A novel operator-independent algorithm for cardiac output ...

European Journal of Echocardiography (2010) 11, 432–437 doi:10.1093/ejechocard/jep233

A novel operator-independent algorithm for cardiac output measurements based on three-dimensional transoesophageal colour Doppler echocardiography Felix Matthews 1*, Thomas Largiade`r2, Patricia Rhomberg 3, Bernd van der Loo 2, Edith R. Schmid 3, and Rolf Jenni 2 1 Surgical Planning Lab, Brigham and Women’s Hospital, 75 Francis Street, Boston, MA 02115, USA; 2Clinic of Cardiology, University Hospital, Raemistrasse 100, 8091 Zurich, Switzerland; and 3Division of Cardiac Anaesthesia, Institute of Anaesthesiology, University Hospital, Raemistrasse 100, 8091 Zurich, Switzerland

Received 11 October 2009; accepted after revision 23 December 2009; online publish-ahead-of-print 27 January 2010

Aims

Cardiac output (CO) measurements from three-dimensional (3D) trans-mitral Doppler echocardiography are prone to error as manual selection of the region of interest (i.e. the site of measurement) is required. We newly developed an automated, user-independent algorithm to select the site of colour Doppler CO measurement. We aimed to validate this new method by benchmarking it against thermodilution, the current gold standard for CO measurements. ..................................................................................................................................................................................... Methods Transoesophageal colour 3D Doppler echocardiographic studies were obtained from 15 patients who also had received a pulmonary catheter for invasive CO measurements. Trans-mitral flow was determined using a novel operand results ator-independent algorithm to automatically select the optimal site of measurement. The operator-independent CO measurements were referenced against thermodilution. A good correlation was found between operator-independent Doppler flow computations and thermodilution with a mean bias of 0.09 L/min, standard deviation of bias 1.3 L/min, and a 26% error (2 SD/mean CO). Mean CO was 4.94 L/min (range 3.10–7.10 L/min). ..................................................................................................................................................................................... Conclusion Our findings demonstrate that CO computation from transoesophageal colour 3D Doppler echo can be automated concerning the site of velocity measurement. Our operator-independent algorithm provides an objective and reproducible alternative to thermodilution.

----------------------------------------------------------------------------------------------------------------------------------------------------------Keywords

Cardiac output † Transoesophageal echocardiography † Doppler flow † Thermodilution

Introduction Transoesophageal echocardiography has been proposed as a method to assess cardiac output (CO) during cardiac surgery. Stroke volume can thereby be determined using two-dimensional (2D) echocardiograms of either the left ventricular inflow or outflow tract. The computations are, however, based on the assumptions of: (a) a constant flow area and (b) an instantaneous peak velocity which is representative of the mean velocity.1 These assumptions do not hold true in reality, as the threedimensional (3D) velocity profile is complex (Figure 1). Furthermore, the instantaneous peak velocity cannot be extrapolated to

mean velocity due to changing velocity profiles.2 Assessment of only one velocity in the left ventricular inflow or outflow tract and determination of the respective diameter from the 2D section view will therefore lead to fundamental errors (in order of magnitude +50%) in the calculation of the volume, a fact that makes assessment of those parameters based on 2D sections rather useless.3 The advent of 3D Doppler echocardiography brought significant improvement. If the 3D distribution of velocity over the crosssectional area of the left ventricular inflow tract is registered with sufficient temporal resolution, then the cross-sectional area can be calculated on the basis of the velocity profile. The spatial

* Corresponding author. Tel: þ1 (617) 732-7389; fax: þ1 (617) 582-6033. Email: [email protected] Published on behalf of the European Society of Cardiology. All rights reserved. & The Author 2010. For permissions please email: [email protected]

Downloaded from https://academic.oup.com/ehjcimaging/article-abstract/11/5/432/2396841 by guest on 14 December 2017

