Copyright © 2005, The European Society of Cardiology
Semi-automatic boundary detection to improve reporting of regional left ventricular function
aDepartment of Cardiology, The John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
bMirada Solutions Ltd., Oxford, UK
Received 14 January 2005; received in revised form 15 May 2005; accepted after revision 1 June 2005.
* Corresponding author. Tel.: +44 1865 741166; fax: +44 1865 220585. jon.timperley{at}orh.nhs.uk
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
Aims The reporting of regional left ventricular function is based on subjective assessment of endocardial motion and thickening and has a significant learning curve. We hypothesized that the use of an semi-automatic boundary detection system generating images with superimposed moving endocardial borders and a fixed end-diastolic reference border could improve the reporting of regional function.
Methods We obtained 58 resting contrast images of 15 patients and using a new boundary detection system (Quamus®), generated images with superimposed endocardial borders. The contrast images, images with additional Quamus borders and Quamus borders alone were assessed by two level 1 and two level 2 echocardiographers. They scored regional function and results were compared to two level 3 experienced stress echocardiography readers.
Results The addition of borders improved the agreement of level 1 echocardiographers (weighted Kappa increased from 0.55 to 0.64) but did not change for level 2 echocardiographers (0.63 to 0.64) and has the potential to be a useful training tool.
Keywords: Automatic boundary detection; Left ventricular function; Contrast echocardiography
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
The assessment of left ventricular regional wall motion is an important part of any echocardiographic study and is the basis of stress echocardiography. The assessment of regional function is based on subjective interpretation of endocardial motion and wall thickening. It has long been recognized that reduced segmental wall motion correlates with ischaemic muscle action.1,2 However, there is a significant learning curve to the assessment of function, and also significant interobserver variability. The ACC/AHA clinical competence statement on echocardiography states "the assessment of segmental wall motion remains one of the most challenging aspects of echocardiographic interpretation".3 The use of intravenous contrast agents improves endocardial definition and reduces interobserver variability when image quality is limited4,5 and second harmonic imaging has been shown to improve endocardial visualization during dobutamine stress echocardiography.6 Despite these advances, there still remains a significant learning curve to regional assessment. A number of new techniques such as Doppler tissue imaging,7 regional strain rate and strain8 have been developed and used to assess regional function. However, regional function is usually reported semi-quantitatively based on a visual assessment. Relatively inexperienced echocardiographers often undertake initial echocardiographic examinations and the education of such echocardiographers is important.
Semi-automatic boundary detection systems have been developed and used for the quantification of global left ventricular function.9,10 We hypothesized that using a semi-automatic boundary detection system (Quamus®) with contrast images, displaying a fixed end-diastolic frame as well as a continuously moving endocardial border throughout the cardiac cycle, could improve the reporting of regional function by relatively inexperienced echocardiographers. We compared reporting of contrast images alone to those with Quamus borders superimposed and Quamus borders alone by two groups of echocardiographers with different levels of experience.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
Study population
The study group consisted of 15 patients requiring a clinically indicated contrast echocardiogram for the assessment of regional or global left ventricular function. Seven of the patients had suffered a previous myocardial infarction. The indication for the study was assessment of ventricular function in patients with known cardiac disease in 7, shortness of breath in 6 and systolic murmur in 2. All patients were in sinus rhythm. Patients with evidence of a cardiac shunt were excluded. All patients gave written informed consent and the local ethics committee approved the study.
