Changes and classification in myocardial contractile function in the left ventricle following acute myocardial infarction

Abstract

In this research, we hypothesized that novel biomechanical parameters are discriminative in patients following acute ST-segment elevation myocardial infarction (STEMI). To identify these biomechanical biomarkers and bring computational biomechanics 'closer to the clinic', we applied state-of-the-art multiphysics cardiac modelling combined with advanced machine learning and multivariate statistical inference to a clinical database of myocardial infarction. We obtained data from 11 STEMI patients (ClinicalTrials.gov NCT01717573) and 27 healthy volunteers, and developed personalized mathematical models for the left ventricle (LV) using an immersed boundary method. Subject-specific constitutive parameters were achieved by matching to clinical measurements. We have shown, for the first time, that compared with healthy controls, patients with STEMI exhibited increased LV wall active tension when normalized by systolic blood pressure, which suggests an increased demand on the contractile reserve of remote functional myocardium. The statistical analysis reveals that the required patient-specific contractility, normalized active tension and the systolic myofilament kinematics have the strongest explanatory power for identifying the myocardial function changes post-MI. We further observed a strong correlation between two biomarkers and the changes in LV ejection fraction at six months from baseline (the required contractility (r = - 0.79, p < 0.01) and the systolic myofilament kinematics (r = 0.70, p = 0.02)). The clinical and prognostic significance of these biomechanical parameters merits further scrutinization.

Keywords: cardiac modelling; contractility; immersed boundary method; machine learning; myocardial infarction; statistical inference.

Conflict of interest statement

We declare we have no competing interests.

© 2017 The Authors.

Figures

Figure 1.

LV model constructions for one healthy control and one MI patient based on in vivo magnetic resonance imaging data. The healthy LV model: (a) LV wall segmentation; (b) reconstructed LV geometry. The MI model (c) short-axis LGE imaging, the infarct region is enhanced with micro-vascular obstruction appearing dark inside the enhanced region; (d) long-axis LGE imaging; (e) LV wall segmentation; (f) reconstructed LV geometry, coloured by MI extent (1: 100% MI, 0: healthy myocardium). (Online version in colour.)

Figure 2.

Optimization of Treq in a healthy heart. (a) Finding optimal Treq by minimizing objective function (equation (2.12)); (b) strain comparison between CMR measurements and values from the LV model Treq. The difference in the strain between the measurements and the model prediction is 0.008 ± 0.02. (Online version in colour.)

Figure 3.

Distributions of age (a), SBP (b), EDV (c), LV [i.e. LVEF] EF (d), CS (e) and GLS (f) in the healthy controls and MI patients at baseline. (Online version in colour.)

Figure 4.

Examples of mechanical modelling of LV dynamics from a healthy control and a MI patient. (a) Deformed LV geometry at end-diastole and end-systole (b) from a healthy volunteer; (c) deformed LV geometry at end-diastole and at end-systole (d) from a MI patient; (e) distributions of systolic active tension and (f) myofibre stress from the healthy volunteer; (g) distributions of systolic active tension and (h) myofibre stress from the MI patient; (i) systolic myofibre strain distribution and (j) twist degree from LV base to apex from the healthy volunteer; (k) systolic myofibre strain distribution and (l) twist degree from LV base to apex from the MI patient. (Online version in colour.)

Figure 5.

Average passive stiffness along myofibres in healthy controls and MI patients. Error bar is the standard deviation. (Online version in colour.)

Figure 6.

Comparisons of Treq (a), Ta (b) and σf (c) between the healthy volunteers and the MI patients. (Online version in colour.)

Figure 7.

Distributions of Tnorma (a), σnormf (b) and Cs (c) between healthy volunteers and MI patients. (Online version in colour.)

Figure 8.

(a,b) C5.0 Decision-Trees for the different datasets D1 (a) and D2 (b). Note that only the first tree for each dataset is shown from an ensemble of boosting trees. The decision tree is the same for dataset D3 as dataset D1 because the same features are selected for the splits, thus not shown here. The selected C5.0 parameters for tree construction are based on the error rate evaluation in electronic supplementary material, figure S7.

Figure 9.

(a–c) Summary of feature relevance scores, obtained with different statistical/machine learning methods. The features are separately ranked in ascending order of importance, taken from the set {0, 1, …, p − 1}, where p is the total number of features. The ranks are accumulated over all methods shown in the legend. These methods are represented by different grey shades in the bars. A higher rank indicates a more important feature. The three panels correspond to the three datasets used in our study; from left to right: D1, D2 and D3.

Figure 10.

(a–c) Summary of classification accuracy for different methods and three datasets D1, D2 and D3 (columns). The area under the ROC curve (AUROC) is calculated from the area under the convex hull (dashed lines) that connects the best performing methods. The specificity and sensitivity score for each method was estimated with LOOCV. Each of the symbols for the univariate logistic regression (univariate LR) corresponds to one feature used as predictor.

Figure 11.

(a,b) Posterior probability plots of Gaussian process—automatic relevance determination (ARD). The two plots show the posterior probability contours given the most important features estimated by GP-ARD from fig. S2: Cs, Treq and Tnorma. The plus symbols correspond to healthy volunteers and the circles to patients with MI. The decision boundary of 0.5 is highlighted with a thick grey line.

Figure 12.

(a,b) Linear correlation analysis between Treq, Cs and LV function recovery at six months after acute-MI. (Online version in colour.)