433

Algorithm for 3D Doppler CO measurements

resolution (voxel size) of the 3D data set is a function of the acquisition probe and is therefore known. Hence, the cross-section of the velocity profile can be calculated by tracing the voxels containing negative Doppler signal over the surface selected as site of measurement.4 Therefore, volumetric flow measurements in a 3D data set need neither assume a constant annulus diameter, nor rely solely on the peak flow velocity for CO computation. Instead, 3D Doppler flow computations can account for complex and varying velocity profiles over the course of the diastole. Today, the cardiologist can compute CO offline using commercially available software packages such as TomTec 4D.5 – 7 The Achilles heel of 3D Doppler volumetric flow measurement is that it requires manual positioning of the site of measurement within the 3D volume. While some authors have demonstrated a good correlation between 3D volumetric flow measurements of the left ventricular inflow tract and thermodilution, they have remained ambiguous as to the criteria for positioning the site of measurement.8 Others have attempted to manually track the movement of the annulus plane frame by frame,9,10 which is impracticable for clinical routine. In our clinical experience, flow measurements vary considerably based on the chosen site of measurement. The objective of this study was therefore to describe the dependence of CO measurement upon the selected site of measurement. Based on those findings, we describe an approach for automated, operatorindependent CO computation. We validate our method by benchmarking Doppler CO with thermodilution.

Methods Study population We examined 15 patients (nine males, six females; 66.8 + 11.0 years old) in the cardiac surgical intensive care unit. All of them had undergone a cardiac operation (13 aortocoronary bypass grafts for coronary artery disease; 3 aortic valve replacements for aortic stenosis) and had received a pulmonary artery catheter. All patients agreed to participate in the study and signed an informed consent prior to the operation. The study was approved by the local ethical committee. Patients with mitral and tricuspital regurgitation were excluded. All 15 patients had high 2D-colour transoesophageal echocardiography image quality.

Pulmonary artery thermodilution

Figure 1 Three-dimensional velocity profile of trans-mitral diastolic flow in early, mid, and late diastole. Note the complex changes of velocity profile during diastole.

Prior to echocardiography, CO was measured by pulmonary artery bolus thermodilution, as described in detail previously.11,12 Thermodilution measurements were recorded by an independent researcher not involved in the echocardiographic studies. Briefly, all patients had a balloon-tipped thermodilution catheter inserted via the right internal jugular vein into the pulmonary artery and connected to a CO monitor. CO was automatically computed after manual injection of 10 mL of 0.9% iced saline into the right atrium. Injections were repeated three times, and the average value was taken. If more than 8% variation was obtained between the three values, two additional measurements were performed and the highest and lowest values were excluded from average calculation.

Downloaded from https://academic.oup.com/ehjcimaging/article-abstract/11/5/432/2396841 by guest on 14 December 2017

434

F. Matthews et al.

Data acquisition and analysis Doppler image acquisition Data acquisition was performed using a transoesophageal TEV5Ms transducer connected to an ultrasound system (Sequoia, Acuson, Siemens Medical, Mountain View, CA, USA) and a TomTec 4D cardio scan terminal (Tomtec Imaging Systems, Unterschleissheim, Germany). The left ventricular inflow tract was shown in an echocardiographic 2D view, so that an ideal colour-coded Doppler flow signal could be obtained. The depth and fan width were adjusted to achieve highest velocity and also to encompass the entire mitral inflow region. The baseline shift was adjusted so that aliasing was avoided. For data acquisition, the ultrasound probe rotated pulse synchronously in increasing 58 steps up to a total of 1808 around the axis of the ultrasound beam. In each probe position, the probe acquired 8– 10 ECG-gated frames in order to cover an entire cardiac cycle. The complete image acquisition sequence could be completed within ,2 min per patient. The operator verified that no fluctuations in heart rate occurred during this time span. Each measurement was repeated three times per patient. A single, experienced operator conducted the Doppler studies on all patients.

Volume reconstruction and flow computation theory The rotational set of 2D echocardiography slices can be reconstructed into a Z-series of slices using linear interpolation as described by Duann et al.13 and subsequently sequenced into a temporal volumetric model (four-dimensional model). The general flow equation found from conservation of matter is a double integration over the velocity field of interest on a given surface (also referred to as region of interest): ðð Flow ¼

V~  dA~

surface

Poulsen and Kim14 demonstrated that the scalar product between velocity vector V~ and normal vector to the small surface dA~ cancel out. This leads to an angle-independent flow computation, in which solely the flow direction is inverted: ðð Flowin ¼

Vm  dA surface

In this study, CO was computed from 3D Doppler flow measurements based on this established general equation, and thus all flow calculations are angle independent. The novelty is that the site of measurement was selected by means of an operator-independent positioning algorithm, as described in the following.

