Remodeling after acute myocardial infarction: mapping ventricular dilatation using three dimensional CMR image registration

Since the advent of primary percutaneous coronary intervention (PPCI) to treat acute myocardial infarction immediate survival has improved but at the expense of a rising incidence of progressive heart failure [1]. Ischemia-reperfusion injury leads to a sequence of events that result in predictable changes to the structure and function of the left ventricle (LV) that may eventually cause congestive heart disease [2]. Our understanding of remodeling is largely based on animal models which show that within the first few days after coronary occlusion there is slippage and stretching of myocytes in the infarcted zone [3]. Late remodeling also involves changes to the non-ischemic myocardium as it adapts to the extra load placed on it by dilating [4, 5]. Current methods of evaluating remodeling rely on measuring global changes in left ventricular volume, however this does not reveal where myocardial enlargement and dilatation occur within the ventricle or how remodeling is influenced by the severity of local ischemic injury. Determining the pattern of left ventricular remodeling in patients and the contribution made by infarcted and remote myocardium to chamber dilatation has importance for evaluating interventions aimed at preventing heart failure [6].

In this paper we apply techniques previously developed in neuroscience research to align and co-register Cardiovascular Magnetic Resonance (CMR) images of the heart to map the effects of remodeling in three dimensions [7]. This framework enables highly consistent comparisons to be made between different CMR sequences obtained at various time-points [8]. By tracking how each point on the myocardial surface changes relative to neighboring points we can build a three dimensional model of how the LV remodels over time. Co-registering structural imaging with late-enhancement sequences also allows the extent of local remodeling to be compared to the degree of ischemic injury present. However, developing these techniques in the heart is challenging as it requires a consistent and accurate approach to co-registering and transforming the cardiac MR images acquired in a longitudinal study.

The aim of this study was to use 3D co-registration techniques to determine how the LV remodels after reperfused acute ST-elevation myocardial infarction (STEMI) and the relationship to ischemic injury.

Baseline characteristics for all patients are shown in Table 1. The baseline CMR was performed at a mean of 2.7 days (range 1 – 7 days) post-PPCI and the follow-up study at a mean of 54 weeks (range 53 – 57 weeks). All subjects demonstrated an enhancing infarct on the LGE images within the vascular territory of the culprit coronary artery corresponding to high signal on T2-weighted imaging. No patients demonstrated co-existing chronic infarctions or infarcts in multiple coronary territories. Mean infarct size was 13.3% (range 1.2 - 34.0%). Changes in left ventricular functional indices are shown in Table 2. Twenty seven (59%) patients had evidence of MVO on the baseline LGE images, and in 16 (35%) patients there was global LV remodeling. At follow-up 39 patients were NYHA class I, 4 patients class II, 2 patients class III and 1 patient class IV.

The mean surface-to-surface distance between manual and automatic segmentations was 1.2 mm ± 0.3 for the endocardium and 1.5 mm ± 0.2 for the epicardium.

In this study we have used image-registration to map in three dimensions where the left ventricular wall dilates in the first year following STEMI and to relate this to the extent of local ischemic injury. Our findings show that following PPCI with standard medical therapy infarct transmurality and MVO have a significant effect on local ventricular wall remodeling while only modest dilatation is observed within non-ischemic myocardium.

Image co-registration is an emerging technique for the statistical analysis and quantification of cardiac motion and geometry [8]. In this study we used a labeled atlas to provide prior knowledge of cardiac anatomy to assist with image segmentation and as a template onto which different MR sequences are co-registered. This approach has been successfully used for determining shape variation of the left ventricle within patient populations [18, 19]. It is also applicable to analyzing dynamic changes in cardiac morphology by using intrinsic signal intensity variations and anatomical features in the heart to track myocardial deformation over time - a process analogous to the explicit tags used in strain imaging [20, 21]. In this work we used normalized mutual information between the myocardial segmentations at baseline and follow-up to map the chronic regional structural adaptations that occur within the left ventricle in response to STEMI. We also co-registered the anatomical images with the infarct segmentation enabling us to compare the severity of local remodeling with the extent of transmural necrosis at each point in the LV. This approach provides an appealing new methodology for assessing temporal changes in left ventricular morphology in relation to local ischemic injury.