Contrast echocardiography
Echocardiography was performed using a Philips Sonos 5500 system (Philips Medical Systems, Andover, Mass) using the S3 probe (1.6/3.2MHz). The low power modality Power Modulation was used with a mechanical index of 0.1 and the focus was set at the level of the mitral valve in the apical views. Gain controls were optimized at the beginning of the study and then left unchanged. The contrast agent SonoVue (Bracco, Milan, Italy) was infused through a cannula in the antecubital vein. The VueJect pump was used which continuously and alternately rotates the syringe through 180° to prevent bubbles from floating to the surface. SonoVue was infused at a rate of 0.7ml/min to a maximum of 15ml. The infusion rate was increased if there was significant destruction or inadequate border delineation and reduced if there was attenuation towards the mitral valve. Single loop cardiac cycles were obtained of the apical 4-chamber, 2-chamber and 3-chamber views and the parasternal short axis view at the mid-ventricular level. These images were stored on magneto-optical disk for off-line analysis. Care was taken to avoid foreshortening the images in the apical views. Left ventricular biplane volumes were calculated using Simpson's method of discs.
Off-line semi-automatic boundary detection
Images were transferred to an off-line personal computer (model P220, Dell Computers UK, Bracknell, UK) for further analysis. A prototype semi-automatic boundary detection software called Quamus was used (Mirada Solutions Ltd, Oxford, UK). This software uses a new approach to spatio-temporal boundary detection based on a phase-based method. Speckle noise corrupts data by introducing sharp changes in the image intensity profile whilst attenuation artefacts alter the intensity of equally significant cardiac structures. This suggests that measures based on phase information may be more appropriate than intensity based derivatives.11 Spatial methods ignore temporal continuity, i.e. that seen during endocardial motion, which can be used to improve the reliability of feature detection. For any given frame there will be a number of spurious artefacts that will have similar shape and intensity as cardiac tissue. However, as opposed to endocardial borders, most of these will not be persistent from one frame to the next. These artefacts are removed by a number of temporally orientated filters. The software weighs the importance of spatial features according to their temporal significance. This is not so the case with Acoustic Quantification based techniques.
Once the image is loaded, a number of approximate points are placed outside the endocardial border on the first frame of the image and in the case of the apical views, three points placed at the margins of the mitral valve annulus and the apex to define set points to divide the borders into six segments. For the short-axis view one point is placed at the junction of the inferior wall and inferior septum. Having drawn a rough estimate of the contour, the software refines the shape to a more accurate location on the endocardial border. Successive frames in the sequence are then tracked automatically using the same matching technique.
The endocardial border of the 4-chamber view is divided into six equal colour-coded segments, three in the septum and three in the lateral wall. In addition, a fixed end-diastolic border is displayed during the cardiac cycle as a reference to the moving colour-coded border (Fig. 1). These images were saved on a personal computer. The same process was repeated similarly for the 2-chamber and 3-chamber views.
|
Regional wall motion assessment
The regional wall motion of the contrast images (Fig. 1A) were assessed by two level 3 echocardiographers experienced in stress echocardiography (J.T., H.B.). They were scored as normal (1), hypokinetic (2), akinetic (3) or dyskinetic (4). Any differences were discussed leading to a single assessment for each of the six regions in the four different views.
The contrast images were then assessed by two level 1 echocardiographers (A and B) and two level 2 echocardiographers (C and D) (Fig. 1A). None of these four echocardiographers had any experience of stress echocardiography reading. An 18-segment model of the left ventricle was used. The myocardium was divided into six segments (two basal, two middle, two apical) in each of the apical views and six regions for the short axis view in the mid left ventricle. The apical long-axis view included assessment of two apical segments and so was a minor modification of the American Society of Echocardiography 16-segment model.12 The echocardiographers then re-assessed the images with the superimposed fixed end-diastolic border and the moving colour-coded endocardial border. This was repeated for all 58 images (Fig. 1B). Two weeks later, each reader interpreted images of the Quamus borders only, with no contrast image visible (Fig. 1C) in a different order to the contrast images and with no patient markings on the images. The images were presented in a random order and not as a set of images for each patient. Therefore, information from the short-axis view could not be used to assess the apical view mid segments and vice versa.