Operator-independent CO computation After bedside image acquisition was completed, raw Doppler image data was transferred from the TomTec 4D terminal to a desktop computer for offline processing (Fujitsu-Siemens Scaleo, Intel Pentium processor, Microsoft Windows XP). We developed

Figure 2 Schematic of an apical two-chamber view in diastole. Cardiac output is calculated at multiple depths by moving the site of measurement (R1, R2, . . . , Rn). LA, left atrium; LV, left ventricle; M, mitral valve plane; arrow, diastolic flow.

a proprietary software package to compute CO requiring no user interaction. First, the above-described volume reconstruction and flow computations were performed with custom Matlab scripts (Matlab 7.0, Mathworks, Natick, MA, USA). Instead of requiring the operator to select the site of measurement manually, the scripts automatically computed the flow velocity profile at various locations within the volume by placing the site of measurement at discrete positions above, through, and below the mitral valve plane. Figure 2 shows a schematic of the mitral valve and the positioning of the site of measurement. A velocity profile was computed for each site of measurement within the same data set. CO in litres per minute was calculated using the velocity time integral described above. For each data set, the CO was plotted in function of the depth of the site of measurement. Within each echo data set, the site of measurement yielding the maximum CO was determined by using the first derivative: Flow(ROI)0 ¼ 0. This maximal CO of each Doppler data set was used for all subsequent data analysis.

Statistical analysis Operator-independent CO measurements were evaluated against thermodilution. Linear regression analysis was performed, and the coefficient of correlation R 2 determined. Computed CO from Doppler studies was plotted against measured CO from thermodilution, along with the 95% confidence interval. The measurements were further evaluated according to Bland and Altman,15 i.e. plotting the difference between the two methods against the average of the same. In accordance with Critchley’s recommendation, mean bias, precision (2 SD), and percentage error (2 SD of bias/mean CO) were calculated. Thereby, a percentage error ,30% represented a good correlation.16 The level of significance was chosen at P , 0.05.

Downloaded from https://academic.oup.com/ehjcimaging/article-abstract/11/5/432/2396841 by guest on 14 December 2017

435

Algorithm for 3D Doppler CO measurements

Figure 4 Operator-independent cardiac output computations based on three-dimensional Doppler echocardiography compared with thermodilution. Linear regression and 95% confidence intervals (CI) demonstrate a good correlation.

Figure 3 Diastolic trans-mitral Doppler flow. White line: plot of cardiac output measurement obtained by positioning the site of measurement at varying depths. The star indicates the depth with the maximum cardiac output of 6.1 L/min.

Results The 15 patients, who had good image quality of echocardiography, were evaluated with the novel operator-independent method. CO measurements showed a strong dependence on the selected measurement depth. Figure 3 shows a typical CO distribution curve as a function of depth. Throughout the examined data sets, the maximum flow was found at the level of the apical half of the fully open mitral valve leaflets. The maximum computed CO measurement showed the best correlation with CO measurements as obtained from thermodilution. In the examined collective, the mean computed CO was 4.94 L/min (range 3.10–7.10 L/min). Linear regression between our user-independent computation and thermodilution yielded a correlation coefficient R 2 of 0.71 (P , 0.001). The mean bias was 0.09 L/min (range 21.09 to 1.10 L/min) and the precision (Bland– Altman limits of agreement, 1.96  SD) was 1.3 L/min, corresponding to a percentage error of 26% of the mean CO. Figure 4 shows the linear regression and Figure 5 the Bland – Altman plot for operator-independent CO computations.

