Elsevier

Medical Image Analysis

Volume 12, Issue 1, February 2008, Pages 69-85
Medical Image Analysis

Myocardial deformation recovery from cine MRI using a nearly incompressible biventricular model

https://doi.org/10.1016/j.media.2007.10.009Get rights and content

Abstract

This paper presents a method for biventricular myocardial deformation recovery from cine MRI. The method is based on a deformable model that is nearly incompressible, a desirable property since the myocardium has been shown to be nearly incompressible. The model uses a matrix-valued radial basis function to represent divergence-free displacement fields, which is a first order approximation of incompressibility. This representation allows for deformation modeling of an arbitrary topologies with a relatively small number of parameters, which is suitable for representing the motion of the multi-chamber structure of the heart. The myocardium needs to be segmented in an initial frame after which the method automatically determines the tissue deformation everywhere in the myocardium throughout the cardiac cycle.

Two studies were carried out to validate the method. In the first study the myocardial deformation was recovered from a 3D anatomical cine MRI sequence of a healthy volunteer and then validated against the manual segmentation of the biventricular wall and against the corresponding 3D tagged cine MRI sequence. The average volume agreement between the model and the manual segmentation had a false positive rate of 3.2%, false negative rate of 2.8% and true positive rate of 91.4%. The average distance between the model and manually determined intersections of perpendicular tag planes was 1.7 mm (1.2 pixel). The same procedures was repeated on another set of 3D anatomical and tagged MRI scans of the same volunteer taken four months later. The recovered deformation was very similar to the one obtained from the first set of scans. In the second study the method was applied to 3D anatomical cine MRI scans of three patients with ventricular dyssynchrony and three age-matched healthy volunteers. The recovered strains of the normal subjects were clearly stronger than the recovered strains of the patients and they were similar to those reported by other researchers. The recovered deformation of all six subjects was validated against manual segmentation of the biventricular wall and against corresponding tagged MRI scans. The agreement was similar to that of the first study.

Introduction

The study of myocardial motion is essential for understanding of both normal heart function and cardiovascular diseases, which are the leading cause of morbidity and mortality in the United States (N.C. for Health Statistics, 2003). While several cardiac imaging techniques have been developed (see Section 1.2), here we present a method for biventricular myocardial deformation recovery from cine MRI. Cine MRI is widely used for analysis of the cardiac function because of its high soft tissue contrast as compared to non-MRI modalities and relatively short acquisition time as compared to other cardiac MRI acquisition techniques. Tagged (Zerhouni et al., 1988, Axel and Dougherty, 1989), phase velocity (Pelc et al., 1991) and DENSE (Kim et al., 2004) MRI contain more information about the myocardial motion than cine MRI. However, they are not as widely used and are typically not contained in existing databases of cardiac MR images, in which case cine MRI remains the only option for the analysis of the myocardial motion. For this reason, the proposed method is suitable for both prospective and retrospective cardiac deformation analysis.

Several methods for imaging the myocardial motion have been developed. They span all standard medical imaging modalities: ultrasound, nuclear-based techniques, computed tomography and magnetic resonance imaging.

Echocardiography is an application of ultrasound to the heart. It is fast and fairly inexpensive. However, the method has a low soft tissue contrast and low signal-to-noise ratio in the far field. Since ultrasound is severely attenuated in air and bone, only limited views of the heart can be obtained. If tissue Doppler imaging is done in conjunction with echocardiography, quantitative values for wall velocity can be obtained, but only in one direction (toward the transducer) and only in selected locations within the myocardial wall. Despite the limitations, echocardiography is the second most frequently performed diagnostic procedure after electrocardiography (Fuster et al., 2001).

Cardiac positron emission tomography (PET) is used for study and quantification of blood flow and metabolism of the heart and it does not provide a direct measure of cardiac motion. Gated single photon emission computed tomography can be used to quantify the biventricular myocardial wall motion and thickening but it suffers from very poor spatial resolution that causes it to have only fair accuracy and reproducibility (Fuster et al., 2001).

Cardiac computed tomography allows for multi-slice imaging of the heart at multiple time points over the cardiac cycle, which can be used for motion analysis. The drawbacks to the technique are that iodinated contrast agent must be used to generate good contrast between the blood and myocardium and the patient must be exposed to a relatively large dose of ionizing radiation (Fuster et al., 2001).

