Introduction to a mechanism for automated myocardium boundary detection with displacement encoding with stimulated echoes (DENSE).

Objective: Displacement ENcoding with Stimulated Echoes (DENSE) is an MRI technique developed to encode phase related to myocardial tissue displacements, and the displacement information directly applied towards detecting left-ventricular (LV) myocardial motion during the cardiac cycle. The purpose of this study is to present a novel, three-dimensional (3D) DENSE displacement-based and magnitude image quantization-based, semi-automated detection technique for myocardial wall motion, whose boundaries are used for rapid and automated computation of 3D myocardial strain.

Methods: The architecture of this boundary detection algorithm is primarily based on pixelwise spatiotemporal increments in LV tissue displacements during the cardiac cycle and further reinforced by radially searching for pixel-based image gradients in multithreshold quantized magnitude images. This spatiotemporal edge detection methodology was applied to all LV partitions and their subsequent timeframes that lead to full 3D LV reconstructions. It was followed by quantifications of 3D chamber dimensions and myocardial strains, whose rapid computation was the primary motivation behind developing this algorithm. A pre-existing two-dimensional (2D) semi-automated contouring technique was used in parallel to validate the accuracy of the algorithm and both methods tested on DENSE data acquired in (N = 14) healthy subjects. Chamber quantifications between methods were compared using paired t-tests and Bland-Altman analysis established regional strain agreements.

Results: There were no significant differences in the results of chamber quantifications between the 3D semi-automated and existing 2D boundary detection techniques. This included comparisons of ejection fractions, which were 0.62 ± 0.04 vs 0.60 ± 0.06 (p = 0.23) for apical, 0.60 ± 0.04 vs 0.59 ± 0.05 (p = 0.76) for midventricular and 0.56 ± 0.04 vs 0.58 ± 0.05 (p = 0.07) for basal segments, that were quantified using the 3D semi-automated and 2D pre-existing methodologies, respectively. Bland-Altman agreement between regional strains generated biases of 0.01 ± 0.06, -0.01 ± 0.01 and 0.0 ± 0.06 for the radial, circumferential and longitudinal directions, respectively.

Conclusion: A new, 3D semi-automated methodology for contouring the entire LV and rapidly generating chamber quantifications and regional strains is presented that was validated in relation to an existing 2D contouring technique. Advances in knowledge: This study introduced a scientific tool for rapid, semi-automated generation of clinical information regarding shape and function in the 3D LV.