Statistical analysis
Continuous variables were expressed as mean±standard deviation. The inter-rater agreement between the Level 1 and 2 echocardiographers with the level 3 echocardiographers was assessed using a weighted Kappa test for agreement. The inter-rater level of agreement was interpreted in the following way13:
| ||||||||||||||
Agreement was also performed when considering regional function as a dichotomous group, either normal or abnormal (including hypokinetic, akinetic or dyskinetic) using the Kappa test for agreement.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
The mean age was 59±15 years, and the study group consisted 14 males and 1 female. Seven patients had a documented history of previous acute myocardial infarction. Mean left ventricular volumes were as follows: end diastolic 188±72ml, end systolic 115±56ml, ejection fraction 41±12% with 40% of the patients having an ejection fraction of less than 35%. Two images (one 4-chamber and one 2-chamber image) were unable to be processed due to poor image quality. Semi-automatic boundary detection using the Quamus software was possible on all other images giving 348 wall regions. Reporting of regional function by the two level 3 echocardiographers was as follows: 189 normal, 97 hypokinetic, 61 akinetic, 1 dyskinetic. All regions were interpretable.
Kappa tables comparing level 1 and 2 echocardiographers to the level 3 echocardiographers are shown in Tables 1–6
. The level 1 echocardiographers had moderate agreement with the level 3 echocardiographers for contrast images (weighted Kappa statistic 0.55) improving to good agreement when using the additional Quamus borders (weighted Kappa statistic 0.64). In comparison, there was no significant change with the level 2 echocardiographers, with good agreement using contrast images and with additional Quamus borders (weighted Kappa statistic 0.63 and 0.64, respectively).
|
|
|
|
|
|
There were similar results when assessing function dichotomously as either normal or abnormal (hypokinetic, akinetic or dyskinetic). Level 1 echocardiographers showed an improvement in agreement from moderate to good by the addition of Quamus borders (Kappa statistic increased from 0.55 to 0.64). In comparison, there was good agreement with the level 2 echocardiographers using contrast images and with additional Quamus borders (Kappa statistic increased from 0.65 to 0.66). Individual results for each reader are shown in Table 7.
|
When looking at images of Quamus borders alone with no underlying contrast image (Fig. 1C), the agreement was moderate for both level 1 and 2 echocardiographers for semi-quantitive scoring of regional function (level 1 weighted Kappa 0.54, level 2 weighted Kappa 0.51). For dichotomous reading the agreement for the level 1 echocardiographers was good (Kappa 0.60) but only moderate for the level 2 echocardiographers (Kappa 0.53).
On removing the apical long-axis segments from the analysis, i.e. using the American Society of Echocardiography 16-segment model, there was no change in the agreement between the groups.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
The assessment of regional left ventricular function is a skill with a significant learning curve, and is based on visual interpretation of endocardial motion and thickening. Shape distortion of the left ventricle caused by cardiomyopathy, ischaemic and valvular heart disease is well recognized14–16 and recently left ventricular shape assessment has been used as a tool in the determination of left ventricular function.17 Semi-automatic boundary detection systems have been used to try and quantify regional and global function using a number of techniques.18–20 To our knowledge, there are no data to date regarding the interpretation of echocardiograms using superimposed endocardial borders with contrast echocardiograms.
The addition of a moving endocardial border and a fixed end-diastolic border improved the reporting of regional function by level 1 echocardiographers from moderate to good, with the Kappa statistic increasing from 0.55 to 0.64. The level 1 echocardiographers using the additional Quamus borders had the same agreement as the level 2 echocardiographers using contrast or contrast with Quamus borders (Kappa statistic 0.63 and 0.64, respectively). This demonstrates the ability of this system of image display to improve regional reporting of inexperienced readers to a higher level, and also its potential as an aid to training. The contours, although only demonstrating endocardial excursion, appear to bring to the attention of the level 1 echocardiographers information on regional function that was missed whilst examining the contrast images alone.