There are five MRI-based acquisition techniques that can be used to analyze cardiac motion: cine, tagged, phase velocity, displacement-encoded (DENSE), and strain-encoded (SENC) MRI. Cine MRI can cover the entire cardiac anatomy and can generate high temporal resolution images over the entire cardiac cycle. It is by far the most frequently used cardiac MRI technique because of its high reproducibility, high soft-tissue contrast, and good spatial and temporal resolution (Fuster et al., 2001). Myocardial tagging can be used to lay down a set of grid lines on the myocardium by modulating the magnetization, and cine images showing the deformation of the tag lines can then be acquired (Zerhouni et al., 1988, Axel and Dougherty, 1989). Phase velocity MRI can be used to acquire velocity of the myocardial tissue (Delfino et al., 2006, Pelc et al., 1991). Displacement-encoded MRI allows for imaging of the myocardial displacement (Kim et al., 2004). Strain-encoded MRI can be used to acquire through-plane myocardial strain (Garot et al., 2004). Tagged, phase velocity, displacement-encoded, and strain-encoded MRI are not widely used and are always acquired in addition to cine MRI.

Recovery of cardiac deformation from images has received considerable attention. Researchers have worked on cardiac deformation recovery methods that use tagged MRI (Tustison and Amini, 2006, Chandrashekara et al., 2004, Deng and Denney, 2004, Rougon et al., 2005, Pan et al., 2005, Huang et al., 1999), phase velocity MRI (Amini and Price, 2001), and echocardiography (Ledesma-Carbayo et al., 2005, Shekhar et al., 2004, Comaniciu et al., 2004, Suffoletto et al., 2006).

Cine MRI has also been used for automated computation of cardiac function. Some of these methods provide only the segmentation of the myocardium without recovering the displacement field (van Assen et al., 2006, Kaus et al., 2004, Pluempitiwiriyawej et al., 2005, Lorenzo-Valdes et al., 2004). This prevents one from computing myocardial strain fields, which are critical for analysis of the cardiac function. In addition, there are methods that track only the endocardial and epicardial surfaces (Montagnat and Delingette, 2005, Bardinet et al., 1996) without computing the displacements within the heart wall. These methods also cannot be used to compute myocardial strains.

In our previous work (Bistoquet et al., 2007) we used an exactly incompressible deformable model of the left ventricle (LV) to recover myocardial motion from cine MRI. While the method is relatively accurate, it uses a curvilinear coordinate system based on the midsurface of the LV wall that does not generalize to multi-chamber topologies. In Medical Image Computing and Computer-Assisted Intervention (MICCAI) conference in 2002, Lorenzo-Valdes et al. (2002) presented an automated method for segmentation and tracking of cardiac deformation from cine MRI. They modeled the frame-to-frame 3D heart deformation using cubic B-splines (as proposed by the same group for the registration of breast MR images (Rueckert et al., 1999)) and obtained the model parameters by maximizing the normalized mutual information (Studholme et al., 1999). The method propagates the segmentation from the first frame to the rest of the cardiac cycle. The authors correlated the volume of the manual segmentation of the myocardium to the one obtained automatically by the method and reported the r value of 0.98 and 0.83, for the case of manual or atlas-based segmentation of the first frame, respectively.

Researchers have developed methods for 3D cardiac deformation recovery from cine MRI that require a considerable amount of user input, i.e. that cannot be considered fully automated. The method by Papademetris et al. (2002) requires the user to manually segment the myocardium in all the slices and all the frames of the image sequence, the method by Remme et al. (2005) requires the user to manually track a set of 3D points throughout the cardiac cycle, and the method by Shen et al. (2005) uses the knowledge of the boundaries between the myocardium and blood pool and the myocardium and surrounding structures, which is equivalent to segmentation. It is desirable that the method automatically recovers the myocardial deformation in all the frames of the cardiac cycle.

The myocardium is a nearly incompressible material. Its constituents are mainly composed of water, which is almost perfectly incompressible. However, the myocardium is perfused with blood, which affects the total myocardial volume over the cardiac cycle. A few independent studies (Yin et al., 1996, Hoffman and Spaan, 1990, Judd and Levy, 1991, Liu et al., 1992, Vergroesen et al., 1987) have been carried out to quantify the change of the myocardial volume over the cardiac cycle. The common conclusion of these efforts is that the total myocardial volume changes no more than 4% during a cardiac cycle. This means that the myocardium is not perfectly incompressible. However, this volume change is relatively small and it is distributed in all three directions. Even in the regions with predominant orientation of the blood vessels, the myocardial tissue does not expand or contract by more than 2% in any direction. For this reason, one can conclude that “incompressibility” is not simply another approach to cardiac deformation recovery; near incompressibility is a physical property of the myocardium that should not be ignored. Thus, any myocardial deformation recovery method that deviates from incompressibility by more than a few percents cannot be correct. It should be noted that near incompressibility itself is not enough to guarantee correct cardiac deformation recovery, i.e. near incompressibility is a necessary but not a sufficient constraint. The proposed method is directly based on this property.

