Automated left ventricle boundary delineation in left ventriculograms

Automated left ventricle boundary delineation

“…The Bayesian pixel classifier labels each pixel in the image frame with one of 3 classes [17], the background, EDnot-ES and ES based on the gray scale vector through systole, shown in Figure 2. The underlying concept is that background pixels retain their intensities throughout systole, ES pixels increase their intensities throughout systole, and ED-not-ES pixels demonstrate a characteristic pattern of gray values over time depending on location as the LV border passes over the pixels during contraction.…”

“…The LV AoV angle and long axis are defined by the AoV end points and apex, shown in Figure 3. This alignment efficiently keeps the LV region information [17]. Those aligned ground truth class region images are stacked one on the top of the other.…”

“…But tracing the LV boundaries from LVG is a tedious job. Although automated boundary delineation (ABD) in LVGs has long been sought [3,4,1,13,7,19,12,18,17], no methods have been reported in terms of the accuracy of volume and ejection fraction (EF) derived from the ABD border, due to the challenges of LVG variation. Moreover, most previous methods were tested on only a limited number of LVG samples so that no consistent accuracy evaluation was available.…”

“…The first stage is a non-parametric regional pixel classifier [12,17]. It segments the LV region from the background and obtains a raw LV boundary.…”

“…It segments the LV region from the background and obtains a raw LV boundary. The second stage is a shape regression [18,17] that corrects the systematic error of the region classifier [8,17]. The last stage is a rejection classifier that flags unreliable results.…”