Manifold-enhanced Segmentation through Random Walks on Linear Subspace Priors

Abstract

In this paper we propose a novel method for knowledge-based segmentation. Our contribution lies on the introduction of linear sub-spaces constraints within the randomwalk segmentation framework. Prior knowledge is obtained through principal component analysis that is then combined with conventional boundary constraints for image segmentation. The approach is validated on a challenging clinical setting that is multicomponent segmentation of the human upper leg skeletal muscle in Magnetic Resonance Imaging, where there is limited visual differentiation support between muscle classes.

The Generalized Log-Ratio Transformation: Learning Shape and Adjacency Priors for Simultaneous Thigh Muscle Segmentation

“…The works of Baudin et al [24], [26], [27] are perhaps closest to this work algorithmically, but the data they use exhibits significantly different properties (Fig. 14).…”

“…, q M }, with q i : Ω → P K , extracting the mean shape and modes of greatest variation. In previous works, PCA was applied directly to training segmentations by treating probabilistic labelings as vectors in R n·K [26], [29]. However, PCA assumes that data lies in an unconstrained vector space.…”

“…However, the detection of enough boundaries to ensure seeds lie in the interior of muscle regions may be difficult. Baudin et al also present a technique for encoding shape models learnt from training segmentations into a random walker (RW) segmentation [25] framework, which benefits from RW's robustness to missing edges [26], [27]. These works exhibit the greatest similarity to our proposed approach, as the segmentation step is convex and it provides a probabilistic solution.…”

Discrete Visual Perception

“…The problem was also considered in its temporal aspect [35], [36], [37] through a combined iconic-geometric approach [37] seeking to determine the optimal deformation along with correspondences for a subset of interest points along this structure. In the context of multi-label segmentation [38] studied two distinct cases of introducing prior knowledge to the random walker graphical model. The first was aiming to position automatically the seeds with respect to the different classes [39] through the optimization of a graph-based objective function.…”

“…The first was aiming to position automatically the seeds with respect to the different classes [39] through the optimization of a graph-based objective function. Later the idea of introducing prior knowledge [38], [40] in this context was studied using statistical models of varying complexity, like mono-modal distributions or linear sub-spaces per class. Multi-class segmentation of striated Muscles in NMR Images was the clinical case being investigated in this context where the problem of optimal parameter setting was also investigated through machine learning methods [41].…”