A limitation of conventional imaging is that assessment of cavity volumes does not reflect the dynamic regional changes in myocardial structure that occur in response to ischemia-reperfusion injury. In animal studies of remodeling there is an acute phase of lateral slippage of the myocardial planes within the infarct leading to progressive left ventricular dilatation [22]. In the later stages of disease non-ischemic myocardium may develop a maladaptive response of ventricular dilatation which ultimately leads to heart failure [5, 23]. Left ventricular remodeling is also determined by factors other than initial infarct size [17, 24] and both infarct transmurality and the presence of persistent MVO influence the development of chronic heart failure [25‐27]. Despite revascularization and medical therapy 35% of the patients in our study developed global remodeling as defined by end diastolic volumes. Our findings indicate that infarcted myocardium plays a much greater role in such chronic remodeling than non-ischemic tissue. Our observation of progressive dilatation in the ischemic territory is consistent with elongation of the infarcted myocardium which typically develops during early post-infarction remodeling [28]. The extent of transmural necrosis is known to influence global remodeling [27] and we observed that local myocardial dilatation occurs when infarct thickness exceeds 50%. This is consistent with findings that acute sub-endocardial infarctions remain viable with the potential for contractile recovery [29]. We also observed the highest rates of local wall dilatation when infarcts had regions of MVO, which is recognized as a predictor of adverse remodeling [16]. We observed a significant decrease in left ventricular mass and wall thickness in the infarcted region over time and this is consistent with transient ischemia-reperfusion edema within the infarct zone [30].

In our cohort remodeling in the non-ischemic territory was modest compared to the infarcted region. The late phase of ventricular remodeling has been shown to be amendable to medical therapy, as chronic administration of angiotensin-converting enzyme inhibitors (ACEIs) are associated with a reduction in the extent of ventricular dilation [4, 31] and prolonged therapy with β-blockers may limit or even reverse remodeling and dysfunction [32, 33]. The great majority of patients in this study were prescribed both ACEIs and β-blockers and our findings suggest that this therapy may be effective in limiting the development of maladaptive remodeling within remote myocardium following successful PPCI.

Image registration offers a more refined model of left ventricular adaptations to infarction than can be derived from global measures of cavity volume. The potential clinical applications include interventional studies of cardio-protective therapies where the physiological effects of remodeling within infarcted and remote myocardium may be separately measured and compared to imaging markers of ischemic injury. The approach may also be helpful for predicting the complications of ischemic remodeling such as ventricular aneurysm formation and the development of ischemic mitral regurgitation. Developments in parallel imaging also make 3D acquisition of both cine images and LGE appealing for registration techniques as they avoid section mis-registration [34‐36].

We do not have inter-study reproducibility data for the registration technique as the patients were only scanned once at each visit. We used the segmentation of the myocardium to guide registration so the final transformation is weighted to geometric changes within the myocardium. We also used an initial non-rigid transformation to limit the influence of local structures other than myocardium. However, as the mutual information between images is not explicitly defined the registration may also be influenced by structures immediately adjacent to the heart. EM segmentation provided an accurate segmentation of the myocardial boundaries, but the influence of signal intensity differences in acute and remote myocardium was not evaluated. Mean wall thickness in infarcted myocardium at follow-up was 8.9 mm and so there were typically at least 4 voxels across the width of the LV to perform the non-rigid registration However, accuracy of registration may be limited by severely thinned myocardium. We did not use follow-up LGE images in the remodeling analysis as we wanted to distinguish expansion of remodeled myocardium from expansion of the zone of necrosis. We also did not segment the T2 images due to the relatively low signal to noise ratio of these sequences.

In our cohort the mean baseline LVEF was not significantly impaired and this is likely to have been influenced by the exclusion of subjects with acutely impaired left ventricular function who were not suitable for CMR within 7 days of PPCI. However, the observation that local remodeling may be observed despite a normal ejection fraction supports the premise that global indices of function may mask the regional effects of ischemic damage. Further research is needed to explore whether local morphological changes affect long-term outcome or are independent predictors of heart failure. GMM has not been formally assessed for infarct segmentation although the methodology is similar to validated techniques for pixel-weighted infarct quantification [37]. Our findings are limited to assessing structural remodeling and we did not measure regional strain or wall thickening to evaluate changes in contractile function as a consequence of STEMI [38]. Further work is also needed to establish the correspondence between image-based transformations and ultra-structural adaptations within the myocardium.

The severity of ischemic injury has a significant effect on local ventricular wall remodeling while only modest dilatation is observed within non-ischemic myocardium. This indicates that revascularization and standard medical therapy may be effective at preventing structural changes in the remote myocardium, and that further limitation of chronic remodeling may depend on therapies directed at modulating ischemia-reperfusion injury.

Co-registration of CMR images acquired in longitudinal studies is an analysis technique which has potential for assessing dynamic changes in ventricular structure in relation to therapeutic interventions.