Section snippets

3D nearly incompressible transformation model

We propose a nearly incompressible model that can represent the deformation of structures with arbitrary topologies, e.g. multi-chamber structures. In this paper we use it to model the two cardiac ventricles.

The model is comprised of a domain of an arbitrary topology and a displacement field defined over the domain. To represent the displacement field we interpolate the displacements specified at a finite number of locations in the reference frame. These locations are referred to as nodes. We

MR protocols and subjects

We performed two studies to evaluate the proposed method. In the first study, we acquired a 3D anatomical cine MRI scan together with a 3D tagged cine MRI scan of a healthy volunteer and then repeated the acquisitions four months later. The volunteer was a 27 year old male subject with no history of heart disease. The purpose of this study was twofold: to perform a 4D validation of the deformation of the biventricular myocardium wall recovered from the anatomical scan against the corresponding

Deformable model

The proposed deformable model has the following properties: it is volumetric, it behaves uniformly in all the regions, its displacement field is C1 continuous, it is nearly incompressible, and it is capable of generating realistic cardiac deformation patterns and normal strains. Fig. 1f shows a deformation pattern typical for ES, which is a combination of radial contraction, circumferential twisting, and longitudinal shortening. Normals strains are shown in Section 3 and discussed in Section 4.2

Conclusion

We have developed an automated method for cardiac deformation recovery from cine MRI based on a 3D deformable model that is nearly incompressible; a desirable property since the myocardium has been shown to be nearly incompressible. The presented method recovers the deformation throughout the myocardial wall of the left and right ventricle, i.e. not only at the endocardium and epicardium. The method was quantitatively evaluated against manual segmentation and against the corresponding

References (46)

  • F. Brookstein

    Principal warps: thin-plate splines and the decomposition of deformations

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (1989)
  • M.D. Cerqueira et al.

    Standardized myocardial segmentation and nomenclature for tomographic imaging of the heart

    Journal of Applied Mathematics and Physics (ZAMP)

    (2002)
  • R. Chandrashekara et al.

    Analysis of 3-D myocardial motion in tagged MR images using nonrigid image registration

    IEEE Transactions on Medical Imaging

    (2004)
  • D. Comaniciu et al.

    Robust real-time myocardial border tracking for echocardiography: an information fusion approach

    IEEE Transactions on Medical Imaging

    (2004)
  • J.G. Delfino et al.

    Comparison of myocardial velocities obtained with magnetic phase velocity mapping and tissue doppler imaging in normal subjects and patients with left ventricular dyssynchrony

    Journal of Magnetic Resonance Imaging

    (2006)
  • X. Deng et al.

    Three-dimensional myocardial strain reconstruction from tagged MRI using a cylindrical b-spline model

    IEEE Transactions on Medical Imaging

    (2004)
  • V. Fuster et al.

    Hurst’s THE HEART

    (2001)
  • J. Garot et al.

    Spatially resolved imaging of myocardial function with strain-encoded MR: comparison with delayed contrast-enhanced MR imaging after myocardial infarction

    Radiology

    (2004)
  • J. Hoffman et al.

    Pressure-flow relations in coronary circulation

    Physiological Reviews

    (1990)
  • J. Huang et al.

    Spatio-temporal tracking of myocardial deformations with a 4-D B-spline model from tagged MRI

    IEEE Transactions on Medical Imaging

    (1999)
  • R. Judd et al.

    Effects of barium-induced cardiac contraction on large and small vessel intramyocardial blood volume

    Circulation

    (1991)
  • D. Kim et al.

    Myocardial tissue tracking with two-dimensional cine displacement-encoded MR imaging: development and initial evaluation

    Radiology

    (2004)
  • S.S. Klein et al.

    Noninvasive delineation of normal right ventricular contractile motion with magnetic resonance imaging myocardial tagging

    Annals of Biomedical Engineering

    (1998)
  • Cited by (0)

    View full text