Estimation of 3D left ventricular deformation from echocardiography☆
Introduction
A fundamental goal of many efforts in the cardiac imaging and image analysis communities is to assess the regional function of the left ventricle (LV) of the heart. The general consensus is that the analysis of heart wall deformation provides quantitative estimates of the location and extent of ischemic myocardial injury. Regional left ventricular deformation can be determined using all of the principal imaging modalities, including contrast angiography, echocardiography, radionuclide imaging, cine computed tomography (CT), and magnetic resonance (MR) imaging. There have been considerable efforts within the medical image analysis community aimed at estimating this deformation from each of these imaging modalities. Much of the effort has been confined to analysis of two-dimensional images or projections of the heart. Although, recently significant effort has been directed at a more comprehensive analysis of left ventricular deformation in all three dimensions.
Left ventricular deformation can be assessed in three-dimensional space using ECG-gated single photon emission computed tomography (SPECT) Shen et al., 1999, Cwajg et al., 1999, Faber et al., 1999, Calnon et al., 1997, Berman and Germano, 1997, Germano et al., 1995, Cooke et al., 1994 or positron emission tomography (PET) Miller et al., 1994, Yamashita et al., 1989, Buvat et al., 1997. However, both of these radionuclide methods have a restricted ability to assess left ventricular deformation, secondary to the limited spatial and temporal resolution of these approaches. These radionuclide methods have involved both count-based Shen et al., 1999, Calnon et al., 1997, Cooke et al., 1994 and geometry-based approaches Cwajg et al., 1999, Faber et al., 1999, Germano et al., 1995.
Cine MR imaging has emerged as a more comprehensive approach to assess myocardial deformation in three-dimensional space (Moore et al., 2000). MR imaging offers improved spatial resolution. Unique to cine MR imaging is the ability to track deformation of myocardial tissue within the wall as well as on the endocardial and epicardial surfaces. However, the analysis of mid-wall myocardial deformation requires special cine MR imaging sequences, including MR tissue tagging Amini et al., 1998, Haber et al., 1998, Kerwin and Prince, 1998, Park et al., 1996, Prince and McVeigh, 1992, Young et al., 1995 and others, or MR phase contrast velocity imaging Pelc, 1991, Duncan et al., 1998, Meyer et al., 1996, Zhu et al., 1997. While these newer MR approaches offer a comprehensive analysis of regional left ventricular deformation, wide application of MR imaging remains limited by cost and the difficulty in routinely applying these MR approaches to critically ill cardiac patients.
Echocardiography offers significant advantages over both radionuclide imaging and MR imaging. Echocardiographic images can be acquired on critically ill patients in an emergency room or at the patient’s bedside in the intensive care unit, and this can be accomplished at a reduced cost. Comprehensive analysis of left ventricular deformation is now feasible using echocardiography, with the advent of newer three-dimensional acquisition systems (von Ramm and Smith, 1990). Recently, commercial software has become available to automatically assess global and regional left ventricular function (Lang et al., 1996). However, these newer automated echocardiographic approaches have not been fully validated. Hence, development of automated analysis of echocardiographic images is attracting an increasing amount of attention in the literature Chuang et al., 1999a, Chuang et al., 1999b, Chuang et al., 2000, Sheehan et al., 1998, Angelini, 1999, Brandt et al., 1999, Montagnat et al., 1999, Jacob et al., 1999. However, none of these methods is capable of estimating dense maps of three-dimensional deformation from echocardiographic images comparable to those obtained from the analysis of MR tagging images.
In this paper we describe, test and present prelinary validation for an approach to estimate the regional three-dimensional deformation of the left ventricle using echocardiography. We use a biomechanical model to describe the myocardium and shape-based tracking displacement estimates on the endocardial and epicardial walls to generate the initial displacement estimates. These are integrated in a Bayesian estimation framework and the overall problem is solved using the finite element method. This method produces quantitative regional 3D cardiac deformation estimates from ultrasound images which up to now was thought to be only possible using magnetic resonance and especially MR tagging. We validate these estimates by comparing them to invasive measurements performed simultaneously using implanted sonomicrometers. The fast improving quality of ultrasound images with the introduction of harmonic imaging (Caidahl et al., 1999) and contrast agents (Porter et al., 1994) should make it possible to obtain even more accurate estimates of 3D left ventricular deformation in the future.
Section snippets
Our approach
We estimate a dense displacement field within a Bayesian estimation framework which consists of a data term and a model term. These are described in Sections 2.1 and 2.2, respectively. The data term captures the image-derived information about the problem. We segment the images interactively and then proceed to extract initial displacement estimates using a shape-tracking approach. We then model the noise in these estimates using a Gaussian noise model. The model term captures our prior beliefs
Animal experiments
To evaluate the efficacy of using image-derived in vivo deformation estimates to measure regional LV function we conducted experiments on fasting, anesthetized, open chest, adult mongrel dogs with approval of the Yale University Animal Care and Use Committee. In this preliminary work, we report results from four animals. The 3DE images were obtained either before (D1 and D2) or after occlusion of the left anterior descending coronary artery (D3 and D4), using the procedure described in Section
Results
The potential of our methodology is illustrated in Fig. 9, which shows a cut through our tracked 3D mesh overlaid on a slice through the original 3DE image data over time. This could be seen as a form of software-derived, 3DE-based ‘tissue tagging’ somewhat in the sense of MR tagging. Note the spreading grid lines near the endocardium on the right as the LV thickens from end diastole to end systole.
