Singular Value Decomposition Applied to Cardiac Strain from MR Imaging for Selection of Optimal Cardiac Resynchronization Therapy Candidates

Abstract

Purpose: To use singular value decomposition (SVD) in heart failure (HF) to reveal primary spatiotemporal strain patterns in the left ventricle (LV), then develop and test a time-independent metric of cardiac dyssynchrony on the basis of the circumferential uniformity ratio estimate (CURE) computed with SVD (CURE-SVD) in both a canine model of HF with or without left bundle branch block (LBBB) and a clinical cohort referred for cardiac resynchronization therapy (CRT).

Materials and methods: The research was approved by the institutional review board and conformed with HIPAA requirements. All subjects provided informed consent. In both the canine model (n = 13) and the clinical cohort (80 CRT candidates; mean age, 65.2 years; range, 18.5-86.9 years), regional strains were derived by using cardiac magnetic resonance (MR) displacement encoding with stimulated echoes. CURE-SVD was compared with the standard CURE (averaged over systolic phases). Statistical methods included the Wilcoxon rank-sum test, Hodges-Lehmann estimator, Bland-Altman test, multivariable logistic regression, and receiver operating characteristic analysis.

Results: In the canine model, the median difference in CURE-SVD (range, 0-1) for LBBB-HF group versus narrow-QRS-HF group (-0.40; 95% confidence interval [CI]: -0.79, -0.31) was similar to that for CURE (-0.43; 95% CI: -0.72, -0.34]). In 80 CRT candidates, CURE-SVD and CURE were highly correlated (r = 0.90; P < .0001). The multivariable model for CRT response with CURE-SVD demonstrated excellent performance without the need for time averaging over cardiac phases (area under the receiver operating characteristic curve = 0.96, P < .0001).

Conclusion: SVD of circumferential strain in HF identifies primary LV spatiotemporal contraction patterns with minimal user input, while the time-independent CURE-SVD parameter has excellent performance in a canine model of dyssynchrony and is strongly associated with CRT response in patients with HF.

(©)RSNA, 2015.

Figures

Figure 1:

A, Circumferential strain and (rank-1 SVD approximation of circumferential strain) plotted over LV spatial segment and time are shown for a patient who had HF with dyssynchrony. Simultaneous stretch and contraction are seen with each cardiac phase. In the line plot, the spatial data from one column of varies greatly over the LV segments. B, In contrast, circumferential strain and are homogeneous for a given cardiac phase in the example of a patient with synchrony. In line plot from one column of , the spatial homogeneity contrasts with the dyssynchronous example. Color spectra = circumferential strain (Ecc) and the rank-1 approximation of Ecc as a function of cardiac phase and spatial segment (color bars indicate how the strain values map to color). Ecc/Max. Ecc = Ecc divided by the maximum value of Ecc, which normalizes Ecc between -1 and 1.

Figure 2:

Box plots show that, A,CURE-SVD approaches one in the narrow-QRS–HF group (NQRS) and in normal animals but is much lower in the LBBB-HF(LBBB) group and, B, a similar pattern is present for CURE in these three groups. * = Significant difference in metrics (P < .05) for the LBBB-HF group versus the narrow-QRS–HF group, + = significant difference in metrics (P < .05) for the LBBB-HF group versus the normal animal group.

Figure 3:

Correlation plot for CURE-SVD and CURE shows a very strong correlation between these two parameters (r = 0.90, P < .0001).

Figure 4:

ROC curve is shown for the multivariable model with CURE-SVD, LV mass index, absence of scar at the LV lead position, and delayed onset of contraction at the LV lead position. AUC = area under the ROC curve.