Discussion We have demonstrated for the first time that, using our novel algorithm, operator-independent selection of the site of measurement for CO determination within the left ventricular inflow tract is feasible with a high degree of precision as compared with thermodilution. Previously published data17 using 2D techniques apparently demonstrating a better precision must be inferior because, owing to the shape of the velocity profile, it is impossible to

Figure 5 Bland – Altman graph of operator-independent cardiac output computations based on three-dimensional Doppler (3DD) plotted against thermodilution (ThD). The difference is plotted against the mean.

calculate stroke volume using only one-gated velocity within the left ventricular outflow tract. Three distinct methods are commonly used to determine CO based on echocardiography: (i) transoesophageal colour Doppler flow measurement of the left ventricular inflow tract,8 (ii) transthoracic or transgastric colour Doppler flow measurement across the left ventricular outflow tract,18 – 20 and (iii) volume computation.21 The measurement of CO plays a central role in cardiology. It is, among others, crucial for the management of patients in intensive care units. CO measurements by means of thermodilution are invasive, time-consuming, expensive, and associated with a higher risk for the patient.

Downloaded from https://academic.oup.com/ehjcimaging/article-abstract/11/5/432/2396841 by guest on 14 December 2017

436

Validation of 3D Doppler flow measurement Our study demonstrates that 3D colour Doppler echocardiography is comparable to thermodilution in determining CO, as evidenced by an acceptable percentage error of 26% (i.e. below the suggested 30% cut-off16). Most importantly, we benchmarked against an already validated bolus thermodilution technique whose accuracy and precision, as previously published, are known.7 Less powerful, but still noteworthy, is a statistically significant correlation found in the linear regression analysis.

Shortfalls of 2D Doppler Conventional 2D echocardiographic measurements of the subvalvular diameter and the velocity time integral assume a flat velocity profile and a circular outflow tract.22 Haugen demonstrated that these assumptions do not necessarily hold true. In the case of measurement in the left ventricular outflow tract, the 2D method does not account for annulus movements, which can result in a missing volume.9 Further, the angle between Doppler beam and flow direction introduces an error in 2D measurements.14 Bettex et al.23 previously demonstrated that trans-mitral 2D stroke volume computation is unreliable.

Advantages of 3D Doppler 3D colour Doppler echocardiography accommodates for these deficiencies and is thus a more reliable technology. It has previously been demonstrated that flow measurements using multiplanar pulsed Doppler ultrasound are independent of the Doppler beam angle.17,19 3D Doppler CO has been extensively studied in both human and animal studies.3,8,22,24,25 However, when examining flow through the left ventricular inflow tract, the authors have remained vague when describing the site of measurement examined. This is intriguing since echocardiography is subject to high inter-observer variability, and since transoesophageal examinations are particularly known for their high operator dependency.26 However, an easy and reproducible method to determine CO through trans-mitral volumetric flow remains to be defined.

Operator-independent CO measurement To avoid inter-operator variability, we sought to automate the CO computation. This is the first study to show that the selection of site of measurement within the reconstructed 3D data set highly influenced the computed CO. Our results showed that the maximum computed CO correlates well with the thermodilution reference method (Figure 4). Experienced cardiologists may possibly select an appropriate site of measurement by referencing anatomical landmarks such as the mitral annulus plane and the mitral leaflets. We, however, demonstrated here for the first time that this process can be fully automated and thereby consistently yield accurate CO results.

F. Matthews et al.

standard of thermodilution. As mentioned above, the precision of our thermodilution technique as reference standard is known. Interestingly, colour 3D Doppler CO computations appeared to underestimate CO in high-flow situations. This phenomenon becomes apparent in the linear regression slope of 0.88, when compared with thermodilution (R 2 ¼ 0.71). Surprisingly, to our knowledge, underestimation of CO in 3D Doppler echocardiography measurement has not been reported by other authors. One possible interpretation is that the thermodilution device we used overestimated CO in high-flow situations. Another, more likely explanation is that 3D Doppler measurements actually underestimate cardiac flow. Our data sets were sampled with 8 –10 frames over the entire heart cycle. In high-flow situations, corresponding to short RR cycles, there is an increased probability of the peak diastolic flow occurring between two frames. This will lead to lower CO computations in patients with high stroke volumes. Sampling rate is primarily an issue with matrix probes, but we expect that this effect will be mitigated as new probes with higher sampling rates become available.

