Skip Navigation

European Journal of Echocardiography 2007 8(6):431-437; doi:10.1016/j.euje.2007.03.002
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Karagiannis, S. E.
Right arrow Articles by Poldermans, D.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Karagiannis, S. E.
Right arrow Articles by Poldermans, D.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

Copyright © 2007, The European Society of Cardiology

Automated coupled-contour and robust myocardium tracking in stress echocardiography

Stefanos E. Karagiannisa, Jos Roelandta, Maleeha Qazib, Sriram Krishnanb, Harm H.H. Feringaa, Radosav Vidakovica, George Karatasakisc, Dennis V. Cokkinosc and Don Poldermansa,*

aDepartment of Cardiology, Erasmus MC, Rotterdam, The Netherlands
bSiemens Medical Solutions, Malvern, PA, USA
c1st Department of Cardiology, Onassis Cardiac Surgery Centre, Athens, Greece

Received 5 August 2006; received in revised form 3 October 2006; accepted after revision 4 March 2007.

* Corresponding author. Department of Cardiology, Room H 921, Erasmus MC, Dr. Molewaterplein 40, 3015 GD Rotterdam, The Netherlands. Tel.: +31104639222; fax: +31104634957. d.poldermans{at}erasmusmc.nl


   Abstract
Top
 Abstract
Introduction
 Medical background
 Automated coupled-contour and...
 Technical considerations for...
 Clinical implications
 References
 
Dobutamine stress echocardiography is a commonly used imaging modality for the diagnosis of coronary artery disease and the detection of myocardial viability. The major limitations are that it is operator dependent and that the analysis is subjective and qualitative resulting in interobserver variability. It is also tedious and time consuming. Consequently, several quantitative approaches have been proposed, such as acoustic quantification and color kinesis but none of these has proved to be fully quantitative. In this manuscript we describe the development of a new, quantitative technique based on tracking of both endocardium and epicardium providing information of endocardial excursion and myocardial thickening, a crucial parameter of wall function evaluation. Preliminary data indicate that the method is practical and feasible, but clinical trials are required to prove whether it will improve the sensitivity and specificity of dobutamine stress echocardiography.

Keywords: Dobutamine stress echocardiography; Subjective; Quantitative techniques; Myocardial thickening


   Introduction
Top
 Abstract
 Introduction
Medical background
 Automated coupled-contour and...
 Technical considerations for...
 Clinical implications
 References
 
Two-dimensional echocardiography is a commonly used non-invasive method for the assessment of left ventricular function. It provides precise information on both global and segmental myocardial function by displaying endocardial motion and wall thickening.

Dobutamine stress echocardiography (DSE) is currently the leading non-invasive imaging technique for the diagnosis and prognosis of coronary artery disease (CAD), the documentation of myocardial viability and for long-term risk stratification.1

However, assessment of endocardial motion and wall thickening from 2D echocardiography images largely depends on image quality and is both subjective and qualitative.2 This may result in interpretation variability.


   Medical background
Top
 Abstract
 Introduction
 Medical background
Automated coupled-contour and...
 Technical considerations for...
 Clinical implications
 References
 
Current limitations of dobutamine stress echocardiography
The sensitivity of DSE for detecting CAD ranges from 81% to 85% with a similar specificity range. These values are less in the presence of single vessel disease.2 Nevertheless, they are similar to single-proton emission tomography (SPECT) and the other non-invasive stress imaging modalities for both the detection of myocardial ischemia and viability (Table 1).2–7


View this table:
[in this window]
[in a new window]

  		Table 1 Comparison of sensitivities and specificities of different stress imaging modalities in detecting CAD (A) or myocardial viability (B). Concluding results from meta-analyses1,3–7

 
The subjective regional wall motion scoring system in DSE is based on using a five point score (1=normokinesis; 2=mild hypokinesis; 3=moderate or severe hypokinesis; 4=akinesis; and 5=dyskinesis), dividing the left ventricle (LV) into 17 segments according to the recommendations of the American Society of Echocardiography.2,8,9 A test is considered positive if new wall motion abnormalities occur (i.e., if wall motion in any segment worsens by 1 grade during testing, with the exception of akinesis becoming dyskinesis).2,8,9

However currently, DSE is an experience dependent technique subject to a high degree of inter- and intra-observer variability. Signal dropout can be the cause of suboptimal images and even lead to misdiagnosis in some patients. The majority of studies that support its use are mainly derived from expert centres10 and show the higher agreement for detection of 3-vessel disease (Fig. 1). Limitations occur in the performance of the test in less expert hands, as the results obtained by stress echocardiograms result from subjective interpretation of wall motion.11 (Table 2). Furthermore, analysis of DSE data is tedious and time consuming.