Both groups had better agreement with the presence of Quamus borders with contrast images compared to Quamus borders alone. This probably is due to echocardiographers, including the inexperienced level 1 echocardiographers, taking information of regional thickening from the underlying contrast images. When looking at Quamus borders alone, level 2 echocardiographers had the worst correlation and had a poorer correlation than the level 1 echocardiographers, probably representing the fact that they use myocardial thickening as major part of their regional wall function interpretation.
This study included patients with significantly impaired function with a mean ejection fraction of 41%, and 40% of the patients having an ejection fraction of less than 35%. We have therefore looked at a population with significant impairment, a population where missing a regional wall motion abnormality may be significant.
Although in this study we have used an 18-segment model with two extra apical long-axis segments, removal of these from the analysis did not affect the results. Such models have been used previously and the borders from the Quamus software are readily available.
|
Limitations |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
In this study, the decision of two experienced observers was used as the gold standard, which is still subject to variability. However, these were resting images and not dobutamine stress echocardiograms, and hence the use of coronary angiography as a gold standard was not used. Although only a small number of patients were included, this still gave 348 regions of which 159 were abnormal.
We have not used parasternal long axis images in this study. The lack of the apex makes semi-automatic tracking using Quamus, which is dependent on the presence of a continuous border very challenging. However, by including the long-axis view we have included all 16 regional segments.
In this study we have only looked at resting images due to the relatively low frame rate possible with power modulation (20–25Hz). With the high heart rates occurring at peak dobutamine stress, the number of frames for semi-automatic detection is small. In view of this and the fact that we aimed at assessing the use of this system as a training tool for relatively inexperienced echocardiographers in whom stress echocardiography training would occur at some time in the future, we did not use dobutamine stress images.
|
Conclusions |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
In conclusion, the use of a semi-automatic boundary detection system with contrast images which presents the reader with a moving endocardial border and a fixed end-diastolic border is an effective aid in the interpretation of regional function by inexperienced echocardiographers. The less experienced reader uses endocardial excursion more than more experienced echocardiographers in the decision-making process for regional function. For these echocardiographers, such a system as used above has the potential to be a useful training tool. The best results were obtained when using Quamus borders in addition to the underlying contrast images, allowing for interpretation of thickening from the contrast and the aid of the boundaries to guide inexperienced echocardiographers to the abnormality.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionLimitationsConclusionsReferences |
|---|
- Kerber R.E., Abboud F.M. Echocardiographic detection of regional myocardial infarction. Circulation (1973) 47:997–1005.
[Abstract/Free Full Text] - Baxley W.A., Reeves T.J. Abnormal regional myocardial performance in coronary artery disease. Prog Cardiovasc Dis (1971) 13:405–421.[CrossRef][Medline]
- Quinones M.A., Douglas P.S., Foster E., Gorcsan J. 3rd, Lewis J.F., Pearlman A.S., et al. American College of Cardiology/American Heart Association clinical competence statement on echocardiography: a report of the American College of Cardiology/American Heart Association/American College of Physicians — American Society of Internal Medicine Task Force on Clinical Competence. Circulation (2003) 107:1068–1089.