The quantitative results are summarized in Table 1. Function in the risk area, which was
Validation using implanted sonomicrometers
In an effort to obtain an independent source of in vivo strain values for validation of image-derived strains, we have developed an independent approach for strain measurement using cubic arrays of sonomicrometers implanted in the canine LV myocardium (see details in (Dione et al., 1997)). The efficacy of this technique was illustrated by additional work (Meoli et al., 1998) that showed that the distances obtained with sonomicrometers compared favorably (r=0.992) with those obtained using the
Conclusions
In this work we have demonstrated that estimates of 3D cardiac deformation can be obtained from ultrasound images. These estimates are generally consistent with values reported in the literature. Further, we validate such estimates of regional deformation directly by comparing them to strains measured concurrently from implanted sonomicrometers. While many problems remain to be solved, such as improving and speeding up the segmentation process, we are confident that this approach has the
Movies
There are three movies included with this paper (available as electronic annexes via www.elsevier.com/locate/media). The first movie, papad1.mov, corresponds to Fig. 9. The second and third movies, papad2.mov and papad3.mov, correspond to the top and bottom parts of Fig. 10, respectively.
Supplementary data
Acknowledgements
The first author would also like to thank professors Turan Onat and Gary Povirk from the Department of Mechanical Engineering at Yale University for many useful discussions. Additional thanks to Farah Janzad, David Meoli, Jason Soares and Jennifer Hu for their help with segmenting the images, processing the sonomicrometer data, tissue processing and surgical preparation respectively. We would like to acknowledge support from the National Institutes of Health under grant NIH-NHLBI RO1-HL44803
References (54)
- et al.
Validation of a new counts-based gated single photon emission computed tomography method for quantifying left ventricular systolic function: comparison with equilibrium radionuclide angiography
J. Nucl. Cardiol.
(1997) - et al.
Three-dimensional echocardiography improves accuracy and compensates for sonographer inexperience in assessment of left ventricular ejection fraction
J. Am. Soc. Echocardiography
(1999) - et al.
Impact of on-line endocardial border detection on determination of left ventricular volume and ejection fraction by transthoracic 3-dimensional echocardiography
J. Am. Soc. Echocardiography
(1999) - et al.
Importance of imaging method over imaging modality in non-invasive determination of left ventricular volumes and ejection fraction: Assessment by two- and three-dimensional echocardiography and magnetic resonance imaging
J. Am. Coll. Cardiol.
(2000) - et al.
Physical and geometrical modeling for image-based recovery of left ventricular deformation
Prog. Biophys. Molec. Biol.
(1998) - et al.
Evaluating a robust contour tracker on echocardiographic sequences
Medical Image Analysis
(1999) - et al.
Cardiac material markers from tagged MR images
Medical Image Analysis
(1998) - et al.
Analysis of left ventricular wall motion based on volumetric deformable models and MRI-SPAMM
Medical Image Analysis
(1996) - et al.
Improved endocardial border resolution using dobutamin stress endocardiography with intravenous sonicated dextrose albumin
J. Am. Coll. Cardiol.
(1994) - et al.
Quantification of regional myocardial wall thickening on electrocardiogram-gated spect imaging
J. Nucl. Cardiol.
(1999)
Coupled B-snake grides and constrained thin-plate splines for analysis of 2D tissue deformations from tagged MRI
IEEE Trans. Med. Imaging
Finite Element Procedures in Engineering Analysis
Evaluation of ventricular ejection fraction, wall motion, wall thickening, and other parameters with gated myocardial perfusion single-photon emission computed tomography
J. Nucl. Cardiol.
A ‘hybrid’ method for measuring myocardial wall thickening from gated pet/spect images
J. Nucl. Med.
New concept in echocardiography: harmonic imaging of tissue without use of contrast agent
The Lancet
Deformable boundary finding in medical images by integrating gradient and region information
IEEE Trans. Med. Imaging
3D brain mapping using deformable neuroanatomy
Phys. Med. Biol.
Determining the accuracy of calculating systolic wall thickening using a fast Fourier transform approximation: a simulation study based on canine and patient data
J. Nucl. Med.
Differentiation of viable and nonviable myocardium by the use of three-dimensional tagged MRI in 2-day-old reperfused infarcts
Circulation
Gated myocardial perfusion tomography for the assessment of left ventricular function and volumes: comparison with echocardiography
J. Nucl. Med.
A Practical Guide to Splines
Left ventricular function and perfusion from gated spect perfusion images: an integrated method
J. Nucl. Med.
Stochastic relaxation, Gibbs distribution and Bayesian restoration of images
IEEE Trans. Pattern Anal. Machine Intell.
Cited by (170)
Parametrized pre-trained network (PPNet): A novel shape classification method using SPHARMs for MI detection
2023, Expert Systems with ApplicationsThree-dimensional speckle tracking echocardiography: The future is now
2018, Revista Portuguesa de CardiologiaA novel methodology for personalized simulations of ventricular hemodynamics from noninvasive imaging data
2016, Computerized Medical Imaging and GraphicsCitation Excerpt :Similarly, ejection parameters and blood kinetic energy obtained from both PC-MRI and 3D+t echocardiography by flow mapping can yield information on the cardiac energetic efficiency (Markl et al., 2011; Yang et al., 2007), and techniques of wall motion tracking are routinely used to infer the distribution of strains and their rate. These are generally based on tagged MRI data (Axel et al., 2005; Chandrashekara et al., 2004; Pan et al., 2005); however, algorithms have also been developed for 3D+t echocardiography and standard MRI Cine sequences (Papademetris et al., 2001; Shi et al., 2013). Despite this progress, however, the accuracy of measurement depends on the temporal and spatial resolution of the images, which is dictated by technical limitations and by the necessity to keep the acquisition times at a minimum.
Digital Image Processing and Automated Image Analysis in Echocardiography
2016, Practice of Clinical EchocardiographyUnsupervised Polyaffine Transformation Learning for Echocardiography Motion Estimation
2023, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- ☆
Electronic Annexes available. See www.elsevier.com/locate/media.