Figure 1
View larger version (8K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 1 Interinstitutional agreement on stress echocardiography for different degrees of coronary artery disease.10

 


View this table:
[in this window]
[in a new window]

  		Table 2 Stress echocardiography limitations

 
Quantitative techniques
Quantitative techniques have been developed to provide objective quantitative and reproducible information on global cardiac and regional wall function during stress.1,2,11–15

The most important are: (1) image enhancement techniques including second harmonic imaging, endocardial border enhancement by transpulmonary contrast agents (Albunex, Laevovist, BY 963) for improving cavity delineation and 3D imaging with or without contrast agents and (2) analytic software improvement, such as automatic contour techniques (acoustic quantification, colour kinesis) and calculation techniques, such as tissue Doppler myocardial velocity derived parameters for analysis.1,12,15

They provide better image quality and hence allow a more accurate analysis resulting in less interobserver variability. However, none of these methods takes into account the thickening of the myocardium.

Color kinesis requires good to excellent 2D-image quality for tracking of the endocardium, and encodes translational and rotational motion of the heart as endocardial motion. Its accuracy is also affected by tachycardia and arrhythmias including atrial fibrillation and frequent ectopic beats.11,12

Endocardial tracking based on tissue Doppler technique is a semi-quantitative technique that can be used for the estimation of global ventricular function but has limitations in the presence of regional wall motion abnormalities mainly because of low temporal resolution.12–14

Consequently the development of more sophisticated quantitative techniques is needed. These should take into account several criteria and should aim at an increase in sensitivity without loss in specificity11 (Table 3). Cardiac ultrasound is a reflection technique. In addition to the fast motion of the heart, respiratory interferences pose additional problems in the analysis.17 A new quantitative technique ideally should take into account several parameters, such as the shape and the motion of both the endocardium and epicardium, providing important information on myocardial thickening.11,15,17


View this table:
[in this window]
[in a new window]

  		Table 3 Criteria for an improved quantitative methodology to stress echocardiography

 

   Automated coupled-contour and robust myocardium tracking
Top
 Abstract
 Introduction
 Medical background
 Automated coupled-contour and...
Technical considerations for...
 Clinical implications
 References
 
Coupled-contour and robust fusion for myocardial tracking is a newly developed analytic software that aims to provide automated on-line quantitative data from images obtained in the setting of DSE. It is designed to track both myocardial layers (i.e. epi- and endocardium) and uses both excursion and thickening of the myocardium to detect wall function abnormalities. Early automated contour detection techniques tracked only the endocardium in high-quality images off-line.18

Preliminary data have shown an at least maintained sensitivity and increased specificity for the detection of CAD. In a few preliminary clinical data standard DSE had a sensitivity of 88% and specificity 75%. When the automated on-line system was used, sensitivity was greater than 80% (p: non-significant), for 80% specificity (p>0.01). These findings are comparable with the current DSE data, however, potentially there are improvements to be expected as the system is less dependent on the quality of the images obtained. Furthermore, while not a true speckle tracking system, the system uses optical flow, and is capable of measuring rotation.16,19,21


   Technical considerations for quantitative wall assessment
Top
 Abstract
 Introduction
 Medical background
 Automated coupled-contour and...
 Technical considerations for...
Clinical implications
 References
 
The new software automatically classifies the myocardial segments as normal or abnormal based on the endocardial excursion and thickening. This process consists of 3 steps: detection, tracking, and feature extraction and classification17,19 (Fig. 2)


Figure 2
View larger version (74K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 2 The principle of automatic myocardial wall tracking. (A) (step 1: border detection): dots in the user defined area represent endocardial detection. (B) (step 2: border tracking): dots in the user defined area represent epi- and endocardial tracking. (C) (step 3: feature extraction and classification in a different patient than A and B): dots are interconnected with lines at both the epi- and endocardium to generate contours used for image quantification.

