220, 2000 (June 13), circulation Electronic Pages:http://circ.ahajournals.org/cgi/content/full/101/23/e215.

[8] P. J. and T. W. J., “A real-time qrs detection algorithm,” Biomed. Eng.,vol. BME-32, no. 3, pp. 230–236, Mar. 1985.

21