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

Figure 1.. (a) Boundary detection with the 3D semi-automated technique starts with a bounding ellipse (outer dotted ellipse) that locates points on the myocardial boundaries based on pixelwise intensity gradients in an inwards, radial direction. Detection starts with the quantized reference image at end-diastole and is continued based on phase-unwrapped displacement information in subsequent timeframes. Dotted red line is the semi-automated epicardial contour before applying the LOESS curve fitting technique and starred red line is the one after the LOESS fit. Dotted green line is the pre-LOESS fit semi-automated endocardial contour and starred green line is the one after the LOESS fit. (b) Examples of pixel intensity-based gradients in the radial direction whose peaks are used to locate epicardial and endocardial boundary points. 3D, three-dimensional.

Figure 2.. Quantization with a Lossy irreversible compression process. The original image shown has been quantized to seven classes of pixels and six thresholds using the non-uniform, histogram-based Otsu’s method.

Figure 3.. Shown are the progressive phase-unwrapping based pixelwise displacements within the myocardium boundaries from end-diastole (dotted lines) through subsequent timeframes (solid lines) untill 48 ms in time. The mean incremental displacements in neighboring pixels (between consecutive timeframes) were applied to determine the direction of motion at each pixel for a new timeframe and the location of the new boundary points determined.

Figure 4.. Shown are the progressive phase-unwrapping based pixelwise displacements within the myocardium boundaries for timeframes between 63 ms and 108 ms, following those in Figure 3.

Figure 5.. Shown are the progressive phase-unwrapping based pixelwise displacements within the myocardium boundaries for timeframes between 123 ms and 168 ms (end-systole), following those in Figure 4.

Figure 6.. (a) The end-systolic boundaries (solid red and green lines) shown in comparison to the end-diastolic boundaries (dotted red and green lines). Finding a control point’s motion at each timeframe occur by tracking pixels with displacements (shown in cyan), whose tails map back to an original boundary control point at end-diastole. (b) Morphology of the two myocardial boundaries between end-diastole and end-systole using the 3D semi-automated methodology.

Figure 7.. Comparison of basal (top), midventricular (middle) and apical (bottom) boundary contours with surface maps of radial strains between the new 3D semi-automated method (left) and the Spottiswoode et al. (right) method. Abbreviations: 3D, three-dimensional; A, anterior; AS, anteroseptal; PS, posteroseptal; P, posterior; PL, posterolateral; AL, anterolateral.

Figure 8.. Comparison of basal (top), midventricular (middle) and apical (bottom) boundary contours with surface maps of circumferntial strains between the new 3D semi-automated method (left) and the Spottiswoode et al. (right) method. Abbreviations: 3D, three-dimensional; A, anterior; AS, anteroseptal; PS, posteroseptal; P, posterior; PL, posterolateral; AL, anterolateral.

Figure 9.. Comparison of basal (top), midventricular (middle) and apical (bottom) boundary contours with surface maps of longitudinal strains between the new 3D semi-automated method (left) and the Spottiswoode et al. (right) method. Abbreviations: 3D, three-dimensional; A, anterior; AS, anteroseptal; PS, posteroseptal; P, posterior; PL, posterolateral; AL, anterolateral.

Figure 10.. Comparison of 3D regionally averaged (a) circumferential, (b) radial and (c) longitudinal strains between the new 3D semi-automated and 2D Spottiswoode et al contouring techniques in N = 14 normal subjects. Strain was computed with the RPIM applied to 3D LV geometries and phase-unwrapped displacements. Abbreviations: 2D, two-dimensional; 3D, three-dimensional; A, anterior; A, apical; AL, anterolateral; AS, anteroseptal; B, basal; MV, midventricular; P, posterior; PL, posterolateral; PS, posteroseptal; RPIM, Radial Point Interpolation Method.

Figure 11.. Surface map of radial strain computed with RPIM on LV geometries, reconstructed from boundaries detected with the newly introduced 3D semi-automated myocardial contouring technique. Abbreviations: 3D, three-dimensional; A, anterior; A, apical; AL, anterolateral; AS, anteroseptal; B, basal; LV, left-ventricular; MV, midventricular; P, posterior; PL, posterolateral; PS, posteroseptal; RPIM, Radial Point Interpolation Method.

Figure 12.. Surface map of circumferential strain computed with the RPIM method on LV geometries, reconstructed from boundaries detected with the newly introduced 3D semi-automated myocardial contouring technique. Abbreviations: 3D, three-dimensional; A, anterior; AS, anteroseptal; PS, posteroseptal; P, posterior; PL, posterolateral; AL, anterolateral; LV, left-ventricular; RPIM, Radial Point Interpolation Method.