Embedding Overlap Priors in Variational Left Ventricle Tracking

Abstract

Abstract-This study investigates overlap priors for variational tracking of the Left Ventricle (LV) in cardiac MagneticResonance (MR) sequences. The method consists of evolving two curves toward the LV endo-and epicardium boundaries. We derive the curve evolution equations by minimizing two functionals each containing an original overlap prior constraint. The latter measures the conformity of the overlap between the nonparametric (kernel-based) intensity distributions within the three target regions-LV cavity, myocardium and background-to a prior learned from a given segmentation of the first frame. The Bhattacharyya coefficient is used as an overlap measure. Different from existing intensity-driven constraints, the proposed priors do not assume implicitly that the overlap between the intensity distributions within different regions has to be minimal. This prevents both the papillary muscles from being included erroneously in the myocardium and the curves from spilling into the background. Although neither geometric training nor preprocessing were used, quantitative evaluation of the similarities between automatic and independent manual segmentations showed that the proposed method yields a competitive score in comparison with existing methods. This allows more flexibility in clinical use because our solution is based only on the current intensity data, and consequently, the results are not bounded to the characteristics, variability, and mathematical description of a finite training set. We also demonstrate experimentally that the overlap measures are approximately constant over a cardiac sequence, which allows to learn the overlap priors from a single frame.

Left Ventricle Segmentation via Graph Cut Distribution Matching

“…Accurate segmentation of the left ventricle (LV) cavity in magnetic resonance (MR) sequences is very important for complete diagnosis of cardiovascular diseases [1], [3]. Manual segmentation of all images is prohibitively time-consuming.…”

“…Therefore, automatic or semi-automatic algorithms are highly desired. Albeit an impressive research effort has been devoted to the LV [1]- [15], current methods are still not sufficiently fast and flexible for routine clinical use, mainly because of the difficulties inherent to MR cardiac images [4]. Existing methods are based, among others, on active contours [1]- [3], [5]- [11], active appearance/shape models [12], [14], and registration [15].…”

“…Generally, the problem is stated as an energy optimization. In this connection, optimization of active contour functionals has been the most prevalent and flexible choice in the literature because it allows introducing a wide range of photometric and geometric 1 constraints on the solution [1]- [3], [5]- [11]. Generally, these constraints reference a sum over the target region or its boundary of pixelwise correspondences between the image and geometric/photometric models learned from a training set.…”

“…Generally, these constraints reference a sum over the target region or its boundary of pixelwise correspondences between the image and geometric/photometric models learned from a training set. Pixelwise photometric information cannot distinguish connected cardiac regions having almost the same photometric profile [1]- [3], for instance the papillary muscles within the cavity and the myocardium (cf. the examples in Fig.…”

“…Therefore, most of existing methods bias the solution towards a finite set of shapes learned a priori. Although very effective in some cases, training-based algorithms may have difficulty in capturing the substantial subject variations in a clinical context [1], [3], [4]. The ensuing results are bounded to the characteristics, variability, and mathematical description of the training set.…”