Limitations of the study We acknowledge a limited number of patients included in the study, in part because of poor image quality postoperatively. Image quality is a key determinant of accurate CO measurements, both in the traditional user-dependent selection of measurement site and with our user-independent computation method. We used rotating 2D transoesophageal Doppler probes. Arguably, this technology is gradually being replaced by 3D matrix probes that immediately generate 3D volumes. However, irrespective of the technology used to acquire the 3D echo data sets, the underlying clinical concern of selecting the appropriate site of measurement over which to measure the volumetric flow remains the same. In our 15 patients, CO ranged from 3.1 to 7.1 L/min; in high output states, it could be possible that, despite baseline shift, the velocities could reach the Nyquist limit. When the ultrasound probe rotates in increasing 58 steps up to 1808 around the axis of the ultrasound beam, duration for data acquisition might be quite long. To reduce the scan time to 50% in cases of clinical instability, one may increase to 108 steps.

Outlook This new technique will allow truly quantitative measurements of blood flow using a non-invasive approach with a minimum burden for the patient. It has the potential to have crucial implications for the treatment of patients, especially in the intensive care unit. We now seek to verify our method using both transoesophageal and transthoracic colour 3D Doppler matrix probes. Confirming our results with larger collectives will also be imperative before routinely deploying this novel method of user-independent CO computation. We further propose that our method be utilized to compute mitral or aortic regurgitation.

Comparison of 3D Doppler with thermodilution

Conclusion

We found that user-independent CO computation based on colour 3D Doppler showed a good correlation with the gold

Through automated selection of the site of measurement, the echocardiographic quantification of CO, shunt lesions, and valvular

Downloaded from https://academic.oup.com/ehjcimaging/article-abstract/11/5/432/2396841 by guest on 14 December 2017

437

Algorithm for 3D Doppler CO measurements

regurgitations will be subject to less user-dependent variability in the near future.

Acknowledgement We are grateful to the Swiss Heart Foundation for the support of this study. Conflict of interest: none declared.

References 1. Berton C, Cholley B. Equipment review: new techniques for cardiac output measurement—oesophageal Doppler, Fick principle using carbon dioxide, and pulse contour analysis. Crit Care 2002;6:216 –21. 2. Berg S, Torp H, Haugen BO, Samstad S. Volumetric blood flow measurement with the use of dynamic 3-dimensional ultrasound color flow imaging. J Am Soc Echocardiogr 2000;13:393–402. 3. Haugen BO, Berg S, Brecke KM, Torp H, Slordahl SA, Skaerpe T et al. Blood flow velocity profiles in the aortic annulus: a 3-dimensional freehand color flow Doppler imaging study. J Am Soc Echocardiogr 2002;15:328 – 33. 4. Rusk RA, Li XN, Mori Y, Irvine T, Jones M, Zetts AD et al. Direct quantification of transmitral flow volume with dynamic 3-dimensional digital color Doppler: a validation study in an animal model. J Am Soc Echocardiogr 2002;15:55 –62. 5. Shahgaldi K, Soderqvist E, Gudmundsson P, Winter R, Nowak J, Brodin LA. Flowvolume loops derived from three-dimensional echocardiography: a novel approach to the assessment of left ventricular hemodynamics. Cardiovasc Ultrasound 2008;6:13. 6. Li XC, Yao GH, Zhang C, Zhang M, Zhang PF, Zhang W et al. Quantification of regional volume and systolic function of the left ventricle by real-time threedimensional echocardiography. Ultrasound Med Biol 2008;34:379 –84. 7. Soliman OI, Krenning BJ, Geleijnse ML, Nemes A, van Geuns RJ, Baks T et al. A comparison between QLAB and TomTec full volume reconstruction for real time three-dimensional echocardiographic quantification of left ventricular volumes. Echocardiography 2007;24:967–74. 8. Lodato JA, Weinert L, Baumann R, Coon P, Anderson A, Kim A et al. Use of 3dimensional color Doppler echocardiography to measure stroke volume in human beings: comparison with thermodilution. J Am Soc Echocardiogr 2007;20: 103 –12. 9. Haugen BO, Berg S, Brecke KM, Samstad SO, Skjaerpe T, Slordahl SA et al. Measurement of volumetric mitral and aortic blood flow based on a new freehand three-dimensional colour flow imaging method. An in vivo validation. Eur J Echocardiogr 2000;1:204 –12. 10. Ge S, Bu L, Zhang H, Schelbert E, Disterhoft M, Li X et al. A real-time 3dimensional digital Doppler method for measurement of flow rate and volume through mitral valve in children: a validation study compared with magnetic resonance imaging. J Am Soc Echocardiogr 2005;18:1 –7.