 
Step 1: detection by database-guided segmentation
The first step is the identification of the endocardium (Fig. 3). The application is based on learned pattern recognition of several shape models. In general there are several different shape models17,19 (Fig. 4).This is done using a technique called Database-Guided Segmentation.17,19 As the name implies, segmentation is guided by a database of cases which attempts to "teach" the system how to detect and segment a border. This is done by first creating a large database of cases which represent the wide variety of patients and disease types typically tested in coronary artery disease. Each of these cases is then contoured by one or more experts, giving the basis of "learning" to the system. The learning set of the described system is based on data from 345 DSE tests delivered from several centres, scored in Erasmus MC, Rotterdam, The Netherlands. Each segment has a score 1–5 (1=normal, 2=hypokinetic, 3=akinetic, 4=dyskinetic, 5=aneurysmal). From this database and domain knowledge of the individual LV contours, a reference set of LV shapes is created. Each reference shape may be the result of averaging shapes of many studies and is referred to as the "learning set". These shapes are subsequently used for matching and recognition of LV contours and shapes during a DSE study.17,19,22


Figure 3
View larger version (80K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 3 (A) The first step is the localization and detection of the endocardium by using multiple LV frames. (B) Dots depict the epi- and endocardial border.

 


Figure 4
View larger version (124K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 4 Images of learning sets showing different long-axis (A) and short-axis (B) left ventricular shape models. The whole application is based on pattern recognition using many different shapes of left ventricles learned from patient studies. Each image is an average of several cross-sections obtained from different DSE studies. Dots interconnected by lines represent the endocardium.

 
Step 2: tracking by robust information fusion
The second step of the process is to track the border. This is called Robust Information Fusion,17,19 and uses learned motion models of the heart to track the endocardium throughout the heart cycle. That is, once the endocardial border is identified in one frame, it needs to be tracked to all of the other frames in the heart. This is done by first recognizing that the specific location of the endocardium as identified by the system has some level of uncertainty (Fig. 5). These measurement uncertainties stem for the echocardiograms from the acoustic dropout (where tissue surface is parallel to ultrasound beam).19–21,23 Tracking is performed by obtaining independently from each model a motion estimate and its uncertainty through optical flow.17,19 These models are again learned from the database of previous studies, just as detection models were learned for detection.


Figure 5
View larger version (91K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 5 These apical four-chamber views are obtained from two different patients undergoing a DSE study. The magnitude of the ellipses shows the uncertainties in motion estimation of the dots representing the endocardium in these examples. The larger the ellipse the bigger the uncertainty. (A) the uncertainties are largest in the distal/septal area while in (B) they are largest in the apex. This means that the system cannot accurately recognize the exact location of a certain landmark point in the next frame.

 
In echocardiographic images the epicardium is often more difficult to identify than endocardium by a software algorithm because of the high-reflecting surrounding structures. So for epicardium tracking, this system utilizes a double contour that can propagate information from the endocardium to guide the localization of the epicardium (or vice versa), and can better follow its location during the cardiac cycle and reduce the chance of crossing.17,19 For this purpose it uses the apex, the papillary muscles and the basal end of septum as landmark points that can be automatically assigned.17 This coupled-contour approach integrates more spatial information, thus can provide more robust tracking of the endocardium and epicardium (Fig. 6) and permits accurate border motion tracking.17,19,20


Figure 6
View larger version (94K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 6 The second step is coupled motion tracking of endo- and epicardial contours. A four-chamber view is shown as an example. The tracking is done by matching (fusing) the information gained by the learning set of the system's subsequent steps. Then the continuous contour estimation of each subsequent frame follows as shown in Fig. 5.

 
Step 3: feature extraction and classification
Feature extraction and classification is the third step of the processing of the data. Once the endocardial and epicardial border have been detected and tracked, a coupled-contour is derived, from which the system extracts a few global (involving the whole LV) and local (involving individual segments visible in the image) numerical features for quantitation.17,19 These features derive from volume changes (segmental volumes by Simpson's rule), velocity, timing, radial strain (thickening) and circumferential strain 17,19–24 (Fig. 7).


