Segmentation of the Right Ventricle Using Diffusion Maps and Markov Random Fields

Automatic segmentation of right ventricle in cardiac cine MR images using a saliency analysis

“…When the DSC measure is used, the proposed method's segmentation accuracy is slightly below (−0.02) the results reported by Ringenberg et al However, in comparison with this approach and other state-of-the-art methods, the present strategy requires no parameter fine tuning nor previous training or a minimal quantity of data. Other methods demand a strong adjustment of parameters 34,46,51 or are atlas based and, in consequence, computationally expensive and data quantity/quality dependent. 26,28,50 Some authors 31 have used the propagation of a manual segmentation to the rest of the RV, but the dependency on the expert is inevitable and a considerable burden.…”

Medical Image Understanding and Analysis

“…Heart shapes were found to form a non-linear manifold, e.g. in [11]. The beating motion of the heart is also complex and nonlinear.…”

“…Although heart deformation seems to have a high number of degrees of freedom, we investigate whether it lies on a lower dimensional manifold. This approach has been successfully used to encode the shape and the appearance of both the heart [11] and highly deformable objects such as the human body [3], with the aim to support tasks such as pose recovery, reconstruction, or tracking.…”

“…Isomap combined with image distance measures representing deformations in MR images provided a representation of non-rigid chest cavity deformation due to breathing [17]. In [11], the authors demonstrated the ability of Diffusion Map to learn the right ventricle's shape variations in order to drive its segmentation. The closest work to our own is that of Gifani et al [5], who used Locally Linear Embedding (LLE) on 2D echocardiography images to model the cardiac cycle of individual subjects and to automatically detect its phase.…”

Shape prior based image segmentation using manifold learning

“…In image analysis, these nonlinear methods have shown their potential in facial recognition, hyperspectral image classification, gait recognition, hand-written character recognition [10] and several medical imaging tasks such as segmentation and registration [11]. For the segmentation task, non linear shape statistics were first introduced with kernel PCA [12], [13] and pursued in the following years with Laplacian Eigenmaps [14] and Diffusion Maps [15], [16]. Etyngier et al in particular were the first to introduce non-linear shape priors into a deformable model framework [15].…”