11. Mihaljevic T, von Segesser LK, Tonz M, Leskosek B, Seifert B, Jenni R et al. Continuous versus bolus thermodilution cardiac output measurements—a comparative study. Crit. Care Med. 1995;23:944–9. 12. Schmid ER, Schmidlin D, Tornic M, Seifert B. Continuous thermodilution cardiac output: clinical validation against a reference technique of known accuracy. Intensive Care Med. 1999;25:166–72. 13. Duann JR, Lin SB, Hu WC, Su JL. Computer system for four-dimensional transesophageal echocardiographic image reconstruction. Comput Med Imaging Graph 1999;23:173 –9. 14. Poulsen JK, Kim WY. Measurement of volumetric flow with no angle correction using multiplanar pulsed Doppler ultrasound. IEEE Trans Biomed Eng 1996;43: 589 –99. 15. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307 –10. 16. Critchley LA, Critchley JA. A meta-analysis of studies using bias and precision statistics to compare cardiac output measurement techniques. J Clin Monit Comput 1999;15:85 –91. 17. Darmon PL, Hillel Z, Mogtader A, Mindich B, Thys D. Cardiac output by transesophageal echocardiography using continuous-wave Doppler across the aortic valve. Anesthesiology 1994;80:796 –805; discussion 25A. 18. Perrino AC Jr, Harris SN, Luther MA. Intraoperative determination of cardiac output using multiplane transesophageal echocardiography: a comparison to thermodilution. Anesthesiology 1998;89:350–7. 19. Schmidt C, Theilmeier G, Van Aken H, Korsmeier P, Wirtz SP, Berendes E et al. Comparison of electrical velocimetry and transoesophageal Doppler echocardiography for measuring stroke volume and cardiac output. Br J Anaesth 2005;95: 603 –10. 20. Parra V, Fita G, Rovira I, Matute P, Gomar C, Pare C. Transoesophageal echocardiography accurately detects cardiac output variation: a prospective comparison with thermodilution in cardiac surgery. Eur J Anaesthesiol 2008;25:135 –43. 21. Kuhl HP, Franke A, Merx M, Hoffmann R, Puschmann D, Hanrath P. Rapid quantification of left ventricular function and mass using transoesophageal threedimensional echocardiography: validation of a method that uses long-axis cutplanes. Eur J Echocardiogr 2000;1:213 – 21. 22. Fisher DC, Sahn DJ, Friedman MJ, Larson D, Valdes-Cruz LM, Horowitz S et al. The mitral valve orifice method for noninvasive two-dimensional echo Doppler determinations of cardiac output. Circulation 1983;67:872–7. 23. Bettex DA, Hinselmann V, Hellermann JP, Jenni R, Schmid ER. Transoesophageal echocardiography is unreliable for cardiac output assessment after cardiac surgery compared with thermodilution. Anaesthesia 2004;59:1184 – 92. 24. Pemberton J, Li X, Karamlou T, Sandquist CA, Thiele K, Shen I et al. The use of live three-dimensional Doppler echocardiography in the measurement of cardiac output: an in vivo animal study. J Am Coll Cardiol 2005;45:433 –8. 25. Pua EC, Idriss SF, Wolf PD, Smith SW. Real-time 3D transesophageal echocardiography. Ultrason Imaging 2004;26:217–32. 26. Schmidlin D, Bettex D, Bernard E, Germann R, Tornic M, Jenni R et al. Transoesophageal echocardiography in cardiac and vascular surgery: implications and observer variability. Br J Anaesth 2001;86:497 –505.

Downloaded from https://academic.oup.com/ehjcimaging/article-abstract/11/5/432/2396841 by guest on 14 December 2017