Figure 7
View larger version (57K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 7 The final step is the feature extraction and classification. Once the endocardial and epicardial borders are detected and their motion is tracked, the system extracts global and segment based numerical features. These are velocity, thickening, timing and volume changes. These numerical features are the classifiers resulting in automatic quantification. Dots and lines in apical four-chamber image represent the coupled-contour of endo- and epicardial border. The graph provides the probability of a normal or abnormal new case (segment) based on numerical feature extraction. The probability curve shifts to the left or to the right according to the number of the features of a new case that match to features previously collected by the learning set of the system and were classified as normal or abnormal using criteria.

 
For each segment of a DSE study, a subset of relevant features is selected for classification. Statistical correlation methods of small number of features are applied for proper feature collection from the DSE study. Using a small number of features not only improves performance time but also results in a more robust and versatile learning set.17,19 The collected features are the classifiers and will provide the final automatic quantification.

Fig. 7 depicts the diagram of system processing, from data to final quantitative features.


   Clinical implications
Top
 Abstract
 Introduction
 Medical background
 Automated coupled-contour and...
 Technical considerations for...
 Clinical implications
References
 
Coronary artery disease is one of the leading causes for mortality and morbidity. DSE could become the primary test as an everyday clinical tool, if a quantitative technique could substitute the eyeball assessment of regional wall motion abnormalities. The above-mentioned technique seems promising as it aims to improve the accuracy and reduce the inter- and intra-observer variability of DSE. It is the first robust automatic coupled-contour myocardial tracking technique that utilizes the thickening of the myocardium during systole. However, the system may not accurately recognize the exact location of a certain landmark point in subsequent frames. Moreover the training set of the proposed image processing technique is set by "expert" results which in turn are subject to a certain degree of observer variability. These could present possible limitations of the proposed system. It remains to be seen whether this image processing technique will be feasible and will provide an accurate interpretation of wall motion abnormalities.


   References
Top
 Abstract
 Introduction
 Medical background
 Automated coupled-contour and...
 Technical considerations for...
 Clinical implications
 References
 

