A simplified approach to discriminate between healthy subjects and patients with heart failure using cardiac magnetic resonance myocardial deformation imaging

Abstract

Aims: Left ventricular global longitudinal strain (LV-GLS) shows promise as a marker to detect early heart failure (HF). This study sought to (i) establish cardiac magnetic resonance imaging (CMR)-derived LV-GLS cut-offs to differentiate healthy from HF for both acquisition-based and post-processing techniques, (ii) assess agreement, and (iii) provide a method to convert LV-GLS between both techniques.

Methods and results: A secondary analysis of a prospective study enrolling healthy subjects (n = 19) and HF patients (n = 56) was conducted. LV-GLS was measured using fast strain-encoded imaging (fSENC) and feature tracking (FT). Receiver operating characteristic (ROC) analyses were performed to derive and evaluate LV-GLS cut-offs discriminating between healthy, HF with mild deformation impairment (DI), and HF with severe DI. Linear regression and Bland-Altman analyses assessed agreement. Cut-offs discriminating between healthy and HF were identified at -19.3% and -15.1% for fSENC and FT, respectively. Cut-offs of -15.8% (fSENC) and -10.8% (FT) further distinguished mild from severe DI. No significant differences in area under ROC curve were identified between fSENC and FT. Bland-Altman analysis revealed a bias of -4.01%, 95% CI -4.42, -3.50 for FT, considering fSENC as reference. Linear regression suggested a factor of 0.76 to rescale fSENC-derived LV-GLS to FT. Using this factor on fSENC-derived cut-offs yielded rescaled FT LV-GLS cut-offs of -14.7% (healthy vs. HF) and -12% (mild vs. severe DI).

Conclusion: LV-GLS distinguishes healthy from HF with high accuracy. Each measurement technique requires distinct cut-offs, but rescaling factors facilitate conversion. An FT-based LV-GLS ≥ -15% simplifies HF detection in clinical routine.

Keywords: cardiac magnetic resonance imaging; cut-off; deformation imaging; early identification of heart failure; global longitudinal strain.

Figure 1

Scatter plot for fSENC- and FT-derived LV-GLS, stratified by entity of HF. Within-group means are shown as point with corresponding 95% CI (error bar). Individual measurements are shown as small dots (jittered horizontally for better visibility). Proposed cut-offs border the coloured regions. Specifically, LV-GLS cut-offs to discriminate between healthy subjects (green area) and those with HF (yellow and red area) were set at −19.3% for fSENC and at −15.1% for feature tracking. Cut-offs to distinguish between HF with mild (yellow area) and severe DI (red area) were set at −15.7% for fSENC and at −10.8% for feature tracking. P-values of tests for differences in mean are given above brackets indicating which HF entities are compared.

Figure 2

ROC curves of the comparisons of FT-based (red) and fSENC-based (blue) LV-GLS for the discrimination between (A, left) healthy subjects and HF patients and (B, right) HF with mild DI and HF with severe DI. Proposed LV-GLS cut-off values have been labelled with white boxes and positions on the ROC curve are marked with diamonds (◆). Corresponding CIs for sensitivity and specificity are represented as shaded boxes. Targeted minimum sensitivity (i.e. 85%) is indicated as a dotted horizontal line. AUROCs for the discrimination between (A, left) healthy subjects and patients with HF and (B, right) HF with mild DI and HF with severe DI are reported for fSENC and FT. P-values correspond to a test of the AUROC being equal to 50% (random classification).

Figure 3

Bland–Altman plots for variability of fSENC-derived and FT-derived LV-GLS. The blue line indicates the mean difference (bias) between methods; the red lines, as indicated, show limits of agreement (95% CI of the differences of the measured values). Corresponding ribbons represent 95% CIs for the respective parameters. The grey dashed line reflects the increase of difference between the two methods for increasing LV-GLS values.

Figure 4

Linear regression model without intercept visualizing LV-GLS derived through FT vs. fSENC. The regression equation (y = 0.76x) is stated in the upper left corner. HF entities of individual patients are represented as different shapes and colours.

Figure 5

Scatter plot for FT-derived LV-GLS, stratified by entity of HF. Cut-off values are rescaled by a factor of 0.76 based on conversion from fSENC to FT. Within-group means are shown as point with corresponding 95% CI (error bar). Individual measurements are shown as small dots (jittered horizontally for better visibility). Rescaled FT LV-GLS cut-offs border the coloured regions. Specifically, rescaled LV-GLS cut-offs of −14.7% and −11.9% were used to discriminate between healthy subjects (green area) and those with HF (yellow and red area) and between HF with mild (yellow area) and severe DI (red area), respectively.

Figure 6

Simplified approach for the identification of HF using fSENC-based, FT-based, and rescaled FT LV-GLS in clinical routine care. Using the cut-offs summarized in the table, the proposed approach can be used to discriminate between healthy subjects, HF with mild DI, and HF with severe DI.