From myocardial infarction to heart failure: role of tissue anisotropy and its prediction through probabilistic neural network

Introduction: : Myocardial infarction (MI) is caused by a partial or complete blockage of blood flow to the myocardium, leading to cardiomyocyte death. Following MI, scar tissue is rapidly added, resulting in changes to the content and structure of the infarcted region [1]. The loss of contractile tissue in the myocardium results in a series of remodeling events occurring at multiple length scales, ranging from alterations to the cellular and extracellular structure at the fiber level to fibrosis and regional thinning at the tissue level to dilation and decompensated function at the organ level. Maladaptive left ventricular (LV) remodeling post-MI is known to contribute to the development of heart failure (HF). Understanding the role of LV free wall (LVFW) biomechanical remodeling in the transition of infarcted hearts to HF will improve prognostic capabilities and intervention strategies. The primary hypothesis of this study is that the loss of LVFW anisotropy in the early stages post-MI is a strong driving factor in the transition between adaptive and maladaptive remodeling in the later stages. In this study, we used a rodent model of MI to investigate the effect of biomechanical remodeling at the tissue level on organ-level function, and their role in the transition from adaptive to maladaptive remodeling, and subsequently, HF.

Materials and

Methods: : A cohort of 75 animals was used in this study. Anteriobasal MI was induced in 8-week-old, male, Wistar-Kyoto (WKY) rats through the ligation of the left anterior descending (LAD) artery. 15 animals were used as control subjects and underwent sham surgery. Animals were sacrificed weekly (n=15 for each group) up to four weeks post-infarct. Echocardiography and invasive catheterization were performed at the terminal timepoint for all animals. Then, the animals were euthanized and the hearts were excised. LVFW specimens were used to perform biaxial mechanical testing along circumferential and longitudinal directions and stretched equibiaxially up to 30% stretch [2], with stiffness being calculated as the slope of the stress-stretch curve at 30% stretch. Myocardial anisotropy was defined as the ratio of circumferential to longitudinal stiffness. A Bayesian neural network (BNN) was trained with the timepoint and hemodynamics indices serving as the input, and ex-vivo stiffness serving as the output. The BNN architecture was chosen to quantify the uncertainty in stiffness prediction, and the subsequent uncertainty in anisotropy prediction.

Results, Conclusions, and Discussions::

Results: In-vivo measurements from echocardiography and catheterization indicated significant changes to the morphology and kinematics of the LV. LV ejection fraction (EF) reduced significantly at 3-weeks post-MI (Fig. 1A). LV end-diastolic volume (EDV) increased post-MI with the change being near significant at 4 weeks (Fig. 1B), indicating progressive LV dilation. The LVFW passive mechanical behavior remained constant along the circumferential direction but stiffened along the longitudinal direction (Figs.1C, D). The increase in longitudinal stiffness, combined with constant circumferential stiffness resulted in decreased anisotropy post-MI, which was near-significant from the 2-week timepoint. This indicated that myocardial anisotropy could detect maladaptation earlier than a global metric such as ejection fraction. Training results from the BNN (Fig. 2A) indicated an average training accuracy of 82% and validation accuracy of 67% (Figs. 2B, C). The average uncertainty in predictions was estimated to be 8.9 kPa and 10.3 kPa along the circumferential and longitudinal directions, respectively.

Discussion: One of the primary findings of this study was the loss of LVFW anisotropy post-MI. In addition, this loss of anisotropy demonstrated a moderate non-linear correlation with the decrease in LV ejection fraction. More importantly, the loss of anisotropy was significant at two weeks post-MI and thus can capture the loss of cardiac function earlier than an organ-level metric such as ejection fraction. In addition to an improved prognostic capacity, the change in myocardial anisotropy at earlier timepoints indicated that changes in lower length-scale mechanics occur in response to insults first, and this change is gradually reflected in organ-level function. In addition, the training results of the BNN indicate that hemodynamics indices can be used to estimate tissue-level stiffness with a high degree of certainty, and this accuracy can be further improved by increasing the number of training points.

Conclusions: In this study, we investigated the behavior of the myocardium post-MI at multiple length scales through in-vivo and ex-vivo measurements. One of the primary findings of this study was the loss of LVFW anisotropy post-MI. The capacity to predict cardiac decompensation prior to the onset of heart failure can significantly improve patient outcomes.

Acknowledgements and/or References (Optional):: RA was supported by the NIH, under grant No. R00HL138288.