  1. Poldermans D., Bax J.J. Dobutamine echocardiography: a diagnostic tool comes of age. Eur Heart J (2003 Sep) 24(17):1541–1542.[Free Full Text]
  2. Picano E. Stress echocardiography. 4th ed. 2003. Springer Verlag.
  3. Fleischmann K.E., Hunink M.G.M., Kuntz K.M., Douglas M.D. Exercise echocardiography or exercise SPECT imaging? A meta-analysis of diagnostic test performance. JAMA (1998) 280:913–920.[Abstract/Free Full Text]
  4. Machac J. Cardiac positron emission tomography imaging. Semin Nucl Med (2005) 35:17–36.[CrossRef][Web of Science][Medline]
  5. Nagel E., Lehmkuhl H.B., Klein C., Schneider U., Frantz E., Ellmer A., et al. Influence of image quality on the diagnostic accuracy of dobutamine stress magnetic resonance imaging in comparison with dobutamine stress echocardiography for the non-invasive detection of myocardial ischemia. Z Kardiol (1999 Sep) 88(9):622–630.[CrossRef][Web of Science][Medline]
  6. Nagel E., Lehmkuhl H.B., Bocksch W., Klein C., Vogel U., Frantz E., et al. Noninvasive diagnosis of ischemia-induced wall motion abnormalities with the use of high-dose dobutamine stress MRI comparison with dobutamine stress echocardiography. Circulation (1999) 99:763–770.[Abstract/Free Full Text]
  7. Bax J.J., Wijns W., Cornel J.H., Visser F.C., Boersma E., Fioretti P.M. Accuracy of currently available techniques for prediction of functional recovery after revascularization in patients with left ventricular dysfunction due to chronic coronary artery disease: comparison of pooled data. J Am Coll Cardiol (1997 Nov 15) 30(6):1451–1460.[CrossRef]
  8. American Society of Echocardiography Committee on Standards, Subcommittee on Quantification of Two Dimensional Echocardiograms. Recommendations for quantitation of the left ventricle by two dimensional echocardiography. J Am Soc Echocardiogr (1989) 2:358–367.[Medline]
  9. Armstrong W.F., Pellikka P.A., Ryan T., Crouse L., Zoghbi W.A. Stress echocardiography: recommendations for performance and interpretation of stress echocardiography. J Am Soc Echocardiogr (1998) 11:97–104.[CrossRef][Web of Science][Medline]
  10. Hoffmann R., Lethen H., Marwick T., Arnese M., Fioretti P., Pingitore A., et al. Analysis of interinstitutional observer agreement in interpretation of dobutamine stress echocardiograms. J Am Coll Cardiol (1996 Feb) 27(2):330–336.[Abstract]
  11. Marwick T.H. Quantitative techniques for stress echocardiography: dream or reality? Eur J Echocardiogr (2002) 3:171–176.[CrossRef][Medline]
  12. Platts D.G., Monaghan M.J. Colour encoded endocardial tracking: the current state of play. Eur J Echocardiogr (2003) 4:6–16.[CrossRef][Medline]
  13. Leischik R., Bartel T., Mohlenkamp S., Bruch C., Buck T., Haude M., et al. Stress echocardiography: new techniques. Eur Heart J (1997 Jun) 18(Suppl. D):D49–D56.[Web of Science][Medline]
  14. Pan C., Hoffmann R., Kuhl H., Severin E., Franke A., Hanrath P. Tissue tracking allows rapid and accurate visual evaluation of left ventricular function. Eur J Echocardiogr (2001) 2:197–202.[Abstract/Free Full Text]
  15. Pellikka Patricia A. Stress echocardiography for the diagnosis of coronary artery disease: progress towards quantification. Curr Opin Cardiol (2005) 20:395–398.[CrossRef][Web of Science][Medline]
  16. Helle-Valle T., Crosby J., Edvardsen T., Lyseggen E., Amundsen B.H., Smith H.J., et al. New noninvasive method for assessment of left ventricular rotation: speckle tracking echocardiography. Circulation (2005 Nov 15) 112(20):3149–3156.[CrossRef]
  17. Zhou X.S., Comaniciu D., Gupta A. An information fusion framework for robust shape tracking. Trans Pattern Anal Mach Intell (2005 Jan) 27(1):115–129.[CrossRef]
  18. Hozumi T., Yoshida K., Yoshioka H., Yagi T., Akasaka T., Takagi T., et al. Echocardiographic estimation of left ventricular cavity area with a newly developed automated contour tracking method. J Am Soc Echocardiogr (1997 Oct) 10(8):822–829.[CrossRef][Web of Science][Medline]
  19. Comaniciu D., Zhou X.S., Krishnan S. Robust real-time myocardial border tracking for echocardiography: an information fusion approach. Trans Med Imaging (2004 Jul) 23(7):849–860.[CrossRef]
  20. Jacob G., Noble J.A., Behrenbruch C., Kelion A.D., Banning A.P. A shape-space-based approach to tracking myocardial borders and quantifying regional left-ventricular function applied in echocardiography. Trans Med Imaging (2002 Mar) 21(3):226–238.[CrossRef]
  21. Chen H., Meer P. Robust fusion of uncertain information. Trans Syst Man Cybern B Cybern (2005 Jun) 35(3):578–586.[CrossRef]
  22. Dydenko I., Jamal F., Bernard O., D'hooge J., Magnin I.E., Friboulet D. A level set framework with a shape and motion prior for segmentation and region tracking in echocardiography. Med Image Anal (2006 Apr) 10(2):162–177. [Epub 2005 Sep 13].[CrossRef][Web of Science][Medline]
  23. Jacob G., Noble J.A., Kelion A.D., Banning A.P. Quantitative regional analysis of myocardial wall motion. Ultrasound Med Biol (2001 Jun) 27(6):773–784.[CrossRef][Web of Science][Medline]
  24. Chen W., Meer P., Georgescu B., He W., Goodell L.A., Foran D.J. Compute image mining for investigative pathology using optimized feature extraction and data fusion. Methods Programs Biomed (2005 Jul) 79(1):59–72.[CrossRef]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?



This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Karagiannis, S. E.
Right arrow Articles by Poldermans, D.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Karagiannis, S. E.
Right arrow Articles by Poldermans, D.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?