[Free Full Text] - Porter T.R. Role of intravenous ultrasound contrast in stress echocardiography. Echocardiography (2000) 17:92–99.[Web of Science][Medline]
- Schnaack S.D., Siegmund P., Spes C.H., Tammen A.R., Theisen K., Angermann C.E. Transpulmonary contrast echocardiography: effects on delineation of endocardial border, assessment of wall motion and interobserver variability in stress echocardiograms of limited image quality. Coron Artery Dis (2000) 11:549–554.[CrossRef][Web of Science][Medline]
- Sozzi F.B., Poldermans D., Boersma E., Elhendy A., Bax J.J., Borghetti A., et al. Does second harmonic imaging improve left ventricular endocardial border identification at higher heart rates during dobutamine stress echocardiography. J Am Soc Echocardiogr (2000) 13:1019–1024.[CrossRef][Web of Science][Medline]
- Gorcsan J. 3rd, Deswal A., Mankad S., Mandarino W.A., Mahler C.M., Yamazaki N., et al. Quantification of the myocardial response to low-dose dobutamine using tissue Doppler echocardiographic measures of velocity and velocity gradient. Am J Cardiol (1998) 81:615–623.[CrossRef][Web of Science][Medline]
- Davidavicius G., Kowalski M., Williams R.I., D'Hooge J., Salvo G.D., Pierre-Justin G., et al. Can regional strain and strain rate measurement be performed during both dobutamine and exercise echocardiography, and do regional deformation responses differ with different forms of stress testing? J Am Soc Echocardiogr (2003) 16:299–308.[CrossRef][Web of Science][Medline]
- Yu E.H.C., Skyba D.M., Sloggett C.E., Jamorki M., Iwanochko R.M., Dias B.F., et al. Determination of left ventricular ejection fraction using intravenous contrast and a semiautomated border detection algorithm. J Am Soc Echocardiogr (2003) 16:22–28.[CrossRef][Web of Science][Medline]
- Mele D., Sclavo M., Seveso G., Aschieri D., Nesta F., D'Aiello I., et al. Impact of tissue harmonic imaging in patients with distorted left ventricles: improvement in accuracy and reproducibility of visual, manual and automated echocardiographic assessment of left ventricular ejection fraction. Eur J Echocardiogr (2002) 4:59–67.
- Mulet-Parada M., Noble J.A. 2D+T acoustic boundary detection in echocardiography. Med Image Anal (2000) 4:21–30.[CrossRef][Web of Science][Medline]
- Schiller N.B., Shah P.M., Crawford M., DeMaria A., Devereux R., Feigenbaum H., et al. Recommendations for quantification of the left ventricle by two-dimensional echocardiography. J Am Soc Echocardiogr (1989) 2:358–367.[Medline]
- Altman D.G. Practical statistics for medical research. (1991) London: Chapman and Hall.
- Douglas P.S., Morrow R., Ioli A., Reichek N. Left ventricular shape, afterload and survival in idiopathic dilated cardiomyopathy. J Am Coll Cardiol (1989) 13:311–315.[Abstract]
- Fantini F., Barletta G., Voegelin M.R., Fantini A., Maioli M., Di Donato M. Abnormalities of left ventricular shape in patients with stable angina. Int J Cardiol (1990) 27:107–116.[CrossRef][Web of Science][Medline]
- Howard R.J., Moe G.W., Armstrong P.W. Sequential echocardiographic-Doppler assessment of left ventricular remodelling and mitral regurgitation during evolving experimental heart failure. Cardiovasc Res (1991) 25:468–474.
[Abstract/Free Full Text] - Arsenault M., Crete M., Bergeron S. Left ventricular shape assessment: a new simple diagnostic tool in stress echocardiography. J Am Soc Echocardiogr (2003) 15:1321–1325.[Web of Science]
- Fujino T., Ono S., Murata K., Tanaka N., Tone T., Yamamura T., et al. New method of on-line quantification of regional wall motion with automated segmental motion analysis. J Am Soc Echocardiogr (2001) 14:892–901.[CrossRef][Web of Science][Medline]
- Chandra S., Bahl V.K., Reddy S.C., Bhargava B., Malhotra A., Wasir H.S. Comparison of echocardiographic acoustic quantification system and radionuclide ventriculography for estimating left ventricular ejection fraction: validation in patients without regional wall motion abnormalities. Am Heart J (1997) 133:359–363.[CrossRef][Web of Science][Medline]
- Lucariello R.J., Sun Y., Doganay G., Chiaramida S.A. Sensitivity and specificity of left ventricular ejection fraction by echocardiographic automated border detection: comparison with radionuclide ventriculography. Clin Cardiol (1997) 20:943–948.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||